Sie sind auf Seite 1von 28

Journal of Sotmd and Vibration (1974) 33(3), 267-294

DYNAMIC STABILITY OF PIPES CONVEYING FLUID

M. P. PAiDOUSSIS AND N. T. ISSID


Department of Mechanical Engineering,
McGill University, Montreal, Quebec, Canada

(Received 4 August 1973, attd hz revisedform 20 October 1973)

This paper deals with the dynamics and stability of flexible pipes containing flowing fluid,
where the flow velocity is either entirely constant, or with a small harmonic component
superposed. An extensive historical review of the subject is given.
In the case of constant flow velocity, the dynamics of the system is examined in a general
way and it is shown that conservative systems are subject not only to buckling (divergence)
at sufficiently high flow velocities, but also to oscillatory instabilities (flutter) at higher
flow velocities. Also presented are some new results for cases of systems subjected to internal
dissipative forces.
In the case of harmonically varying flow velocity, the equation of motion derived here
exposes an error in a previous derivation. Stability maps are presented for parametric
instabilities, computed by Bolotin's method, for pipes with pinned or clamped ends, as well
as for cantilevered pipes. It is found that the extent of the instability regions increases with
flow velocity for clamped--clamped and pinned-pinned pipes, while a more complex
behaviour obtains in the case of cantilevered pipes. In all cases, dissipation reduces the
extent of, or entirely eliminates, parametric instability zones.

1. INTRODUCTION
A series of experiments by Aitken [1 ], reported in 1876, on travelling chains and elastic cords,
illustrating the balance between motion-induced tensile and centrifugal forces, is perhaps
among the earliest work pertinent to the study of dynamics of flexible pipes conveying fluid.
Certainly some aspects of the dynamics of such systems have been known for a long time, such
as the considerable resisting lateral force that must be exerted by one holding a fire-hose when
the discharge rate is high.
Similarly, the first observation must have been made long ago of the peculiar spontaneous
motions imparted to the free end of a rubber pipe, such as might be used to water the lawn,
by a sufficiently high flow rate. Evidently, this was first recognized as a self-excited oscillation
by Marcel Brillouin in 1885, but his work on the subject remained unpublished. Bourri~res,
one of Brillouin's students, was the first to undertake a serious study of the dynamics of
flexible pipes conveying fluid. In a remarkable paper [2], published in 1939, Bourri6res
examined the oscillatory instability of cantilevered pipes conveying fluid, both theoretically
and experimentally. He derived the correct equation of motion and, although unable to
obtain analytically the critical flow velocity for the onset of the oscillation, he determined
most of the salient characteristics of the phenomenon. Unfortunately, this valuable paper
was evidently '~lost", in the sense of being unknown to those who have since undertaken
work in this area--until now.t
Interest in the subject was reactivated in 1950, in connection with the study of vibration of
the Trans-Arabian pipeline, by Ashley and Haviland [4]. Later, Feodos'ev [5] derived the

"["The senior author is grateful to Professor A. Fortier of the University of Paris who recalled Bourri~res'
work in the course of a seminar delivered by the author in France last year [3].
267
)68 M.P. PAIDOUSSISAND N. T. ISSID
equation of motion in full for a pipe conveying fluid, and analysed the case of a pipe with
simply-supported ends. The same problem was studied independently by Housner [6] using
a different approach. Feodos'ev and Housncr found that for sufficiently high flow velocities
the pipe may buckle, essentially like a column subjected to axial loading! A subsequent,
elegant and more general study by Niordson [7] led to the same equation of motion and to
essentially the same conclusions regarding stability of pipes with simply-supported ends.
Long [8] was the first, after Bourri~res, to be concerned with cantilevered pipes conveying
fluid, although he also considered other boundary conditions. His method of solution was
applicable to relatively small flow velocities, considerably below the threshold of oscillatory
instability, of the existence of which he seemed to be unaware. Nevertheless, he did perceive
and confirm experimentally that, in contrast with those of simply-supported pipes, forced
motions of cantilevered pipcs arc damped by internal flow in the range of flow velocities
considered.
Handclman [9] presented an analytical method in which the character of the eigcnvalues of
the problem is determined from the structure of the differential equation of motion without
determining specific solutions. Also, Heinrich [I0], Bolotin [I 1] and Hu and Tsoon [12]
studied various aspects of the problem. More recently, Movchan [13] retrieved the condition
of stability for a simply-supported pipe conveying fluid by means of Liapunov's direct
method.
In all the abovc studies, excepting Bourri~res', the only form of instability discovered was
buckling. It was not until 1963 that Gregory and Pa~doussis [14, 15] showed theoretically
and experimentally that, at sufficiently high flow velocities, cantilevered pipes are subject to
oscillatory instabilities (flutter) rather than buckling (divergence). Howcvcr, the existence of
oscillatory instabilitids was fully anticipated in two outstanding papers by Benjamin [16, 17]
concerned with the dynamics of articulated pipes (consisting of rigid tubes connected by
flexible joints) conveying fluid, which is a discrete rcpresentation of the continuously flexible
system: a cantilevered system of articulated pipes was found to be subject to oscillatory
instability. Benjamin was the first to perceive that the dynamical problem is independent of
fluid friction, and he predicted analytically the existence of oscillatory instability of canti-
levered pipes conveying fluid, both effects later being confirmed by Gregory and Paidoussis'
work.
Nevertheless, the dynamical behaviour of articulated and continuously flexible pipes
conveying fluid is not strictly analogous. Thus, Benjamin [16] found that buckling instability
is possible in the case of a vertical cantilevered system, where gravity is operative, if the fluid
is sufficiently heavy; on the other hand, Pa~doussis [18] found that vertical, continuously
flexible pipes are never subject to buckling. This paradox was clarified by Paidoussis and
Deksnis [19].
The stability of tubular cantilevers conveying fluid (neglecting gravity forces) was further
discussed by Nemat-Nasscr, Prasad and Herrmann [20], with emphasis on the effect on
stability ofvelocity-depcndcnt forces, such as dissipative and Coriolis forces; they showed that
such forces may destabilize the system, an effect also reported prcviously by Gregory and
PaXdoussis [14, 15]. Subsequent papers by Herrmann [21 ] and Herrmann and Nemat-Nasser
[22] stressed the connection between the problem of instability of a cantilever conveying fluid
and the more general problem of instability of a cantilever subjected to a "follower"-typc
force at the free end: i.e., a force retaining the same angular disposition relative to the free
end in the course of small motions of the cantilever. Recently, Wiley and Furkert [23] con-
sidered the problem of a beam subjected to a follower force acting within the span, where the
force is generated by a fluid jet attached to the beam and fed by an infinitely flexible supply
line; thcy found that either buckling or oscillatory instabilities, or both, may occur, depending
on the boundary conditions.
STABILITY OF PIPES CONVEYING FLUID 269
A recent study by Stein and Torbiner [24], which is mainly concerned with infinitely long
pipes conveying fluid, brings to the fore a correction to the equation of motion as derived in
some of the studies discussed so far. This correction, which was first recognized by Heinrich
[10] and by Hu and Tsoon [12] independently, and in a slightly different context by Haringx
[25] earlier still, arises from the effect of internal pressure and may become significant for
sufficiently high pressures. Naguleswaran and Williams [26] studied this aspect of the
problem both theoretically and experimentally. It is of interest that pipes with both ends
supported may buckle even at very small flow velocities by the action of internal pressure.
Thurman and Mote [27] presented a non-linear analysis for a pipe with simply-supported
ends conveying fluid, using a perturbation technique. They found that, in determining the
natural frequencies of the system, the importance of non-linear terms increases with flow
velocity, so that the range of applicability of linear theory becomes more restricted as the
flow velocity increases.
More recently, Chen [28] studied the stability of a pipe conveying fluid with the upstream
end clamped and the downstream end constrained by a linear spring, so that the boundary
conditions are intermediate between clamped-free and clamped-pinned; accordingly, both
buckling and oscillatory instabilities are possible in general, depending on the spring constant.
Paidoussis and Denise [29, 30] studied the dynamics of very thin elastic pipes conveying
fluid, by utilizing thin-shell theory to describe the motions of the pipe and potential flow
theory to obtain the fluid forces. They analyzed both cantilevered pipes and pipes with
clamped ends. They found that in addition to instabilities in the beam modes of the system
(corresponding to those found previously by beam theory for thicker pipes), instabilities in
the shell modes are also possible, as verified by their experiments. Of particular interest was
the finding that thin pipes with clamped ends are not only subject to buckling (divergence)
but also to coupled-mode flutter. Similar theoretical results were obtained later by a different
analytical method by Weaver and Unny [31] in the case of simply-supported shells.
In all the studies discussed above, the flow velocity was taken to be steady. Recently, Chen
[32] examined the stability of simply-supported pipes conveying fluid with a flow velocity, U,
which has a time-dependent harmonic component superposed on the steady velocity, Uo,
such that U-- Uo(l + llcoscot). He found that parametric instabilities are possible in such
cases, and determined the boundaries of stability-instability regions; moreover, he found that
combination resonances are also possible. These instabilities are akin to those experienced
by a column subjected to an end load of the type P = Po(1 + llcoscot).
Thus we see a remarkable development of the subject in the twenty years Since the Trans-
Arabian pipeline was observed to vibrate presumably as a result of internal flow. However,
the impetus .for most of these studies did not come from a desire to solve a practical problem,
but rather from one or more of the following considerations: (i) the physical problem was
inherently intriguing; (ii) the problem offered scope for interesting mathematical manipula-
tion; (iii) the problem represents one of the few cases where a (non-conservative) follower
load is physically realizable. The force of this last point becomes obvious upon recalling the
very considerable theoretical interest in the stability of columns subjected to a tangential
.follower load at the free end [21], even when no means was evident for producing it.t Now,
internal flow exiting from the free end of a cantilever does generate a tangential load. This
being a physically easily realizable system--as compared to using solid-fuel rockets--it
offers a rare opportunity for combined theoretical-experimental study in the general area of
dynamics of elastic systems subjected to non-conservative forces, and explains, to some extent,
the large volume of research activity on this topic. Quite recently, however, increasing in-
t Thus in 1961 Timoshenko and Gere [33], upon presenting the analysis leading to the critical load for this
problem, state that " N o definite conclusion can be made (as yet) regarding the practical value of this result,
since no method has been devised for applying a tangential load to a column during bending".
270 M. P. PAIDOUSSIS AND N. T. ISSID

