Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Reprinted from May 1988, Vol.

110, Journal of Pressure Vessel Technology

L. F. Waring
Research Assistant.

D. S. Weaver
Professor and Chairman. Department of Mechanical Engineering, McMaster University, Hamilton, Ontario, Canada

Partial Admission Effects on the Stability of a Heat Exchanger Tube Array


An experimental study is reported of the effects of partial admission on the fluid elastic stability of a heat exchanger tube array. The array geometry was a parallel triangular configuration with a pitch ratio of 1.47. Tests were conducted in a wind tunnel with uniform flow over from 33 to 100 percent of single span tubes. In these experiments, the flow location was also varied from center-span to the end supports. Additionally, tests were conducted with uniform flow over one span of two and three-span tube arrays. These results are compared with theoretical predictions.

Introduction Heat exchanger tube arrays in cross flow are subjected to a variety of vibration excitation mechanisms [1]. Of these, fluid elastic excitation has received the most attention as this is the mechanism most likely to cause short-term failure of the tubes. The most commonly used approach for designing against fluidelastic instability is based on a semi-empirical equation developed by Connors [2]

y /( md\ 1/2 (1) pcPJ where V/fd is the dimensionless "reduced" velocity at which instability occurs, md/pd2 is the mass-damping parameter, and K is an empirical "stability constant'' which depends on tube array pattern and pitch. In many heat exchanger configurations, the flow velocity distribution across the tubes may not be uniform as is assumed in equation (1). In order to account for a nonuniform velocity distribution, V'=V)p(x)t the tube vibration mode shape, <j>(x), must be taken into consideration. Several authors have developed theoretical modifications of Connors' equation to include these effects [3-5]. While the flow velocity definitions and specific forms of the equations presented by these researchers vary somewhat, they all may be shown to have the form V _/m6\"* fd (x)dx (2)

providing that the tube mass per unit length, m> is constant. Connors [4] stated that this "effective velocity concept" had been verified experimentally, although no details were given. On the other hand, Franklin and Soper [3] reported exContributed by the Pressure Vessels and Piping Division for publication in the Manuscript received by the PVP Division, January 14, 1988.
J O U R N A L OF P R E S S U R E V E S S E L T E C H N O L O G Y .

periments using cantilevered tube arrays with flow blockage at various locations. They found that the theory was 21 percent unconservative when one third of the span measured from the cantilever tip was blocked off. In contrast, the theory was found to be conservative when 50 percent blockage was applied at the root. Franklin and Soper concluded that, while the theory is elementary, the predicted change in critical velocity is in the correct direction and the error should not be too great as long as the blockage is modest. Other than the few tests discussed in the foregoing, no systematic experimental evaluation of equation (2) has been reported in the open literature. At the same time, there is now substantial evidence to suggest that Connors' equation (1) may not be capable of representing the stability threshold of tube arrays over the whole range of the mass-damping parameter. Chen [6] has presented a thorough review of the available experimental data and has shown that the exponent of the massdamping parameter depends on tube array geometry as well as the value of the mass-damping parameter itself. In particular, this exponent is generally different from the value of 0.5 specified by Connors' theory. As this same theoretical approach is used to develop the expression for nonuniform flow distribution given by equation (2), it is reasonable to ask whether the exponent of the flow distribution and mode shape factor should be equal to 0.5. The purpose of the research reported in this paper was to systematically determine the effects of partial admission on a tube array. The tests were conducted in a wind tunnel varying both percent admission and flow location on the tube span. Experiments were also conducted with flow over one span of two and three-span tubes. The results are compared with the theoretical predictions of equation (2). Experimental Facility and Instrumentation The experiments were conducted in a wind tunnel with an octagonal test section which is 0.616 m across the flats. The
Transactions of the ASME

