Journal of Vibration and Acoustics

Technical Briefs
where is uid density and c the speed of sound. The scaling laws can be derived using dimensional analysis which takes into account the connection between geometrical parameters, the characteristic ow velocity and the turbulence scale. Various experimental data have partly proved and partly rejected the proposed scaling laws. Experimental results by Prek and Novak 7 show that two geometrical parameters have an inuence on the sound producing mechanism. Following this idea, the appropriate functional form of the governing equations can be determined statistically using multiple nonlinear regression analysis.

Scaling Laws of Hydrodynamic Noise Generation for a Simple Fluid Valve Model
Matjaz Prek
Faculty of Mechanical Engineering, University of Ljubljana, Askerceva 6, SI-1000 Ljubljana, Slovenia

Noise-source mechanisms for the simple uid valve model were observed and compared with predicted mechanisms. A comparison of the predicted mechanisms with the experimental data indicated that the scaling laws could be further improved by a reexamination of the inuence of the impact parameters. This was done by an extended analysis of the inuence of geometrical parameters on acoustical characteristics. Additional parameters were determined statistically using nonlinear multiple regression analysis of the experimental data. The estimated correlation between the observed and predicted values for the applied geometrical and hydrodynamic parameters are good. S0739-3717 00 01303-9

Experimental Procedure and Results

A clear acoustic model is needed in order to understand the relevant processes taking place in the valve. The valve model presented a ow control element with three important geometrical parameters: valve stem length a, inside tubing diameter d and valve opening s. Water was used as the uid medium and the valve model operated under nonwhistling and noncavitating conditions. A Bruel & Kjaer type 8103 miniature hydrophone was used to measure uid-borne sound pressure. The output signals from the hydrophone were routed through a signal-conditioning preamplier Bruel & Kjaer type 2635 , further through an FFT analyzer Bruel & Kjaer type 2032 , and then converted for advanced data processing using appropriate PC software 8 . The experimental results were examined statistically using nonlinear multiple regression to determine the joint effect of a and d, together with ow velocity v , on noise generation. For this purpose the sound pressure is dened in accordance with scaling p laws as follows: v X1 a X2 d X3 p (2)

Introduction
The governing equations describing the hydroacoustical processes in uid throttling devices were obtained by Fuchs 1 , continuing the previous work of Fuchs and Michalke 2 . A dimensional investigation of the principal equations, as is shown by Fuchs 3,4 , was used to analyze the turbulent excitation of a uid mass. The inuence of the turbulence structure is dened by a proportional factor in the form of a characteristic length, as suggested by Fuchs 5 . Different operating conditions and the inuence of geometrical parameters on noise are described by variations in characteristic length. A comprehensive dimensional analysis with derived scaling laws is given by Fuchs 6 . In this dimensional analysis only a monopole sound generating mechanism is considered. The acoustic generation mechanism refers to the propulsion of mechanical energy from the ow. Therefore, a dimensional investigation of the principal equations which were introduced by Fuchs 1 may help to explain the impact of different parameters on noise formation. Equation 1 is derived in accordance with this analysis; the turbulence excitation of the uid mass leads to the resulting uid-borne sound power P. P is proportional to a scale of characteristic ow velocity v and turbulence scale l at the narrowest contraction area such that: P
v4 l c
2

In Eq. 2 , the exponents Xi represent the impact of the observed parameter on noise generation. Since the dimensional analysis is based on the characteristic ow velocity, the measurement results are normalized with respect to the ow velocity in the narrowest cross-section. As was pointed out in a previous parametrical study, the sound pressure can be dened using the following expanded equation in order to include the combined inuence of different parameters: v X1 a X2 d X3 v p
a/d X4

(3)

where the last term includes the weighted inuence of the combined geometrical parameter. By regression analysis of the experimental data, the following equation was obtained and describes the sound pressure level for applied geometrical and hydrodynamical parameters as: S PL 71,1 51,7 log v 0,61 a/d log v 65,2 log d 28,1 log a (4)

(1)

Contributed by the Technical Committee on Vibration and Sound for publication in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received Sept. 1999; revised Feb. 2000. Associate Technical Editor: R. L. Clark.

In addition, the last term in Eq. 4 incorporates a new impact factor connected with valve stem geometry. This factor denes how the turbulence-generated noise source is inuenced by valve geometry and causes the transition between the monopole and Transactions of the ASME

330 Vol. 122, JULY 2000

Copyright 2000 by ASME

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

rameter was introduced, in this case s/d. The expected variation of the characteristic turbulence length was explained by this parameter. No pronounced inuence on the noise-generating mechanism was observed for the nearly closed valve, in contrast to measurements with a wide open valve. 4 An interesting consequence of the measurement results is the inuence of the valve diameter on the noise-generating mechanism. With a wide open valve, when d s, the valve diameter also inuenced the noise source. This happened in a connection with valve stem length described as the combined geometrical parameter a/d and denes the transition between the monopole and dipole type of noise-generating mechanism.

