Sie sind auf Seite 1von 56

Prog. Aerospace Sci. Vol. 26, pp. 169-224, 1989.

Printed in Great Britain. All fights reserved

0376--0421/89 $0.00+0.50 1989 Maxwell Pergamon Macmillan plc

PHENOMENA AND MODELLING OF FLOW-INDUCED VIBRATIONS OF BLUFF BODIES


GEOFFREY PARKINSON

University of British Columbia, Vancouver, Canada (Received 17 February 1989)

CONTENTS
1. SCOPE 1.1. Forms considered 1.2. Other forms 2. PHYSICAL CHARACTERISTICS 2.1. Section afterbodies 2.1.1. Importance of afterbody 2.1.2. Afterbody effects for rectangular sections 2.2. Wake vortex phenomena 2.2.1. Issues of experimental methods 2.2.2. Wake vortex effects on oscillating circular and D-section cylinders 2.3. Combined effects 2.3.1. Parameters of aeroelastic and hydroelastic vibrations 2.3.2. Hydro- and aeroelastic phenomena 2.3.3. Phenomena from forced-vibration experiments 3. MODELLING GALLOPING 3.1. Quasisteady theory 3.2. Two-dimensional applications 3.3. Extension to three-dimensional problems 4. MODELLING VORTEX-INDUCED AND COMBINED VIBRATIONS 4.1. Flow-field models 4.2. Nonlinear wake oscillator models 4.2.1. Vortex-induced vibration models 4.2.2. Combined vibration models REFERENCES 169 169 170 171 171 171 173 176 176 177 188 188 189 195 198 198 203 206 209 209 210 211 217 222

1. SCOPE
1.1. FORMS CONSIDERED

There are many forms of flow-induced vibration of elastic bodies, and many aspects of scientific and engineering interest in the vibrations. The fluid dynamicist is challenged by puzzling phenomena of unsteady, separated flow. The applied mathematician finds new employment for the theory of differential equations. The professional engineer usually wishes to learn how to prevent or eliminate the vibrations. All of this interest has generated a large and growing literature, which in turn has led to several surveys. Recent surveys include a book by Blevins (1977), and articles by McCroskey (1977), Sarpkaya (1979), and Ikarman (1984). Blevins' useful book, intended primarily for the practising engineer, treats a comprehensive list of topics relevant to civil, mechanical and marine engineering. In it he provides less detail of the fluid mechanics of the phenomena than of the analysis of actual vibrations, an emphasis in keeping with the aims of the book. He does not, however, treat classical flutter of lifting surfaces in streamline flow, important to the aeronautical engineer and naval architect. McCroskey's article does describe recent research in flutter as well as in many of the topics treated by Blevins, and in it he gives considerable insight into the fluid mechanics. Sarpkaya's article deals exclusively with vibrations of the circular cylinder, either vortex-induced or forced, under two-dimensional conditions. In it he discusses experimental data from several laboratories, and considers several of the theoretical models proposed for vortex-induced vibrations. Of these, he devotes the most space to a description of his form of numerical flow-field model, in which the transverse exciting force is produced by the
169

170

G. PARKINSON

unsteady pressure field of a large array of potential line vortices. Each is free to move under the influences of the rest of the flow, and they represent the shear layers bounding the actual wake. This type of model will be discussed briefly in a later section of this article. In his article, Sarpkaya performs a useful service by considering in some detail the correct roles for, and measurement of, added mass and fluid damping effects in flow-induced vibration. Bearman's more recent review also deals with vortex-induced and forced vibrations of circular cylinders, but he gives equal attention to square-section cylinders and the article essentially covers aspects of experimentally observed vortex-shedding phenomena. Bearman's views on these issues will also be considered in a later section. The present review, like Sarpkaya's and Bearman's, is narrow in scope and treats in detail only the transverse vibrations of single long bodies of bluff section in steady incident flow normal to their span. Here a bluff section is understood to mean one from which the flow separates, producing two shear layers bounding a relatively broad wake, and steady incident flow is meant to rule out any organized transient or oscillatory character, although random turbulence may be allowed. Of course, such turbulence will directly cause transverse buffeting vibrations, but these are not considered here, and only indirect effects of incident turbulence are examined. Buffeting vibrations have a considerable literature, largely in wind engineering contexts (see Blevins, 1977), with the widespread use of rational statistical concepts owing much to the pioneering paper by Davenport (1961). Recent developments in analytical methods of predicting fluctuating surface pressures on bodies in turbulent flow, such as the paper by Durbin and Hunt (1979), give promise of further progress. With buffeting eliminated from further consideration, the remaining forms of transverse vibration are galloping and vortex-induced, and the modelling of these forms and their combinations will be examined in some detail in the following sections. The review can thus be considered an updated and expanded version of an earlier survey, Parkinson (1974). It is appropriate to consider these two forms of vibration together, since they have many features in common: 1. Both occur as transverse vibrations of single long bodies. 2. Both can occur for any noncircular bluff section with an appreciable afterbody (defined as the part of the section downstream of the separation points). 3. Both occur in steady incident flow normal to the body span. 4. Both occur at a frequency close to a natural frequency of the elastic body. 5. Both are of nearly harmonic waveform in air flow, and typically show little random amplitude modulation. 6. Both behave as nonlinear oscillators. 7. Both result from interaction of the wake with the section afterbody. However, despite all of the above common features, the two forms are essentially different. Galloping is an instability phenomenon, a self-excited vibration in the conventional sense in that the exciting force vanishes with the motion. It is potentially catastrophic since the amplitude increases continuously with flow velocity above a critical value. Vortex-induced vibration, on the other hand, is generally considered to be a form of nonlinear resonance although Ericsson (1984) takes a different view. Vibrations occur only over a limited range of flow velocities.containing the velocity at which the wake vortices from the stationary body form at the body natural frequency, and the resulting vibration amplitudes are self-limiting.

1.2. OTHER FORMS Before commencing a detailed study of these two forms of vibration, and of the body-flow interactions that produce them, it seems appropriate to at least mention some of the other important forms of flow-induced vibration. For a single body, turbulent buffeting has already been mentioned. Vortex-induced vibrations in line with, rather than transverse to, t h e flow can also occur, and studies by King et al. (1973), Griffin and Rarlaberg (1976), Currie and Turnbull (1987), and a review by Naudascher (1987) are of particular interest. Vibrations

Flow-inducedvibrationsof bluffbodies

171

of submerged piles and other marine structures caused by the oscillatory loads of wave action can lead to fatigue failures, and these phenomena are being studied in many laboratories, usually under idealized conditions. Sarpkaya (1976) has pioneered the use of the U-tube water tunnel in such studies, and the results of significant recent work can be found in papers by Maull and Milliner (1978) and by Bearman and Graham (1979). Sarpkaya (1985) gives a useful review. A single body capable of elastically-restrained motion in two or more degrees of freedom may be subject to flutter instability involving at least two of them. The theory is welldeveloped for bodies in streamline flow, and Garrick (1976) gives a worthwhile review. Although the flow past them is usually partly separated, suspension bridge decks may be subject to this type of classical flutter instability, and designers must ensure that the critical wind speed above which the flutter would occur is out of the design range. However, a purely torsional instability, closely similar to transverse galloping, is also a possibility for suspension bridge decks, and was in fact the cause of the destruction of the first Tacoma Narrows Bridge in 1940. Scanlan (1979) discusses these and other aeroelastic problems of suspension bridges in an important review. Some forms of flow-induced vibration involve more than one body. One important form involving two bodies is wake galloping, in which one body lies in the wake of the other and oscillates in a large-amplitude elliptical orbit. It occurs in electric transmission lines using bundled conductors, where it is referred to as subspan oscillation, and it can also occur for twin towers, as noted by Cooper and Wardlaw (1971 ). Although wake galloping continues to be widely studied, the emphasis has been on the analysis of the dynamics of motion in two degrees of freedom, and the detailed nature of the aerodynamics causing the phenomenon has been rather neglected (Tsui, 1977; Price and Piperni, 1988). If more than two bodies are involved in flow-induced vibration, as in the piping systems of nuclear reactors and other heat exchangers, the multiplicity of variables and the complex geometry in which the turbulent, usually separated flows occur make progress in understanding and analysis of the problems very difficult. There are several classes of vibration possible in such systems and Paidoussis (1979) has presented a thorough and critical review of current approaches, supplemented by an extensive collection of case histories from industry. He provides a useful update to his review of the field in Paidoussis et al. (1989). Finally, the scope of the article is further restricted in that there is no attempt to review and discuss everything written in recent years on galloping and vortex-induced vibrations. Instead, the review is selective with a view to providing a central core of observed phenomena and ideas for modelling them. In commencing a study of any form of flow-induced vibration, a necessary first step is to try to learn as much as possible about the flow whose interaction with the body or bodies is causing the vibration. In each form of vibration to be considered in this article, only a single long body of bluff section is involved, and the effects of the separation of the flow from the section dominate the vibrations that occur. Some important physical characteristics relating the body cross section and the separated flow, and some resulting effects in flow-induced vibration, are discussed in the next part of the article. Following parts deal with theoretical methods of modelling the vibrations.

2. PHYSICAL CHARACTERISTICS 2.1. SECTIONAFTERBODIES

2.1.1. Importance of Afterbody


A long elastic body of bluff cross section will not exhibit significant galloping or vortexinduced vibrations unless the section has an appreciable afterbody, and if it has one, its size and shape will strongly influence the characteristics of both types of vibration. Consider, as examples, the semicircular or D-section, studied by Brooks (1960) among others, and the

172

G. PARKINSON

/ /////

i
2

~ 1

FIG. 1. Effects of afterbody shape on galloping. Experimental results: O, Brooks (1960); &, Slater (1969).

symmetrical structural right-angle section, studied in idealized form without radiused corners by Slater (1969). With incident flow along the axis of symmetry of the section, neither section will gallop and only the angle section displays vortex-induced vibrations, of negligible amplitude, when oriented with the curved surface of the D-section or the vertex of the angle facing upstream. Thus oriented, the D-section has no afterbody, since the flow separates at the edges of the flat face, and the angle section has only the projection of the squared-off ends of the legs. In contrast, when the sections are turned 180 with respect to the incident flow, both will exhibit strong galloping and vortex-induced vibrations, but with different characteristics, arising mainly from the differences in their afterbodies. Typical galloping behaviour for each section is shown in Fig. 1, with galloping amplitude Ys plotted against incident flow velocity U, each in a dimensionless form defined later. The curves for both sections show the characteristic nearly linear increase of amplitude with flow velocity at higher velocities, but the behaviour is different at lower velocities. The D-section is a hard oscillator, i.e., it will not gallop from rest, but at any flow velocity requires an initial amplitude to be exceeded before the amplitude will increase to the stable galloping value. The variation of this initial or unstable amplitude with flow velocity is shown by a dotted curve. The angle section, on the other hand, is a soft oscillator, and will gallop from rest at all flow velocities above a critical value, although there is a range of flow velocities in which two stable amplitudes are possible, separated by an unstable amplitude. The lower stable amplitudes would be reached from rest, and the upper would be reached if the flow velocity were to be lowered from above this range while the body was galloping. These differences in galloping characteristics arise from the fact that the downstream corners of the ends of the angle section interfere with the separated shear layers as the section oscillates, while nothing similar occurs with the curved afterbody of the D-section. Both of the experiments of Fig. 1 were conducted in the same wind tunnel, using models of about the same size, under approximately two-dimensional conditions. The damping of the elastic system was higher for the angle section experiments. In Fig. 1, the vortex-induced vibrations of both sections, which occur near U = 1, are omitted for clarity. A dramatic example of the influence of afterbodies on vortex-induced vibration is found in a paper by Toebes and Eagleson (1961). They made water tunnel tests on four fiat plates of 50.8 mm chord and 6.3 mm thickness elastically mounted for torsion about the leading edge. The leading edges were rounded, and the plates were initially aligned with the incident flow, so that it remained attached up to the blunt trailing edges. Two of the plates had trailing edge afterbodies, one semicircular and one triangular. The other two had none, one being cut off square and the other being given a reentrant notch. The first two showed vortex-induced oscillations of relative maximum amplitude 300 and 200 respectively, while the last two, to the same scale,

Flow-induced vibrations of bluff bodies

173

showed maximum amplitudes of only 40 and 5, respectively. Thus, although the afterbodies of the first two plates in each case represented only 6% of the chord whose pressure loading could contribute to the oscillation, their effect was so great that the average maximum amplitude of the two plates without afterbody was only 9% of that of the two plates with afterbody. 2.1.2. Afterbody Effectsfor Rectan#ular Sections Since afterbodies clearly have a strong influence on galloping and vortex-induced vibration characteristics, it is of interest to examine the effect of systematic variation of afterbody shape on some of these and other related characteristics. This has been done in several laboratories for the rectangular section of streamwise dimension d and transverse dimension h. The entire rectangle is the afterbody, since the flow separates from the upstream corners, unless d/h is sufficiently large for the flow to reattach to the sides, in which case the flow finally separates at the downstream corners and there is no afterbody. Phenomena of reattachment of the initially separated flow are central to the characteristics of galloping, the vibration form most intensively studied for rectangular sections. Figure 2 shows some effects of systematic variation of rectangular afterbody aspect ratio d/h on fluid dynamic characteristics related to galloping, and on galloping behaviour itself. Some effects of turbulence intensity in the incident flow are also shown. Turbulence intensity i is the ratio of the r.m.s, velocity fluctuation in the stream direction to the mean stream velocity. In Fig. 2(a), the variations of base pressure coefficient Cpb and Strouhal number S with d/h are shown for stationary prisms in smooth flow. In addition, the variation of Cpb is shown for a

(a)

~
-C%
I

\/-c%

0.2

s
03

0.12

!
I I I

(b) 04 F ~ I ~ 02
0

d h Tyi:O

I
I

I
I O. 12 u.u~

(C)
0.12

006

d/h

FIG. 2. Effects of rectangular afterbody aspect ratio and turbulence intensity; (a)base pressure and Strouhal number, (b)soft galloping amplitude, (c)reattachment and galloping. Experimental data from Brooks (1960), Hoerner (1965), Laneville (1973), Nakamura and Tomonari (1977), Novak (1974), Smith (1962), and Washizu et al. (1978).

174

G. PARKINSON

limited range of d/h in turbulent flow of 12% intensity. Pressure coefficient Cv is defined by

Cv _ P-- Poo p/2I/2'


where p is the pressure at the point of evaluation, p~ and I/are the free stream pressure and velocity, respectively, and p is fluid density. The base pressure Pb is that on the downstream face of the rectangle. Since it is not quite constant over the face because of secondary flows, values quoted are for the centre of the face unless otherwise noted. Strouhal number is defined by f~h

S-- T ,

wheref~ is the vortex formation frequency in one shear layer from the stationary body. The base pressure curves are taken from data by Nakamura and Tomonari (1977) for d/h < 1.1, supplemented by values inferred from Brooks (1960) and Hoerner (1965). The Strouhal number curve is from Brooks (1960), with additional values, indicated by dashed curves and uncertainty bars, from Washizu et al. 0978). The most notable feature of Fig. 2(a) is that in the range 0 < d/h < 1.0, Cvb undergoes large changes and S changes very little, while the reverse is true in the range 2.0 < d/h < 3.0. For the first range, Bearman and Trueman (1971) argue that at first increasing d reduces the size of the base 'cavity' from which entrainment of fluid takes place by the shear layer forming a discrete vortex, and the base pressure is accordingly lowered. Also, if d is small enough not to cause interference with the process of vortex formation from the shear layer, the resulting vortex as it first reaches full strength will lie closer to the afterbody base as d increases, again contributing to the lower base pressure. Moreover, the lack of interference with the shear layer would explain the small effect on S. The above trends with increasing d cannot continue indefinitely because a sutiiciently large d will begin to interfere with the inward-curving shear layer and force changes in the process. Thus Cvb reaches a minimum at about d/h = 0.62. Beyond this, Bearman and Trueman suggest that the interference from the longer afterbodies forces the vortex formation further downstream, thus raising the base pressure, and also making the shear layers more diffuse prior to vortex formation which, following Gerrard (1966), would cause a decrease inf~ and provide an explanation for the observed gradual drop in S. The other abrupt changes, in the range 2.0 < d/h < 3.0, clearly correlate with reattachment of the shear layers on the afterbody sides, with final separation occurring at the downstream corners. The jump in S arises from the drop in lateral spacing of the two shear layers on reattachment. Because of the natural configuration of the wake vortex system, a smaller lateral spacing corresponds to a smaller streamwise spacing of the vortices, and therefore to a higher formation frequency f~, and thus a higher value of S. The above discussion relates to the variation of Cpb and S with d/h in smooth flow. In turbulent flow, as suggested by Gartshore (1973) and again by Hillier and Cherry (1980), the intensity of the incident turbulence creates increased entrainment of fluid by the separated shear layers, thus thickening them and thereby promoting, as d/h is increased for the rectangular sections considered here, first interference between the shear layer and the trailing edge corner of the section and then reattachment of the shear layer at this trailing edge at lower values of d/h than for smooth flow. The scale, or eddy size, of the incident turbulence does not seem to be important. The first effect is seen in Fig. 2(a) as a reduction and shift to lower d/h of the negative peak in Cvb. The expected second effect would be a jump to higher values of S at a lower d/h than in smooth flow. No data on S in turbulent flow was available for Fig. 2(a), but the data from Washizu and his co-authors (1978) seem to have some of the expected behaviour, although their wind tunnel flow was stated to have a turbulence intensity of only 0.3%. Their data show two values of S in the range 2.0 < d/h < 2.8, the lower values agreeing with those of Brooks (1960), while the upper values approach those found by Brooks to occur only for d/h > 2.8. Possibly some lowfrequency unsteadiness in the flow was causing intermittent reattachment for d/h > 2.0.

Flow-induced vibrations of bluff bodies

175

The argument that the first effect of incident turbulence intensity as d/h is increased is interference between the shear layer and the trailing edge corner at lower d/h than in smooth flow is supported by another experiment by Nakamura and Tomonari (1977), in which they mounted a fence, or spoiler, 10% h in height at the trailing edge of each section tested in smooth flow, and the resulting variation of Cpb with d/h virtually duplicated the curve of Fig. 2(a) for 12% turbulent flow. The main purpose of this section of the article is to relate fluid dynamic behaviour caused by afterbodies to cylinder flow-induced vibration characteristics, and for this first consider Fig. 2(b), in which the maximum ratio of soft galloping amplitude to flow velocity, in dimensionless form Ys/U, is plotted against d/h for several values of i. The inset sketch defines some of the physical variables. The dimensionless forms of transverse displacement and flow velocity are defined by

Y = y/h,

U = V/COnh,

where con is the undamped natural circular frequency of the elastic system of the oscillating body. ~ is the steady-state amplitude of oscillation. The smooth-flow data are taken from Brooks (1960) and Smith (1962). The turbulent-flow data are taken mainly from Laneville (1973), supplemented by information from Novak (1974) and Nakamura and Tomonari (1977). The smooth flow data are the most complete, since tests were carried out for 13 values of d/h, ranging from 0.375 to 4.8. None of the sections with d/h < 0.75 would gallop from rest. For 0.75 < d/h < 3.0, vigorous soft galloping occurred, with (Ys/U)m,x decreasing with increasing d/h, and for d/h > 3.0, no transverse galloping at all could be induced. A glance at Fig. 2(a) shows that the two boundaries for soft galloping correlate closely with the abrupt changes in Cpb and S discussed previously. At the lower boundary, d/h = 0.75, the link between the beginning of soft galloping and the occurrence of abrupt changes in Cpb is undoubtedly the interference between the shear layer and the trailing edge corner of the rectangular section. Soft galloping, as will be shown in more detail in a later part of the article, is excited by the creation of an unbalanced pressure distribution on the afterbody sides when the section is given a small transverse velocity. This unbalance requires a secondary flow between the two shear layers and the body, and such a secondary flow presumably gains enough strength to trigger soft galloping only when d/h has become large enough to cause the above interference. At the upper boundary, d/h = 3.0, the link between the end of galloping and the occurrence of a jump in S is clearly the reattachment of the shear layers at the section trailing edge, so that the section no longer has an afterbody, and becomes immune to transverse galloping. The decrease in (Y~/U)max with increasing d/h is also determined by reattachment effects, for at an instant in the galloping cycle, as shown in the inset sketch, the section has a transverse velocity )) so that the relative velocity is at angle of attack a and one shear layer lies closer to the section than the other. The near shear layer creates higher suction on the adjacent side than is occurring on the other side, and this differential pressure gives the exciting force, but for a value of a = as at which the near shear layer actually reattaches at the trailing edge, a positive pressure gradient is created and, at a slightly higher value of a the exciting force becomes a damping force. As d/h is increased, the value of a s will obviously decrease, thereby limiting the values of)) in the galloping cycle. Thus ))max and therefore ITem,, decreases as d/h increases. From the discussion of effects of turbulence in connection with Fig. 2(a), it might be expected that galloping behaviour in turbulent flow would be similar to that in smooth flow, but shifted to lower values of d/h, and this is seen to occur in Fig. 2(b). The data are far less complete than for smooth flow, so that soft galloping boundaries are rather tentative, as indeed are the amplitudes for d/h < 0.75. Nevertheless, it can be seen that as turbulence intensity in the incident flow is increased, a soft galloping section becomes weaker and eventually stable, and a hard galloping section becomes soft. Figure 2(c) emphasizes the relationship between galloping and reattachment by superimposing curves of constant aR on an i-d/h map showing galloping zones. The a s data are

176

O. PARKINSON

from shadowgraph experiments by Laneville (1973) on several stationary rectangular sections of different d/h tested through a range of a in smooth flow and in two turbulent flows of different intensity. In these experiments the shear layers were made visible by heating a metal ribbon embedded along the centre-line of the front face of the rectangular section. The boundary layer flow was heated by the ribbon before separating, and the resulting temperature gradients in the shear layers made them visible in shadowgraphs. Experiments on stationary sections were appropriate as a part of studies of galloping because in typical examples of galloping the reduced frequency (which can be taken as the reciprocal of U ) is low, order 10-1 or less, and flow conditions can be assumed quasi-steady, so that a section at an instant in the galloping cycle, as in the inset sketch of Fig. 2(b), at apparent angle of attack a, can be considered to experience the same flow field and fluid dynamic loading as it would if held stationary at a in a steady flow. In Fig. 2(c) the right boundary of soft galloping, at which the section becomes stable, is close to the curve aR = 0, but not coincident with it. Rather, the galloping boundary approximately coincides with the curve a R = 3, perhaps indicating that if aR is 3 or less, the transverse exciting force is too weak for galloping to occur, even in a lightly damped system. The left boundary of soft galloping, at which it becomes hard, so that the section requires a sufficient initial transverse velocity before galloping will occur, lies close to the curve aR = 16, again implying a link between the flow phenomena leading to reattachment and the nature of the galloping. In this section of the article, attention has been focussed on the effects of afterbody shape on flow-induced vibration. In the next section, the subject will be the wake vortex systems formed from the shear layers, and their interactions with the oscillating afterbody.