194/Vol. 110, MAY 1988

wind tunnel has screens and flow straighteners downstream of the octagonal bell mouth followed by a 4:1 octagonal contraction. This produces a flat, low turbulence intensity, velocity profile in the test section. The test section itself was specially designed for these experiments with sides fabricated from 19-mm-thick acrylic plate. The tube bundle was a 7-row-deep parallel triangular array of 12.7-mm-dia tubes with a pitch ratio of 1.47. There were 17 centrally located tubes which were flexibly mounted. These were surrounded by relatively rigid acrylic rods to fill out the test section, including half-tubes at the test section sides. Figures 1(a) and (b) show photographs of the test section. The flexibly mounted tubes had brass support plates which were machined such that the tubes had effectively circular knife edge supports. These support plates could be moved axially so that the tube span between supports could be varied. Thus, the flexibly mounted tubes were exposed to a uniform flow velocity across a length equal to the width of the test section. Fixing the supports at various locations outside of the test section permitted both varying the percent of a tube span exposed to flow and, for a given tube length, varying the location of the flow along the length. Figure 1(a) shows the centrally located flexibly mounted tubes and the tube support plate for the case of 100 percent admission, i.e., a tube length equal to the width of the test section. Figure 1(b) shows a case where the tube length exceeded the test section width. In such cases, a long flexible corrugated pipe surrounded the tubes outside of the test section walls and sealed the test section so that air could not be drawn into the wind tunnel along the tubes. Note also the cruciform splitter downstream of the tube bundle which was used to prevent the strong acoustic resonance observed at some flow velocities. In all experiments, the tubes were mounted vertically and supported axially by a centrally located pin at the lower end so as to approximate a support with zero rotational stiffness. The original set of tubes used were acrylic with an outside diameter of 12.7 mm and a wall thickness of 1.6 mm. However, in some of the tests with large spans, the static deflection due to aerodynamic drag was significant and the validity of the results was questionable. Therefore, tests were also conducted with stainless steel tubes with an outside diameter of 12.7 mm and a wall thickness of 0.25 mm. The dynamic response of both sets of tubes was obtained using a nylon carrier with orthogonally mounted miniature accelerometers. This carrier fitted snugly inside the tubes and could be located at tube midspan to maximize the output signals. Frequency analysis was conducted using a Spectral Dynamics 375 Fourier Analyzer. Flow velocity, Vg1 was measured using calibrated miniature hot wire probes placed in the minimum gap between fixed tubes in the array. This location was chosen in order to minimize uncertainty in the velocity measurements as the upstream velocity, Vu, was rather low in some of the experiments. For higher flow velocities, the upstream velocity was measured using both a pitot-static probe and a hot wire probe. The agreement found was excellent and, for the velocity range tested, the gap velocity was found to be a linear funcNomenclature a = length of tube exposed to uniform flow = 0.616 m d = tube diameter = 12.7 mm ft = tube natural frequency in /th mode K - stability constant L = total tube length m = tube mass per unit length
Journal of Pressure Vessel Technology

(b)

tion of upstream velocity, Vg = 3.44 Vu. This corresponds to a Reynolds number range based on gap velocity and tube diameter of approximately 3 x 103 < Re < 1.1 x 104.

Fig. 1

Photographs of wind tunnel test section with tube bundles

n = exponent of mass-damping parameter PAF = partial admission factor S = distance between tube supports (span length) u = distance from center of tube span to outside edge of flow field

b = logarithmic decrement of damping *,<*) = mode shape function of /th mode Hx) = flow distribution function p = fluid density MAY 1988, Vol. 110/195