dustrial interest has developed in the dynamics of pipes conveying fluid, in conjunction with
determining the response of such systems to arbitrary force fields, with application to heat
exchangers, liquid-fuel rocket piping and nuclear reactor coolant channels.
In this paper we reconsider the dynamics of pipes conveying fluid with steady flow velocity
and present some new and rather unexpected results. We also reconsider the case of a harmonic-
ally varying flow velocity, rederiving the pertinent equation of motion and correcting an
error in Chen's [32] formulation, and extend!ng the analysis to boundary conditions other
than simply-supported.

2. THE EQUATION OF SMALL LATERAL MOTIONS

2.1. DERIVATIONOF THE EQUATIONOF MOI'ION


The system under consideration consists of a uniform tubular beam of length L, internal
perimeter S, mass per unit length m, and flexural rigidity El, conveying incompressible fluid
of mass per unit length M, flowing axially with velocity U. The cross-sectional flow area is A,
and the fluid pressure, measured above the atmospheric, is p. It is assumed that, although
flexible, the material of the tubular beam is such that its nominal dimensions do not alter
significantly by the effects of internal pressure and frictional drag. The pipe axis in its un-
deformed (equilibrium) state coincides with the x-axis.

i" j ::'y

1, u

Figure I. A vertical flexiblepipe conveyingfluid.

Consider a case where the pipe is hanging vertically, so that the x-axis is in the direction
of gravity (Figure 1). One can assume, without loss of generality, that free motions of the
pipe take place exclusively in one arbitrary plane, say the (x,y)-plane, and consider small
motions y(x, t).
Consider now a small element of the pipe, of length cSx, and the corresponding element of
the enclosed fluid, of volume 6~, as shown in Figure 2. The rate of change of momentum
over 6~O may be written as

d~l
dtl3--_-ffS [~- '1 (V'V) V] p d91' (1)

where V is the instantaneous velocity of a small element d~B within 6~B. Assume that radial
variations in the flow velocity are small and that secondary flow effects are negligible--which
STABILITY OF PIPES CONVEYING FLUID 271

was actually presupposed in the assumption ofplanar motions: i.e., assume that one has plug
flow, and hence that V may bewritten as
Ov Oy.
V = U + --:-'O=t Ui + ~-t]. (2)

Moreover, upon assuming t h a t y and ay[at are small and that motions are of long wavelength
such that ay/Ox is small, equations (I) and (2) after some manipulation yield

d933 - - M--~
OU 6xi + M 0 U O.O_l(a)'+ U~16xj" (3)
dt ~ + Ox]~Ot ox]
Hence, for the fluid element of Figure 2(a), force balances in the x - and y-directions yield
Op Oy au
--A ~x -- qS + Mg + F~x -- M ~ = 0, (4)

and

F4,4~ p y=0, (5)

where q is the shear stress on the internal surface of the pipe and Fis the transverse force per
unit length between pipe wall and fluid.

9 MgSx A(p ~78x)

(a) (b)
Figure 2. Forces and moments acting on elements of the fluid and of the pipe, (a) and (b) respectively.

Similarly, for the pipe element (Figure 2(b)) one obtains


aT
-~x + qS + nlg - F ~x
a3' =0, (6)

O--"x+ F + Ox k Ox] + qS'~x - n'-3-iT - c~-[ = O, (7)

a~/t ( a _).a3y
a=- ax = " E*~7+L- l~xSxa' (8)

where Tis the longitudinal tension, Q is the transverse shear force in the pipe, dr/is the bending
moment, E* the coefficient of internal dissipation which was assumed to be viscoelastic and
272 M . P . PAIDOUSSIS AND N. T. ISSID

of the Kelvin-Voigt type, and c is the coefficient of viscous damping due to friction of the
pipe with the surrounding stationary fluid medium; terms of second order magnitude were
neglected in accordance with the Euler-Bernoulli beam approximation for small lateral
motions.
Combining equations (5), (7) and (8) one obtains

E*g+E Ig- ax (r-pA) +M (9)


and adding equations (4) and (5) yields
0 OU
~x(T-pA) = M-~- (M + re)g,

which integrated from x to L becomes


OU
(T-PA)Ix=L- ( T - pA) = M y f ( L - x) - (M + m) g(L - x). (10)

The tension at the downstream end of the pipe, i.e., at x = L, will be zero, unless the pipe is
subjected to an externally applied tension, which is denoted by •. The pressure at x = L will
also be zero, since it is measured above the atmospheric, unless the fluid does not discharge to
atmosphere, in which case one may assume that there is an internal pressure equal to p, at
x = L, over and above that expended to overcome friction. If the downstream end is not free
to slide axially, nor completely free, internal pressurization induces an additional tensile
force, which for a thin pipe is equal to -2vpA, where v is Poisson's ratio [26]. Accordingly,
equation (10) may be written in general in the form
T - pA = T - pA ( 1 - 2v3) + [(M + m)g - m (0 U/Ot)] (L - x), (11)
where 6 = 0 signifies that there is no constraint to axial movement of the downstream end,
and 6 = 1 if there is.
Now, substituting equation (11) into equation (9) gives the equation of small lateral
motions:

02y OZy 0)' Oy 02y


• 2 + 2MUo--~t + (M + ,7,)g ~x + c ~ - + (M + , 7 , ) ~ = 0. (12)

This equation is more general than that utilized previously by the first author for the study of
cantilevered pipes conveying fluid [18] in several respects. Firstly, the effects of pressurization
and external tension have been included here--for a cantilever ~ = 0 and p = 0, in any case.
Secondly, and more importantly, the flow velocity here has not been assumed to be steady,
so that the above equation should also apply to cases where the flow contains harmonic
components; this is accounted for by the additional term M(OU[Ot)(L - x)(O2y]Ox2). In an
earlier study by Chen [32], on the dynamic stability of simply-supported pipes conveying
fluid, this term is notably absent from the equation of motion; instead, a term M(O U[Ot)(Oy]Ox)
is retained from the expansion of the lateral acceleration term, which as has been seen here
cancels out when equations (9) and (10) are combined. Effectively, it is presumed in reference
[32] that the equation of motion for steady flow applies also for unsteady flow, simply by
substituting Uo(1 + pcoscot) for U. Hence, Chen's equation of motion does not take account
of the longitudinal acceleration term M(OU]Ot) in equation (4) and is therefore erroneous.
STABILITY OF PIPES CONVEYING FLUID 273
Alternatively, Chen's formulation may be regarded as neglecting the axial movements of the
pipe (contraction), as shown in Appendix I. Moreover, the work presented here is more
general by the inclusion of terms involving ~r, fi and g.
It is recognized, nevertheless, that equation (12) is only approximately correct, in view of
the assumptions made. Perhaps the most restrictive of the latter is the assumption of plug
flow which may be considered to hold approximately, only for fully-developed turbulent
flow and if au[at is small both in amplitude and frequency.

2.2. THE BOUNDARY CONDITIONS


Next consider the boundary conditions. The following generalized set may be used:

OaY T~O~-Y+koy=O, x 2 - C O OOY=o


E I ~02Y x at x = O,
E10x 3 - ox

Oay _ Oy OZy Oy
E I-~--~ - T-~x - kL y = O, E I-~x2 + CL o---f= O at x = L, (13)

from which all the standard boundary conditions may be retrieved accordingly as ko, kL, Co
and cL are either zero or infinite.t

2.3. DIMENSIONLESSPARAMETERS
The problem may be expressed in dimensionless terms by defining the following dimension-
less quantities:

(]1,t1112 1~1 M + m 7"L2


u= -~1 UL, fl=M+m' r = .E-----f--Lag, F= El'

ft.4 L 2 cL 2 ko L a kL t a
17= E1 ' • 1/2' h'o= E1 " ~cl E1

.t
COL 2 ~l
eLL 2
t<o = (14)
E1 ' KI= E1
Substituting equations (14) into equations (12) and (13) one obtains the dimensionless
equation of motion

~a--~-~+~-7 u~-r+n(l - 2v~) + [ B ' ~ - ~ (1-0 x

a2 q 2[W" O2 il . all+ aq a ~q
• ar + = o, (15)

and the dimensionless boundary conditions

03q FO~l!+Koll= 0211 K~Oq=O at 4=0,


a~ a or ar ~ - "

az q + K' oq
~ F Oq Kll1 = ~ - --=0 at 4=1. (16)
04a 04- 042 104
1"At a free end ?'= O.
274 M . P . PA1DOUSSIS AND N. T. ISSID