Fig. 1 Observed versus predicted values of the sound pressure level in dB re 1 Pa for small valve opening s d

References
1 Fuchs, H. V., 1983, Generation and Control of Noise in Water Supply Installations: Part 1: Fundamental Aspects, Applied Acoustics, 16, pp. 325 346. 2 Fuchs, H. V., and Michalke, A., 1973, Introduction to Aerodynamic Noise Theory, Progress in Aerospace Sciences, 14, pp. 227297. 3 Fuchs, H. V., 1985, Sound generation in liquid uid valves, Fraunhofer Institut for Bauphysik, IBP Report BS 118, Stuttgart, Germany. 4 Fuchs, H. V., 1985, Sound generation in liquid uid valves, Fraunhofer Institut for Bauphysik, IBP Report BS 119, Stuttgart, Germany. 5 Fuchs, H. V., 1972, Space correlations of the uctuating pressure in subsonic turbulence jets, Journal of Sound and Vibration, 1, pp. 7799. 6 Fuchs, H. V., 1993, Generation and Control of Noise in Water Supply Installations: Part 2: Sound Source Mechanisms, Applied Acoustics, 38, pp. 5985. 7 Prek, M., and Novak, P., 1999, Untersuchungen von Armaturengeraeuschen mit Ventilmodell und direkten Messungen des Wasserschalls, Bauphysik, 21, pp. 7783. 8 Prek, M., and Novak, P., 1998, Sound Generation Measurements in Hydraulic Networks, Proceedings, 3rd International SITHOK Congress, Maribor, Slovenija, pp. 309318.

dipole type of noise source. As the valve stem is kept the same or lengthened and the valves diameter is narrowed, the turbulence eddies becomes affected by a solid wall. This transition of the driving mechanism from the uctuating masses to the uctuating surface pressure is dened as a valve geometry-dependent parameter, with a weighted impact on the characteristic uid velocity. A second set of experiments were done with a nearly closed valve. Contrary to the previous measurements, where the turbulence scale varied between l a and l d, here it varies between l s and l d. The two parameters used were the valve opening and valve stem inside diameter. For analysis, the proposed sound pressure was dened as: v X1 d X2 s X3 p d s
X4

(5)

where the parameter (d/s) was used for transferring the characteristic turbulence length between l d and l s. Figure 1 shows the results of the multiple nonlinear regression on the experimental data. Equation 5 gives the functional dependency of the sound pressure level: S PL 104,2 31,5 log v 23,4 log d s 63,7 log d 22,3 log s (6)

Minimizing the Truncation Error in Assumed Modes Models of Structures

S. O. Reza Moheimani
Senior Lecturer, Department of Electrical and Computer Engineering, University of Newcastle, NSW 2308, Australia

and explains 97 percent of the observed variance. In Eqs. 5 and 6 , the inuence of the valve stem length is omitted since the results showed that it has no pronounced inuence either on the characteristic length or on the noise-generating mechanism. This was clearly evident when both sets of measurements were done with a wide open valve (s d) and an almost closed valve (s d). Under these two limiting geometrical conditions, the cleareast insight into noise generation is achieved.

Robert L. Clark
Professor, Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC 27708-0300

Summary
1 Parametrical analysis enables a verication of scaling laws using the experimental data. Variation of a single geometrical parameter can be made in connection with ow characteristics or in connection with other geometrical parameters. 2 The experimental results support ideas about the combined inuence of geometrical parameters on noise generation. Since, from the experiments, it was easy to detect the impact of the length and diameter on sound generation, a new measure, i.e., the ratio a/d, was implemented in order to better explain this dependency. The appropriate functional form of the equation required an additional geometrical parameter, which was determined statistically using multiple nonlinear regression analysis. 3 A scaling law was experimentally derived for a nearly closed valve. As for the wide open valve, an additional geometrical paJournal of Vibration and Acoustics The assumed modes approach is a widely used technique in modeling of distributed systems. Such models often consist of a large number of modes. For controller design purposes these models are simplied by truncating the modes that lie out of the bandwidth of interest. Truncation can alter zeros of the system. This paper presents a method of minimizing the truncation error by adding a second order term to the truncated model. This extra term is determined such that the in-bandwidth error is minimized in an optimal H2 sense. The technique is extended to multivariable systems. S0739-3717 00 02003-1
Contributed by the Technical Committee on Vibration and Sound for publication in the JOURNAL OF VIBRATION AND ACOUSTICS. Manuscript received March 1999; revised March 2000. Associate Technical Editor: B. Yang.