Table 1 Datum case data Mm


m/pd2 b (K M ) cr (m/s) Acrylic 347 31.9 0.24 3.26

Stainless steel 378 92.0 0.07 6.83

PERCENT ADMISSION

a. L

1 0 Q

FLOW LOCATION

INL/2

2_2u ~a
L

Experimental Procedure The original length of the flexibly mounted tubes used in the experiments was L848 m. Since the test section width was 0.616 m, this tube length permitted the study of a single-span tube with 33 percent admission or the study of a tube with multiple spans, the flow over one of them. When tests were completed with this tube length, the tubes were shortened by a predetermined amount and another set of experiments run. This process was repeated until the tubes had a length equal to the test section width. This was taken as the datum case as there was 100 percent admission, i.e., uniform velocity across the entire tube length. For each tube length and support configuration, the experimental procedure was the same. At some low flow velocity, several minutes were permitted for a steady-state response to be achieved and measurements were then taken of flow velocity and tube response amplitude and frequency. Qualitative observations were also made of tube response both visually in the bundle and on an oscilloscope screen. The flow velocity was then incremented and the process repeated until tube-to-tube clashing was clearly evident. This was usually preceded by tube rattling in the support plates as some small clearance was necessary to permit alteration of the span length between supports. The overall tube response amplitude was obtained at each flow velocity by computing the square root of the sum of squares of the two accelerometer output signals. A plot of this overall tube response against flow velocity was used to obtain the stability threshold. In most experiments, the onset of instability was sudden and, therefore, the stability threshold well defined. However, in some cases the stability threshold was not the same for all tubes in the bundle. This is undoubtedly due to the slightly loose and imperfect tube supports as well as manufacturing tolerances on the tubes and supports. At any rate, the response of several tubes in the third, fourth and fifth rows was monitored and the experiments for each tube length repeated up to ten times in order to establish the uncertainty in the stability threshold. Experimental Results and Discussion Datum Case and Partial Admission Factor. When the entire tube length is subjected to a uniform flow velocity, the velocity distribution function \p (x) is equal to unity and equation (2) reduces to equation (1). This may be taken as the datum case and used to obtain the partial admission factor, PAF, for the other configurations. Thus, for uniform flow over a tube span Ll<x<L2t the partial admission factor in the /th mode is given by (PAF),=
L fd

CROSS FLOW VELOCITY, V

>

Fig. 2 Schematic of tube with definitions of percent partial admission and flow location on tube span

_L 1

2,00
< Q.

A C O
O < 1,50 in

is
<

<
CL

1.00 0.05 PERCENT ADMISSION x 100 L 100

Fig, 3 Comparison of theory and experiment for flow at mid-span on single-span tubes

}
J D a t u m case J

Jo 4jbc)dx
i

(3)

_f2 <j>Ux)dx assuming that the damping in the L various configurations does not change substantially. This was found to be true for the single-span experiments with the average logarithmic decrement of damping for acrylic and steel tubes equal to 0.24 and 196/Vol. 110, MAY 1988

0.07, respectively. The experimental partial admission factor is then found by dividing the reduced velocity stability threshold for a given configuration by that for the datum case. The theoretical PAF is found by computing the integrals of the right-hand side of equation (3). The data for the datum cases are given in Table 1. Mid-Span Partial Admission. A series of experiments was conducted using various lengths of tubes with their supports located such that the flow was centered on the span. The definitions for percent admission and flow location are illustrated schematically in Fig. 2. The wind tunnel test section height is denoted by a, the span length by L, Thus the percent partial admission is given by 100 (a/L). The tube bundle lengths varied from 1.846 m down to 0.616 m and, since the wind tunnel test section height was 0.616 m, the partial admission varied from 33 to 100 percent. In the experiments discussed in this section, the flow was centered on the span, and hence, u = a/2. The results of these experiments are summarized in Fig. 3. The bars on the experimental points represent the 95 -percent confidence limits computed from 5 to 10 repetitions of the experiments. For the case of 100 percent admission, the partial admission factor is equal to unity by definition and this represents the datum case. The experimental datum case
Transactions of the ASME

LL <

s
(a) ! _ (b) 4 0 % ADMISSION

a.

cc O O <

I ..L
f N=1

(b,ia=i (c,Ll
f N = 2

f N=3

f N = 3 3

,d!Li

1 o <
<
CL

2.0 1.0 2.0

Fig. 5 Schematic of tube showing various tests on multiple-span tubes of equal span lengths

r P
1

$
n = 0.5

t
I

1
m

!
J c ) 44%

1
ADMISSION

" 1
o

T -

;
n = 0.5

n=0.2

,1

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 FLOW LOCATION, (2u-a)/L
Fig. 4 Comparison of theory and experiment for the effects of flow location on single-span tubes