2.2. WAKE VORTEX PHENOMENA

2.2.1. Issues of Experimental Methods There is an enormous and growing literature on aspects of organized vortex formation in the wake of bluff bodies. Afterbody effects on Strouhal number were discussed in the previous section, and it will be assumed here that the reader is familiar with the basic characteristics of a Karman vortex street (Karman, 1911) formed behind a stationary bluff body under two-dimensional conditions. The most popular bluffbody for study has been the circular cylinder, but even for such a simple geometric shape the organized wake characteristics have proved very complex, as an important review some years ago by Morkovin (1964) testifies. When the body is set in transverse oscillation the complexity of the wake characteristics increases, and two schools of thought about the most useful form of experiments have developed. One school with a long history (Meier-Windhorst, 1939), to which the author has belonged for many years (Parkinson and Brooks, 1961), attempts to simulate flow-induced vibrations directly. Typically, a rigid length of bluff cylinder is mounted in a wind tunnel or water channel on an external support system with adjustable linear springs and system damping so that transverse flow-induced vibrations can develop under nearly twodimensional conditions. Flow velocity and elastic system parameters are controlled, while everything else of interest is measured or photographed as a function of time. Measured quantities include cylinder displacement and oscillation frequency, cylinder surface loading magnitudes, frequencies and spectral properties; and wake frequencies and spectral properties. Phase measurements are made when synchronization occurs. Flow patterns made visible by various techniques are photographed. The other school, also with a long history (den Hartog, 1934) and wide membership (Smirnov and Pavlihina, 1957; Bishop and Hassan, 1964), also mounts a rigid length of bluff cylinder in a wind tunnel, water channel or towing tank on an external support system, and creates nearly two-dimensional flow conditions. However, in this case the support system is a mechanism which produces forced transverse oscillations of the cylinder in which amplitude, frequency and waveform (usually harmonic) are controlled. Only cylinder

Flow-inducedvibrationsof bluffbodies

177

surface-loading properties and wake properties need to be measured. Flow patterns are photographed as for free oscillations. Arguments can be made for and against both approaches. The first approach, by simulating in the laboratory the actual vibrations observed in the field, gives direct evidence of the nonlinear interactions between excitation and response that are central to the problem. On the other hand, there are more quantities to measure, all of which tend to change as flow velocity is varied, so interpretation of results is more difficult. The second approach provides simpler experiments under greater control, but not all of the data they produce represent the characteristics of flow-induced vibration. Whether a cylinder in forced oscillation with a given amplitude, frequency, waveform and flow velocity represents the characteristics of a possible flow-induced free oscillation of the same cylinder must be determined from the direction of the energy transfer between cylinder and flow, which in turn must be deduced from the data. This can be done and the characteristics of the equivalent steady-state flow-induced free oscillation of the cylinder with a linear elastic system can be predicted from the forced-oscillation data. Staubli (1983) has done this with the intent of reproducing vortex-induced oscillation characteristics of a circular cylinder measured by Feng (1968). However, there are some qualifications in the use of forcedvibration data in this way. Elastic system damping, particularly when it is low enough to permit large-amplitude flow-induced vibrations may be motion-dependent rather than constant, with significant consequences for the vibration. Berger (1987) points out that forced vibration experiments block this effect and cannot reveal what he finds to be an important stability mechanism. Forced vibration experiments also would not be suitable for studying the transient build-up of flow-induced vibrations, discussed for galloping in Section 3, nor would they be suitable for studying large-amplitude hydroelastic vibrations, in which the strongly nonlinear fluid forces produce nonharmonic waveforms, as found by Bouclin (1977). Let us now consider the results of some experiments by both methods.

2.2.2. Wake Vortex Effects on Oscillating Circular and D-section Cylinders In this sub-section of the article, the wake vortex phenomena are restricted to those produced by oscillating circular and D-section cylinders because on the one hand the cylinder shapes are of interest and comparison of their wake characteristics is useful, and on the other hand their afterbodies do not interfere with the separated shear layers and the subsequent vortex formation, so that only vortex-induced vibration from rest will occur for elastically-mounted cylinders, and galloping is not an issue. Wake phenomena for squaresection cylinders in which principles of galloping and vortex resonance appear in combination will be considered in the next section. Several writers on wake vortex phenomena of oscillating cylinders, including Sarpkaya (1979), Zdravkovich (1982), Staubli (1983), Bearman (1984), Berger (1987), OngSren and Rockwell (1988) and Williamson and Roshko (1988), have referred to the studies of flowinduced vibration of circular and D-section cylinders by Feng (1968), and in some cases used his results as a comparison model. It seems appropriate to begin this section by describing his experiments and presenting some of his results. Feng's experiments, like those of Brooks (1960), Smith (1962), and Laneville (1973), referred to in Section 2.1, were carried out in the closed-circuit low speed wind tunnel of the Department of Mechanical Engineering at the University of British Columbia. It produces a very uniform flow with turbulence level below 0.1% through a test section 0.686 m high by 0.914 m wide. The circular and D-section cylinders were of 7.62 cm diameter and 0.686 m length, of light aluminum and plastic construction with aluminum skin of 0.56 mm thickness. Each was instrumented with 0.64 mm diameter surface pressure taps around the mid-span cross section, 32 of them on the D-section. Each tap was connected to a length of 1.68 mm inside diameter polyethylene tubing. These tubes were brought out through the end of the cylinder, so that each could be connected to a pressure transducer mounted outside the wind tunnel.

178

G. PARKINSON

The cylinders were mounted on the vertical centre-line of the test section by thin fingers extending through narrow slots in the tunnel floor and ceiling. These fingers were attached to aluminum tubes supported in cylindrical air bearings. This mounting gave low friction, but restrained the cylinders to transverse oscillation, normal to the cylinder axis and the wind direction. Additional damping of viscous type could be applied by a magnetic damper operating on the aluminum tubes, and cylinder displacement could be measured by a transducer in which the aluminum tube interfered between primary and secondary coils of an air core transformer. The mounting system and damping and displacement instrumentation were designed by Smith (1962) for his study of galloping. The two cylinder models described above were built for a preceding study by Ferguson (1965). The acoustic level pressure transducer used was the Barocel, made by Datametrics. It was mounted externally, so it was necessary to calibrate the output signal to account for attenuation and phase lag of the pressure signal during transmission through the connecting tubing. Calibration consisted essentially of producing a known pressure signal at the input end of an equivalent tube and pressure tap configuration. To determine the system damping and natural frequency for transverse oscillations in the absence of appreciable aerodynamic forces, a thin streamlined model of the same mass as each cylinder and its attachments was mounted in place of the cylinder, and the decay of its free oscillations from a realistic initial amplitude in its own plane in still air was recorded. Figure 3 shows a summary of some of the results of Feng's experiments in the form of the stability diagram introduced by Scruton (1965). In the experiments, the nonaerodynamic damping was set at various levels up to the maximum available by the magnetic damper, and the range of wind speeds over which each cylinder oscillated with Ys> 0.01 was determined, as was Y,,,. in the range. The figure shows for each cylinder the upper and lower boundaries Y, = 0.01 plotted in the units used by Scruton, but in terms of the notation of this paper, 2~U

\ ~__.OT ~ O_Uy ....


\'\ I= 6 o,/ '~

108

06

\\. ~
\
-

04 i).~

02

08

u--D
8--

ou
06

,"
7-04 I>~

6~

02

5(

S5

I
2w~In

rO

E[o. 3. Stability diagramsand maximumvibration amplitudesfor circular and D-sectioncylinders:Experimental data from Eeng(1968).

Now-induced vibrations of bluffbodies and F,m.~ plotted against

179

2n(fl/n) (now increasingly known ph21


2m

as the Scruton number), where

fl = fraction of critical viscous damping, n= mass parameter,

roll =

mass per unit length of oscillating system.

From Fig. 3 it is seen that the D-section cylinder oscillations are much harder to suppress by increased damping than those of the circular cylinder. Figures 4, 5, 6, 7 show phenomena of the vortex-excited oscillation over a range of wind speed for the minimum and maximum damping levels used for each cylinder. In the experiments, the damping level was set for the cylinder and the wind speed increased in small steps. At each wind speed the vortex frequencyfv was measured from the pressure signal from a surface tap, at the transverse diameter for the circular cylinder, and near it for the D-section cylinder. If the cylinder oscillated, Ys and cylinder frequency f~ were measured from the displacement transducer signal. With the cylinder still oscillating the wind speed was increased by the next step, and the process was repeated. Whenfv was captured byf~, there followed a range of wind speed over which a phase angle ~b between the pressure and displacement signals could be measured. The angle ~bis presented as the angle by which the suction at the tap mentioned above leads the displacement in the direction of that tap (this is the same as the phase angle by which the transverse exciting force leads the displacement). The process was continued to the highest wind speed giving appreciable oscillation, and then repeated for decreasing wind speeds, always with the wind speed increment or decrement being set while the cylinder was oscillating. For the minimum damping levels

--

-- 5 0

vv

vv

2 --

O ~1

i
I

--

o_

,'tlf !
~,

--04

rr,

',~
--02

o-

~
)6

"~'~P~'08

I
{

I~
~2

"A~

^0 14

FIG. 4. Vibration phenomena for circular cylinderwith low damping. Data from Feng (1968). ~ = rq;f, = O; f~ = ~7; I~y,= ~; Y, = A; Ysfrom rest = ~.
3PAS 2 6 : 2 - E

180

G. PARKINSON

--

100

15

50

/
s:0.198 05

u-'O
0.6

04

0.2

J~i{A~ ~'{i ~^
08

^l ,', ^
{ U

l
I 2.

'0
14

FIG. 5. Vibration phenomena for circular cylinder with high damping. Data from Feng (1968). ~ = i~;f~ = O;

= v: ~.= i .
o ~ / IO0

D U--,.- T ?

o -i ~
o 50

/,o

~A

--

-- 04

~"

P
12

-oz

0.8

"

"" ~

14

FIG. 6. Vibration phenomena for D-section cylinder with low damping. Data from Feng (1968). ~b = [~;f~ = O; f~ = W; Ys = A; Y, from rest = @.

Flow-induced vibrations of bluff bodies IOO

181

u -"D-r9
.50

~o

o o

oO o

o o o

O~ -- 04

A~

7.

0J \
&

I U

12

~0

FIG. 7. Vibration p h e n o m e n a for D-section cylinder with high damping. Data from Feng (1968). ~ = O;.~ = O;

only for each cylinder, the measurement of Y, was made again over the complete wind speed range, with the cylinder starting from rest at each wind speed. In Figs 4-7 Y,,fv/f,,f~/f, and are plotted against U (f, is the system undamped natural frequency). A reference line corresponding to the known Strouhal number for the stationary cylinder is included on each figure. (S = 0.198 for the circular cylinder in the range of Reynolds number under consideration, from 104 to 5(10)4. S = 0.135 for the D-section.) Figure 4 also shows a plot of exciting force coefficient Cy. vs. U. This was done only for the low-damping case for the circular cylinder because it seemed to be the most interesting, and because the experimental procedure was very time-consuming. Cy. is the steady-state amplitude of oscillatory exciting force coefficient Cy, defined by Fy Cy = p/2V2h , where Fy is the transverse exciting force per unit span created by the vortex wake. For these experiments a second circular cylinder model was built, of the same dimensions and construction as the first, but fitted with a spanwise line of 17 pressure taps, which could be rotated and set at any circumferential position. In each run at a given wind speed and steadystate cylinder vibration amplitude, the r.m.s, fluctuating pressures from the 17 taps were recorded for a given angular setting, and the process was then repeated in subsequent runs for increments of angular position until all positions were covered. The r.m.s, value of the transverse exciting force for the given wind speed and cylinder amplitude was then calculated by circumferential and spanwise integration of the pressures, and this was reduced to coefficient form and multiplied by ~ to produce equivalent mean peak amplitude values, presented as ~'y, in Fig. 4. Only the three data points shown were measured in this way. In addition, the values of U at which abrupt jumps in Cy. occurred

182

G. PARKINSON

FIG. 8. Transient effects on cylinder and surface pressure amplitudes during lock-in. Upper traces: surface pressure at transverse tap, f, = 9.04 Hz. Lower traces: cylinder displacement,f~ = 9.04 Hz. (a) U = 1.01, (b) 1.01 < U < 1.06, (c) U = 1.06, (d) 0.97 > U > 0.94. Data from Feng (1968).

were determined from oscilloscope traces of fluctuating pressure at a transverse tap as the wind speed was slowly varied during cylinder oscillation (see Fig. 8). The dotted curve in Fig. 4 drawn through the data points and including the jumps is therefore somewhat speculative. Figures 4-7 all display four important characteristics of vortex-excited oscillation-capture over a discrete wind speed range offv byfc, which remains nearly constant and close tofn, increase of ~bwith U during capture, Ys~,axin the capture range, and reversion Offv to nearly its stationary cylinder value outside the capture range. There are, however, some differences in the circular cylinder behaviour from that observed in experiments by others, and the D-section cylinder shows some characteristics different from those of the circular cylinder. Figure 4 is directly comparable to Fig. 9 of Ferguson and Parkinson (1967), since both are for the same circular cylinder with the same mounting, and without magnetic damping added. Examination of the two figures shows that the measurements off~ andf~ are nearly identical, that the capture range is the same, and that the variation of ~ with U is in close agreement where the present data with the cylinder started from rest at each wind speed are used. When, however, the wind speed is increased with the cylinder oscillating, it is seen that a much higher ~m,~ is reached at about the same value of U. Beyond this value of U, which is near the middle of the capture range, Ys drops sharply to the value that was reached from rest at that wind speed. If the wind speed is then decreased, the 'rest' values of Ys are obtained until a second jump back to the higher values occurs, so that a clockwise oscillation hysteresis loop results, as indicated by the arrows on Fig. 4. The variation of ~b with U in Fig. 4 is from approximately 0 to 100 and also displays a hysteresis loop, counterclockwise. (The phase values in Fig. 9 of Ferguson and Parkinson (1967) are now thought to be incorrect, although their trend is correct.)

Flow-inducedvibrationsof bluffbodies

183

FIG.9. Circularcylindervibrationphenomenabeyondlock-inrange.DatafromFerguson(1965).U = 1.15.Upper trace: cylinderdisplacement,f = 9.0 Hz. Lowertrace: surfacepressureat transversetap,f, = 13.0Hz. In Fig. 5, the results of applying relatively high magnetic damping to the circular cylinder are shown. Ysm,~and the capture range are much reduced, and no hysteresis loop appears in the ~ data, but the frequency and phase variations are very similar to those of Fig. 4. In both Fig. 4 and Fig. 5, Ysmaxoccurs near the middle of the capture range, and appreciable oscillation continues at wind speeds beyond the capture range. In Fig. 6, showing the results of the D-section cylinder tested without magnetic damping, there is seen to be a difference in the relative position of the capture range compared with the circular cylinder results of Fig. 4. In Fig. 4, capture first occurs as f~ for the stationary cylinder closely approachesf~. In Fig. 6, however, capture occurs whenf~ is only 78% off~, and in fact the capture range ends, rather than begins, approximately where f, for the stationary cylinder equals f~. For the D-section, appreciable amplitude ~ is found only inside the capture range, and ~m,~ occurs at the end of the range, rather than in the middle, as for the circular cylinder. When the wind speed is increased, then subsequently decreased again, with the cylinder oscillating, the clockwise oscillation hysteresis loop for Y, is observed. The angle ~ was not measured with decreasing wind speed. The range of ~b is smaller than for the circular cylinder. None of the amplitudes reached as a result of the vortex excitation were high enough to trigger the hard galloping characteristics discussed in Section 2.1. Figure 7 shows the result of applying relatively large magnetic damping to the D-section cylinder. Y~,,x and the capture range are not reduced by nearly so much as they were in Fig. 5 for the circular cylinder, which had less damping applied. All the phenomena of Fig. 6 are repeated, except that the range of ~bis further reduced, and no hysteresis loop is present. The comparison between results for the circular cylinder and the D-section is interesting because their afterbody shapes are nearly the same. However, there are two important differences in the separating flows. First, the shear layers leave the D-section at 90 to the stream direction, whereas they leave the circular cylinder at about 10. Second, the D-section has fixed separation lines at the edges of the fiat front face, whereas the separation lines on the circular cylinder are boundary-layer controlled and suffer spanwise variations. Both differences may be used to explain the stronger vortex-induced oscillations of the D-section indicated in Fig. 3. The first wake vortices to form from the D-section shear layers may be in

184

G. PARKINSON

a position to exert a stronger transverse pressure loading on the parent body than those forming from the circular cylinder shear layers, and the fixed separation line oftbe D-section promotes better spanwise correlation of the wake vortex formation, again creating conditions for a stronger pressure excitation of the body. Feng made wake hot-wire measurements which confirmed that spanwise correlation of velocity fluctuations was considerably higher for the oscillating D-section cylinder than for the oscillating circular cylinder. Finally, two other phenomena in Feng's results require comment. Figure 8 shows photographs of oscilloscope traces from the circular-cylinder experiments. In each photograph, the upper trace is of the fluctuating pressure at the transverse tap and the lower trace is of the cylinder displacement. In both traces the oscillation frequency is synchronized at 9.04 Hz, slightly below the system natural frequency of 9.10 Hz. Figure 8(a) is taken at U = 1.01, near the condition for maximum cylinder amplitude in the upper branch of the hysteresis loop. In Fig. 8(b) U is slowly increased from 1.01 to 1.06, and the amplitude of the fluctuating pressure is seen to drop abruptly to a much lower value, while the cylinder amplitude then decreases slowly to a new lower steady-state value on the lower branch of the hysteresis loop. In Fig. 8(c), U is constant at 1.06, and both traces show constant amplitude with little modulation, whereas the high-amplitude pressure trace of Fig. 8(a) and 8(b) showed considerable modulation. In Fig. 8(d), U is slowly decreased from 0.97 to 0.94 and here the amplitude of the fluctuating pressure is seen to rise abruptly to a considerably higher value, without much modulation, while the cylinder amplitude increases slowly to a new higher steady-state value on the upper branch of the hysteresis loop. These photographs demonstrate that the cause of the hysteresis is in the fluid mechanics, rather than in the elastic system. The other intriguing phenomenon is the persistence of cylinder oscillation of considerable amplitude at wind speeds beyond the upper end of the synchronization range, as shown in Fig. 4. This was also observed in the earlier experiments by Ferguson (1965), and Fig. 9 shows a photograph taken at U = 1.15 of two of his oscilloscope traces of the same type as in Fig. 8. In Fig. 9, the lower trace gives the signal from the transverse pressure tap which is at 13.0 Hz with large random amplitude modulation, whereas the upper trace gives the cylinder displacement signal which is at 9.0 Hz and has no amplitude modulation. The explanation for this is still being sought. Staubli (1983) has carried out an interesting study, in which a circular cylinder performed forced harmonic transverse vibrations while the mounting rig was towed at constant velocity through otherwise still water. An attempt was made to achieve two-dimensional flow conditions, and towing velocity, vibration amplitude, and vibration frequency were controllable variables. The fluid forces on the cylinder were measured. The transverse force at the vibration frequency was reduced to coefficient form and its amplitude CLo and phase angle tk were presented in Staubli's paper in three-dimensional plots vs. Ysand So, defined as the Strouhal number for the oscillation frequency. Staubli then used these results to attempt to reproduce Feng's Ys-U and tk-U plots of Figs 4 and 5. His method was to assume a linear elastic system with a harmonic forcing function, and the system constants supplied from Feng's data. Then the method of harmonic balance (see Section 4.2.1) was used to obtain two simultaneous equations for the limit-cycle solutions. In this approach, CLo and ~ were considered known from the forced-vibration plots, and the unknowns Y~ and oscillation circular frequency o~c (using the notation of the present paper) could be found by iteration. Staubli's calculated curves agreed well with Feng's plots, particularly over the lower half of the synchronization range of U in each case. Over the upper half (roughly beyond U for ~m..) the agreement was not nearly as good in either case. In the low-damping case of Fig. 4, the calculated curve reproduced the hysteresis loop but greatly underestimated Y~ and overestimated ~bfor its lower branch. For values of U beyond the synchronization range the calculated Y, was too low by about a factor of 10. Staubli commented that these low calculated values of Y, were at least partly a result of the neglect of any exciting force component at the stationary-cylinder Strouhal frequency. This certainly would be true for the results beyond the synchronization range, but in addition to seeking explanations of why