3. DYNAMICS OF PIPES WITH STEADY FLOW


3.1. PIPES WITH BOTH ENDS E1TttER PINNED OR CLAMPED

Consider first the dynamics of a pipe conveying fluid, with simply-supported e n d s - - t o be


referred to here as a pinned-pinned pipe. This is the simplest case from the analytical point
of view and it was the first to be thoroughly investigated.
It has been found previously [7] that by the action of the equivalent "compressive" force
MU2(a2y/ax2)--by analogy to a column subjected to a compressive l o a d - - t h e natural
frequencies of the pipe decrease with increasing flow velocity, but remain real in the absence
of dissipative forces. With increasing flow, the natural frequencies of all the modes vanish in
turn, indicating the onset of buckling in the corresponding modes of the system.
The critical flow velocities may be found particularly easily in this case by application of
Euler's method of equilibrium. One may presume that an equilibrium position ~/(~) exists
near the trivial position of equilibrium tl(O = 0. With ~ = ), = 0 for simplicity, r/(O must
satisfy equation (15) with the time derivatives eliminated: i.e.,
~ q , a 2 r/
0r I- v - ~ S - = O, (17)

where vz = u 2 + / / ( l - 2v3) - F. The above equation admits solutions


4-
r/(~) = • Aj exp (i:0 O,
./=t

with ctl = ct2 = 0, cts = v, cq = - v . Hence, rewriting the solution in the form

tl(~) -- A1 + A2 ~ + B~ cos v~ + B2 sin v~, (18)

and substituting into the boundary conditions, equations (16), with ~Co-- iq = 0% ~c~= ~c'1= 0,
for non-trivial solution one obtains a vanishing determinant which yields

v4 sin v = 0. (19)

Since v = 0 is a trivial solution, one obtains v = j ~ , where j = l, 2 . . . . . This, of course, is


nothing else than Euler's result; thus in the case o f u = / 7 = 0, one h a s t - F =- -~L2/EI = (jg)2.
Physically, one may regard v2 as an effective compressive load. Alternatively, and more
appropriately to the case of a flowing fluid, V2(021][0~2) may be regarded as a generalized
"centrifugal" force; when this force overcomes the flexural restoring force, the pipe buckles.
Similar behaviour was found to obtain in the case of a clamped-clamped pipe conveying
fluid. The critical values of v for buckling are found to be given by v -- 2~, 8.99 .... 4r~. . . . ,
corresponding to the zeros of 2(1 - cosy) - vsinv = 0.
It was decided to re-examine the dynamical behaviour of these systems by obtaining
solutions of equation (12) in its full form, i.e., by employing the so-called " d y n a m i c " or
"kinetic" method (as defined in reference [34]), with some rather startling results. The methods
of solution and computation are essentially as described in references [35] and [18] and will
not be discussed here further.
Some calculated frequencies for pinned-pinned cylinders with ~ = 7 = F = / / = 0 are
shown in Figure 3, for fl--0.1 and 0.5. The real and imaginary parts of the dimensionless
frequencies, Re(co) and Im(co), are plotted in Argand diagrams, with the dimensionless flow
velocity, u, as the parameter.

t It is recalled that here Tis a tension, so that -~rrepresents a compressive load.


STABILITY OF PIPES CONVEYING FLUID 275

In Figure 3(a) it is noted that, with increasing flow, the frequency ofthe first mode diminishes
and eventually vanishes at u = re, which is the first critical flow velocity for buckling; similarly,
the frequency of the second mode vanishes at u = 2r~. However, at slightly higher u, the loci
of the first and second modes coalesce on the [Im(o)]-axis and leave the axis at symmetrical
points, indicating the onset of coupled-mode flutter. To the authors' knowledge this theoretical
result has not been reported hitherto.

52 1 I II i Il i I ,7 I ' I//I I

- 8~<~ ist and 2ndmodes


7.
3rdmode~
IT; 6.e,,,_. ~ G S ~ Volues o~u /

~lT2"tr 6265 6-125 ~/ 6,2 0 85 $ ,/,2


~l,~--~---;. ..__. < .~.//. 9 <"~t -9 "'
tTlT- 25
4 ~l ~ 6575 c. = Ist mode

-16 if'6 "~.7


/ ~_- lstond 2rid mOSes
/ :8

-52 I i I I I } I ." i I 071, i


0 4 8 12 58 42 86 90

(o)
E

32 , I I I I I // I I It/I 1//1 l

5~ 9 Io.
~6 <.U_____ 9.4.%._.~ . ~
9.,5/
,,
8.297 . / / mode
!]{625
.A,r ~ it./ aa
1/943
1__l_e~ /96~99"45953./"
- A--I95
9-25
9 ~/
29 1 4 90

T_,',I I mode ,,? 9 2 mode


41T!~ 5 . 7 5 " ~

-52 t I I I I I // I t !;t t I//I t


0 4 8 24 2858 4086 90
Re(~)
(b)

Figure 3. Dimensionless complex frequency diagrams of pinned-pinned pipes f o r / ' = H = ~ = ~. = 7 = 0


and ( a ) fl = 0-1 a n d ( b ) fl = 0 " 5 . The loci that actually lie on the axes have been drawn slightly off the axes but
parallel to them for the sake of clarity. - t - - e - , first mode; -I1--11-, second mode; - h , - - J , - , third mode;
-III--e--II-, combined first and second modes.

In Figure 3(b) it is noted that, once again the first-mode frequency vanishes at u = re.
However, u = 2r~ does not correspond to buckling in the second mode, but rather it is the
point where the system regahzs stability in its first mode! At slightly higher u the first- and
second-mode loci coalesce orl the [Re(o)]-axis, once again indicating the onset of coupled-
mode flutter (at u_~ 6"3). With increasing u the real part of the frequency of oscillations
eventually vanishes, at u - 9.41. Then, by a similar process, coupled-mode flutter re-occurs
at zt "~ 9.51, involving the third-mode locus.
276 M.P. PAIDOUSSIS AND N. T. ISSID

Figure 4 shows the effect of viscoelastic damping on the stability of the system of Figure
3(b). It is noted that the complex behaviour of the loci is greatly simplified; however, the
system does not regain stability after the onset of buckling in the first mode. It is also noted
that there is no longer symmetry of the loci about the [Re(oJ)]-axis in this case.
Similar results were obtained in the case of clamped-clamped pipes, as shown, for instance,
in Figure 5. The system buckles in its first mode at u = 2rr and then regains stability at u " 8.99.
At lt_~ 9-3 the first- and second-mode loci coalesce indicating the onset of coupled-mode
flutter. Also, there occurs flutter involving the third mode at higher flow velocities.
#
24 I II I I I I I I I "~1 I I Hi

8 ~.....~_ e
).11 I ~ 9 9 9 3 rd mode
16 " ~12
t_i2

8 - ~.Srl
41i~
3
a mode
"3 25 6 25
ll" 5 2 0
o 1/ #t
3 25 _ __
6 32
~~ 6o- 3 5
4
-8
55
o7
\
-16 I I I I I I I ill I IliJ---
0 4 8 12 56 4086 88
Re ({al

Figure 4. Dimensionless complex frequency diagram of a clamped-clamped pipe for F = H = z = 7 = 0 ,


/7 = 0-5, ~ = 5 x 10 -3.The loci that actually lie on the [Im(og)]-axis have been drawn off the axis but parallel
to it for the sake of clarity.

I I I //i I I 9 I I I I i I II I

16 I / l l ~

li 0

8 ilTle8' / I ~9 C o m b i n e d ,., and 2nd modes

II ;~
i ! . ~e # .z~ modo /
*~93 9 z 9 2
9, 9.:> 2 o / iJLi

I I I /,i I I I I i1 i I r I
4 i6 20 24 60 6~' U20
Re(M

Figure 5. Dimensionless complex frequency diagrams of a clamped-clamped pipe for F = H = z = = = y = 0 ,


/7 = 0"5. The loci that actually lie on the axes have been drawn slightly off the axes but parallel to them for the
sake of clarity.
STABILITY OF PIPES CONVEYING FLUID 277