reduced threshold velocities used for subsequent analysis were the average of ten repetitions with the maximum variation being 4 percent for the case of the acrylic tubes and 8 percent for the stainless steel tubes. It is thought that the variation was greater for the latter because of the relatively greater proportion of the damping for the steel tubes being due to rattling at the supports. It is noted that the 95-percent confidence limits for the 33 and 40-percent admission experiments are large. These tubes were quite long and the steady flow drag forces produced noticeable tube deflection. Thus, the flexible tube array geometry exposed to the flow is modified, expecially in relation to the surrounding, relatively rigid tube array. This probably accounts for the variation in threshold velocity observed for different tubes in the array and for a given tube from experiment to experiment. These variations were substantially smaller for shorter tubes (admission >40 percent). Also noteworthy in Fig. 3 is that the data for 40, 44 and 67 percent admission is well above the trend of the other data. At the same time, the results for the 44 and 67-percent admission cases were quite repeatable as shown by the small 95-percent confidence bands. It is felt that this scatter is due to the loose tube supports. The tube support plates were drilled and slightly loose so that they could be slid along the tubes to adjust the span length. The solid lines are theoretical predictions, the upper line based on Connors' equation with the theoretically predicted exponent n~0.5 (equation (3)). As noted from the experiments of Weaver and El Kashlan [7] this exponent may be appreciably less than 0.5 for parallel triangular arrays; and, for the sake of comparison, a theoretical curve for n-0.2 is also shown. While there is substantial scatter in the experimental data, it is clear that the theoretical prediction with an exponent of 0.5 gives the correct trend and is conservative over the range of partial admission tested. This result agrees with the findings of Franklin and Soper [3]. The theory is conservative when the partial admission occurs over that part of the tube span which has the largest deflection.
Journal of Pressure Vessel Technology

Partial Admission Location Effects. A series of experiments were conducted for different percent admissions in which the location of the flow was varied along the span. Referring again to Fig. 2, the flow location is defined by the distance from the center of the span to the center of the flow field, w a/2. Normalizing this with respect to half the span length, L/2, gives a flow location parameter which varies from zero when the flow is mid-span to 1 - a/L when the flow extends to one of the tube supports. The results of these experiments are summarized in Fig. 4. Again the error bands on the experimental points represent 95-percent confidence limits and theoretical lines are given for exponents in equation (3) of = 0.5 and /i = 0.2. As in the experiments discussed in the foregoing, the scatter in the data is thought to be due to tube support looseness and, perhaps, static deflection. However, two trends are clear. Firstly, the theory seems to overpredict the importance of flow location. In all three cases, the increase in partial admission factor as the flow location progresses from mid-span to tube support is less than theoretically predicted. Secondly, the importance of the theoretical overprediction of flow location effect increases with decreasing percent admission. Thus, the theoretical prediction is always conservative for a 44-percent admission while it is unconservative for 33 percent admission when the flow location parameter exceeds about 0.4, i.e., as the flow location approaches the tube support. Again, this agrees with the experiments on cantilevers by Franklin and Soper [3] who reported a 21-percent unconservative prediction for 67 percent admission applied at the cantilever root. It is to be noted that the theoretical prediction with an exponent n-Q.2 better fits the trend of the experimental data and is always conservative. Multiple Span Experiments. Several tests were conducted with tubes on multiple supports such that there were two or three equal length spans. The configurations tested and definitions used are shown in Fig. 5. Figure 5(a) gives the single-span datum case, while 5(b) shows the two-span case. When there are 3 spans, there are two different spans over which the flow may be located as shown in Figs. 5(c) and (d). It is easy to show that for these multiple, equal span length cases, equation (3) reduces to (4) (~) where N is equal to the number of equal span lengths and it is assumed that the critical mode corresponds to the lowest natural frequency. In single-span experiments, this is a reasonable assumption. However, for multiple spans the mode which becomes unstable first (the lowest critical flow velocity) may be a higher mode, and each mode will have its own partial admission factor, (PAF),.
V fd / D a t u m case

(PAF) N =

(-)