Flow-induced vibrations of bluffbodies

185

Staubli's calculated ~ values were so low, one might also ask again why Feng's experimental Ys values were so high in this range. Another interesting result of Staubli's calculations was the prediction of a (presumably) stable-unstable closed loop of higher-amplitude oscillations in the high-damping case of Fig. 5. If it exists it would have to be reached by providing a sufficient initial amplitude to trigger the hard oscillatory characteristics, so it was not discovered by Feng. As a final comment on the comparison of the two sets of results it is worth noting that the agreement would probably not have been as good if Feng's experiments had been in water flow and Staubli's in air flow, since the waveforms in Feng's relatively high-amplitude cylinder vibrations would have been nonharmonic because of the strong nonlinearities, similar to those calculated and observed by Bouclin (1977), as described in Section 4.2.2. The occurrence of oscillation hysteresis involving jumps in Cy., Ys, and ~ is the most intriguing effect observed in vortex-induced vibration of circular and D-section cylinders, and considerable effort has been expended in several laboratories in seeking an explanation. The theory of nonlinear oscillations admits the possibility of oscillation hysteresis occurring because of nonlinearities in the spring force or damping terms of the differential equation for the elastic system. These possibilities have seemed plausible since experiments on flowinduced vibration of circular cylinders have involved different types of elastic system in different laboratories, and not all such laboratories have reported the occurrence of oscillation hysteresis. However, the evidence now seems sufficiently conclusive that the jumps and hysteresis occur because of conditions in the separated flow, not in the elastic system. There is the direct evidence from Feng's results, such as Fig. 8, and the fact that the existence of jumps and hysteresis in flow-induced vibrations of circular cylinders can be inferred from the results of forced-vibration experiments by Bishop and Hassan (1964), Bearman and Currie (1979), and Staubli (1983). What, then, happens in the flow to produce these effects? It would clearly be very difficult to design experiments using conventional local probes, such as hot wires, to discover the larger-scale, even global, changes occurring in the separated flow. On the other hand, flow visualization techniques offer the prospect of photographs of a time sequence of instantaneous views of the oscillating body and a considerable portion of its wake from which useful deductions could be made. In an interesting paper, Zdravkovich (1982) made deductions of this kind from analysis of the results of earlier flow-visualization studies of oscillating circular cylinders by den Hartog (1934), Meier-Windhorst (1939), Angrilli et al. (1974), and Griffin and Ramberg (1974). He found that in the range of U for synchronization off~ withf~ two modes of vortex shedding existed with different timing in the oscillation cycle, one for lower and the other for higher values of U, and he inferred that this could explain the hysteresis effects. Others have followed this lead. Recently two important papers have appeared reporting on separate but similar flowvisualization studies designed (among other objectives) to reveal details of the changing vortical structure of the near wake of an oscillating circular cylinder in the range of U for synchronization. Both studies involved forced harmonic transverse vibrations of a rigid cylinder in water, with controllable Y~,f~, and U under approximately two-dimensional conditions and in the same Reynolds number range. In both studies, Y~,f,and U were held constant for each experimental run. In one of the studies, Ong6ren and Rockwell (1988), the cylinder was mounted vertically in the centre of a free-surface water channel with its top end clamped to the vibration mechanism. Flow visualization was from photographs of time lines created by hydrogen bubble generation from a horizontal wire upstream of the cylinder at mid-span. Experiments were carried out on cylinders of triangular and square section, in addition to those on the circular cylinder. In all experiments, Y~was held constant at 0.13 and the Reynolds number was usually 855. In the other study, by Williamson and Roshko (1988), the vertical circular cylinder moved along and oscillated about the centre plane of a towing tank, whose travelling carriage supported the cylinder and contained the vibration mechanism. Flow visualization was from photographs of patterns created by aluminum particles on the water surface. Experiments

186

G. PARKINSON

were carried out over a wide range of values of Y, and of vibration wavelength 2, related to cylinder frequency f~ by

k = v/f~,
where V is the velocity of the cylinder in the towing tank. The Reynolds number range was from 300 to 1000. In the reported results for the circular cylinder from the two studies there is much common ground, but there are some significant differences. For example, there is considerable interest in the possible occurrence of subharmonic resonances in flow-induced ~1 vibration, i.e., local synchronizations under conditions where fc",~fv., where N is an integer. In each study one such subharmonic synchronization is found, but whereas (~ng6ren and Rockwell report a 1/2-subharmonic, Williamson and Roshko report a 1/3subharmonic. Williamson and Roshko, in their interpretation of their flow-visualization results, deal directly with the issue of oscillation hysteresis and provide an explanation for it in terms of a change in the configuration of the vortex wake resulting from a change in the timing of vortex shedding in the oscillation cycle as the wavelength 2 changes. Figures 10 and 11, adapted from their paper, give inset line drawings of the relevant vortex configurations, denoted by codes 2S and 2P. Quoting from the paper, "the designation 2S means that in each half cycle a vortex is fed into the downstream wake, like the natural Karman vortex shedding; 2P means the formation of vortex pairs which convect laterally outwards from the wake centreline". Figure 10 is a map of zones containing different vortex patterns near the fundamental synchronization region on a graph of Y, vs. 2/h, which is used as a convenient control variable by Williamson and Roshko. Their figure also includes an auxiliary abscissa scale offv,/fc, which is replaced in the present Fig. 10 by a scale of U. Since in free vortexinduced vibration fc ~f,, it follows that

2/h ,~ 2nU,
and the U-scale is helpful in relating the forced-vibration results of Fig. 10 to free vibrations. It can be seen in Fig. 10 that the curves do not extend to Y,=0, since a small threshold amplitude, perhaps Y, ,~ 0.05 (Koopmann, 1967), must be exceeded before synchronization occurs.

16
14

12

7,
O~ 06 04 02

I
0 I ,2 3 4 5 6 7 8 9 I0

I
t
16

k/h

1
0

I
GZ

[
0.4

I
0.6

I
Ol 8

I
I

I
I l2

I
I ,4

FIG. 10. (From Williamson and Roshko, 1988.) M a p of vortex synchronization patterns near the fundamental lock-in region for the circular cylinder.

Flow-induced vibrations of bluff bodies


"2P"mode

187

Two ~ r l e x pairs eoh ycte

ol /
X/n
~'2S"mode
Hot'moil street-type woke

FIG. 11. (From Williamson and Roshko, 1988.)A schematic variation of phase ~ with wavelength Ato demonstrate

the possibilityof oscillation hysteresisbeing caused by an overlap of the regions where the '2S' and '2P' modes occur. In Fig. 10, the boundary between zones 2S and 2P is noted as a 'critical curve', which for amplitudes Ys~<0.6 lies between the two dashed curves I and II. These represent the boundaries of the range of forced oscillation hysteresis observed by Bishop and Hassan (1964), but similar curves could be drawn from the free-oscillation results of Feng (1968). Williamson and Roshko provide detailed evidence and arguments to relate the occurrence of this oscillation hysteresis to the configuration mode change between 2S and 2P across the critical curve. In summary, they find that the acceleration of the body during the first half of each vibration half cycle leads to the roll-up of four regions of vorticity per cycle, but that the timing of actual shedding of the vortices changes with the cycle wavelength 2. Below a critical value of 2/h, two vortices of the same sign coalesce in each half cycle to create a Karman vortex street-type wake, the 2S mode. Above this critical value coalescence does not occur and the four vortices arrange themselves as two vortex pairs, the 2P mode. At the critical value of ,~/h, which varies with ~ as can be seen in Fig. 10, a single and more concentrated vortex sheds at the rear of the cylinder each half cycle, and Williamson and Roshko denote this condition as 'resonant synchronization'. They relate it to the occurrence of force maxima in Bishop and Hassan's results. Their arguments on the effect of wavelength on the timing of vortex shedding lead them to the conclusion that the mode jump from 2P to 2S will occur at a lower value of,~/h than the reverse jump. Thus a hysteresis loop is created, as shown for phase angle ~bin Fig. 11,just as observed by Bishop and Hassan for forced oscillations, and by Feng for free oscillations. The nature of the mode change is also consistent with the observed change in magnitude of ~y. across the jump, since one would expect smaller surface forces on the cylinder from the vortex pairs of the 2P mode than from the single vortices of the 2S mode. Thus the explanation offered by Williamson and Roshko for the occurrence of the oscillation hysteresis is very convincing. It was mentioned earlier that Williamson and Roshko reported the existence of a region of 1/3-subharmonic resonance. This occurred at the low-amplitude end of a mode zone coded 2 S + 2 P located near 2/h=15 (which corresponds to fc "-'x "" 3fv.). The coding denotes a configuration with vortex pairs forming at the top and bottom of the body trajectory, as in

188

G. PARKINSON

the 2P mode, but with additional single vortices of appropriate sign between each pair. Williamson and Roshko point out that the 2S + 2P configuration has the proper symmetries for stability, whereas a configuration that would lead to a 1/2-subharmonic resonance would not. Therefore they would not expect to find 1/N-subharmonics where N is even, but they would expect them where N is odd. In fact, Durgin et al. (1980) have found a strong 1/3subharmonic resonance in experiments on free oscillations of a circular cylinder. 2.3. COMBINED EFFECTS In this section phenomena observed for oscillating cylinders of square section are presented and discussed as representative of noncircular sections that can experience both vortex-induced and galloping oscillations from rest. Pure vortex-induced effects have been covered sufficiently in the previous section and pure galloping phenomena are relatively straightforward and predictable, and are covered along with the modelling theory in Section 3, so that this section deals with observations of combined effects of vortex resonance and galloping. 2.3.1. Parameters of Aeroelastic and Hydroelastic Vibrations I-Icre we consider the physical parameters of aeroelastic and hydroelastic vibrations of long bodies of square section immersed in flows normal to their span and steady upstream except for low to moderate turbulent fluctuations. The elastic system properties are supplied externally, usually in the form of linear springs and viscous-type damping, and the bodies are rigid. Also, the incident flows in the laboratory are generally nearly uniform so that conditions can be taken as two-dimensional. Then the oscillatory displacement Y can be modelled by the differential equation I?+ 2flI;'+ Y = CynU 2, (2.1)

for a linear elastic system excited by a fluid force. In Eq. (2.1) all quantities are in dimensionless form, as defined previously. Equation (2.1) serves for all of the forms of hydroelastic and aeroelastic vibrations considered here, with only the nature of exciting force coefficient Cy and the range of values of other parameters changing from one application to another. As a result, the equation can be used as a basis for a preliminary discussion of the different forms. First, an examination of the order of magnitude of its terms gives useful information. Parameter fl, the fraction of critical viscous damping for the elastic system, is typically of 0(10)- 2 in the field, where the natural damping of long metal cylinders is usually involved, and of 0(10) - 3 to 0(10) - 2 in the laboratory, where it is of interest to explore the effects of very low system damping. These values would apply in both aero and hydroelastic oseiUations. For mass ratio parameter n, however, usual values in aeroelastic oscillations will range from 0(10)-4 to 0(10)-3, while values in hydroelastic oscillations will range from 0(10)-2 to 0(1) because of the 800: I ratio of water to air density. For the forms of excitation considered Cy is of 0(1), and the relevant range for flow velocity U is from 0 to 10. Thus, for aeroelastic oscillations, in which Y can be of 0(1), Eq. (2.1) can be thought of as basically an equation for harmonic oscillations ~r+ y = 0, with the solution Y = Y cosz, (2.3) and amplitude Y determined, not by initial conditions, but by a balance between small exciting and damping forces. If, as in the theories to be considered in Sections 3 and 4, Cy is modelled as a nonlinear function of 1;'(the quasi-steady theory of galloping), or is regarded as being itself governed by a separate nonlinear differential equation (the wake oscillator theory for vortex-induced oscillations), the above balance between damping and excitation (2.2)

Flow-induced vibrations of bluffbodies

189

is found through the theory of weakly nonlinear differential equations, and the stable oscillations represent limit cycles. On the other hand, for hydroelastic oscillations with Y of 0(1), only the damping term in Eq. (2.1) is small, so that its effect can be exPected to be negligible, while the remaining terms suggest a departure from harmonic oscillations and the appearance of strongly nonlinear effects. Furthermore, the relationship between two important parameters of the theories of galloping and vortex-induced oscillation, Uo and Ur, changes when hydroelastic oscillations are being considered. It will be shown in Section 3 that Uo is the flow velocity above which galloping will occur, given by 2/7
Uo nA1

(2.4)

where A1 is the initial sloPe of the experimental curve of C r vs. a, the angle of attack of the section held stationary in the flow. Ur is the flow velocity at which the wake Strouhal frequency for the stationary cylinder equals the natural frequency of the elastic system, and is given by
1

u, = 2 ~ s '

(2.5)

where S is the Strouhal number of the section. Now S is of 0(10)- l for sections capable of both vortex-induced oscillation and galloping, so for those sections Ur is close to 1.0 in either air or water flow, from Eq. (2.5). For those same sections, A1 in Eq. (2.4) is of 0(1), so that for aeroelastic galloping Uo can range from below U, (for low/7 and high n) to much greater than Ur (for high/7 and low n). On the other hand, for hydroelastic galloping Uo will always be less than Ur for realistic values of /7, because of the high values of n. This has important implications for hydroelastic effects. In aeroelastic galloping, it will be shown in Section 3 that stable oscillations occur only for U > Uo, and at high U amplitudes ~ increase with U towards an asymptotic variation
= sU,

(2.6)

where sloPe s is determined by the shaPe of the section and the turbulence intensity of the flow. In vortex-induced oscillations, on the other hand, large amplitudes occur only if U is close to Ur, and Ym,. is of 0(1). It follows that for aeroelastic systems capable of both galloping and vortex-induced vibrations, the two forms can occur quite indePendently when Uo >> Ur, whereas for hydroelastic systems the two forms must interact for values of U near U,, since Uo is always less than Ur. Interaction effects can of course also occur in aeroelastic systems, since Uo can be near or less than Ur, and observations of such effects for both hydro- and aeroelastic systems are considered in the next sub-section. 2.3.2. Hydro- and Aeroelastic Phenomena In this sub-section the results of some hydroelastic and aeroelastic exPeriments on cylinders of square section in the author's laboratory are compared with one another and with the results of similar hydroelastic exPeriments by Bokaian and Geoola (1983, 1984). Maximum Reynolds numbers in the different experiments ranged from 5,000 to 17,000 and the water and air flows used were quite uniform and of low (< 0.1%) to moderate (7%) turbulence intensity. Examples from the aeroelastic experiments were chosen with damping fl low enough to make Uo < Ur, as in all the hydroelastic experiments. In Fig. 12, dimensionless cylinder amplitude ~ and frequency fi, defined by
=

~o/~.

= f~/fn,

are plotted against U from four experiments on flow-induced oscillations of square-section cylinders. Two experiments (Bearman et al., 1987) were in a low turbulence wind tunnel (one exPeriment with turbulence grid installed), one exPeriment (Bokaian and Geoola, 1983) was

190

G. PARKINSON

(a)

oi

o0

O ~ l "~

o
-

o;O oo/ / /1 /,
o5
u

'

l/'i

y . ./ "

/<,,,

t
U,

6 U

I0

(b]
| .,. . . . . ~ .....

I 0.6-0 ! !

t U,

4 U

I0

FIG. 12. Vibrationphenomenaof square-sectioncylinders.(a) Y,vs. U, (b) fl' vs. U. Aeroelasticexpts: Bearmanet al. (1987) - - - , . . . . . . . Hydrcelastic expts.: Bouclin (1977) .... , Bokaian and Geoola (1983); theory: Bouclin (1977),quasi-steady O, wake oscillator <>, Strouhal line . . . . , regime boundary (after Bearman and Luo, 1988). . . . .

in a water channel with moderate turbulence, and one in another water channel (Bouclin, 1977) with relatively low turbulence. The data for each experiment arc shown as single curves, since scatter was small. Points from numerical calculations representing two different theories, to be considered in Sections 3 and 4, are also shown. The curves of Y, vs. U all lie close together a~d display a nearly linear increase of Y, with U from zero near Ur. Superimposed on the nearly linear trend, a small regular wave-like modulation can be seen clearly in three of the curves and may be present in the fourth, from Bokaian and Gcoola (1983), although there arc too few data points to be certain. As to the ordering of the curves, the highest is in water flow with relatively low turbulence and the lowest is in air flow with moderate turbulence. For the frequency variations, in the aeroelastic experiments f~ ~ 1.0 to the accuracy of measurement, so only the results of the hydroelastic experiments are plotted in Fig. 12. For these the curves of f~ vs. U are very similar and lie close together. For U near Ur the curves lie close to the line f~, = 2 n S U , representing the Strouhal frequency for the stationary cylinder, and fl increases with U, becoming constant for higher U at values close to 0.9. In addition to these strong oscillations, starting near U, and growing with U, in one of the hydroelastic experiments shown in Fig. 12 there was observed an additional weak but clearly identifiable oscillation near 1/3U,, with f~ near 1.0. This was also observed in experiments on the other two cylinders of square section tested in the program reported in Bouclin (1977). The effect of the mass and damping parameters n and/7 on the hydroelastic results is of considerable interest. It is convenient to divide the flow velocity range into two parts. For U E 2 amplitudes Y become of 0(1), so that the damping term 2/1I7 in Eq. (2.1) can be neglected and with the quasi-steady assumption that Cy is a function of Y / U only, the

Flow-induced vibrations of bluff bodies

191

equation can be transformed to

~" + r ' = f U '2,

(2.7)

where Cy = Cyof(}'/U), Y ' = CyonY, U' = CyonU. Thus all hydroelastic galloping amplitude-flow velocity data for a given cylinder section should be collapsable to a single curve independent of system parameters, provided that flow conditions (Reynolds number, turbulence) are the same, or nearly so. The factor Cyo is introduced to allow for differences in Reynolds number, based on the observation that curves of Cy vs. a in experiments (Bokaian and Geoola, 1983; Bearman et al., 1987) on cylinders of square section in flows with about the same turbulence intensity but of considerably different Reynolds number had the same shape over the same range of a, but differed in ordinate. The hypothesis is tested for cylinders of square section in Fig. 13, where data from five different cylinders and a quasi-steady theory (to be considered later) are compared. The data were obtained in three different flow channels with flows of low to moderate turbulence intensity covering different Reynolds number ranges. In view of these qualifications, the degree of collapse of the data can be considered a satisfactory verification of Eq. (2.7). Galloping frequency ~ for U > 2 is also dependent on n and can be seen in Fig. 12 to approach a constant value which increases with decreasing n. For U 52, Y is small enough that all terms of Eq. (2.1) are significant and the ratio fl/n (proportional to the Scruton number) becomes an important parameter. Here it is interesting to examine two characteristics that are distinctly different from their counterparts in aeroelastic galloping--the critical speed above which galloping occurs, here denoted by Ui to distinguish it from Uo as given by Eq. (2.4), and the departure of the oscillation frequency from the system natural frequency, here measured by the function (1 - f l ) . In Fig. 14, Ui and (1 - f~r), where O r is the value off~ for U = {Jr, are plotted vs. fl/n for four of the experiments examined previously in Fig. 13 for effects of higher flow velocity. In addition calculated points from a fluid oscillator theory (Bouclin, 1977) to be described later, are shown on the figure. It can be seen that the experimental data points in each case plot on a smooth curve giving physically reasonable trends. As fl/n becomes large the trends are consistent with an approach of Ui to Ur and of(1 - O r ) to zero, limiting values typical of cases of aeroelastic galloping for which fl/n is high in comparison with hydroelastic values, but low enough to make U~ = Ur, as in the two aeroelastic examples of Fig. 12. As fl/n --* 0, the trend of Ui suggests a curve starting from the origin. Thus far the discussion of observed hydro- and aeroelastic phenomena for cylinders of square section has focussed on the cylinder response characteristics. For an understanding of the phenomena it is of course also necessary to study the characteristics of the excitation,

03

02

y,

J
w

J(

~(

0.,

02

04

0.6

U'

FIG. 13. Reduced amplitude vs. reduced flow velocity for square-section cylinders. Theory: Bouclin ( 1 9 7 7 ) Expts.: Bouclin (1977), n = 0.033 = ; 0.052 = A; 0.093 = ~7; Santosham (unpublished), n = 0.097=X; Bokaian and Geoola (1983), n=0.061 = FO,

192

G. PARKINSON

Ur---iP 12
0 0

Ub 081

0.4

I ",~,r
0 0
O.2

0.4

0.6

B/n
FIG. 14. Initial galloping parameters for square-section cylinders. Expts.: Bouclin (1977) Q, A, V; Bokaian and Geoola (1983) D. Theory: Bouclin (1977) .