It is worth noting the difference in behaviour between systems of low and high mass ratio,
p, which holds true for both pinned-pinned and clamped-clamped pipes. For small values of/}
the system buckles in both the first and second modes before the onset of ftutter (Figure 3(a)).
For larger values of fl (Figures 3(b) and 5), the system does not buckle in the second mode
prior to the onset of flutter.
These results are of fundamental importance, as it was previously thought [14, 16] that
oscillatory instabilities are not possible when both ends of the pipe are supported, the
system being inherently conservative in such cases. Although the existence of coupled-mode
flutter for such systems was evidently unknown, nevertheless, similar results have recently
been obtained in the case ofvery thin pipes conveying fluid, the dynamics of which was studied
by using shell theory to describe the motions ofthe pipe [30, 31]. Until now it was thought that
these results were peculiar to cylindrical shells conveying fluid, while being impossible to
obtain when using beam theory [36].
It is of interest that a pinned-pinned or clamped-clamped column subjected to an end load
is not subject to these coupled-mode instabilities; this was confirmed by conducting similar
calculations to those above, but setting fl -- 0, in which case the end load is equal to MU 2, or
to u z in dimensionless terms. It is evident, therefore, that the existence of oscillatory instabili=
ties is connected to the presence of Coriolis forces which are operative only when ~ # 0. Indeed
this is the key to an explanation of the unexpected results obtained above, as will be discussed
below.
The first point which needs to be explained is why the "critical" flow velocities obtained
by Euler's method of equilibrium do not always correspond to thresholds of instability, as
indicated by the results obtained by the dynamic method (Figures 3-5). The explanation to
this is obtained upon realizing that the system at hand, being subject to non-working Coriolis
forces, is a gyroscopic conservative system. Accordingly, the method of equilibrium only
gives the first critical point of instability (the first critical flow velocity), which corresponds
to buckling in the first mode of the system [34]; beyond this point, the system ceases being
positive definite, and the dynamic method alone is capable of yielding all the higher critical
velocities and their correct physical interpretation (i.e., whether the system loses or regains
stability at a certain critical flow velocity). Indeed, this is as good a case as any for demonstrat-
ing the pitfalls in using the method of equilibrium for assessing stability of other than non-
gyroscopic conservative systems.
A second point of interest is related to the well known stabilizing effect of gyroscopic
forces, e.g., in connection with the whirling of shafts. Here this effect is demonstrated clearly
in Figure 3(b), where after the onset of buckling the Coriolis forces stabilize the system prior
to the onset of coupled-mode flutter. This effect is even more pronounced at higher 8, the
Coriolis forces being proportional to B1/2.
The third point that needs be discussed is the very existence of flutter (oscillatory instabili-
ties) in conservative systems. It is well known that non-gyroscopic conservative systems are
not subject to flutter [34]; (this was demonstrated here by the calculations in which the Coriolis
forces were eliminated by setting fl = 0). However, flutter was widely believed to be impossible
also in gyroscopic conservative systems. For the problem at hand this was supported by
energy considerations (see, e.g., references [16] and [18]). Thus, the work done by the fluid
forces in the course of small motions of the pipe over a period t is

AIV=- MU(p2 + Upy')l~dt- 89 M[p2 +(Uy')2ll;dx. (20)


0 o

The first term, which is the work done by the non-conservative part of the fluid forces (cf..
equation (A1)), clearly vanishes for the conservative systems under consideration here, by
278 M . P . PAIDOUSSIS AND N. T. ISSID

virtue of the boundary conditions. The second term also vanishes if the motions are periodic,
such that y(x, t) = y(x, t + t) for 0 < x ~<L. Therefore, in the course of such periodic motions
one must have A IV = 0, --i.e., the system can neither absorb nor expand energy--in contrast
to a non-conservative system (cantilevered pipe) where the first integral may be either positive
or negative. Hence, as the fluid is the only possible source of energy, one might arrive at
the paradoxical conclusion that flutter in the conservative systems here under consideration
is impossible.

/f/ / .,o

,a,C~_ ,2 30 ~ ' - F ~ / ,2 /

/\ /
iof

10~&9

I A a -..,',, 2 ~\ tz ,a "r
oi_/_:_%_o . ,, _ _ : o ~ . - - ~ , ~.o

| _
-Lol , , "~, , , "~'~
0 I0 20 30 40 50 60 70

Re(~)
Figure 6. Dimensionless complex frequency diagrams o f a cantilevered pipe for fl = 0"65, 7 = 10, z = 0;
, ct = 0; . . . . , ct = 0.0189. The diagrams on the left-hand side o f the figure display the behaviour of the
loci on the llm(a0]-axis.

On closer examination, the above argument simply implies that the critical condition for
flutter cannot be a point of neutral stability--in contrast to the conditions for a non-conserva-
tive (cantilevered) pipe system. However, the critical condition for coupled-mode flutter, as
opposed to single degree-of-freedom flutter (Figure 6), is not a point of neutral stability as
it corresponds to coincidence of two frequencies and, hence, to growing oscillations of the
form y(x, t) =f(x)(a + bt)exp(kot). Therefore, the proper conclusion from the above argu-
ment is that, apart from divergence (buckling), the gyroscopic conservative system can only
be subject to coupled-mode flutter.
Recently, Shieh [37] has arrived at similar conclusions for gyroscopic conservative systems
in general, and illustrated the existence of coupled-mode flutter by an example from gyro-
dynamics involving a shaft rotating with angular velocity P.. under an axial compression P.
The equation of motion being
Ely I" + Py" + M(.~; - 2Qp - 02) ') = 0,
the mathematical similarity between that problem and the one discussed here is obvious.

3.2. CANTILEVERED PIPES


The dynamics of cantilevered pipes conveying fluid has been treated extensively in the past
(see, e.g., references [14, 15, 18, 22]). The character of free motions of the system may be
described with the aid of Figure 6 showing the loci of the natural frequencies of the lowest
few modes of a typical system in the a-plane as functions of ,. First consider the dynamics
of the undamped system (c~= 0). It is seen that small flow velocities (a) reduce the frequency
STABILITY OF PIPES CONVEYING FLUID 279
of oscillation (real components of to), and (b) have a damping effect on free motions of the
system (Im (to) > 0). As the flow velocity is increased, the damping effect associated with some
of the modes is diminished, and for a sufficiently high flow velocity vanishes in the case of one
of the modes (at u = 12.88); further increase of flow precipitates oscillatory instability (flutter)
in that mode.
At the critical flow velocity, it, = 12.88, the character of the system changes and the rate of
work done by the non-conservative forces,
dW
"dr =-MU(p2 + Upy')[x=L, (21)

changes sign from negative to positive; i.e., for u~ - e the system loses energy and motions
are damped, while at u, + e the system gains energy and motions are amplified.

4~ 3\ ' X\,3 ' ' ' ' ' -

30- .

~1o ~ Io

aa ,=9
~o ~ : __Mode 3"

,o-. ~ 6~1------~%,~ Mog3- ~ ".~, =


pli~r"~e~a -".e.~,...ik~k)de2+ e.,.._ fldode2- Z3 "~l2:

o
-r"
go "~o ,~.,, "..._
" ;o
ModeI- / -,2~0 ' 8 ~ .

-I0 I I I I I~ I
I0 20 30 40 50 60 70
Re(w)
Figure 7. Dimensionless complex frequency diagrams of a hysteretically damped cantilevered pipe for
fl= 0.65, 7 = 10, ~ = 0, and hystereticdamping coefficientIt = 0-I.

Figure 6 also shows the effect of viscoelastic dissipation, of the Kelvin-Voigt type, on the
dynamics of the system. Note the dramatic increase of damping with mode number at zero
flow veloci{y. With increasing flow, the loci of individual modes are completely changed by
the inclusion of dissipation. But the overall character of free motions is not radically altered.
However, oscillatory instability in this case occurs at u = 9.85, well below the critical value
when dissipative forces are absent. This destabilizing effect ofdissipative forces is a character-
istic of non-conservative gyroscopic systems [18, 20, 34].
Next consider the dynamics of the system when the dissipative forces are purely hysteretic.
If one assumes solutions to the equation of motion of the type
ll(~,z ) =Y(Oe I'=,
hystcretic damping corresponds to material properties such that 7to = p, where p is a constant
independent of to, called the hysteretic damping coefficient. This representation is valid only
if the system executes harmonic or nearly harmonic motions, implying that co must be nearly
wholly real [38]. Hysteretic damping approximates the dissipative mechanism in metals and
in certain types of rubber over frequency ranges of practical interest.
280 M.P. PAIDOUSSISAND N. T. ISSID
In this case if ico is a root of the equation of motion, the complex conjugate of io~ is not
necessarily a root. Hence, the symmetry of the root loci about the [Im (co)l-axis is destroyed
and it now becomes necessary to show the root loci on the left-hand of the frequency plane
also.
A typical case is shown in Figure 7, where p = 0.1. (The values o r e = 0.0189 and It = 0.1
give identical logarithmic decrement for the first mode at zero flow velocity.) Note that the
effect of hysteretic dissipation is not as severe as that of viscoelastic damping at small flow
velocities. For higher flow velocities, however, the behaviour of the mode-loci becomes quite
different from the previous cases; it is of particular interest that there would appear to exist
apparent discontinuities in the values of Im(og) as the [Im(o~)]-axis is crossed, if only the
positive [Im(og)]-plane were considered. It is also noteworthy that it is the locus of the first
mode starting with a negative Re(o)) that eventually becomes unstable. Finally, it must be
recalled that, in accordance with the limitations to the validity of the hysteretic dissipation
model referred to above in this section, only the portions of the loci near the [Re(o~)]-axis
will have physical significance.

I9 I ~121 l I i I i // 1 i i
e/" /
ezsLrV8 2oV / l

8, 2 ' ~ 7 [---
I.., II/; ~'~.-95 3 " mode ,,
I ~5 ~ ,o \.9 f--,~o-..~
7~s "%.6 ,, \ /
] V 8 l 6 S7if* l _

' r ~. 63.4 ,4 8 "~'~-~-"c_k


@0 6 8~ 8 6875 9
/ o..~ V.,=
,\
-I0 [ I I I I i~ ~15 I I// I I I
0 I0 20 30 40 50 60

Re(w)
Figure 8. Dimensionless complex frequericy diagrams of a cantilevered pipe for y= 10, ~= ~ =0;
- - - , .8 = 0.2; ,/i'= 0.3.

Finally, Figure 8 shows the complex frequency loci for cantilevered pipes with ? = 10 and
p = 0.2 and 0.3, respectively. Here it is the second mode that eventually becomes unstable.
The results of this figure will be discussed further in section 4.5.