MAY 1988, Vol. 110/197

triangular tube array with a pitch ratio of 1.47. The experiments studied the effects of varying percent admission with the flow centered on the tube span, the effects of varying flow location along the span for various percent admissions and the effects of multiple spans. While there was some scatter in the results, some general conclusions can be drawn. 1 For percent admissions greater than 30 percent on singlespan, simply supported tubes, the theory is conservative as long as the flow is centered on the span, i.e., where the tube deflection is greatest. 2 The theory overpredicts the effects of flow location on single-span, simply supported tubes. This is especially true for Fig. 6 Comparison of theory and experiment for PAF of muitiple-span smaller percent admission cases where the theory is unconsertubes of equal span length. Data points and theoretical predictions corvative when the flow is located towards the tube supports. respond to instability in the first mode. 3 The theory does not appear to work well for multiple span tubes with flow over only one span. In the 2 and 3-span cases tested, the theory was substantially unconservative. In M(x)dx 17#JNU.... one case, the theory failed to predict the correct mode of instability. (PAF), = N=l9 29 . 4 Based on these experiments, the theory would appear to be a rather rough approximation to be used when the percent exceeds 40 percent and is directed over the tube (5) admission mid-span. More systematic experiments are required to proIn this case, a conservative error is introduced as the damping perly evaluate the theoretical predictions for the effects of is expected to increase with the number of spans. multiple spans and nonuniform flow velocity distribution. The experimental results for cases (b) and (c) are shown in Fig. 6. In both cases, the mode excited corresponded to the The data in Chen's paper [6] shows that the stability lowest natural frequency. It is seen that the theory with an ex- behavior of parallel triangular arrays deviates most from Conponent of 0.5 is substantially unconservative while the nors' equation. Thus, it is reasonable to assume that these theoretical prediction with an exponent of n = 0.2 fits the ex- conclusions are applicable to other tube array geometries and perimental data very well. parameter ranges. Unfortunately, there is insufficient data at Case 5(d), in which the flow was over the end span of a present to provide quantitative design guidance for those cases 3-span tube, was not plotted in Fig. 6 because the critical mode in which the theory was found to be unconservative. was found to correspond to the second natural frequency. It is seen from equation (5) that the critical velocity for mode /, Vh is proportional to the product of the natural frequency of Acknowledgment The authors gratefully acknowledge the financial support of mode i,fi9 and the partial admission factor of mode i, (PAF),. the Natural Sciences and Engineering Research Council of Thus, Canada. (6) For case 5(d), the natural frequencies and theoretical (PAF), are given by References 0 5 1 Paidoussis, M. P., " A Review of Flow-Induced Vibrations in Reactors and 1st mode fl(PAF)l = 32.5 x(3) -56.3 (7) Reactor Components," Nuclear Engineering and Design, Vol. 74(1), 1983, pp. 2nd mode / 2 (PAF) 2 = 41.5x(2.03) 05 = 59.1 (8) 31-60. 2 Connors, H. J., "Fluidelastic Vibration of Tube Arrays Excited by Cross While these products are nearly the same for both modes, one Flow," ASME Symposium on Flow-Induced Vibration in Heat Exchangers, York, 1970, pp. 42-56. would expect instability in the first mode as the product New 3 Franklin, R. E., and Soper, B. M. H., "An Investigation of Fluidelastic Inft (PAF), is lower in this case. However, the experiments stabilities in Tube Banks Subjected to Fluid Crossflow," HTFS Report always showed instability in the second mode. Furthermore, AERE-R8708, HTFS RS 192, AERE Harwell and National Engineering the experimental partial admission factor computed according Laboratory, April 1977. 4 Connors, H. J., "Fluidelastic Vibration of Heat Exchanger Tube Arrays," to equation (5) was found to be (PAF) 2 = 1.040.03. Thus, ASME Journal of Mechanical Design, Vol. 100, 1978, p. 347-353. the theory not only failed to identify the critical mode, but is 5 Pettigrew, M. J., Sylvestre, Y., and Campagna, A. O., "Vibration also very unconservative. Analylsis of Heat Exchanger and Steam Generator Designs," Nuclear Engineering and Design, Vol. 48, 1978, pp. 97-115. Conclusions 6 Chen, S. S., "Guidelines for the Instability Flow Velocity of Tube Arrays Journal of Sound and Vibration, Vol. 93(3), 1984, pp. 439-455. A systematic set of experiments has been conducted in a in Cross-flow," 7 Weaver, D. S., and El Kashlan, M., "The Effect of Damping and Mass wind tunnel to examine the effects of partial admission on the Ratio on the Stability of a Tube Bank," Journal of Sound and Vibration, Vol. critical flow velocity for fluidelastic instability of a parallel 76, 1981, pp. 283-294.
NUMBER O F SPANS,

HA
^f\d)

D a t u m case

198/Vol. 110, MAY 1988

Transactions of the ASME

You might also like