/
/ J 0/

s"

~v 4

0050~0 0 0 0
o ooo

/ /

f
0 ,o 0=$o o = O+ 0 0 0 o

mo~b$

I
U

I 8

I I0

U~

FIG. 15. Wake vortex frequencies vs. flow velocity for square-section cylinders. Expts.: Bouclin (1977) O, Parkinson and Santosham (1967) A, Bokaian and Geoola (1984) II, + , 41,; Theory: Bouclin (1977) ~ ; Strouhal line - - - .

and therefore of the flow. In some of the hydroelastic experiments examined here (Bouclin, 1977; Bokaian and Geoola, 1984) pertinent wake measurements were made using hot film anemometers, while in some of the aeroelastic experiments wake measurements were made with a hot wire anemometer (Parkinson and Santosham, 1967), or measurements were made of fluctuating surface pressure on the cylinder (Bearman et al., 1987). Figure 15 shows a sample of measured wake frequencies ~ , of square-section cylinders obtained by spectral analysis, plus the stationary-cylinder Strouhal line and some points calculated numerically from the fluid oscillator theory of Section 4.2. First some data not shown in the figure should be mentioned. Data from the smallest of the three cylinders whose tests are described in Bouclin (1977) is shown in Fig. 15, since for it these tests cover the same Reynolds number range as those of Bokaian and Geoola. However, for this cylinder actual flow velocities were lower than for the other two, and as a result wake signals were too weak to be detected for U much below U,. For the other two

Flow-induced vibrations of bluff bodies

193

cylinders wake signals were measured at D~ = 0.5 and 1.0 over the range of U near 1/3Ur in which the low-amplitude cylinder oscillations at 12 = 1.0 were observed, as mentioned previously. For U near Ur, wake signals could be obtained for the small cylinder, and Fig. 15 shows that in this range D~ is synchronized with 12 (shown in Fig. 12), initially with both following the Strouhal line and then, still synchronized, both curving away towards a constant value less than, but close to fl -- 1, corresponding to the cylinder natural frequency. This trend is also suggested by the data of Bokaian and Geoola although few wake measurements are provided. However, for U ~>2 the trends in Fig. 15 from the two sets of hydroelastic data are quite different. In Bokaian and Geoola (1984), two strong signals are found at each flow velocity, one at the constant cylinder frequency 12 and the other at the stationary cylinder Strouhal frequency, proportional to U. On the other hand, in Bouclin (1977) no signal at 12 was detected for U>~2, and no continuous trend following the Strouhal line. Instead, as U increased nearly constant signals at harmonics of 12 were observed over ranges of U which sprang from or straddled the Strouhal line. Harmonics at 3[~, 4f~, and 512 are shown in Fig. 15, but Bouclin also detected signals at 212 for the other two square-section cylinders tested. Corresponding results from the aeroelastic experiments cited are of interest. Figure 15 also shows the wake signal reported for a square-section cylinder by Parkinson and Santosham (1967) in which the frequency follows the Strouhal line over the full range of wind speed tested, whereas only nearly-constant frequency near 212 was detected for a rectangular-section cylinder with d/h= 2. The fluctuating surface pressure measurements during aeroelastic vibrations of a square-section cylinder reported by Bearman et al. (1987) indicated organized wake phenomena corresponding to all of the above observations. Strong signals were obtained at 12, 312, and 512; and at the Strouhal frequency (although the relative strengths changed with U), and weak signals at 212 and 412. Also, it should be noted that there seems to be a correlation between the occurrence of these frequency harmonics and the wave-like modulation of cylinder amplitude in Fig. 12. Figure 16, taken from Bearman et al. (1987) shows an example of measurements of the r.m.s, value C~ of the fluctuating transverse force coefficient, the signal at body frequency 12, denoted by Cy,, and the observed harmonics C'y2 - C'y,. For these, only the maximum value of each spectral peak was measured, and not the integrated value as would be required for

1.4--

12

c;,
0.8

C;N
0.6

0.4

i
I
;.

",~
~

+ c~
"~,
k c,

Y~

0.2

d
I
I 2

~"~
3

c;~
5

J
U

FIG. 16. Force coefficients and harmonic components in acroelastic vibrations of a square-section cylinder. Data from Bearman et al. (1987) for case with low turbulence and low damping.

194

G. PARKINSON

precise values. The results are therefore estimates of the relative importance of each harmonic, and their approximate relationship to the spectrally complete value of C'y. Where broad peaks in the spectrum occurred, no measurements of the harmonics were made. Figure 16 presents values for the case from Fig. 12 in which the flow turbulence level was very low ( < 0.1%) and the system damping was low enough that U0 was less than Ur, so that vibrations began at U,. The figure shows a general trend of C'y decreasing with increasing U and that most of the side force exerted on the vibrating body occurs at (or very close to) the body frequency at least up to U ~ 5. Near U = 3, where a kink occurs in the corresponding Ys vs. U plot of Fig. 12, C'y3 rises sharply, indicating the presence of a strong force component at about three times the body frequency. At the upper limit of the amplitude range for the apparatus (Y, ~ 1.4), occurring for low damping and low turbulence at U ~ 5, it can be seen in Fig. 16 that C'r3 is rising sharply showing now the strong presence of the fifth harmonic of body motion in the force. The even harmonic terms C'r2 and C'y, remain low throughout. Thus the data of Fig. 16 seem to confirm the sequential synchronization of the wake vortex system with harmonics of the body motion shown in Fig. 15, but also demonstrate the dominant effect of the odd harmonics. The phase angle ~ between the body displacement Y and the force coefficient at the body frequency C'y, is plotted vs. U in Fig. 17, also from Bearman et al. (1987), for three experiments at progressively higher values of system damping 8, all in low-turbulence flow. The values of ~b are seen to increase progressively with U, although considerable scatter is evident. The quasi-steady theory of galloping to be developed in Section 3 of the paper is assumed to be valid for high U, and in it Cy is assumed in phase with body velocity, so that ~b = 90 . The trend in Fig. 17 is consistent with an approach to ~b = 90 at high U. Another result from Bearman et al. (1987) is deserving of comment. In all but their first test run (low damping, low turbulence) a search was made for spectral peaks other than at the body frequency and its harmonics, and in all cases such a peak was found corresponding to the Strouhal frequency for the stationary cylinder. The corresponding Strouhal number was remarkably constant considering that it was measured during oscillations with amplitudes as large as Ys = 1.3. For high damping cases the excitation at the Strouhal frequency was

80

70

6
~0 o

6o o

60

5O --

9 /
4O

9 99

30

20

I
U

I0

FIG. 17. Phaseanglein aeroelasticvibrationsof a square-sectioncylinder.Data from Bearmanet al. (1987)for lowturbulence cases. Damping:low Q, medium (3, high ~3.

Flow-inducedvibrationsof bluffbodies

195

much stronger than at the body frequency, and S from individual data points was constant within 3 %. For medium damping cases the Strouhal and body-frequency excitations were of the same order and S was constant within 4%. For low damping cases the Strouhal excitation was weaker than that at the body frequency, and S was constant within 7%. The overall mean value was S = 0.132, very close to the value of 0.133 measured for the same cylinder at rest in the same Reynolds number range in low-turbulence flow.

2.3.3. Phenomena from Forced-vibration Experiments The results in the previous sub-section demonstrate the complexity of free-vibration phenomena for the square-section cylinder, arising from its ability to exhibit both vortexinduced and galloping oscillations, separately or in combination. This ability arises in turn from the size and shape of the section afterbody, which interferes during oscillation with the process of vortex formation from the separated shear layers. A consequence of the complexity is a corresponding difficulty in interpreting the results and finding explanations for the phenomena. This complexity can be reduced simply by reducing the number of dependent variables in the problem, and this is the motivation for forced-vibration experiments. In the experiments in Section 2.3.2, U was the independent variable while/~ and n were controlled constants (although the effective value of fl is difficult to determine). Dependent variables included the cylinder variables Y, f~, and waveform (at least for hydroelastic vibrations), and the flow variables C'y, C'FN, tar, and $. In addition there are bounding values in the results which require interpretation, such as Y==,=for vortex-induced vibrations and Ui and s for galloping. On the other hand for forced-vibration experiments either U or ~ is the independent variable; with the other and ~ as controlled constants, the waveform is harmonic and only the flow variables are dependent. The quantities fl, n, Y=. . . . Ui, and s are not involved. Several investigators have carried out forced-vibration experiments on square-section cylinders, including Otsuki et al. (1974), Nakamura and Mizota (1975), Bearman and Obasaju (1982), Obasaju (1983) and Bearman and Luo (1988). It is useful to examine the forced-vibration results obtained by Bearman and his students because they relate closely to free-vibration results from the author's laboratory and contain some interesting correlations of forced- and free-vibration data. A brief history of the investigations that led to these correlations will perhaps help to avoid confusion. Wawzonek (1979) carried out a study of combined effects of free vortex-induced and galloping oscillations of square-section cylinders under two-dimensional conditions in lowturbulence flow. In these experiments Uo was varied from below to well above Ur, and the variation of Y=with U was measured for each Uo. Some interesting combined effects were observed, but the information was incomplete since fluid forces were not measured. This led to a collaborative effort with improved apparatus which finally resulted in Bearman et al. (1987), discussed in Section 2.3.2. Meanwhile, in his own laboratory, Bearman and his students pursued complementary forced-vibration studies. Obasaju used results from these studies to estimate free-vibration characteristics of the square-section cylinder for velocities near U,, much as Staubli (1983) did for the circular cylinder. Figure 18, taken from Obasaju (1983), shows three points calculated from forced-vibration data superimposed on a ~ - U plot of Wawzonek's results, with good agreement. Recently Bearman and Luo (1988) have reported on a further forced-vibration study of the square-section cylinder, in which the principal aim was to measure the r.m.s, transverse force coefficient C'y and its spectral components over a wide range of U, including the range in which galloping and vortex-shedding forces may interact. For measuring C'y a pressureaveraging manifold of a type proposed by Surry and Stathopoulos (1977) was used, and since this same method had been employed in the free-vibration experiments of Bearman et al. (1987), the two sets of force data were more directly comparable than usual. Figures 19 and 20, adapted from Bearman and Luo (1988), show plots of C'yvs. U and ~bvs. U, each for two constant values of :Ps that were within the range observed in the freeJPAS 2 6 : 2 - F

196
08

G. PARKINSON

0.7 06 05 04
03

0 o 0
O o

Q o

cPo
oo

02

c~qb

I0

12

14

16

18

2~U

FIG. ! 8. (From Obasaju, 1983.)Comparisonof Y,vs. U data for vibratingsquare-sectioncylindersfrom aeroelastic experiments by Wawzonek (1979): O, El, and forced-vibrationexperiments by Obasaju (1983): O, converted to aeroelastic form.

2 25
2-I 75

Ur

4,
I

8
I

12 I

16 I

20 I

24

(o)
--~

150 --,

c;

12'3
I

0 ;'5 O50

"

Y(bll
I ,.,

I
I ..,

2.2 /,,,
Cv
i

,~ |
13
I

I
l ~'~f, v 4 ~

1
ipl O O O

pw

o7

U,

V,
4

12

FIG. 19. (From Bearman and Luo, 1988.) C'yvs. U in forced vibrationof a square-sectioncylinder.(a) Y, = 0.25, (b) g,=0.675. Data points transposedfrom aeroelastic experiments by Bearman et aL (1987): +. vibration experiments. Phase angle ~b is again the angle by which the component of force at the body frequency leads the displacement. The first things to note in the figures are the points marked + on the plots (one each in Figs 19(a), 19(b); three each in Figs 20(a), 20(b)). These were taken from the free-vibration data plots in Bearman e t al. (1987) of C'y, ~ and Y, vs. U, reproduced in the present paper in Fig. 16, Fig. 17, and in part in Fig. 12(a). The agreement between the two sets of data is remarkably good. This is particularly impressive at

Flow-induced vibrations of bluff bodies


120

197

=- (a)

90--

60--

30--

o-U
- 30 --

-60--

-9o
120 - -

I (b)

-e.
90--

0 60

30

-30

-60

-90

~l 2

I0

f2

14

P6

18

Ur

FIG. 20. (From Bearman and Luo, 1988.) ~ vs. U in forced vibration of a square-section cylinder. (a) Y, = 0.25 (b) Y,--0.675, Data points transposed from aeroelastic experiments by Bearman et al. (1987): +.

the lower values of U, where both C'y and 0 experience rapid changes with U and sharp peaks and troughs occur. It is noteworthy that the transferred point in Fig. 19(b) lies at the local Cy peak in the forced-vibration data occurring in the range of U for local synchronization at 3f~ (see Fig. 15), and this point happens to be transferred from the corresponding C; peak in the free-vibration data of Fig. 16. Since the free- and forced-vibration results are in good agreement, what can be learned about the free flow-induced vibrations from the forced-vibration experiments? First, the phase measurements of Fig. 20 demonstrate why no free vibrations occur for U < Ur ~ 1.2, since 0 is negative. Moreover, the value of U at which ~ becomes positive increases with F,, so that not only will free vortex-induced oscillations occur only in the part of the primary synchronization range for which U _> Ur, but their amplitude is severely limited, even for low system damping. Next, Figs 19 and 20 reveal an interesting form of organization of the wake interactions with the cylinder. At a velocity Urn> Ur force coefficient C'y has a minimum Cym~,0.75, independent of F,, although Bearman and Luo's complete test results show that Um increases finearly with ~ from approximately U, at ~ = 0. At large values of U, C'y approaches its stationary-cylinder value and ~ approaches 90, both consistent with the assumptions of the quasi-steady theory of galloping to be developed in Section 3. Bearman and Luo suggest defining three ranges of U as indicated in Fig. 19(b): (i) U < Urn, (ii) Um< U < U,, where U, corresponds to C'y,~1.4 and is also found to increase linearly with F~, (iii) U > Us. In range (i), defined as the "multiple lock-in' regime, the angle of attack a of the flow velocity relative to the oscillating cylinder (see Section 2.1) reaches high values because of the relatively low U for a given F,. This causes strong interactions between the shear layers and the sides of the square section, including intermittent reattachments, which in turn leads to drastic modifications of the vortex shedding process. Flow visualization l~hotographs taken

198

G. PARKINSON

of square-section cylinders during forced vibration in range (i), such as Fig. 20(b) of Bearman and Obasaju (1982) and Fig. 12(e) of Ong6ren and Rockwell (1988), demonstrate these effects, which are undoubtedly responsible for the 'multiple lock-ins'. Of course, for a given Ys, only the part of range (i) for which Ur < U <Um is relevant for free flow-induced vibrations, but this seems to include the portion of Ys-U space containing the low-damping curves of Fig. 12(a), so it seems clear that the scalloping superimposed on these curves is a consequence of the 'multiple lock-ins'. (A line representing the variation of Um with Y~, taken from Bearman and Luo (1988), is superimposed on Fig. 12(a), thereby defining the boundary between regimes (i) and (ii) on the Y~-U plot.) In range (ii), defined as the 'lift-recovery' regime, amax in the oscillation cycle is smaller because of the higher values of U for a given ~ , so the afterbody interference with the shear layers is diminished, although still present. The oscillation characteristics correspond qualitatively, if not quantitatively, to those of galloping in free oscillations. In range (iii), defined as the 'small-incidence quasi-steady' regime flow conditions during oscillation are quantitatively compatible with the quasi-steady assumptions on which the galloping theory of the next section is based.

3. MODELLING GALLOPING
3.1. QUASI-STEADYTHEORY

It is appropriate to consider the modelling of galloping before that of vortex-induced vibrations because the theory of galloping is not controversial and it will be necessary to describe only one model in detail to give an adequate picture. It was shown in the previous section that pure galloping occurs at high values of U, or low values of 1/U, the reduced oscillation frequency, and this is the classical condition for the application of a quasi-steady theory for a body oscillating in a flow. The name of den Hartog (1930) is usually associated with the quasi-steady force criterion for galloping instability, and Sisto (1953) has applied nonlinear methods like the ones described here to the stall flutter of airfoils. Scruton (1960) has determined galloping amplitudes of bluff cylinders from empirical force data by a graphical method, and Novak, in a series of papers (1969, 1970, 1972), has extended the nonlinear model for two-dimensional flow described here to continuous elastic systems, such as cantilevered towers, and turbulent shear flows. He has also proposed a 'universal response curve' which would permit the prediction of galloping characteristics of a long body of specified bluff cross section from a dynamic wind tunnel test of a scale model, without the necessity of static force measurements. The form of theoretical model presented here was developed by the author and published in Parkinson and Smith (1964). The theory is a two-dimensional quasi-steady model, in which it is assumed that the bluff section, at an instant during the galloping cycle, (see inset of Fig. 2(b)) experiences the same force Fy that it would if held stationary at the angle of attack a and incident flow velocity Vr I arising from the transverse cylinder velocity $. The cause of the force Fy is the pressure differential on the sides of the section afterbody created by the proximity of the separated shear layer on the lower, or windward, side of the section as shown in Fig. 2(b). As mentioned the validity of this assumption requires that the reduced frequency of galloping be very small, or equivalently, that U ~ 1. With the usual linear elastic system for the model, Eq. (2.1) can be employed, with Cy here given by static force measurements as a function of a. Since no analytical theory capable of predicting it is yet available, (see, however, Section 4.1 for comments on current numerical methods) the experimental variation of Cy with a (and therefore with tan a) is approximated by a polynomial in tan a = -- = -V U'

How-induced vibrations of bluff bodies


06--

199

0.~

Cy

o o 0.2-0

o
0

I0

15

Fie. 21. Cy vs. a for square-section cylinder. Data from Parkinson and Smith (1964): expt. O, polynomial - -

such as the 7th degree polynomial used for the square section in Fig. 21, for which the measurements were made in a uniform, low turbulence air flow at a Reynolds number of 2.2 (10)4:

where A 1, A3, A s , AT, are positive constants. We will follow the argument of Parkinson and Smith (1964), from which Fig. 21 is taken, through for the square section. If Eq. (3.1) is substituted in Eq. (2.1) the result can be written:

= ltf(~'),

It = n A 1 ~ 1 for air f

Equation (3.2) has the form of a weakly nonlinear autonomous differential equation capable of solution by the first approximation method of Krylov and Bogoliubov (Minorsky, 1962, Ch. 14). For/~ = 0, the solution is the familiar Y = Y cos r, I;" = - Y sin ~, (3.3)

where Y is constant. The method assumes that for/~ ~ 0, but small, the solution can be obtained as a series in powers of #: Y = Ycos z +/tY1 ( Y,, z)+/t~ ]'2( Y,, ~)+ . . . (3.4)

where Y is now a slowly varying function of ~. (In general, there is also a slowly varying phase angle, but this vanishes in the first approximation for Eq. (3.2).) In most applications, as here, only the first approximation is needed to give an accurate estimate of the oscillation phenomena. In the first approximation, Eqs (3.3) are retained, but Yis considered a function

200

G. PARKINSON
x.

of

If Eq. (3.2) is multiplied by 18, the result can be written ld( 2d* y 2 + 182) = #f(18). 18, (3.5)

or, using Eqs (3.3), ld7 2


2 dx = - #f(~

sin x) Y sin ~.

(3.6)

It can be seen from Eq. (3.6) that d y2/dz is very small, so that the change in Y is negligible over one oscillation cycle. Accordingly, it will be satisfactory to replace the right side of Eq. (3.6) by its average over one cycle, thereby eliminating harmonics: d f2 d~ /~ ~

f;"

f ( - f sin 0 Y sin v dr.