3.3. CONDITIONSOF STABILITY


In connection with stability of a given system, from the practical viewpoint, of most
concern is knowledge of the limiting operating conditions which may be utilized without
incurring instability.
In the case of pinned-pinned and clamped-clamped pipes these critical conditions are
associated with buckling in the first mode. For stability, the values of u, 17 and - F should be
sufficiently small for v2 = u 2 +/-/(I - 2v~5)- F to be less than the values given in Table I.
These relationships are independent of ~, ]~ and x. The negative values of ? indicate that the
flow is vertically upwards, while ? = 0 applies either to horizontal systems or to very stiffand
relatively light vertical systems.
It is of interest that the difference in critical flow velocities for clamped-clamped and pinned-
pinned pipes is a factor of 2, only if? = 0. For ? > 0 this factor diminishes significantly.
STABILITY OF PIPES CONVEYING FLUID 281
In practice, misalignment and non-uniformity of pipes and supports tend to precipitate
buckling earlier than indicated above, as shown for example by Naguleswaran and Williams
[26]. Accordingly, for industrial design purposes it would be advisable to utilize a safety
factor in the above relationships.

TABLE I
Critical condition of stability
Values of v
Pinned-pinned pipes Clamped-clamped pins
-10 2.17 5.87
-5 2.71 6.08
0 n 2n
5 3.51 6.48
I0 3.83 6.67
50 5.56 7.97

In the case of cantilevered pipes the critical conditions are associated with oscillatory
instability. In this case the critical flow velocity depends on the parameters ~, z, fl and 7;/1 and
F are zero if the flow discharges to atmosphere. The effect of fl, z and 7, and of hysteretic
damping, p, on stability has been investigated previously [14, 15, 18]; the effect of ~ on
stability is illustrated in Figure 8. However, it should be noted that the effect of ~, similarly
to those o f ~ and of p, is not always destabilizing; in some cases, depending mainly on fl, the
system is stabilized by dissipative forces. In practice the destabilizing effect of dissipative
forces, when operative, is not very severe because the values ofcq # and z are small for metal
or rubber pipes surrounded by air; moreover, this effect is partly or completely overshadowed
by non-linear effects tending to result in critical flow velocities above those obtained by linear
theory. (Indeed, experiment~ indicate that in the vicinity of the critical flow velocity there is a
small unstable limit cycle in the phase plane about the position of equilibrium and a larger
stable limit cycle.) Consequently, the values of the critical flow velocities of the undamped
system as given in reference [18] may safely be used as conservative design criteria.

4. DYNAMICS OF PIPES WITH HARMONICALLY PERTURBED FLOW


4.1. GENERALFORMULATION
Consider the case where the flow velocity is perturbed harmonically: i.e., assume that

u = uo(l + p cos mr). (22)

In accordance with the assumption of plug flow made in the derivation of the equation of
motion, the excitation parameter, p, must be small.
Substituting equation (22) into equation (15) one obtains
a s ~1 a 4 tl a2q +
~ + ~-~ + [Uo2(l + p cosor) z -- F + 17(1 - 2vfi) - 7 - fl,n uo po~ sin toz]

. a 2t 1 , arl+x a~l+a2r I
aztl+2[3"Zu~176176176
+ (7 + #uz UoIt~ sin coz) ~ ~5" ~z -~z2=0" (23)
282 M . P . PAI, D O U S S I S A N D N . T . I S S I D

In practice pulsations in a circulating system are not likely to be purely harmonic. However,
arbitrary, small velocity perturbations may be decoupled into harmonic components as a
Fourier series, so that the analysis to follow may be regarded as applying to a single Fourier
component in such cases.
The aim here is to obtain the regions of parametric instability, upon presuming they exist,
for systems which are stable when the flow is entirely steady. Thus we shall not be concerned
in this paper with the amplitude of the resulting oscillations; this will be discussed in a forth-
coming paper.

4.2. ANALYSIS
Perhaps the simplest method of determining the regions of instability in the (p,00)-para-
meter space is Bolotin's method [39]. It should be mentioned at this stage that this method
does not yield the instabilities associated with combination resonances, the treatment of
which is deferred to the forthcoming paper referred to above.
Before Bolotin's method carl be applied, the continuous system must be rendered discrete,
which may be accomplished by application of the Ritz-Galerkin method. Accordingly,
express
t1(r z ) = Z ~b,(r (24)
t

where q,(z) are the generalized co-ordinates and ~br(~) are the eigenfunctions of a beam with
the same boundary conditions as the pipe under consideration. Thus, since the ~b,(r satisfy
all the boundary conditions of the system, they are admissible functions. It is presumed that
the series above may be truncated at a suitably high r in the normal way.
By substituting equation (24) into equation (23), multiplying the resulting equation by ~b,
and integrating it from 0 to 1, in view of the orthonormal property of the admissible functions
utilized, the following matric equation is obtained:
ij + (F + 2fl 1/2 uo(l + fl cos r q + {A + [u~(l 4- It cos 00z) z - ~ - fill2 Uolt00 sin 00r - F +

+ H(I - 2w5)] C + [y + fll/2 Uo//cosin mr] D + yB} q = 0, (25)

where q is the vector {q~,q2 . . . . }r, A is a diagonal matrix with elements ).~, ).~ being the ith
dimensionless beam eigenvalue, F is a diagonal matrix with elements ~2~ + • and B, C, D
are matrices whose elements b,,, c,,, d,,, respectively, are given by
1 1 1

0 0 0

the primes denoting differentiation with respect to 4. The values of b~,, e,, and de, are given in
Table 2 for various boundary conditions.
One can now proceed to apply Bolotin's method. Two types of parametric instability may
be distinguished, the so-called primary and secondary instabilities. For an undamped column
subjected to a harmonically perturbed end load the primary instabilities occur at 200,/00 =
1, 3, 5 . . . . . as/~ --> 0, where con are the natural frequencies of the column; the secondary
instabilities occur at 200J00 = 2, 4, .... Of all these the most important is that corresponding
to co/con= 2, the so-called principal primary instability; a principal secondary instability
(00/00n = 1) may similarly be defined.
To obtain the primary instabilities one writes

q= ~ {a, sin (89 + b, cos (89 (26)


k-1.3.5 . . . .
v
+

I i
i
C
e~

+
I
L .G-
(N

v
I

"7
I

I
L I

L.

"~' I 7_,
L I
~ I

e..,

C
Q I

"ft.
&
C
284 M . P . PA]DOOSSIS AND N. T. ISSID

which substituted into equation (25), after some manipulation, yields

{{ [ kc~ [kt~
k=1.3.5 ....

+[A+(u~+ 89 sin ~ +

+ -~T) h,+ {F+2~"~.,o~).. +

+ [~ + {,,o~ + ~,,~,,~- ~ + . { , - % , ) - ~)~ + ~o + ~ . ] N c o u r T ) +

+ _ p,,2 Uol,Bbk + l,u~Ca, + 89 '/z UoIwg(D - C)bk/sin ~ tox) +

2 toz +{ 88

+ {88 2 't 2 Cak} sin


(k4)
T toz + {Ill" zrg Cb,} cos mr = 0. (27)

Upon expanding this equation and collecting terms in cos~o~T, sin89 cos89 etc., the
coetficients of which must vanish independently, one obtains a matrix equation of the form

... G33 G3t G32 G3~ ... a'3


~" * . . . . . . . . . . . . . . . . . . . . . . . . . 1
................. i "'"

9 .. G23 G2t G22 G2.~ ...


................ ,1

9 .. G43 G4t G42 G44 ... b3


........................... /~.

which is of infinite order. The Gsk are coefficients ofak or bk in the equations for sin(89 or
cos(89 where o d d j ' s are associated with sin(89 and evcnj's with c o s [ ~ ( j - l)tor];
similarly, odd k's are associated with a~ and even k's with bk-x.
As shown by Bolotin, the boundary equation ofthe instability regions is obtained by setting
the determinant of the matrix of the Gjk equal to zero. Of course the determinant is of infinite
order, but it belongs to the class of normal determinants and is therefore absolutely convergent.
STABILITY OF PIPES CONVEYING FLUID 285

Hence, the boundaries of instability may be obtained approximately by equating to zero the
determinant shown circumscribed in dashed lines--this will be called the k = 1 approxima-
tion, which necessarily yields only the principal region of instability; a better approximation,
as well as higher regions, would be obtained if the determinant involving all the terms shown
in equation (28) is used--this will be called the k = 3 approximation; and so on. Of course, the
series ofequation (24) must be truncated at an adequately high r, which defines the order of Gjk.
Now, the secondary instabilities may be obtained by expressing
q = ~ {a k sin (89 + bk cos (89 (29)
k=O, 2, 4.. . . .

which substituted into equation (25) yields once again equation (27), but with the summation
now being over k = 0, 2, 4 . . . . . Collecting terms as before one is led to a matric equation
equivalent to equation (28) but with a vector {... a4, a2, bo, b2, ba.... }r, and thence to a vanish-
ing determinant which yields the boundaries of secondary instabilities.