(3.7)

If the function f from Eq. (3.2) is inserted in Eq. (3.7), the integration yields

=.A1

U(3.8)

Equation (3.8) completely determines the galloping behaviour of the bluff cylinder of square section in two-dimensional uniform air flow. It can be integrated directly, but a preliminary qualitative analysis of its form and its relationship to the behaviour of the system in the phase plane (Y, 18) may be useful. The form of Eq. (3.8) is dR dT = at R - a3 R 2 + as R 3 - a7 R 4 = F (R), (3.9)

where R = ~2 and at, a3, as, aT, are functions of the system parameters. Stationary oscillations correspond to (dR/d~) = 0, and are therefore given by R = 0, and by the real positive roots of the cubic a t - a 3 R + a s R 2 - a T R a = 0. (3.10) In the phase plane (Fig. 22), R = 0, the initial position of equilibrium, is a singular point at the origin called a focus. The positive roots Ri of Eq. (3.10) define trajectories in the phase plane called limit cycles, here concentric circles of radius ~, = x/~i. The polar angle 0 in the phase plane is #oven by tan 0 . . . . Y
so that --

18

tan ~,

(3.11)

dO d,

- 1,

(3.12)

and a stationary oscillation of unit circular frequency (real time circular frequency ton) and amplitude Yis represented in the phase plane by a point rotating clockwise with unit angular velocity on a circle of radius Y,t. The stability of equilibrium and of limit cycles is determined by investigating the tendency of the oscillator to return to the original focus or limit cycle after a small displacement 6R. The pertinent function is therefore d6 R/dT, and from Eq. (3.9) this is given by

dtSRdz ~=~,= F'(Ri)6R,

(3.13)

where the prime indicates differentiation with respect to R. It follows that the sign of F'(Ri) determines the stability of the limit cycle, which is unstable, stable or neutrally stable according to whether F'(Ri) is positive, negative or zero. Here

F'(R) = at - 2 a 3 R + 3asR 2 --4a~R 3.

(3.14)

Flow-induced vibrations of bluff"bodies

201

F I R ) ~

(o)

RI~///R 2 /R3

~,~'~

(b)

FIG.22. Phase andcorresponding planes functions in theoryof galloping F(R) (Parkinson Smith,1964). and
The stability of the focus at R = 0 is determined by the sign of
F'(O) ffi a t ffi n A l

U-

(3.15)

from Eqs (3.14) and (3.8). If B is negligibly small, the sign is that of A1 which is positive for the square section, showing it to be unstable at rest at any wind speed in the absence of system damping. Now
AI=

~Ca'[=o' 2~

and this criterion for instabifity is equivalent to the one proposed by Glauert (1919) for autorotation of a stalled airfoil and later by Den ttartog (193(I) for the galloping of electric transmission lines. If ~ > 0, the square section is unstable only for
U > U o = nA 1,

(3.16)

so that galloping can be eliminated below a given wind speed by sufficiently increasing the system damping. IfEq. (3.16) is satisfied and a~ > 0, then it is clear that there must be at least one limit cycle corresponding to a positive root R t of Eq. (3.10), since aa, as, a7 are all positive from Eqs (3.8) and (3.1) and the left side of Eq. (3.10) will therefore be negative for large R. This limit cycle may be the only one, in which case it is stable, since the slopes of F ( R ) at successive real roots will alternate in sign, unless a double root intervenes. The phase plane then shows (Fig. 22(a)) an unstable focus encircled by a stable limit cycle, a system with soft self-excitation designated ~ ~ , following Minorsky 0962), Ch. 7. The corresponding graph of F(R) is also shown in Fig. 22(a). Under these conditions the system oscillates spontaneously from rest with increasing amplitude until a stationary oscillation of amplitude Y , l = x / ' ~ l is attained.

202

G. PARKINSON

Another possibility is that all three roots of Eq. (3.10), R 1, R2, Ra are real and positive, say R3 > R2 > R1. With a 1 > 0 (U > Uo) this means that R1 and Ra represent stable limit cycles separated by an unstable limit cy/cle, corresponding to R2. This system is designated ~ / ~ 6 e , and the phase plane an d graph of F(R)are shown in Fig. 22(b). The transient oscillations, not shown, appear o/fi the phase planes as tightly coiled spirals away from the focus and unstable limit cycle, 0/nd towards the stable limit cycles. Inspection of Fig. 22(b) suggests the remaining possible systems. R 2 and R 3 could coalesce to a double root, giving a system designated q/Se(q/Se), intermediate between those of Figs 22(a) and 22(b). Alternatively, R~ and R2 could coalesce to a double root, giving a system ~'(6ad//)J, intermediate between that of Fig. 22(b) and a final possible system in which R 3 is the only real positive root. The phase plane for this system would be like that of Fig. 22(a). F o r a square prism with given mass and damping parameters, n and fl, the choice among these possibilities depends on the reduced wind speed U as parameter. This is an example of bifurcation theory in nonlinear oscillations. The coefficients a 1, 03, as, a7 all depend on U, and as it passes through critical values, the phase portrait of the system changes from one of the forms described above to another. The dependence of the system on the parameter would commonly be shown on an (R, U) diagram. Here it is convenient to use a ( Ys, U) diagram, giving stationary amplitude plotted against wind speed directly (Fig. 23). The curves predict that the system has oscillation hysteresis. At values of U < Uo, Ys = 0 and the phase portrait is a stable focus. Small oscillations die out and the cylinder remains at rest. At U = Uo the system forks into a form whose phase portrait is like Fig. 22(a), ~Sf, with the amplitude of the stable limit cycle, Ys~, increasing with U, as shown by the curve 0 12 in Fig. 23. Between Ut and U 2, however, the phase portrait has become that of Fig. 22(b), q/6eq/Sf, with the intermediate forms ~6e(q/6e) at UI, and //(~//)Se at U 2. Between U o and U 2 the lower limit cycle Ysl is reached from rest. For U > U2 the form becomes ~/SP with the single stable limit cycle of amplitude Y~3reached from rest. If the wind speed is then decreased to a value between U~ and U 2 while the prism is still oscillating, the upper limit cycle is the one reached, until for U < U~ the amplitude drops to that of the lower cycle again. Thus, the path 012343510 containing the hysteresis loop 12351 is traversed on the (I7s, U) diagram. The unstable limit cycle in the range Ux < U < U2 discriminates between the stationary oscillations on the basis of initial amplitude. If the initial amplitude is less than Y~, the stationary amplitude reached is Y~. If it is greater than Y~, the stationary amplitude is Y~. Two other important characteristics of the ( ~ , U) variation can be seen by equating the left side of Eq. (3.8) to zero, and dividing through by nA1U~2: (1

n~fltU)-(4 A t ) ~ - ) + \8A1 ](~--~-~)4 - \(35A7~(uY----~)6= 0. \ 3A3"~{~'x2 (5A5~ ~]

(3.17)

"6r

"'~2
. . . . . .

//

oLi
Uo

Ul

U2

FIG. 23. ]7 vs. U curvesfrom galloping theory for square-sectioncylinder,showing bifurcations and oscillation S hysteresis (Parkinson and Smith, 1964).

Flow-induced vibrations of bluff bodies

203

It is already known that one stationary amplitude Y,3 exists for U > U2. Equation (3.17) shows that as U ~ oo, ~/U becomes independent of U, so that the ( Y,, U) curve becomes asymptotic to a line through the origin of slope s. Moreover, it can be seen from Eq. (3.17) that a curve of (nA1/2[3)Y, or YJUo plotted against (nA1/2fl)U or U/Uo would be independent of n and fl so that if the theoretical curve of Fig. 23 were plotted in this form it should be universal for the square section, and experimental values of ~ and U for oscillations of square prisms conducted at various levels of n and fl should collapse on this single curve when also plotted in this form. It remains to consider the transient oscillations quantitatively by integrating Eq. (3.8). With some rearrangement of terms, it can be integrated to
-A~ =

fj

oZ(Z-Zl)(Z-Z2)(Z-Z3)"

dZ

(3.18)

where A - 35nA7 U 64 ' Z = U' Z0 = \U-J' Zi = i = 1, 2, 3;

Y0 is the initial amplitude and Y,t are the amplitudes of the limit cycles discussed above. For U1 < U < U2 the Z i are all real and the integral is -Az= where
1

ln~(Z ~-l/z'z2z3(z-z1 )z"23(Z-Z2 )Z2"~(Z-Z3 \Zo-Z, \Zo-Z2 \Z~-Z3,]

Zi,jk = Zi(Z i_Zj)(z i - z k)"


For Uo < U < U,, or U > U2, one of the Z~, say Z3, is real and the other two are complex conjugates ZI, 2 = P + iq. The integral then takes the form (3.20)

-AT~In{(~o)-I/Z3(P2+q2':Z-Z3 3)
-~ (p2-Z3p-q2) {tan-' q(p2 -t-q2)[(p __Z3)2 + q2-]
( ~ ) tan-' (~-~)}. (3.21)

Equations (3.19) and (3.21) give the form of the amplitude-time variation of the prism oscillations, and permit an estimate of the time to build up to a stationary oscillation. This must be defined arbitrarily as, for example, the time to build up from Yo to 0.99 Y,1, since the theory gives an asymptotic approach to the focus and limit cycles. In the next section, some examples of the quantitative application of the theory to prisms of square section are given.

3.2. TWO-DIMENSIONALAPPLICATIONS Predictions from the theory of Section 3.1 for the square section in smooth air flow are compared with experimental results in Figs 24 and 25. The experimental data in these figures and in Fig. 21, used to determine the polynomial coefficients A i, were obtained by Smith (1962). The test cylinder was of 2.54 cm square section and the dynamic test arrangement was as described for Feng's experiments in Section 2.2.2. Mass parameter n was 0.00043 and four different values of damping parameter fl were used, ranging from 0.0011 to 0.0037. The

204

G. PARKINSON

incident air flow was uniform and of low turbulence intensity, and test conditions were closely two-dimensional. In Fig. 24 the universal form of plot ~/Uo vs. U/U o is used, and the theoretical asymptotic and stable and unstable limit cycle curves are shown. Experimental points for the three higher values of fl are plotted. Data for the lowest value o f / / i s omitted, since the corresponding Uo is only 1.54 U,, and this places the data in Bearman's transition zone (ii) of Fig. 19, rather than in zone (iii) where the quasi-steady galloping theory should be quantitatively applicable. (The values of Uo relative to U, for the data plotted in Fig. 24 are indicated by the three marks with appropriate symbols along the abscissa, denoted UdUo. ) It can be seen from Fig. 24 that the predictions of the theory are verified. The critical wind speed Uo, the asymptotic slope s, the existence of oscillation hysteresis, and the collapse of the data onto the calculated stable limit cycle curves are all closely realized by the experimental results. Figure 25 shows two comparisons of theoretical experimental transient amplitude-time curves. The first comparison, curves 1 in Fig. 25 corresponding to point 1 in Fig. 24, is made
12--

08-

S
t~
04-

/ / / /
I Ur/Uo

4
2

-4.
3

U/Uo

FIG. 24. Universal amplitude-velocity characteristic for galloping square-section cylinder (from Parkinson and Smith, 1964). Theory: stable , unstable ....... , asymptote - - - . Experiments at three values of ~: O, A, ~7. Points 1 and 2 correspond to curves of Fig. 25: +.

,:,I]

,,C
ii

%.oo5

300

600 Cyctes ~

900 r

t20o

FIG. 25. Transient amplitude increase in galloping of square-section cylinder (from Parkinson and Smith, 1964). Curves 1 and 2 correspond to points I and 2 in Fig. 24. Theory: . . . . . Expt.:

Flow-inducedvibrationsof bluffbodies

205

for a wind speed well removed from the region of oscillation hysteresis. The theoretical and experimental curves are matched at an initial amplitude Yo = 0,05, since this seemed characteristic of the start of galloping when the cylinder was released by turning on the air supply to the air bearings. The two curves are in good agreement, except that the theoretical curve has a slight hump at about 150 cycles, not observed on the experimental trace. This is a residual effect of the lower limit cycle that would occur at a lower wind speed, and its absence from the experimental trace is probably due to the low damping level of this test, fl = 0.0012; this again brings the wind speed range in which oscillation hysteresis would otherwise occur at that fl too close to Ur. The large number of cycles required for the oscillation to become stationary, about 300, justifies the assumptions of the nonlinear analysis. The other comparison, curves 2 in Fig. 25 corresponding to point 2 in Fig. 24, is made for a wind speed just beyond the range of oscillation hysteresis. The parameters of the theoretical curve used here are slightly different from the ones used for Fig. 24. Because of the sensitivity of conditions in this a-egion, they were adjusted so that the stationary amplitudes agreed and the experimental wind speed lay just beyond the knee of the theoretical ~ - U curve. Point 2 in Fig. 24 is shown in the position it would occupy if it, rather than the theoretical parameters, had been adjusted. The theoretical curve in Fig. 25 is seen to agree well with the observed trace, quite accurately showing the plateau arising from the residual effect of the lower limit cycle, and giving a good estimate of the build-up time, again very long in terms of cycles. The engineering importance of galloping arises from its occurrence in the natural wind, which is highly turbulent. The investigation of the effects of turbulence in the incident flow on galloping is therefore of interest, and a reasonable starting point for an investigation in the wind tunnel is the introduction of homogeneous isotropic turbulence into the flow incident on a bluff cylinder of constant cross section. The mean flow conditions therefore remain closely two-dimensional, and the theory of Section 3.1 can again be applied, under the condition that the experimental static variation of Cy with a used in the theory is measured in the same turbulent wind in which the galloping is observed. Laneville and Parkinson (1971) give results from such an investigation. It is found that the effect of turbulence integral scale is small, within the range tested, but that there is a consistent effect of turbulence intensity, such that increasing intensity makes a soft oscillator weaker, and even stable for a sufficient intensity, while a hard oscillator tends to become soft. In all cases, the qualitative galloping characteristics predicted by the theory from the Cy-acurves were verified by the actual galloping behaviour, and respectable quantitative agreement was also obtained for cylinders of square section and rectangular section with d/h = 2. These are both soft oscillators in smooth flow, and the paper shows theoretical curves and experimental points for turbulence intensities of 6.7%, 9.0%, and 12.0%. Both sections become progressively weaker oscillators with increasing intensity, and the rectangular section in fact becomes completely stable in turbulence of 12.0% intensity. All of the above behaviour is consistent with the remarks in Section 2.1.2 about the importance for galloping characteristics of shear layer reattachment on the afterbody sides, where increasing turbulence intensity is considered to cause reattachment on the windward side at a lower value of a for a given sectional shape. The remaining two-dimensional application of the quasi-steady theory considered here is to hydroelastic galloping of the square section. The experimental phenomena were considered in Section 2.3, because it was shown there that combined effects of wake vortex and galloping excitation could always be expected in hydroelastic oscillations, since Uo < Ur for realistic elastic system parameters, and the oscillations would therefore lie in Bearman's zone (i) (Fig. 12(a)). Nevertheless, for sufficientlylarge U the quasi-steady assumptions should give reasonable results even in zone (i). Accordingly Bouclin (1977) adapted the theory of Section 3.1 to hydroelastic galloping of the square section, and applied it for U > 2. Since the flow in the open channel in which Bouclin's experiments were conducted had low turbulence intensity, he used Smith's data and the polynomial approximation of Fig. 21 for Cy vs. a and therefore, as in the theory for the aeroelastic galloping, substituted Eq. (3.1) into Eq. (2.1) to produce

206

G. PARKINSON

the hydroelastic equivalent of Eq. (3.2). Here, however, mass parameter n was much larger than for the aeroelastic case, so that Eq. (3.2) could no longer be regarded as weakly nonlinear and the first approximation method of Krylov and Bogoliubov could no longer be used. Instead, Bouclin used a numerical method involving a fourth order Runge-Kutta algorithm (Conte and Boor, 1972), in which both amplitude Ys and frequency t) could be solved for. Another departure from the aeroelastic model was the inclusion of an appreciable fluid mass with the elastic system mass by taking the system mass and natural frequency to be constant at the values measured during free oscillations of the cylinders in otherwise still water. Sarpkaya and Isaacson (1981) have properly taken issue with this practice (which they find to be common practice in nonlinear oscillator modelling) on the grounds that any fluid effect should appear naturally from the fluid terms in the oscillator equation, and that there is no theoretical basis for assuming that the added mass should be the same during oscillations in flow with vortex shedding as in still water oscillations. Certainly one cannot quarrel with their first objection, which is physically correct. However, the objection is ignored and the practice commonly followed because it is convenient to have the still-water natural frequency as the system reference frequency, since it is easily measured and is approached as a limit (both theoretically and experimentally) for small flow velocities and cylinder displacements; i.e., tl ~ 1 for small U and Y. Further, their second objection about the lack of theoretical basis is not strictly correct. The exact solution of the linearized problem of an airfoil in plunging oscillation (hydrofoil in heaving oscillation) in a uniform flow involves a potential flow model with a semi-infinite vortex wake (Karman and Sears, 1938). Nevertheless the solution for the lift on the foil contains an isolated term arising from the foil surface boundary condition equal to the negative product of the foil acceleration and the still-fluid added mass. Moreover, the other term in the lift solution, which represents the effect of the foil circulation and the vortex wake, approaches the quasi-steady lift as the reduced frequency becomes small (U becomes large). Thus, the dynamical equation for the foil becomes exactly equivalent to Eq. (2.1) in Bouclin's quasi-steady model. Points from the numerical calculation of ]7 and f~ are included in Fig. 12. The theoretical s amplitudes were actually calculated as r.m.s, values for the nonharmonic waveforms of the strongly nonlinear oscillations, and then multiplied by x ~ for presentation as conventional effective peak amplitudes ~ . It can be seen that the Y~ values agree well with Bouclin's experimental data, but lie above the other hydroelastic data in Fig. 12(a), from Bokaian and Geoola (1983). This is probably because of the higher turbulence intensity of the flow in the latter experiments, which would produce Cy values giving smaller oscillation amplitudes. The agreement of the calculated t) values with the experimental data is less satisfactory, particularly for lower values of U. This is largely a consequence of the dependence of fl on the phase angle ~ between exciting force and displacement, not accounted for in the quasisteady theory where ~b = 90. Indeed, when the agreement between calculated and measured values offl is examined for all 3 cases that Bouclin dealt with, it is found to be fairly good at the highest values of U, where the quasi-steady theory is most applicable. It was~shown in Section 2.3.2 that for hydroelastic oscillations with U > 2 the system damping term in Eq. (2.1) could be neglected and the remaining terms transformed to the universal form of Eq. (2.7), implying that all amplitude-velocity data for a given sectional shape should be collapsable to a single curve independent of system parameters. This was shown in Fig. 13 to be satisfactorily verified for data from 5 different square-section cylinders. The figure also contains the curve calculated in the collapsed coordinates from the quasi-steady theory, and it is seen to represent well the 'single curve' mentioned above.

3.3. EXTENSIONTO THREE-DIMENSIONALPROBLEMS It was pointed out in the previous section that galloping is important in engineering because of its occurrence with elastic bodies exposed to the natural wind, and that a first step in studying galloping under natural wind conditions in the laboratory was to examine the

Flow-induced vibrationsof bluff bodies

207

effects on galloping of turbulence in the incident wind, still under essentially twodimensional conditions. Although this is helpful it is ultimately desirable to simulate directly in the laboratory the interaction between a real elastic body and the real wind, both threedimensional. Fortunately, if the body is long and of constant or gradually varying section the problem can be simplified since it is not necessary to completely abandon two-dimensional methods. The aerodynamicist can consider a strip theory, recalling the precedent in which the aerodynamic theory for a wing of high aspect ratio incorporates the two-dimensional theory for the airfoil sections along the span (Prandtl, 1921). By a similar method Sullivan (1977) has incorporated an adaptation of the two-dimensional theory of Section 3.1 into a quasi-steady theory for the galloping of towers of constant square section in the natural wind, and his method and results are considered in the following paragraphs. Novak (1972) had previously considered such problems, and his formulation allowed for the lateral deflection mode of the tower, and for the variation with height of the wind speed. However, he used a lateral force coefficient averaged over the height of the tower. Sullivan's method uses Novak's formulation but extends it to incorporate the variation with height z of the sectional lateral force coefficient Cy. He solves only for the limit cycles of the transverse oscillations of the tower models, and uses the energy balance requirement that the net energy input to the oscillation from the wind per cycle must equal the energy dissipated per cycle by the system damping. This leads to the equation 0= {Cy(Z, } ' ) n V 2 - 2 f l ~ ' } d Z d v , (3.22)

where Z = z/l, and I is the height of the tower. Other quantities are as defined for previous two-dimensional systems, except that now U, Cy, and Yare functions of Z. Equation (3.22) is the three-dimensional equivalent of Eq. (3.7) with its left side set equal to zero. The time-average wind speed variation with height above the ground U ( Z ) can be approximated accurately by a power law U(Z) = UtZ ~ (3.23)

where U~ is the value of U at the height of the tower and ~ is a constant. The transverse oscillatory displacement Y can again be assumed haxmonic in time and at one of the elastic system natural frequencies. Its spatial variation from zero at the ground to Yzat the top of the tower is assumed to be given by a mode shape determined by the elastic properties of the tower. In the experiments Sullivan conducted to test the theory he used rigid tower models elastically pivoted at their base, and in that case Y, ];" are given by Y(Z, "r) = ~ Z cos z, ];'(Z, r) = - ~ Z sin z, (3.24)

where ~ is the tip amplitude. Equations (3.24) correspond to Eqs (3.3) in the twodimensional analysis. The quasi-steady nature of the theory again enters through the assumption that Cy can be determined as a function of a from static tests on the same body in the same incident flow, and then used in the dynamic equations through the relation tan a = ~ , and a polynomial curve fit to the experimental Cy data. The difference here is that Cy varies with Z as well as a, so that sectional static measurements are made (from sectional pressure distributions in Sullivan's experiments). Sullivan's equivalent of the two-dimensional Eq. (3.1) is then

//]~r,2r-1

(]~')2S ]~

In addition to the dependence on Z, Eq. (3.25) differs from Eq. (3.1) by including even powers for greater flexibility in curve fitting and by including the sign ofeach term in the value ofAi.

208

G. PARKINSON

So that Eq. (3.22) can, like Eq. (3.7), be integrated by quadratures the variation of the Ai with Z is also curve-fitted. Sullivan approximated the experimental Cy(a, Z) relation by

Cy(a,Z)=

5 i=1

2 j=0

~ tanai ~

AuZJ,

(3.26)

using a 5 x 3 matrix of constants A~j, from the curve fitting. As in the two-dimensional analysis the result of integrating Eq. (3.22) gives a universal curve of amplitude vs. wind speed for all towers of the same section, aspect ratio, and mode shape exposed to winds of the same velocity profile, independent of size and mass and damping parameters. The curve is conveniently presented as ~(n/{3)vs. Ut(n/fl). Stability of the limit cycle oscillations is determined by a method similar to that used in the twodimensional analysis. Figure 26 shows the results of Sullivan's static tests and Fig. 27 the corresponding results of the dynamic tests and the calculation of the theoretical galloping behaviour of one of the tower models used. The experiments were carried out in a large boundary layer wind tunnel with a test section 2.44 m wide, 1.58 m high, and 24.4 m long. The thick turbulent boundary layer flow, intended to simulate the natural wind, was grown over a floor artificially roughened by metal angle sections, and at the test location, 20.7 m from the upstream end of the test section the boundary layer was about 0.71 m thick and in mean velocity profile, turbulence intensity profile, and spectral properties represented a typical suburban natural wind to a scale of 1 : 550. The tower model was 0.71 m high and of 5.1 cm square section. In Fig. 26, giving the variation of Cy with tan a for 5 values of Z, it can be seen that the variation changes dramatically with height, growing from negative (stable) values near the base to maximum positive (unstable) values at about three-quarter height, then falling off approaching the top. These results suggest that the use of an averaged lateral force coefficient (Novak, 1972) would lead to inaccurate predictions of galloping behaviour. Figure 27 displays the experimental galloping data and two theoretical curves for the tower in the universal form. The data are seen to collapse quite well onto the theoretical curve calculated from the sectional Cy data of Fig. 26 using Eq. (3.26). The other theoretical curve, using averaged Cy values, seriously underestimates the galloping amplitudes. The data with the highest value of fl/n demonstrates the oscillation hysteresis predicted by the theory, and appears to lie in the equivalent of Bearman's zone (iii), whereas the other three sets of data, with lower values of fl/n bringing the system closer to vortex resonance as

0.4

A N

0.3

C
0.2

v v o

0.~
o o
""O . o ~ o A o A O

0 n

[]

-01

3
-0.2

0
A

I
0.05

I
0.10

I
0.15 020

I
0.25

I
050

I
0.35

I
0.40

v EI

Ton a

FIG. 26. V a r i a t i o n of sectional force coefficient with height of section on square-section tower. D a t a from Sullivan (1977). Z: 0.089 = I I , 0 . 4 0 6 = E3, 0 . 6 9 2 = V, 0 . 8 3 5 = O , 0.973 = A.

Flow-induced vibrations of bluff bodies

209

a>~ 05

~ o3
o E '~ 0.2

O.t

&A

t, I
oDe ...,..,aw

o I
05

I
1.5

I
2

Utrnl~'

DimensionLess wind v e l o c i t y

(O t n/~)

Fit3.27. Universal plot of galloping amplitude vs. wind speed for square-section tower, from Sullivan (1977). Experiments for different ~8/n: O, I1, A, O: Theory: sectional forces 1, average forces 2.

indicated by the marks on the abscissa, depart from the theoretical prediction at low values of U~(n/fl) and appear to lie in the equivalent of Bearman's zone (ii). Generally, then, the quasi-steady theory of galloping is seen to give good predictions under these turbulent, three-dimensional conditions.

4. M O D E L L I N G VORTEX-INDUCED AND COMBINED VIBRATIONS


4.1. FLOW-FIELD MODELS

In modelling flow-induced vibrations the properties and characteristics of the elastic system can usually be taken as well known and understood, so that in systems considered to be governed by Eq. (2.1) the left side of the equation can be assumed to be an accurate simulation of the actual system (with perhaps some reservations about the damping term). Ideally one would like to achieve the same degree of simulation for the right side by solving for the function Cy from a flow-field model. This, after all, is what has been possible for about 50 years in airfoil or hydrofoil fluid-elastic problems for which Cy could be calculated by linearized potential-flow theory (Karman and Sears, 1938). Unfortunately, however, for bluff body oscillations the excitation is primarily from pressure loading on the afterbody, and complete analytic flow-field models such as those available for the unseparated flows past streamlined shapes do not exist for" the wake flows involved. However, in recent years the increasing power of computers has led to an enormous growth in the study of numerical flow-field modelling, now universally labelled by the acronym CFD, for computational fluid dynamics, and some success has been achieved in problems of, or relevant to, flow-induced vibration. A thorough discussion would require a complete paper and is beyond the scope of this article, but some features of recent work on CFD models will be described briefly in the following paragraphs. The first important class of models to appear has been that of two-dimensional discretevortex potential flow models, in which the separated shear layers from a bluff body are simulated by large numbers of potential point vortices. The flow develops as each vortex moves in time steps with a velocity determined by all other influences but its own. Complex variable methods are used, so that kinematic variables are the complex potential and complex velocity, related to the pressure by the unsteady Bernoulli equation, and forces are calculated from the unsteady Blasius equations. Some investigators have used conformal

210

G. PARKINSON

mapping to achieve simple field boundaries (Clements and Maull, 1975), whereas others have employed numerical surface-singularity methods to deal directly with the physical boundaries (Utsunomiya et al., 1988). Initially these models were more successful in predicting kinematic properties of the wake vortex systems, such as the Strouhal number, than in predicting dynamic effects on the bluff bodies. The difficulties arise from the empiricism needed to adjust the ideal potential flow to correspond more closely to the real separated flow. In particular, empirical mechanisms are introduced in the models to deal with the artificial production of each vortex, with the simulation of the effects of viscous dissipation in the real flow, and with the avoidance of vortex instabilities not representative of the actual shear layer behaviour. Further, the scale of computation is limited by simplifying the vortex systems as they get far enough downstream. As the models became more elaborate their simulation of dynamic effects improved. Sarpkaya and Shoaff (1979) devised a discrete-vortex model for the vortex-induced transverse oscillation of a circular cylinder which reproduces many of the characteristics observed experimentally, but not the oscillation hysteresis observed by Feng (1968), perhaps because the model's vortex production mechanism does not correspond to that observed by WiUiamson and Roshko (1988) and described in Section 2.2.2. Investigators are now beginning to apply discrete-vortex modelling to more difficult bluff body problems, and Utsunomiya et al. (1988) have studied flows around oscillating rectangular sections, including the square. They show calculated results for the variation of phase angle ~b with U for the square section that are in good agreement with experimental data such as that of Bearman and Luo (1988) in Fig. 20. Despite its required empiricism, discrete-vortex modelling has proved its value and it is worthy of further development and application. The ultimate flow-field model is of course a solution of the Navier-Stokes equations, and this has been the target of the majority of CFD studies. Until recently such studies have not been particularly useful for problems associated with flow-induced vibrations because of their limitation to relatively low Reynolds numbers. Now, however, it has become possible to compute flows in the Reynolds number range 10" to 105, the range of much of the experimental data and certainly high enough for realistic models of flow-induced vibration phenomena, so progress can be anticipated. It is of course simpler to compute numerical solutions for stationary bodies and such solutions could be useful in providing Cy vs. a data for the quasi-steady galloping theory of Section 3.1 or the combined-oscillation theory of Section 4.2.2. Tamura and Kuwahara (1988) have obtained numerical solutions for the aerodynamic behaviour of a cylinder of constant square section in a uniform flow at a Reynolds number of 10". They deal with the unsteady incompressible Navier-Stokes equations in both their two- and three-dimensional forms, and do not employ a turbulence model. Instead, they use a finite-difference method with a generalized curvilinear coordinate system and a third-order upwind scheme to achieve numerical stability at higher Reynolds numbers. They compute streamlines, pressure distributions and lift and drag forces at a = 0 and 15 and compare their two- and three-dimensional computations with each other and with experimental data from other laboratories. Their results relevant to flow-induced vibrations are that their values of lift and drag computed from the three-dimensional equations correspond to Cy values in good agreement with experimental data such as that of Fig. 21. This indicates that galloping characteristics could now be predicted completely without empirical inputs by using the quasi-steady theory of Section 3.1 supplemented by static Cy vs. a curves computed numerically from Navier-Stokes codes.

4.2. NONLINEAR WAKE OSCILLATOR MODELS

In this final section an alternative approach to solving Eq. (2.1) that has become popular in recent years is described, and applications are considered. Instead of attempting to determine Cy from a flow-field model, as in the previous section, it is assumed that the

Flow-induced vibrations of bluffbodies

211

physical characteristics of the bluff body flow are such that Cy can be regarded as a nonlinear oscillator representing wake vortex effects and governed by a second differential equation which must be solved along with Eq. (2.1). The method has been controversial because of the speculative nature of the second differential equation, but its support appears to be growing in the 1980s, including encouraging comments in the summaries of two specialists' meetings on bluff body flows, 'Euromech 119' (Bearman and Graham, 1980) and a 'Workshop on Bluff-Body Near-Wake Instabilities' (unpublished), organized in connection with the 1986 Annual Meeting of the American Physical Society, Division of Fluid Dynamics. 4.2.1. Vortex-induced Vibration Models Many different mathematical models of vortex-induced transverse vibration of bluff (usually circular) cylinders have been proposed. In this article only the class of nonlinear wake oscillator models will be discussed in detail. In this class solutions are sought for two coupled ordinary differential equations, one being Eq. (2.1) for cylinder displacement Y, )r+ 2ill;'+ Y = CynU 2, (2.1)

and the other a nonlinear equation for transverse force coefficient Cy. The first of these models, which spawned the subsequent work, was devised for the circular cylinder by Hartlen and Currie (1970). They attribute the concept to Birkhoff (Birkhoff and. Zarantanello, 1957), and their idea for employing it to the experiments of Bishop and Hassan (1964). Their version of the second equation is .. . n "3 Cy - AD~ Cy + ~ Cy + D~ Cy = Ol ?, (4.1)

where f~ = U/U, = 2rrSU = f~s/f, and A, B, D are empirical constants. The equation is of the general van der Pol type (Minorsky, 1962), and it was constructed to be as simple as possible while meeting the requirements of 3 situations: flow past a stationary cylinder ( I;" = 0), forced vibration of a cylinder ( l;" is a known harmonic function), and vortexinduced vibration of a cylinder ( I;" is controlled by Eq. (2.1)). This is ingenious and does much to answer the main criticism of this kind of modelling--the amount of empiricism in it. The critics point out that with enough empirical constants models can be developed to simulate most experimental phenomena and that there is as a consequence little to be learned from the process. This is unfair in the present case, since all of the empirical constants in Eq. (4.1) can be determined from experiments of a different class than those with whose results the values calculated from the model are to be compared. Thus, if the same circular cylinder is exposed to the same flow in all three situations listed above, data from experiments in the first two situations can be used to determine the empirical constants A, B, D, which can then be used to predict characteristics of vortexinduced vibration, the primary objective. Also, it should be pointed out that, inasmuch as the phenomena of vortex-induced vibration of a circular cylinder are well documented, it is not the prediction of phenomena but a contribution to the understanding of observed phenomena that is sought from the solutions to the mathematical model. The process of determining the system constants in Eq. (4.1) and then applying it and Eq. (2.1) to the solution of the problem of vortex-induced vibration begins as follows. If the cylinder in a flow is restrained from moving, so that the right side of Eq. (4.1) vanishes, the equation represents the oscillating pressure force on a stationary bluff cylinder. Elementary theory of weakly nonlinear equations, as in Section 3, then gives the result (which of course was designed into Eq. (4.1)) that Cy oscillates at the stationary-cylinder Strouhal frequency f,., and that the amplitude is given by C,o = x / ~ 3 n . (4.2) Experimental values of (S'y can then be used to determine the ratio A/B for insertion in the o analysis of the corresponding problem with the cylinder oscillating. This leaves two of the three constants, say B and D, to be determined.
JPAS 2 6 : 2 - G

212

G. PARKINSON

If the cylinder in a flow is oscillated mechanically, with a specified constant amplitude and frequency, only Eq. (4.1) is needed, since 1;"is known and Minorsky (1962) shows (p. 439) that when the impressed circular frequency coc is sufficiently near vortex frequency o~v the solution of Eq. (4.1) predicts the lock-in of coy to ~oc. Hartlen and Currie applied this solution to a comparison with experimental data by Jones (1968), with good qualitative results. However, Sarpkaya (1979) cites criticism of this comparison on the grounds that the experimental cylinder amplitudes were too low for lock-in to have occurred. The most difficult test of the theory of course is for vortex-induced oscillation of the cylinder, requiring a solution of Eqs (2.1) and (4.1). Hartlen and Currie concentrate their attention on steady-state oscillations in the lock-in range, and these will now be considered. Under these conditions it can be assumed that Y and Cy are given by Y = I?s sinflz, Cy = Cysin (Dz+O). (4.3)

where all quantities are as previously defined in this article. If Eqs (4.3) are substituted into Eq. (2.1) and the resulting coefficients of sin flz and cos ['lz are equated to zero to satisfy the equation, two new equations appear; (1 _ ~ 2 ) Ys = nU2Cys COS*, (4.4) (4.5)

2fillY s = nU 2Cys sin*,


and if Eq. (4.5) is divided by Eq. (4.4) the result is
1 -D

tan*-

2 2fl~"

(4.6)

Equations (4.1) and (4.3) to (4.6) permit direct solution for unknowns D, *, (Tys, Ys as functions of flow velocity U (or of [2s = 2~SU) and system parameters fl, n, S, (Tyo (or A), B, D. Before examining details of the solution it is worth noting that useful information about synchronized vortex-induced vibrations can be obtained simply by inspection of Eqs (4.4) and (4.5). Thus, for vortex-induced vibrations in air flow, n is 0(10)-3, U is 0(1) during lock-in, and Y, and Cys can be expected to be of the same order of magnitude. Therefore, it follows from Eq. (4.4) that D must be very close to unity. Then, from Eq. (4.5), ~ is proportional to 1/fl and, since flmin is 0(10) -3 and Cy, is 0(1), it follows that Ysmazis 0(1). Turning now to the solution of the equations for the Hartlen-Currie model, Eqs (4.3) are substituted into Eq. (4.1), and coefficients of sin flz and cos ~z are equated to zero to satisfy the equation. (In this method of harmonic balance the higher harmonics resulting from the expansion of the nonlinear trigonometric terms are discarded.) The two resulting equations are put in terms of Ys and [2 through the use of Eqs (4.4) and (4.5), and Ys is then eliminated from them to give a final equation for frequency ratio fl,

D
=

~/

nDflfl 2
1-- 2n2S2 [(1 _fl2)2 + 4f12f12] (4.7)

From Eq. (4.7) it can be seen that f] _< D~ if D > 0. With i) given by Eq. (4.7), Y~, Cys, and tp can be solved for from the other equations, and Hartlen and Currie did this using values of n, fl, (3yo, and S corresponding to the data of Ferguson (1965) for the circular cylinder. Parameters B and D were chosen to provide the best fit to the data. These theoretical solutions are shown in Fig. 28 and they can be seen to give good qualitative and fair quantitative agreement with the lower branches of Feng's curves in Fig. 4. (These lower branches represent values for steady-state oscillations reached from rest at each wind speed, as in Ferguson's measurements using the same cylinder and mounting.) However, there was clearly room for improvement. In particular, the oscillation hysteresis observed by Feng and shown in Fig. 4 was not reproduced even qualitatively. Hartlen and Currie commented on this, and noted that their model is in fact capable of generating two real solutions over a range of wind speeds, but at the time they did not think the issue was worth pursuing. Nevertheless the oscillation hysteresis is a real and interesting effect, and

Flow-induced vibrations of bluff bodies


ioo

213

4,"

5O

1.01

f
099

0,6
Cy=
0.4

Y%2
o

0.8

J
u

O9

I
I

FIG. 28. Predictions of f~, ~, Cy,, Y, vs. U for vortex-induced vibration of a circular cylinder during lock-in by the theory of Hartlen and Currie (1970) using experimental parameters of Ferguson (1965).

subsequent studies of nonlinear oscillator modelling have sought to reproduce it. There is indeed a second real stable solution of Eqs (2.1) and (4.1) over a range of wind speeds for certain values of the system parameters, but it is unfortunately not representative of the actual experimental behaviour, since the lock-in frequency obtained in the solution is close to the stationary-cylinder Strouhal frequency instead of the cylinder natural frequency, and the predicted cylinder amplitude is extremely small. It is in fact like a forced vibration at the Strouhal frequency away from resonance. Of the subsequent studies of nonlinear oscillator modelling of vortex-induced vibrations of the circular cylinder, most have been unsuccessful in predicting the oscillation hysteresis, and none has been completely successful. The approach has usually been to modify either Eq. (2.1) or Eq. (4.1), or both, in order to change their nonlinear characteristics. Some studies have pursued the idea that the hysteresis is caused by nonlinearities in the elastic system of Eq. (2.1), either in the springs (Currie et al., 1972) or in the system damping (Wood, 1976), and indeed some plausible results were obtained from the studies. However, as discussed in Section 2.2, the experimental evidence is overwhelming that the origin of the amplitude hysteresis is in the fluid mechanics and therefore, if the two-equation wake oscillator model is to predict oscillation hysteresis successfully, it must be mainly through changes in Eq. (4.1), or through changes in the assumed form of solution, or both. Oey et al. (1975) have considered changing the assumed form of solution of Eqs (2.1) and (4.1) by proposing a combination-oscillation model of the type discussed by Stoker (1950). In this model Cy is assumed to be the sum of two components, one oscillating at the cylinder frequency, representing lock-in, and the other oscillating approximately at the stationarycylinder Strouhal frequency. The model is capable of producing jumps between two stable oscillation amplitudes, and although Oey et al. did not produce quantitative results their paper attracted some attention, and Wood (1976) pursued the idea. He found that the model is useful for studying conditions at the ends of the lock-in range of flow velocities, but does not seem capable of producing two stable oscillations of different amplitude, each harmonic

214

G. PARKINSON

at a frequency very close to the system natural frequency, over a wide band offlow velocities in the middle of the lock-in range, the behaviour that Feng observed. The idea of modifying Eq. (4.1)to produce better resultshas of course occurred to many investigators, since the equation was speculative to begin with. Skop and Griffin (1973) introduced a term in Cy3 intoEq. (4.I),equivalent to a nonlinear 'spring'in the equation but, although thisgives more flexibility matching theory and experiment, itdoes not lead to a in reafisticmodel of oscillation hysteresis.Following a suggestion by Landl (1975), W o o d (1976) investigatedthe effectsof replacing the odd-power cubic damping terms of Eq. (4.I by ) an odd seventh-power sequence; i.e., terms in Cy, ~y3, ~ys, ~ . He was able to obtain a solution for l7, and for Cy,, each with two stable amplitudes separated by an unstable amplitude over the lock-in range of flow velocities, all at a frequency close to the elastic system natural frequency, and with values in the correct range. However, the detailed shapes of the amplitude curves as functions of U were unrealistic, and no corresponding phase jumps were predicted. This is a consequence of the nature of the equations, as Can be seen from Eqs (4.6) and (4.7) for the Hartlen-Currie model. Equation (4.7), which determines frequency fl, does not involve the damping terms of Eq. (4.1) through their parameters A, B. This remains true if higher-order damping terms are used, as by Wood, so the same curves of f~ vs. U and, through Eq. (4.6), of ~ vs. U are predicted as for the original Hartlen-Currie model. Wood, therefore, felt the improvement was only marginal, and proceeded to consider the other models mentioned above. The nonlinear oscillator models considered so far do not employ any flow-field theory in their derivation. Some investigators have attempted to put the modelling on a more physical basis by deriving the equations using semi-empirical flow-field theory. Two models of this type are by Iwan and Blevins (1974) and by Tamura and Matsui (1979). In the first model the equivalent of Eq. (4.1) is a differential equation, again of van der Pol type, in a 'hidden' flow variable defined so that its time derivative represents a weighted average of the transverse velocity within a two-dimensional control volume for which a momentum balance is sought. Parameters of the model again are fixed by matching data from circular-cylinder experiments. In the second model a different version of a concept used by Nakamura (1969) is employed. The concept is that of representing the effects of the vortex wake of a circular cylinder by a finite cavity attached to the cylinder and in torsional oscillation. In the equivalent of Eq. (4.1) for this model, the variable is angular position a, and the nonlinear term comes from an assumed oscillatory length of the cavity, which creates a nonlinear spring effect in the estimated aerodynamic restoring torque of the flow about the cylindercavity combination. Since the nonlinear term is linear in d and quadratic in a, the equation is again of the general van der Pol type. The two models are quite similar mathematically, and both give quite good simulations of experimental data, including some from Ferguson (1965) and Feng (1968), but neither predicts the observed oscillation hysteresis. The most successful improvements on the original Hartlen-Currie model have been made by Berger (1978, 1984(a), 1987). In the 1978 paper he presented the results of a more elaborate and more carefully planned modification of the damping term in Eq. (4.1),
C,+f*C_.,+D~C, = O]",

(4.8)

where f * is an even-power polynomial in Cy. Equations (2.1) and (4.8) are solved as before by assuming steady-state oscillations in the lock-in range of U (or f~) given by Eqs (4.3), and applying the method of harmonic balance. As in the original Hartlen-Currie model the unknowns are Y,, Cy,, [2, and 4, and fl and ~ are again given by Eqs (4.7) and (4.6) since the damping term in Eq. (4.8) is not involved, as was mentioned in connection with Wood's model. However, the equations for Cy, and Y, are now more flexible and Berger's achievement is to have recognized what is needed for a realistic simulation and moulded the equations accordingly. One of the relations resulting from the process of harmonic balance equates f * , a polynomial function of Cy to f * , a function off~ 2 known as a function of U z, (or D~) through Eq. (4,7). In the application of his model to circular-cylinder data from Feng (1968), Berger used fH* in the form

fr~ = a*(C.zy~-C~o)(C~-b*C-.6~ + c*C4y,-d*C2ys + e*).