4.3. CALCULATIONS
A computer program was written which determines the boundaries of instability in the
~ , Og)-parameter space for any given system. The program starts with a given It and determines
the Og'sfor which the determinant arising from equation (28), or its equivalent for Secondary
instabilities, vanishes; this is done by finding the points where the determinant changes sign
as o9is incremented, starting with to = 0. Then with these values ofo9 used as reference values,
the corresponding og's for p + t~It are determined in a similar manner. (The zeros of the
determinant may also be found more precisely by a subroutine utilizing a modified secant
method.) The prograrn continues until the instability regions are traced for the desired range
ofp.
It was found that, typically, the instability boundaries associated with the first mode of
pinned-pinned and clamped-clamped pipes could be determined with adequate precision by
using the k = 1 approximation for the principal primary region and the k = 2 approximation
for the secondary, and truncating the series of equation (24) at r = n = 2. For cantilevered
pipes, on the other hand, it was found necessary to use typically k = 3 and k = 2, respectively,
and n = 5. Generally, the higher the flow velocity, or p, or the mode number, the higher n
and k have to be. Most of the calculations presented here were performed with k = 3 or 2
and n = 5. A more detailed discussion of convergence and of the numerical analysis is given
in reference [40].
It will not be attempted to present here theoretical results covering all possible physical
systems, in view of the large number of parameters involved. Rather, only a sufficient number
of calculations will be presented to illustrate the general behaviour of the system.

4.4. RESULTS FOR CLAMPED-CLAMPED PIPES


Figure 9 shows the regions of parametric instability in the range 0.6 < o9/o9ol < 5.7 for a
clamped-clamped pipe (uo = 2, fl~/2 =0.5, 7 = 10, F = / / = ~ = • where Ogox is the
first-mode natural frequency at zero flow. If instead of O9/O9ol,the ordinate O9/O9,had been
utilized, where O9, is the actual natural frequency for the mode concerned at u = 2, then the
principal primary regions of all the modes would begint at O9/O9,= 2 (for p = 0+), the second
primary region at O9/O9,= 2/3, etc.; the principal secondary region at O9/O9,= l, and so on.
With the ordinate used here, associated with the first mode are (i) the principal primary
region beginning at O9/O9ol~ 1.90, (ii) the principal secondary at O9/O9o~~ 0.95, and (iii) the
second primary region beginning at O9/O9o~~ 0.64. Similarly, associated with the second mode,
t By " b e g i n n i n g " o f an instability region is meant the intercept on the ordinate.
286 M.P. PA1DOUSSISAND N. T. ISSID
the corresponding regions begin at o~/o9ol N 5.24, 2.62 and 1.75, respectively. Associated
with the third mode are the (i) principal secondary region beginning at o9/mol = 5.13, and the
second primary region at co/coo~ = 3.42. The uppermost region in the figure is associated with
the fourth mode. It is noted that the principal primary regions o f instability are by far the
most extensive in the parametric plane.
Figure 10 shows the effect o f viscous and viscoelastic dissipation on parametric instability
for the same system. (The value of~z used in Figure 10(b) was calculated so that the viscously

5.8 I I I I o"

................... .-,---r."
56

5-4

5-2
"-A -i.~-A - - A --A --A,,=L-- - - A , ~-

50

48

9~ 3-4
3
2-7
2-6
2'0~

1"8
.%.-=_

o.o"
0.9
o7"
0.6
I I l I d
0 01 0.2 0-3 0.4 0.5
/.t

Figure 9. Parametric instability boundaries for a clamped-clamped pipe (uo=2, pit2=0-5, 7 = 10,
T'=/7 = = = ~ = 0). The system is unstable within the triangular regions with apex on the ordinate, and stable
without. Boundaries associated with the first mode, ; with the second mode, - - - ; with the third mode,
-A,--A-; with the fourth mode, - e - - e - .

5-8 i J ~ I ~ _ I I I I

56 .e~'__~ __~_~
54 ~ / ~ ' ~
5"2 ~
50

4.8
3.4
2-7
26
2.084
1"8
IO
09
0-7
06
1 1 I I I I 1 I
0 0 I 0"2 0-3 0"4 0'50 O'l 0"2 0-3 04 05
/z

(O) (b)

Figure 10. Parametric instability boundaries for the same system as in Figure 9, (a) with viscous damping
(~ = 0.4), and (b) with viscoelastic damping (~ = 7.54 x 10-4). Regions associated with the first mode, - - ;
with the second mode, - - - - ; with the third mode, -h,__A-; with the fourth mode, - o - - o - .
STABILITY OF PIPES CONVEYING FLUID 287
and viscoelastically damped systems have the same logarithmic decrement in the first mode at
Uo = 0.) It is seen that dissipation diminishes the extent of instability regions, such that origin-
ally narrow instability regions are eliminated altogether. Moreover, it now becomes necessary
that a finite p be used for instabilities to occur. It is noted that viscoelastic dissipation has a
more pronounced effect on the higher modes, as might have been expected, so that all but
the principal primary regions of instability are eliminated. These results establish that the
principal regions ofinstability are by far the most important and the ones likely to be observed
in practice. Accordingly, further calculations will be confined to these regions.
Figure 11 shows the effect of flow velocity and viscous damping on the principal regions
of instability associated with the first mode of a clamped-clamped pipe. It is noted that as the
flow velocity increases the regions of instability are displaced downwards, which reflects the

I I I i

2"0

1"8
Uo:3

1"6
JJ

IO t,,b=I

t.,'O: I. 8

' ;~5-~-~__~.~_.,
0-8 I I f f
0"1 0"2 0-3 0.4 05

F i g u r e 11. The effect o f flow velocity, Uo, and viscous damping on the principal instabilities associated with
the first m o d e o f a clamped-clamped pipe. ( F = / 7 = = = 0, flt/2 = 0.2, 7 = 10) for three values o f Uo.
x = 0; - - - - - , ~ = 0.2; - - - , x = 0-5.

i I i i

...i
20 ~.~2_._ ......

-.~ " ~ ~ "------o ~ 0""

1.0

08 I I l I
o o.i o.z 0-3 0-4 0 5

F
Figure 12. T h e effect o f f l on the principal instabilities associated with the first mode o f a clamped-clamped
pipe ( / " = / 7 = c( = x = 0, Uo = 2, 7 = 10). , flu2 = 0'2; - - - , ,Btl= = 0"5; - e - - e - , #=/2 = 0.8.
288 M. P. PAIDOUSSISAND N. T. ISSID
decrease of the first-mode frequency with flow (cf. Figure 5). It is also noted that the regions
of instability become broader with increasing flow, and that the effect of damping is corre-
spondingly attenuated. Thus at uo = 3 the primary region of instability is very little affected
by fairly large dissipation (• = 0.5), and the system is subject to parametric instability even
when IL < 0.1.
Finally, Figure 12 shows the effect offl on parametric instabilities. It is seen that with increas-
ing fl the regions of instability become broader; it is perhaps worth mentioning that with the
equation of motion used by Chen the width of the unstable regions appears to be independent
of ft. The unstable regions are also displaced downwards with increasing fl, which reflects the
lowering of the natural frequencies as fl increases for this particular flow velocity; this effect
is shown in Figure 3, where col for 13= 0.5 is lower than for ,8 = 0.1 at u = 2 (although to the
scale of this figure the difference is only just discernible).
It is noticed that the lower boundary of the principal primary instability associated with
fit/2= 0.8 in Figure 12 is not straight; this is a characteristic of cases where the instability
boundary concerned is close to another region of instability (which is not shown here for the
sake of simplicity). This also applies to the primary region for Uo = 3 in Figure 11.
The results shown here are for one specific value of?, and F --- H = 0. However, they are
typical of results obtained with other values of F, H, and ?, including ? = 0.

4.5. RESULTSFOR CANTILEVEREDPIPES


AS seen in section 3.1 free motions of cantilevered pipes are damped by steady flow below
the critical value, even in the absence of dissipation. Consequently, parametric instabilities
are not possible for all flow velocities--as had been the case with Clamped=clamped pipes.
Moreover, the instabilities are selectively associated with only some of the modes of the
system; thus in all the calculations performed, at least for relatively low flow velocities, no
instabilities associated with the first mode were ever found.

i i i i t ~

22
.f"
\\ -

I.B

P~
14

1.0

06

I I I I
0 0.I 0.2 05 0-4 0.,5

Figure 13. Parametric instability boundaries for a cantilevered pipe (~= x= 0, ,B=0.2, 7= I0). - - -
Uo=4"5;-----,uo=5"5; 9 , tto=6.
STABILITY OF PIPES CONVEYING FLUID 289
Figure 13 shows the parametric instability regions for a cantilevered pipe with fl = 0.2,
y = I0, ct = x = 0 for uo = 4.5, 5-5 and 6 in the range ta/too2 < 2.4. (Mode loci under steady
flow conditions for this system are shown in Figure 8.) For uo ~<4 it was found that no para-
metric instabilities could occur, at least for the range of It considered. The large regions of
instability in the middle of the figure are the principal primary regions associated with the
second mode, while at the bottom is a principal secondary region which occurs for Uo = 6
only. The small regions at the top are principal secondary regions associated with the third
mode. It is noted that (i) a finite and quite substantial value of it is necessary to induce para-
metric oscillations, (ii) this value of p decreases with increasing flow velocity, and (iii) the
instability regions are more extensive at the higher flow velocities. These results are somewhat
similar to those of Figure 11 for the damped clamped-clamped pipe with ~. = 0.5. Unlike for
the clamped-clamped pipe, however, damping in this case (arising by the action of the
Coriolis forces) is intimately connected with the dynamics of the system; consequently, its
effect on parametric instability is not uniform, nor easily predictable, as will be seen further
below.