(4.9)

Flow-inducedvibrations of bluffbodies

215

Equation (4.9) automatically satisfies the requirements for the existence of limit cycles, and for undamped oscillation of Cy with frequency D, and amplitude Cy0 when the cylinder is stationary ( ]~ = 0). Berger evaluated the positive constants a*-e* by forcing agreement at three points with the amplitudes and trends of Feng's data for the low-damping experiment of Fig. 4. The results are shown in Figs 29 and 30, in which Feng's data from Figs 4 and 5 are re-plotted. In Fig. 29, Berger's theoretical curves for [~ and Y, are seen to b in excellent agreement with Feng's data. The agreement for f~ is not surprising, since here the theory is unchanged from that of Hartlen-Currie. For Y, the degree of agreement is remarkable, even allowing for the amount of empirical fitting of the f * polynomial. The oscillation hysteresis is correctly modelled, both qualitatively and quantitatively, and the corresponding result for the high-damping case in Fig. 30 is even more impressive, since the model correctly predicts Y, with no oscillation hysteresis with only damping parameter ~ changed from the lowdamping case. Cy values are not plotted in Figs 29 and 30, but Berger's calculated values for the upper stable limit cycle arc appreciably lower than Feng's one measured value. Correspondingly, the calculated values of ~ are higher than the measured values for the high-amplitude limit cycle. The major failure of the model is its inability to predict the phase hysteresis observed by Feng. This of course, like the solution for f~, is not surprising since ~ is given by Eqs (4.6) and (4.7) as in the Hartlen-Currie model. However, the failure made Berger dissatisfied with the model to the extent that he felt 'the results are completely misleading' (Berger, 1984(b)), despite the success in reproducing most of the observed effects of the oscillation hysteresis. He therefore re-examined the governing equations, (2.1) and (4.8) and produced a modification of them (Berger, 1987). He noted that the linear coupling term in Eq. (4.8), D ]~, might be

I00"

1.5

5o- q~

o~

~o\.. f,"

VV V ~ ' ~ p ~ o ~ e e g t ~ q $

V V VV

,,o/

/.,"~'~s:o.198
0.6

05

04

v,
-0.2
Aa& & IA . ~
~6
0.8 u I J8 I. 2 &AAA

I.

FIo. 29. Comparisonof theoretical predictions(Berger, 1978)with experimental data (Feng, 1968)for vibration phenomena of circular cyfindersduring lock-in. Low damping case; Theory: stable , unstable...... ;.Expt.: = D,.~ = O,h = V, Y, = A, f, from ix~st= A.

216

G. PARKINSON

Jan than
f / /

iO0 o

1.5 -

I I

I" /"

/ /

-50"

-- Oo
-I';I"

I"
-<0.6

0..5

O4

%
O2

%.6

AI AA

0.8

I
U

1.2

FI(3.30. Comparison of theoretical predictions (Berger, 1978) with experimental data (Feng, 1968) for vibration phenomena of circular cylinders during lock-in. High damping case; Theory: ; Expt.: ~b = [],f~ = O,f~ = ~ ,

?,=~.

more suitably replaced for such a system by a nonlinear term, so that in the new model the constant D becomes a polynomial in ]72. More importantly he recalled that in Feng's experiments the elastic system damping parameter fl had been shown to be motiondependent (Wood, 1976), so in the new model the constant fl also becomes a polynomial in I~2 The important difference in the behaviour of the two models emerges from stability analysis, by methods similar to those of Section 3.1. In the model with constant D and fl no instability occurs near ~b = 90 or at higher values of 4, nearer 180 . Instability in this model is caused exclusively by a negative slope of the curve offH* VS. Cys given by Eq. (4.9). In the model with variable D and fl, limit cycles for which ~blies in the range q~~ 90 + 9 are found to be unstable, as arc limit cycles with ~ between 0 and 9 , and with ~b> 120. Thus the model predicts phase hysteresis, with stable limit cycles having phase angles either in the range 9< ~ < 81 , or in the range 99< ~ < 120. These predictions arc in quite good agreement with Feng's phase measurements, particularly for the more important ranges 81< q~< 99 and ~ ~>120 , where there are almost no data points. It is interesting that Feng observed only phase hysteresis in the high-damping experiment, which tends to confirm the relative independence of the two hysteresis mechanisms in Berger's oscillator model. This helps one to accept the premise that phase hysteresis can originate in the elastic system while the excitation and cylinder-amplitude hysteresis originate in the flow. Also, Berger points out (as mentioned in Section 2.2.1) that forced vibration experiments would not reveal effects such as the phase hysteresis which originate in the elastic system. However, this puts Berger's results at odds with the forced-vibration results of Fig. 11. In the papers cited Bvrger also investigated the modelling of vortex-induced oscillations of the D-section cylinder, again using Feng's results for experimental data inputs and

Flow-induced vibrations of bluff bodies

217

prediction comparisons. The main interest here is the shift to lower values of U of the lock-in range relative to the stationary-cylinder Strouhal line, and the resulting complications in the modelling. Results of the investigation (Berger, 1978) were promising. More details of the modelling for both the circular cylinder and the D-section are given in Berger (1984(a)). In summary, the methods of nonlinear wake oscillator modelling have been shown to be capable of reproducing through the solution of systems of ordinary differential equations, most of the phenomena of the vortex-induced vibration of bluff cylinders. The nature of these solutions is then an aid to better understanding of the phenomena. Most of the models discussed have represented only the range of flow velocities over which cylinder and vortex frequencies are synchronized. No model has provided an explanation for the interesting occurrence shown in Fig. 9, and observed in the experiments of Ferguson (1965) and Feng (1968)--the persistence of cylinder oscillation of considerable amplitude and with little modulation at wind speeds well beyond the upper end of the synchronization range, where the only significant excitation appears to be at the stationary-cylinder Strouhal frequency. Perhaps a form of combination-oscillation model, or possibly a model employing the concept of asynchronous excitation (Minorsky, 1962), would lead to useful results. In the next and final subsection the application of the method of wake oscillator modelling to combined effects of vortex-induced and galloping vibrations will be discussed. 4.2.2. Combined Vibration Models In Section 2.3 interesting phenomena involving combined effects on flow-induced vibrations of wake vortex resonance and galloping excitation were discussed. In attempting to model these phenomena mathematically the class of wake oscillator models is an obvious choice because of the successful versions described earlier for treating galloping and vortexinduced vibration separately. In his attempt to model these combined effects observed in his hydroelastic experiments on square-section cylinders described in Section 2.3.2, Bouclin (1977) employed a simple combination of the galloping model of Section 3.1 and the Hartlen-Currie model of the previous section, so that his governing equations are Y+2/~ ~"+ Y = nU 2 { Cy( ]"/ U ) + Cv }, and -(4.10)

{2

Cvo- 3 \f~ s

Cv+fl~C, = 0 8 ,

(4.11)

where Cy is taken to be the same function of (I~/U) as in the galloping theory of Section 3.1, also used in Bouclin's quasi-steady model described in Section 3.2. Cv is now the notation for the wake vortex excitation coefficient, whose stationary-cylinder amplitude becomes C,o, G is an empirical constant, and other quantities are as previously defined. Equation (4.11) is formally equivalent to Eq. (4.1) if A in Eq. (4.1) is eliminated using Eq. (4.2), except that G in Eq. (4.11) would then correspond to Bf~s and would be proportional to the flow velocity, whereas Bouclin takes G to be a constant. As discussed previously, constants G and D can be determined from the results of forcedvibration experiments, on a cylinder of the same section and in a flow of the same quality. In this procedure Eq. (4.10) is not involved, since I8 becomes a known harmonic function in Eq. (4.11), which is then a weakly nonlinear equation for C~ of the forced van der Pol type, capable of analytic solution by standard methods. (The strong nonlinearity for hydroelastic oscillations is in Eq. (4.10).) G and D appear as parameters in the solution functions and can be chosen to give good agreement between two predicted and measured effects. Equations (4.10) and (4.11) are then complete and ready for solution for flow-induced hydroelastic oscillations and for comparison of predicted and measured effects for those as yet untested cases. For the solution, because of the strong nonlinearity of Eq. (4.10), it is again necessary to employ numerical methods, and Bouclin used a system subroutine for differential equations on the IBM370. Since the C, function contains a number of frequencies, a Fast Fourier Transform method was used to determine them.

218

G. PARKINSON

In applying Eqs (4.10) and (4.11) for his wake-oscillator model to the square-section cylinder for comparison with his own hydroelastic experiments, Bouclin measured parameters co., n,/~, and S directly. As discussed in Section 3.2, natural frequency co, and n were determined from still-water free osdllation tests. For elastic system damping parameter 8, however, the inclusion of hydrodynamic effects was avoided by replacing the cylinder in freeoscillation decay tests by an equivalent streamlined bar giving the same system natural frequency. S was measured for each cylinder with the cylinder held stationary in the flow. A value of C,o was inferred from study of aerodynamic data from several laboratories, covering a wide range of Reynolds numbers, turbulence intensities, and square cylinder length aspect ratios. Considering the low Reynolds numbers, moderately low turbulence, and probable relatively poor spanwise correlation in Bouclin's experiments, a value C,o = 0.70 was considered suitable. G and D were determined by comparing the predictions of the forcedoscillation model described above with data from two sets of aerodynamic experiments on square-section cylinders forced harmonically (Otsuki et al., 1974; Nakamura and Mizota, 1975). Good determinations could be made because the model solutions displayed the same features as the experimental data, and the values used were G = 0.05, D = 1.75. Even better agreement would have been found with the data of Bearman and Obasaju (1982) had it been available at the time of Bouclin's calculations. With all parameters in Eqs (4.10) and (4.11) determined, calculations from the theoretical model were made for the three square-section cylinders used in the hydroelastic experiments. Results are plotted in Figs 12, 14 and 15. In Fig. 12 theoretical amplitudes Y, were actually calculated as r.m.s, values to agree with measured values for the nonharmonic waveforms of the strongly nonlinear hydroelastic oscillations. In addition to these calculated values, all measured r.m.s, values were then multiplied by x/~ for presentation as conventional effective peak amplitudes. In Fig. 12 it can be seen that the wake-oscillator model predicts values of 17 in good agreement with the comparable experimental values for U > 2, whereas predicted values of f/show a decreasing trend with increasing U not observed experimentally. For U ~<2, the theory correctly predicts the occurrence and frequency, but overestimates the amplitude, of the oscillation observed near U = 1/3 U~. It also predicts the observed start of the build up of the strong oscillations at Ui < Ur, the corresponding abrupt drop off/from 1.0 to f/, as the cylinder is forced to oscillate at f / = f/, = ~ , and the departure of f/ and f/,, still synchronized, from the Strouhal line, but it overestimates the value of U at which this occurs. In Fig. 14, the wake-oscillator theory is seen to predict values of Ui in excellent agreement with hydroelastic experimental values for four different square-section cylinders, and values of (1 - f / , ) giving the observed qualitative variation with [l/n, but quantitatively too low by about a factor of 2. In Fig. 15 the theory, in addition to its predictions of D~ synchronized with f/for U < 2, described above, correctly predicts relatively strong wake signals at f/and 3f/coexisting with signals at f/,. No signals at 2f/or 4f/were predicted. Generally, the wakeoscillator theory appears to give very good predictions of hydroelastic phenomena for the square-section cylinder. The greatest errors in its predictions are for cylinder frequency f / a t flow velocities U beyond the value at which the synchronized wake and cylinder frequencies depart from the Strouhal line. Here f/ is too high initially and at higher U eventually becomes too low. This appears to be a consequence of an inaccurate prediction of phase angle ~ between excitation and response. The results of two other investigations of the application of the methods of wakeoscillator modelling to the combined effects of wake-vortex excitation and galloping have been reported recently (Corless, 1987 or Corless and Parkinson, 1988; Tamura and Shimada, 1987). Again the vibrating bluff body is the two-dimensional cylinder of square section, but generally in these two studies the fluid is air although Corless does make one comparison of the results of his analytical method with the numerical calculations of Bouclin (1977) for hydroelastic vibrations in the resonance region. In both studies the two governing differential equations are mathematically very similar to Eqs (4.10) and (4.11), although in the model produced by Tamura and Shimada the variable in the equivalent of Eq. (4.11) is again the angular displacement a of an assumed

Flow-induced vibrationsof bluffbodies

219

wake cavity, as in the model for the circular cylinder by Tamura and Matsui (1979), previously described. When their model is applied to forced vibrations the equations are relatively simple to solve, and their predicted variation of the amplitude and phase of the exciting force with wind speed is in good agreement with several sets of experimental data (Otsuki et al., 1974; Ito et al., 1972; Ito et al., 1975). When their model is applied to flowinduced vibrations, Tamura and Shimada resort to numerical calculations by a Runge-Kutta method, and their predicted variation of cylinder amplitude with wind speed is in good qualitative agreement with experimental data by Wawzonek (1979). Corless, on the other hand, employs the recently fashionable analytic method of multiple time scales (Bender and Orzag, 1978; Kevorkian and Cole, 1981, Nayfeh and Mook, 1981) in solving the governing equations for the flow-induced vibrations. For the two governing equations Corless uses Eq, (4.10) unchanged, including the polynomial dependence of Cy on (l;'/U). His second equation, for Cv, is a slight modification of Hartlen and Currie's Eq. (4.1),
v - A [ I s C + ~ C4A ~ ~ "3 +[1,2 C, = D~'+E~',

(4.12)

incorporating Eq. (4.2) to eliminate B, and adding a term E Y. He also allows A, D and E to be functions of D~. In most of the applications, A is constant as in the Hartlen-Currie model, whereas D and E are taken to be proportional to 1/co and 1/o2, respectively, in order to produce good agreement at the quasi-steady limit, as U 0 / U r ~ oo. The method of multiple time scales, applied to the problem of solving Eqs (4.10) and (4.12), makes use of the fact that both are weakly nonlinear ordinary differential equations because of the small parameters n in Eq. (4.10) and A in Eq. (4.12). The method produces a solution that is asymptotic to the exact solution as n--* 0 and A--. 0. It has the advantages that stability analysis is automatic, a practical error estimate is available by taking the solution to enough terms, and the form of the solution is not assumed but emerges from the analysis. The time scales chosen are the fast time T and the slow times zk,
T = f~mt,

(4.13) k = 1, 2, 3, (4.14) (4.15)

Zk = #k T, where

D,m= 1 +/ztr 1 +#za 2 +#3a 3 + . . . .

and the detuning parameters ai are to be determined. The small parameter g is related to the small physical parameters n and A in Eqs (4.10) and (4.12) by taking # = ~ and defining g = AD~/g. This choice is consistent with the actual relative magnitudes of n and A, and leads to simpler calculations. Equations (4.10) and (4.12) now take the form + Y= g2(r 1~ - r3 ];" 3 + r5 ~ 5 _ r7 ~ 7 + sC v), (~v+n~Cv = #g(~, - #~,t~, + DI;'+ E Y, a where #g = At),, #~ = 4A/3C2of~s,
/.t2rl = h A l ( U - - U o ) ,

(4.16) (4.17)

#2s = nU 2, and #2r i = nU2-iAi for i = 3, 5, 7.

The solution will take the general form


Y = Yo(T, T1, T2, ~3, T4,)+#YI(T, zl, T2, T3)+#2y2(T, ~I, z2)+ . . . , Cv = Co(T, ~1, ~2, z3, z4,)+ #CI(T, ~1, 2, T3)+ . . . .

(4.18) (4.19)

In the method of solution Eqs (4.18) and (4.19) are substituted in Eqs (4.16) and (4.17), which as a result are replaced by a sequence of simpler equations at each order. (In fact only the time scales T, et, and z2 and equations up to order/~2 are needed in the present problem.) Equation (4.16) gives rise to the first 0(1) equation, the first 0(#) equation, and the first 0(# 2) equation. Equation (4.17) provides the corresponding second sequence equations. The 0(1) equations are solved first, then the 0(g) equations, then the 0(#2) equations.

220

G. PARK1NSON

Of any given order, the first equation is solved before the second, as the second depends upon the solution of the first, but not vice versa, so they are effectively uncoupled. Each equation after the 0(1) equations gives rise to an anti-secularity equation (the requirement to eliminate terms in which the time appears as a multiplying factor, such as t cos t) for determining the dependence of the previous terms on the higher-order time scales. The solution of Eqs (4.16) and (4.17) can be divided into several cases: (1) (2) (3) (4) f~s = f~s = ~s = f~s ~ I +/K~ 1, the case of vortex resonance, 3 +/~f~3, the case of subharmonic resonance, + / ~ 1 / 3 , the case of superharmonic resonance, 1, 3, 1/3, the case away from any significant resonance.

In fact there are more resonances, but none of any significance. The first result of the method of multiple scales is to give the form of the solution. In all cases, the forms of solution for Yo and Yt are Yo = I70 cos ~-~m El = ]71 COS~'~mt" t, In case 1, where ~s = 1 +#t~ 1 it turns out that 17o = 0 and the forms for Co and C1 are Co = Co cos (f~mt + ~o), C1 = C1 cos (llmt + W1) + ~ In case 2, 3, 4 the form for C O is Co = (70 COS(Osf~mt+W0) (f~2_ 1~ COSf~mt (D~--1~ sinD~t,

Co

sin (3tlmt + 3Wo).