I I I I I I I I
;~.81
48 26]
44 241

40 /~*/" &,,..&~ A - 221


36 /.~ :- 201
r 181
%,
16J

[4

20 ~ . = ~ - IF, =

I0
,6 . "d,S~" ."-
08
O8 O6
04 I I q"~-II"~-- 04
01 0 ? 03 04 05 0 OI 02 03 04 05
/1
(o) (bl
F i g u r e 14. P a r a m e t r i c instability b o u n d a r i e s for a cantilevered pipe (et = ~ = 0, fl = 0"3, 7 = 10). (a) P r i m a r y
instability r e g i o n s ; (b) s e c o n d a r y instability regions. - e - - e - , U o = 6; . . . . , u o = 7"5; - A - - A - , Uo = 8;
-I--I-, uo = 8"5; - - , Uo ----8"6875.

Figures 14(a) and (b) show, respectively, the primary and secondary instability regions, for
the range of frequencies shown, of a system with fl = 0.3, y = 10, ~ = y. = 0. (The steady-flow
mode loci for this system are shown in Figure 8.) The three uppermost regions of instability in
Figure 14(a), for tto = 6, 7.5 and 8, are principal primary regions associated with the third
mode, while the two large regions in the middle, for Uo = 8.5 and 8.6875, are a mixture of
principal primary regions associated with the second and third modes. By reference to Figure
8 it is noted that for uo = 8.5 and 8.6875 the real frequencies of oscillation of the second and
third modes are quite close.
Finally, the smaller regions at the bottom of Figure 14(a) may similarly be divided into the
following two groups: (i) the regions for uo = 6, 7.5 and 8 are mixtures of principal primary
regions associated with the second mode and of second primary regions associated with the
third mode; (ii) the regions for Uo = 8.5 and 8.6875 are mixtures of second primary regions
associated with the second and third modes. This fusion of the regions of instability is shown
particularly well in the cases ofuo = 8, 8.5 and 8.6875, where each of the regions is formed of
290 M.P. PAIDOUSSIS AND N. T. ISSID

two interlinked distinct zones, the upper o f which is related to the second mode and the lower
to the third mode.
In Figure 14(b) the upper region (Uo = 6) is the principal secondary one associated with t h e
third mode, while the remaining regions are all mixtures o f principal secondary regions
associated with the second and third modes. The upper areas o f the latter are associated with
the third m o d e and the lower areas with the second, except for Uo = 8.6875 where no such
distinction m a y be m a d e ; this is attributed to the proximity o f Re(o92) and Re(cos) for
Uo = 8.6875 as shown in Figure 8.
One interesting aspect o f the results o f this section is that no parametric instabilities occur
at low values ofuo, where the d a m p i n g effect o f the mean flow is small. It is recalled, however,
that both d a m p i n g (due to Coriolis forces) and parametric excitation are proportional to Uo
for a constant # ; hence, for a given Uo the magnitudes o f the parametric excitation and o f
d a m p i n g are predetermined, and as uo is decreased they both decrease proportionally, t
Thus one c a n n o t say a priori for what range o f Uo parametric instabilities may occur, if at all,
for a given range o f #. In this connection it is noteworthy that the regions o f parametric
instability appear to be associated with modes that display local minima in Im(o~) in the
complex frequency plane (Figure 8) and, moreover, at values o f uo close to or above these
minima.

4.6. COMPARISON ~,VITH PREVIOUS WORK

As mentioned earlier Chen has already analyzed the problem at h a n d in the case o f pinned-
pinned pipes, and has presented some theoretical boundaries o f instability [32]. However, as
shown in this paper, his equation o f m o t i o n is in error. Accordingly, it Would be o f interest to
c o m p a r e the results obtained by Chen's and this corrected theory.

I I I I I I I 2.0 I i I I
I-8
1"6
1"6
1.2
'~ 1.4
08
1.2
08 04

0'6 I I I I I I I I I I l
0-1 0.2 0"3 0-4 0 0-1 0"2 O3 0-4 0,5

(a) (b)

Figure 15. Comparison of the parametric instability regions as obtained by this theory, , and by the
theory of reference [32], - - - - - ; (a) for a pinned-pinned pipe (Uo= 0'6n, Btt2 = 0"8, T = / 7 = ~ = ~r = y = 0);
(b) for a cantilevered pipe (uo = 6, B = 0"2, y = I0, ~ = u = 0).

Figure 15(a) shows the principal regions o f instability associated with the first mode o f a
pinned-pinned pipe (Uo = 0.6~, fl1/2 = 0.8, 1, = F = 11 = ~ = ~ = 0) and equivalent results are
shown in Figure 15(b) for a cantilevered pipe (Uo = 6, fl = 0-2, y = 10, ~ = ~ = 0). As m a y be
seen, Chen's theory generally underestimates the extent o f the instability regions; however, it

t A different behaviour is obtained if the excitation force can be varied independently of damping. Thus
calculations in which an external axial load Focos tor was imposed and u kept constant showed that parametric
instabilities may then occur even for small u.
STABILITY OF PIPES CONVEYING FLUID 291

does predict their general shape, and this was the reason for not undertaking any further
calculations involving pinned-pinned pipes. Evidently neglect of the axial movements of
the pipe--or incomplete accounting of fluid acceleration--is not as serious as may have been
supposed. Nevertheless, it should be added that the error incurred by Chen's theory generally
increases with both uo and ft.
5. CONCLUSION
In this paper a rather extensive literature review was undertaken on the topic of stability
of pipes conveying fluid, in the context of general stability theory.
The dynamics of pipes conveying steady flow was re-examined, and it was shown that a
pipe with both ends supported, which is a conservative gyroscopic system, is not only subject
to divergence (buckling) but also to coupled=mode flutter at higher flow velocities. A canti-
levered pipe, on the other hand, which is a circulatory system, is only subJect to flutter of the
single degree-of=freedom, as opposed to coupled-mode, kind. The existence of flutter in
conservative systems was discussed and it was shown to be associated with the gyroscopic
(Coriolis) forces; moreover, it was seen that flutter in such systems can only be of the coupled-
mode kind.
Finally, some design criteria were proposed for the maximum allowable flow velocity,
from the point of view of stability, for clamped=clamped, pinned=pinned and cantilevered
pipes.
The next topic investigated was the dynamics of pipes containing fluid the velocity of which
is harmonically perturbed. It was shown that the equation of motion as derived previously
by Chen is in error, as it neglects the axial acceleration ofthe fluid associated with the imposed
velocity perturbations; this error may result in underestimating the parametric instability
zones appreciably.
It was shown that pipes with both ends supported (conservative system) are subject to a
multitude of parametric instabilities in all their modes, although dissipation eliminates most
of these instabilities, excepting the principal ones. The higher the dissipation, the higher has
to be the amplitude of the harmonic component of flow for parametric instability to occur.
However, the effectiveness of dissipative forces in this respect diminishes with increasing
flow velocity, so that appreciably damped systems may be subject to parametric instabilities
even when the pulsation is quite weak, provided that the flow velocity is sufficiently high.
Cantilevered pipes (circulatory system) are also subject to parametric instabilities, pro-
vided that the flow velocity is not too small. The gyroscopic (Coriolis) forces in this case do
work in the system, and they play a role somewhat analogous to that of dissipative forces.
It was found that only some ofthe possible instability zones exist in this case and are associated
with specific modes (or combinations of modes) and specific ranges of flow velocities.
Of course, the principal new findings of this work must be tested by experiment. Work is
underway to examine whether coupled=mode flutter of damped=damped pipes is indeed
possible as predicted by this linear theory, or whether non-linear effects render it impossible.
Also, experimental work is underway in conjunction with the parametric instabilities caused
by harmonic velocity perturbations. Freliminary experiments indicate qualitative agreement
with the behaviour predicted here by theory. The results of the experimental investigation
will be reported in the near future.

ACKNOWLEDGMENTS
This work was supported financially by the National Research Council of Canada (Grant
~ A4366), the Defense Research Board of Canada (Grant ~ 9550-47)andAtomic Energy
of~Canada i(Whiteshell Nuclear Research !Establishment), ,whose .assistance is gratefully
acknowledged.
292 M.P. PAIDOUSSISAND N. T. ISSID

The authors are also grateful to Paul des Trois Maisons for undertaking the calculations
for the damped cantilevered pipes in steady flow.