E17o

O17o

and it can be seen that there is a component of the excitation near the Strouhal frequency ~ and another component near the body frequency of unity (in the normalized variables). The forms for C1 involve elaborate complex functions in these cases. The next results needed from the method are solutions for the amplitudes, phase angles, and detuning parameters, and finally the automatic stability analysis determines which solutions represent stable limit cycles. Complete details of the analysis, and comparisons with other methods tried, are given in Corless (1987). Figures 31 and 32 show results calculated by the method for cases 1, 4, and 2 in comparison with experimental data from Bearman et al., (1987). The calculated curves are solutions to 0(#) of the model equations, so the ordinates represent the amplitudes of ( Y0 +/~ Y1). These are presented as r.m.s, values for comparison with the experimental r.m.s. displacements. The abscissa variable is 1)s = 2rcSU. In Fig. 31 the agreement of calculated and measured values is seen to be quite good, particularly for the system with low damping, the most important test of the model and the method. One anomaly is the 'overshoot' of the calculated solution for case 1 where it joins the calculated solution for case 4. This is the consequence of the values of 171 becoming too large to meet the assumptions of an 0(#) calculation. It can be seen in Fig. 31 that the calculated curve for the system with medium damping (denoting the three damping levels represented in the figure as low, medium and high) has two branches, representing oscillation hysteresis and two stable limit cycles as in Fig. 24. The prediction is qualitatively correct, but the experimental apparatus used could not accommodate amplitudes high enough to demonstrate the upper limit cycle. This is also true of the high-damping system, for which even the calculated upper branch is beyond the limit of the figure. Figure 32 again shows the low-damping system data, now being compared with calculated curves using modified parameters A, D and E with different dependence on f~. In this calculation of the equations of case 4 and case 2 (the case 1 calculations near resonance are not affected) there is a useful result for case 2, with a definite 'kink' in the curve representing the effects of subharmonic resonance, that did not appear from the calculations of Fig. 31. This kink corresponds to the one in the experimental data at slightly lower values

Flow-induced vibrations of bluff bodies

221

OE

],,' I
/'
[
I~ I

0.80.7-

o;.i
a a

/ :

.I

0.6-05--

/o / /
I A

Y
04-0:5--

iIo
y I" I~ z

I" A

po o

1:7,"'"
I

(12-O.l~ 0

(i.
--~
~ 4

i, =
I
5 6 7 8
9

I
I0 II

FIC. 31. Comparison of theoretical predictions (Corless, 1987) with experimental data (Bearman et al., 1987) for aeroelastic vibrations of square-section cylinders in low-turbulence flow. Low dampin~ Theory , Expt. O; Medium damping: Theory . . . . . , Expt. [3; High damping: Theory-, Expt. A.

0.9 m

0.8--

0.7--

oO/
o ooo

0.6--

o.5
0.4

0.3

0.2

0.1

Fzo. 32. Comparisons of theoretical predictions (Corless, 1987) with experimental data (Bearman et al., 1987) for aeroelastic vibrations of square-section cylinders in low-turbulence flow, including subharmonic resonance. Low damping: Theory , Expt.O.

of O~, and represents an effect noted in both flow-induced and forced vibrations of squaresection cylinders, as discussed in Section 2. The overall agreement with the experimental data, however, is not as g o o d in Fig. 32 as in Fig. 31 because of the changed model parameters. Finally, Corless used the method of multiple scales to produce analytical results to compare with the numerical results of Bouclin (1977) for hydroelastic oscillations with much

222

G. PAP,KINSON

larger values of n than in the above aeroelastic studies. This has led him to estimate the relative advantages of the analytical and numerical methods at different values of n, and hc feels that the numerical method is preferable only when n = 0(0.I) or larger. For smaller n the method of multiple scales will provide cheaper results with adequate accuracy.

REFERENCES
BEARMAN,P. W. (1984) Vortex shedding from oscillating bluff bodies, Ann. Rev. Fluid Mech. 16, 195-222. BEARMAN,P. W. and CURRIE, I. G. (1979) Pressure-flUctuation measurements on an oscillating circular cylinder, J. Fluid Mech. 91, 661-677. BEARMAN,P. W., (3ARTSHORE,I. S., MAULL, D. J. and PARKINSON,G. V. (1987) Experiments on flow-induced vibration of a square-section cylinder, J. Fluids Struct. 1, 19-34. BEARMAN,P. W. and GRAHAM,J. M. R. (1979) Hydrodynamic forces on cylindrical bodies in oscillating flow. In: Proc. 2nd Int. Conf. B.O.S.S., London, 309-322. BEARMAN,P. W. and GRAHAM,J. M. R. (1980) Vortex shedding from bluff bodies in oscillatory flow: a report on Euromech 119, J. Fluid Mech. 99, 225-245. BEARMAN,P. W. and Luo, S. C. (1988) Investigation of the aerodynamic instability of a square-section cylinder by forced oscillation, J. Fluids Struct. 2, 161-176. BEARMAN,P. W. and OBASAJU,E. D. (1982) An experimental study of pressure fluctuations on fixed and oscillating square-section cylinders, J. Fluid Mech. 119, 297-321. BEARMAN,P. W. and TRUEMAN,D. M. (1971) An investigation of the flow around rectangular cylinders. Imp. Coll. Aero. Rep. 71-15. BENDER,C. M. and ORZAG, S. A. (1978) Advanced Mathematical Methods for Scientists and Engineers, McGrawHill, New York. BERGER, E. (1978) Some new aspects in fluid oscillator model theory. In: Proc. 3rd Coll. Ind. Aero. Aachen, 2. BERGER,E. (1984a) Zwei fundamentale aspekte wirbelerrcgter schwingungen. Herrmann Fottinger Inst. Be. 343/12, T.U. Berlin. BEROER, E. (1984b) Private communication. BEROEP,, E. (1987) O n a mechanism of vortex-excitedoscillationsof a cylinder.In: Proc. ?th Int. Conf. Wind Eng., Aachen. BIRKHOFF, (3. and ZARANTONELLO, E. H. (1957) Jets, Wakes and Cavities,Academic Press, N e w York. BISHOP, R. E. D. and HASSAN, A. Y. (1964)The lift drag forceson a circularcylinderoscillating a flowing fluid, and in Proc. R. Soc. London A277, 51-75. BLEVlNS, R. D. (1977) Flow-Induced Vibration,van Nostrand Reinhold, N e w York. BOKAIAN, A. R. and GEOOLA, F. (1983) O n the crossflow response of cylindricalstructures,Proc. Inst.Cir. Eng. 75, 397--418. BOKAIAN,A. R. and GEOOLA, F. (1984) Hydroelastic instabilities of square cylinders, J. Sound Vib. 92, 117-141. BOUCLIN,D. N. (1977) H ydroelastic oscillations of square cylinders. M.A.Sc. thesis, University of British Columbia. BP,OOKS,N. P. H. (1960) Experimental investigation of the aeroelastic instability of bluff cylinders. M.A.Sc. thesis, University of British Columbia. CLEMErCrS, R. R. and MAULL, D. J. (1975) The representation of sheets of vorticity by discrete vortices, Prog. Aerosp. Sci. 16, t29-146. CONTE, S.. D. and Boop,, C. (1972) Elementary Numerical Analysis, 2nd Edn, McGraw-Hill, New York. COOPEP,, K. R. and WARDLAW,R. L. (1971) Aeroelastic instabilities in wakes, In: Proc. 3rd Int. Conf. Wind Eft. Build Struct. Tokyo, 647-655. COP,LESS,R. M. (1987) Mathematical modelling of the combined effects of vortex-induced vibration and galloping. Ph.D. thesis, University of British Columbia. CORLESS,R. M. and PARKINSON,G. V. (1988) A model of the combined effects of vortex-induced oscillation and galloping, J. Fluids Struct. 2, 203-220. Cup,P,IE, I. G., HAP,TLEN,R. T. and MAP,TIN,W. W. (1972) The response of circular cylinders to vortex shedding. In: Syrup. Flow-ind. Struct. Vibr., Karlsruhe, 128-142. CURP,m, I. G. and TUP,NBULL,D. H. (1987) Streamwise oscillations of cylinders near the critical Reynolds number, J. Fluids Struct. 1, 185-196. DAVENPORT,A. (3. (1961) The application of statistical concepts to the wind loading of structures, Proc. Inst. Cir. Eng. 19, 449-472. DEN HAP, TOG, J. P. (1930) Transmission line vibration due to sleet, Trans. A.I.E.E., 49, 444. DEN HARTOG,J. P. (1934) The vibration problems in engineering, Proc. 4th Int. Cong. Appl. Mech., Cambridge, 36-53. DUP,BIN,P. A. and HUNT,J. C. R: (1979) Fluctuating surface pressures on bluffstructures in turbulent winds: further theory and comparison with experiment. In: Proc. 5th Int. Conf. Wind Eng., Fort Collins, 491-507. DUP,OIN,W. W., MAP,ell, P. A. and LEFEBVRE,P. J. (1980) Lower mode response of circular cylinders in cross-flow, J. Fluids Eng., 102, 183-190. ERICSSON, L. E. (1984) Maximum crossflow response of a circular cylinder, a non-resonant flow phenomenon. A.I.A.A. Paper 84--022. FENG, C. C. (1968) The measurement of vortex-induced effects in flow past stationary and oscillating circular and D-section cylinders. M.A.Sc. thesis, University of British Columbia. FEP,OUSON,N. (1965) The measurement of wake and surface effects in the subcritical flow past a circular cylinder at r~st and in vortex-excited oscillation. M.A.Sc. thesis, University of British Columbia.

Flow-induced vibrations of bluff bodies

223

FERGUSON,N. and PARKINSON,G. V. (1967) Surface and wake flow phenomena of the vortex.excited oscillation of a circular cylinder. J. Eng. Ind. 89, 831-838. GARRICK, I. E. (1976) Aeroelasticity-frontiers and beyond, J. Aircr. 13, 641-657. GARTSBORE, L S. (1973) The effects of free stream turbulence on the drag of rectangular two-dimensional prisms. U. West Ont. B.L.W.T. Rep. 4-73. GERRARD, J. H. (1966) The mechanics of the formation region of vortices behind bluff bodies, J. Fluid Mech. 25, 401-413. GLAUERX, H. (1919) The rotation of an aerofoii about a fixed axis. R. and M. 595. GRIEEIN, O. M. and RAMBERG, S. E. (1976) Vortex shedding from a cylinder vibrating in line with an incident uniform flow, J. Fluid Mech. 75, 257-271. HARTLEN, R. T. and CURRIE, I. G. (1970) Lift oscillator model of vortex-induced vibration, J. Eng. Mech. EMS, 557-591. HILLIER, R. and CHERRY, N. J. (1980) The effects of stream turbulence on separation bubbles. In: Proc. 4th Coll. Ind. Aero., Aachen, 145-157. HOERNER, S. F. (1965) Fluid-dynamic drag. Published by author. ITO, M., MIYATA, T. and FWISAWA, N. (1975) The characteristics of the aerodynamic forces acting on a square cylinder in vibration. In: Proc. 30th Ann. Meet. J.S.C.E., 1-208. ITO, M., MIYATA,T. and MORIMITO, Y. (1972) Measurement of aerodynamic forces on a cylinder of square crosssection. In: Proc. 2nd Syrup. Wind Eft. Struet. Japan, 159-166. IWAW, W. D. and BLEVINS, R. D. (1974) A model for vortex-induced oscillation of structures, J. App. Mech. 41, 581-586. VOW KARMAN, T. (1911) Uber den mechanismus des widerstands, den ein bewegter korper in einer flussigkeit erfahrt. Gott. Nachr. Math. Phys. K1. 509-519. Also, (1912) 547-556. Vow KAR~AN, T. and SEARS, W. R. (1938) Airfoil theory for non-uniform motion, J. Aero. Sci. 5, 379-390. KEVORKIAN, J. and COLE, J. D. (1981) Perturbation Methods in Applied Mathematics, Springer-Verlag, New York. Kiwc, R., PROSSER, M. J. and JoHws, D. J. (1973) On vortex excitation of model piles in water, J. Sound Vib. 29, 169-188. KOOeMANW, G. H. (1967) The vortex wakes of vibrating cylinders at low Reynolds numbers, J. Fluid Mech. 2& 501-512. LANDL, R. (1975) A mathematical model for vortex-excited vibrations of bluff bodies, J. Sound Vib. 42, 219-234. LANEVILLE, A. (1973) Effects of turbulence on wind-induced vibrations of bluff cylinders. Ph.D. thesis, University of British Columbia. LAWEVXLLE,A. and PARKINSOW,G. V. (1971) Effects of turbulence on galloping of bluff cylinders. In: Proc. 3rd Int. Conf. Wind Eft. Build Struct., Tokyo, 787-797. MAULL, D. J. and MILHWER, M. G. (1978) Sinusoidal flow past a circular cylinder. Coastal Eng. 2, 149-168. MCCROSKEV, W. J. (1977) Some current research in unsteady fluid dynamics, J. Fluids Eng. 99, 8-38. MEIER-WINDHORST, A. (1939) Flutter oscillations of a cylinder in uniform liquid flow, Mitt. HydE. Inst. Tech. Hochsch. Munich, 9, 3-39. MIWORSKV, N. (1962) Nonlinear Oscillations, van Nostrand, New York. MORKOVlW, M. V. (1964) Flow around circular cylinders--a kaleidoscope of challenging fluid phenomena. In: A.S.M.E. Syrup. Fully Sep. Flows, 102-118. NAKAMURA, Y. and MIZOTA, T. (1975) Unsteady lifts and wakes of oscillating rectangular prisms. Proc. A.S.C.E, J. Eng. Mech. EM6, 855-871. NAKAMURA, Y. and TOMONARI, Y. (1977) Galloping of rectangular prisms in a smooth and in a turbulent flow, J. Sound Vib. 52, 233-241. NAUDASCHER, E. (1987) Flow-induced streamwise vibrations of structures, J. Fluids Struct. 1, 265-298. NAVFEH, A. H. and MGOK, D. T. (1981) Nonlinear Oscillations, Wiley-Interscienee, New York. NOVAK, M. (1969) Aeroelastic galloping of prismatic bodies. Proc. A.S.C.E., EMI, 115-142. NOVAK, M. and DAVENPORT,A. G. (1970) Aeroelastic instability of prisms in turbulent flow. Pro. A.S.C.E., J. Eno. Mech. EMIl, 17-39. NOVAK, M. (1972) Galloping oscillations of prismatic structures. Proc. A.S.C.E., EMI, 27-46. NOVAK, M. (1974) Galloping oscillations of prisms in smooth and turbulent flows. In: Syrup. Flow-Ind. Struct. Vibr., Karlsruhe, 769-774. OBASAJU,E. D. (1983) Forced-vibration study of the aeroelastic instability of a square-section cylinder near vortex resonance, J. Wind. Eno. Ind. Aero. 12, 313-327. O~EY,H. L., CURR1E, I. G. and LEUTHEUSSER, H. J. (1975) On the double-amplitude response of circular cylinders excited by vortex shedding. In: Proc. 4th Int. Conf. Wind Eft. Build. Struct., London, 233-240. ONGOREW, A. and ROCKWELL, D. (1988) Flow structure from an oscillating cylinder. Part I: mechanisms of phase shift and recovery in the near wake, J. Fluid Mech. 191, 197-224. OTSUK1, Y., WASHIZU,K., TOM1ZAWA,H. and OHVA, A. (1974) A note on the aeroelastic instability of a prismatic bar with square section, J. Sound Vib. 34, 233-248. PAIDOUSSlS, M. P. (1979) Flow-induced vibrations in nuclear reactors and heat exchangers. Practical experience and state of knowledge, In: Proc. Syrup. Pract. Exp. Flow-lnd. Vibr. Karlsruhe, 1-38. PAI'DOUSSIS, M. P., PRICE, S. J., NAKAMURA, T., MARK, B. and NJUKI, W. (1989) Flow-induced vibrations and instabilities in a rotated-square cylinder array in cross-flow, J. Fluids Struct. 3. PARRINSON, G. V. (1974) Mathematical models of flow-induced vibrations of bluff bodies. In: Syrup, Flow-induced Struct. Vibr., Karlsruhe, 81-127. PARKIWSOr~,G. V. and BROOKS, N. P. H. (1961) On the aeroelastic instability of bluff cylinders, J. App. Mech. 28, 252-258. PARKlWSON, G. V. and SANTOSHAM,T. V. (1967) Cylinders of rectangular section as aeroelastic nonlinear oscillators. In: A.S.M.E. Vibr. Conf., Boston, Paper 67-VIBR-50. PARKINSON, G. V. and SMITH,J. D. (1964) The square prism as an aeroelastic nonlinear oscillator. Q. J. Mech. Appl. Math. 17, 225-239.

224

G. PARKINSON

PRANDTL, L. (1921) Applications of modern hydrodynamics to aeronautics. NACA Rep. 116. PRICE, S. J. and PIPERN1, P. (1988) An investigation of the effect of mechanical damping to alleviate wake-induced flutter of overhead power conductors, J. Fluids Struct. 2, 53-71. SANTOSHAM,T. V. (unpublished) Notes on water tunnel tests on cylinders of square and rectangular section. SARPKAVA,T. (1976) Vortex shedding and resistance in harmonic flow about smooth and rough circular cylinders at high Reynolds numbers. T.R. NPS-59SL76021, Naval Postgrad. Sch., Monterey, SARPKAVA,T. (1979) Vortex-induced oscillations. A selective review, J. Appl. Mech. 46, 241-258. SARPKAYA,T. (1985) Past progress and outstanding problems in time-dependent flows about ocean structures. In: Prec. Int. Syrup. Sep Flow Marine Struct. Trondheim, 1-36. SARPKAVA,T. and ISAACSON,M. (1981) Mechanics of Wave Forces on Offshore Structures, van Nostrand Reinhold, New York. SARPKAYA,T. and SHOAFF, R. L. (1979) A discrete vortex analysis of flow about stationary and transversely oscillating circular cylinders. T.R. NPS-69SL79011, Naval Postgrad. Sch. Monterey. SCANLAN,R. H. (1979) On the state of stability considerations for suspended-span bridges under wind. In: Prec. Syrup. Pract. Exp. Flow-Ind. Vibr. Karlsruhe, 595-618. SCRUTON,C. (1960) The use of wind tunnels in industrial aerodynamic research. N.P.L., Aero. Rep. 411. SCRUTON,C. (1965) On the wind-excited oscillations of stacks, towers and masts. In: Prec. Syrup. Wind Eft. Build Struct., N.P.L., Teddington, 798-832. SIMPSON, A. (1971) On the flutter of a smooth circular cylinder in a wake, Aero. Q., 22, 25-41. SISTO, F. (1953) Stall flutter in cascades, J. Aero Sci. 20, 598-604. SKOP, R. A. and GRIFFIN,O. M. (1973) A model for the vortex-excited resonant response of bluffcylinders, J. Sound Vibr. 27, 225-233. SLATER, J. E. (1969) Avroelastic instability of a structural angle section. Ph.D. thesis, University of British Columbia. SMIRNOV, L. P. and P^VLmlNA, M. A. (1957) Vortical traces for flow around vibrating cylinders. In: I.M.A. Communications Meet., Moscow. SMITH, J. D. (1962) An experimental study of the aeroelastic instability of rectangular cylinders, M.A.Sc. thesis, University of British Columbia. STAUSLI,T. (1983) Calculation of the vibration of an elastically mounted cylinder using experimental data from forced oscillation, J. Fluids Eng. 105, 225-229. STOKER, J. J. (1950) Nonlinear Vibrations. Interscience, New York. SULLIVAN,P. P. (1977) Aeroelastic galloping of tall structures in simulated winds. M.A.Sc. thesis, University of British Columbia. SURRY,D. and STATHOPOULOS,T. (1977) An experimental approach to the economical measurement of spatiallyaveraged wind loads, J. Ind. Aero 2, 385-397. TAMURA,T. and KUWAHARA,K. (1988) Numerical study of aerodynamic behavior of a square cylinder, J. Wind Eng. (Jpn) 37, 261-270. TAMURA,Y. and MATSUI,G. (1979) Wake-oscillator model of vortex-induced oscillation of circular cylinder. In: Prec. 5th Int. Conf. Wind Eng., Fort Collins, 2, 1085-1094. TAMURA,Y. and SHIMADA,K. (1987) A mathematical model for the transverse oscillations of square cylinders, In: Prec. Int. Conf. Flow-Ind. Vibr. Windermere, 267-275. TOEBES,G. H. and EAGLESON,P. S. (1961) Hydroelastic vibrations of flat plates related to trailing edge geometry, J. Basic Eng. 83, 671-678. TsuI, Y. T. (1977) On wake-induced flutter of a circular cylinder in the wake of another, J. App. Mech. 44, 194-200. UTSUNOMIVA,H., NAOAO, F. and UENOVAM^, H. (1988) Study of flows around rectangular cylinders by finite vortex sheets, J. Wind Eng. ( Jpn) 37, 271-280. WASHIZU,K., OHVA,A., OTSUK1,Y. and FUJI1,K. (1978) Aeroelastic instability of rectangular cylinders in a heaving mode, J. Sound Vib. 59, 195-210. WAWZONEK,M. A. (1979) Aeroelastic behaviour of square prisms in uniform flow. M.A.Sc. thesis, University of British Columbia. WILL1AMSON,C. H. K. and ROSHKO,A. (1988) Vortex formation in the wake of an oscillating cylinder, J. Fluids Struct. 2, 355-381. WOOD, K. N. (1976) Coupled-oscillator models for vortex-induced oscillation of a circular cylinder. M.A.Sc. thesis, University of British Columbia. ZDRAVKOVICH,M. M. (1982) Modification of vortex shedding in the synchronization range, J. Fluids Eng. 104, 513-517.

Das könnte Ihnen auch gefallen