REFERENCES
1. J. AITKEN1878 Philosophical Magazine, Series V, 5, 81-105. An account of some experiments
on rigidity produced by centrifugal force.
2. F.-J. BOURm~RES1939 Publications Scicntifiques et Techniques da )~linistdre de l'Air, No. 147.
Sur un ph6nom6ne d'oscillation auto-entretenue en m6canique des fluides r6els.
3. M. P. PAYDOUSSlS1972 Mechanical Engineering Research Laboratories Technical Note 72-2,
McGill University. Vibration of tubes containing flowing fluid.
4. H. ASHLEYand G. HAVmAND1950JournalofAppliedMechanics 17, 229-232. Bending vibrations
of a pipeline containing flowing fluid.
5. V. P. FEODOS'EV1951 Inzhenernyi Sbornik 10, 169-170. Vibrations and stability of a pipe when
liquid flows through it.
6. G. W. HOtJSNER 1952 Journal ofApplied 3Iechanics 19, 205-208. Bending vibrations of a pipe
line containing flowing fluid.
7. F.I. NIORDSON1953 Kungliga Tekniska H6gskolans Handlingar, No. 73. Vibrations of a cylindrical
tube containing flowing fluid.
8. R. H. LONG, JR. 1955 Journal of Applied Mechanics 22, 65-68. Experimental and theoretical
study of transverse vibration of a tube containing flowing fluid.
9. G. H. HANDELMAN1955 Quarterlyof Applied Mathematics 13, 326-330. A note on the transverse
vibration of a tube containing flowing fluid.
10. G. HEttCRICH1956 Zeitschrifif~r angewandte Mathematik undMechanik 36, 417--427. Vibrations
of tubes with flow.
11. V. V. BOLOTtN 1956 Trudy Moskovskogo Energeticheskogo Instituta; No. 19, 272-291. End
deformations of flexible pipelines.
12. H.-H. Hu and W.-S. TSOON 1957 Proceedings of Theoretical and Applied Mechanics (India),
203-216. On the flexural vibrations of a pipe line containing flowing fluid.
13. A. A. MOVCHAN1965 Journal of Applied Mathematics and Mechanics (Prikladnaia Matematika i
Mekhanika) 29, 760-762. On the problem of stability of a pipe with fluid flowing through it.
14. R. W. GREGORYand M. P. PAiDOUSSlS1966a Proceedings of the Royal Society (London) A, 293,
512-527. Unstable oscillation of tubular cantilevers conveying fluid. I. Theory.
15. R. W. GREGORYand M. P. PMDOUSSm1966b Proceedings of the Royal Society (LondotO A, 293,
528-542. Unstable oscillation of tubular cantilevers conveying fluid. II. Experiments.
16. T. BROOKEBENJAMIN1961aProceedings of the RoyalSocicty (London) A, 261,457--486. Dynamics
of a system of articulated pipes conveying fluid. I. Theory.
17. T. BROOKEBENJAMIN1961b Proceedingsof the Royal Society (LondolO A, 261,487--499. Dynamics
of a system of articulated pipes conveying fluid. II. Experiments.
18. M. P. PMDOUSSlS1970 The Journal of)llechanical Engineering Science 12, 85-103. Dynamics of
tubular cantilevers conveying fluid.
19. M. P. PATDOUSSlSand E. B. DEKSNIS1970 The Journal of Mechanical Enghleerhtg Science 12,
288-300. Articulated models of cantilevers conveying fluid: the study of a paradox.
20. S. NEMAT-NASSER,S. N. PRASADand G. HERRMANN1966 American Inst#ute of Aeronautics attd
Astronautics Journal 4, 1276-1280. Destabilizing effect of velocity-dependent forces in non-
conservative continuous systems.
21. G. HERRMANN1967 Applied Mechanics Reviews 20, 103-108. Stability of equilibrium of elastic
systems subjected to non-conservative forces.
22. G. HERRMANNand S. NEMAT-NAssER1967 lnternationalJoltrnalof Solids and Stmctures 3, 39-52.
Instability modes of cantilevered bars induced by fluid flow through attached pipes.
23. J. C. WmEYand R. E. FORKERT1972 Journal ofthe Enghleering Mechanics Division,Proceedings
of the American Society of Civil Engineers 98, 1353-1364. Beams subjected to follower force
within the span.
24. R. A. STEINand M. W. TORBINER1970 Journal of Applied Mechanics 92, 906-916. Vibration of
pipes containing flowing fluids.
25. J. A. HARINGX 1952 Philips Research Report 7, 112-118. Instability of thin-walled cylinders
subjected to internal pressure.
26. S. NAGULESWARANand C. J. H. WILLIAMS1968 The Journal of Mechanical Engineering Science
10, 228-238. Lateral vibration of a pipe conveying fluid.
STABILITYOF PIPES CONVEYINGFLUID 293
27. A. L. THURMANand C. D. MOTE JR. 1969 Journal of Engineering for Industry, Transactions o f
the American Society of ~Iechanical Enghteers 91, 1147-1155. Non-linear oscillation of a
cylinder containing flowing fluid.
28. S.-S. CHEN 1971 American Society of Mechanical Engineers, Paper 71-Vibr-39. Flow-induced
instability of an elastic tube.
29. M. P. PAiOOUSSISand J.-P. DENISE 1971 Journal of Soundand Vibration 16, 459-461. Flutter of
cylindrical shells conveying fluid.
30. M. P. PAiDOOSSISand J.-P. DENISE 1972 Journal of Sound and Vibration 20, 9-26. Flutter of thin
cylindrical shells conveying fluid.
31. D. S. WEAVERand T. E. UNNY 1973 Journal of Applied Mechanics 40, 48-52. Dynamic stability
of fluid-conveying pipes.
32. S.-S. CHEN 1971 Journal of the Engineering Mechanics Division, Proceedings of the American
Society of Civil Engh~eers 97, 1469-1485. Dynamic stability of a tube conveying fluid.
33. S. TIMOSHENKOand J. M. GERE 1961 Theory of Elastic Stability. New York: McGraw Hill Book
Co.
34. H. ZIEGLER1968Prbtciplesof Structural Stability. Waltham, Massachusetts: Blaisdell Publishing
Co.
35. M. P. PAiDOUSSlS 1966 Journal of Fluid Mechanics 26, 717-736. Dynamics of flexible slender
cylinders in axial flow--I. Theory.
36. M. P. PATDOUSSlS1972 Proceedings of lUTAM]IAHR Symposium on Flow-induced Structural
Vibrations (Karlsruhe), Paper G3. Stability of tubular cylinders conveying fluid.
37. R. C. SHIEH1971 hzternationalJournal of Non-Linear Mechanics 5, 495-509. Energy and varia-
tional principles for generalized (gyroscopic) conservative problems.
38. R. E. D. BISHOPand D. C. JOHNSON1960 The Mechanics of Vibration. Cambridge: Cambridge
University Press.
39. V. V. BOLOTIN1964 The Dynamic Stability of Elastic Systems. San Francisco: Holden Day Inc.
40. N. T. ISSID 1973 M.Eng. Thesis, Department of Mechanical Engineering, McGill University.
Parametric instabilities of tubes conveying fluid.
In a short historical re~,iew, such as the one undertaken in the Introduction, it is perhaps inevitable
that some important contributions are omitted, either by oversight or for brevity. Among those
which should not have been left out are references [41] and [42] given below; reference [43], a most
interesting experimental paper, has appeared after the writing of this paper.
41. H. L. DoDos JR. and H. L. RUNVAN1965 NASA TechnicalNote D-2870. Effect ofhigh velocity
fluid flow on the bending vibrations and static divergence of a simply supported plate.
42. C.D. MOTEJR. 1971 Journal of the Engineering Afechanics Division,Proceedingsof the American
Society of Civil Engineers 97, 645-656. Non-conservative stability by finite element.
43. H.-S. Lzo and C. D. MoTE JR. 1973 American Society ofMechanicalEnghleersPaper 73-DET-118.
Dynamic response of pipes transporting fluids.

APPENDIX I

THE EQUATIONOF MOTIONBY HAMILTON'SPRINCIPLE

Benjamin [16] has shown that the statement of Hamilton's principle for a pipe conveying
fluid, in the absence of dissipative forces, is given by
t2 r2

6 f Ldt- f M U ( R + U'O'3Rdt=O, (AI)


I1 I1

where R is the position vector of the downstream end of the pipet measured from the un-
deformed straight position, and 9 is a unit vector tangential to the free end. The Lagrangian
is L = Tx + 7"2 + Vl + V2, where 7"1 and V~ are the kinetic and potential energies associated
with the pipe, and 7"2 and 1/2 are the corresponding quantities for the enclosed fluid.

t In this Appendix dots and primes denote differentiation with respect to t and x, respectively.
294 M.P. PAIDOUSSIS AND N. T. ISSID

Small motions about the position of equilibrium may be expressed in terms of the lateral
displacement y and the corresponding axial contraction in the x-direction by c, which is
given by

c=
0
f (ds - dx) ~- 89j, (),,)z dx.
0

Clearly, i f y .,. O(t), then c ~ O(t2). One may now write R = --CLi + YLJ, where i a n d j are unit
vectors in the x - and y--directions, respectively. Also ~ = icosy~. + jsiny~. = i + y/~j + O(~2).
Substituting these expressions into equation (A1) one obtains
t2 12

j (L + MUZcL)dt - ~ MU(pL + UyL) 6yLdt = 0, (A2)


1'1 tl

correct to O(e2).
Now, the case of a pipe with neither pressurization nor external tension will be considered
for the sake of simplicity. One has
L

eL = 89f (y,)2 dx,


0
L .Lx

VI= 89 f (y')2dx + ~ m g f f (y')2dxdx,


0 0 0
L x

Vz = 89 f f(y)Zdxdx,
0 0
L

Tl = ~ m j p 2 d x ,
0
L

T, = 89 + M f (89 U.Py'- Ud)dx, (A3)


0

where the last expression was obtained by using the resultant of the fluid velocities in the
y-direction, .f, + Uy', and in the x-direction, Ucosy' - ~ ~ U[I - .~(y,)2] _ ~:, neglecting
terms smaller than O(e2).
By applying the usual variational techniques the equation of small lateral motions is
obtained:

Ely t* + MU2 y " + M O ( L - x) y" - (M + m)g[(L - x ) y ] ' + 2MU.p' + ( M + re)j; = 0. (A4)


If axial movements were neglected, i.e., ifthe term UOin the expression for T2 in equation (A3)
were eliminated, then the term M(I(L - x)y" in the equation of motion is replaced by M(ly',
and this corresponds to the equation derived by Chen [32]. However, there is no justification
for doing so, as proper accounting of the order of magnitude of the various terms dictates.

Das könnte Ihnen auch gefallen