JULY 2000, Vol. 122 331

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Introduction

The assumed modes approach has been used extensively throughout the literature to model dynamics of distributed systems. Such systems include, but are not limited to, exible beams and plates 1 , slewing beams 2 , piezoelectric laminate beams 3 and acoustic ducts 4 . Dynamics of each one of these systems is described by a particular partial differential equation. In the assumed modes approach, the solutions of these partial differential equations are practically based upon a nite set of terms in the expansion; however one is always faced with the tradeoff between model order number of terms and model delity convergence of the solution . In control design problems, one is often only interested in designing a controller for a particular frequency range. In these situations, it is common practice to remove the modes which correspond to frequencies that lie out of the bandwidth of interest and only keep the modes which directly contribute to the low frequency dynamics of the system. This model is then used to design a controller. The performance predicted by such models and control system designs typically exceeds the practical performance which can be achieved in the laboratory. This is mainly due to the fact that although the poles of the truncated system are at the correct frequencies, the zeros can be far away from where they should be. Therefore, it is natural to expect signicant differences between predicted performance based upon truncated models and achieved performance based upon experimental systems since the closed loop performance of the system is largely dictated by the open loop zeros. In a recent paper, the second author demonstrated the effect of out of bandwidth zeros on the low frequency zeros of the truncated model 5 . He also suggested that the effect of higher frequency modes on the low frequency dynamics of the system can be captured by adding a zero frequency term to the truncated model to account for the compliance of the ignored modes. In 6 , the rst author showed that the effect of truncated modes on the low-frequency dynamics of the structure can be captured in an optimal way. In particular, 6 shows how a feed-through term that minimizes the H2 norm of the error system can be found. Moreover, the same procedure is extended to multivariable structural models. In this paper, we extend the works of 5 and 6 to allow for the effect of truncated modes to be captured by a second order term. Moreover, we will attempt to determine the resonant frequency and the gain of this term in a way that the H2 norm of the error system is minimized.

is that at lower frequencies, one can ignore the effect of dynamical response of higher order modes since they are small in comparison with the force responses. Although an approximation, reference 5 shows that K is a good representation of the effect of higher order modes on G N (s). Indeed, it can be shown that this choice of K brings the error to zero at 0. In reference 6 , the rst author showed that K can be chosen in a way that the H2 norm of the error system, i.e., G(s) G N (s) is minimized in the bandwidth of interest. It was also shown that this choice of K could result in a higher error at 0. However, the error at higher frequencies within the bandwidth of interest is lower with this choice of K. In this paper we choose to replace the constant feed-through term with a second order resonant term whose resonance frequency lies out of the bandwidth of interest. In other words, we try to capture the effect of out of bandwidth modes on the inbandwidth dynamics of the truncated model using a second order 2 system such as: M (s) K/s 2 where c and c is the highest frequency of interest. Hence, we approximate G(s) by G s GN s M s . (2)

Our objective is to choose K and function is minimized, e K,

2 2

such that the following cost G s W s

2 2.

G s
2

(3)

Here f (s) 1/2 f ( j ) d where f (s) is a rational function. Moreover, G and G are dened as in 1 and 2 and W(s) is an ideal low-pass weighting function with its cut-off frequency c chosen to lie within the interval c ( N , N 1 ). That is, W( j ) 1 for c c and zero elsewhere. To this end, it should be clear that K and chosen to minimize 3 , will minimize the effect of out of bandwidth dynamics of G(s) on G (s) in an H2 optimal sense. It is easy to see that 3 is equivalent to e K,
i N 1

Fi s2
2 i

K s2
2

W s
2

(4)

The fact that W is chosen to be an ideal low-pass lter with its cut-off frequency lower than the rst out-of-bandwidth pole of G, guarantees that 4 will remain nite. It is straightforward to show that 4 is equivalent to e K, 1 2
c

Fi
c

2 2

Problem Statement
Let us consider the transfer function of a exible structure: G s
i 1

i N 1

2 i

2K
2 2

Fi
i N 1 2 i 2 2

K2
2 2

Fi s2

d . (5)

2. i

(1)

In a typical control design scenario, the designer is often interested only in a particular bandwidth. Therefore, an approximate model of the system is needed that best represents the dynamics of the system in a particular frequency range. A natural choice in this case is to simply ignore the modes which correspond to the frequencies that lie out of the bandwidth of interest. For instance, if N is equivalent or larger than the highest frequency of interest, N 2 one may choose to approximate G(s) by G N (s) i 1 F i /s 2 i . A drawback of this approach is that the ignored higher order modes may contribute to the low frequency dynamics in the form of distorting zero locations. In reference 5 , the second author suggested a way of dealing with this problem. The idea that was put forward in 5 is to allow for a constant feed-through term in G N (s) to account for the compliance of omitted higher order modes of 1 . That is, to approximate G(s) by G (s) G N (s) 2 F i / N . The logic behind this choice of K K where K i N 1 332 Vol. 122, JULY 2000

The problem that we face, then, is to minimize e(K, ) subject to the constraint c . Since the rst part of e(K, ) is independent of K and , this optimization problem is equivalent to nding K and that minimize the following cost function. K, e 1 2
c

K2
2
c

2K
2 2 2 2 i N 1 2 i

Fi
2

d . (6)

Therefore, we intend to solve the following optimization problem. min

K R,

K, e
c

(7)

It is straight-forward, but tedious to nd an analytic expression for . e Transactions of the ASME

Copyright 2000 by ASME

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

 K, e

1 2

K2
2 2

c 2 c

K2 ln 2 3 1 ln
c c

N c c

2K 1
i i i c c

G s
i 1

1 s
2 2 i

Hi

1 s2
2

(11)

Fi
i N 1 2 i 2

ln

. (8)

The optimization problem dened by 7 and 8 is a nonconvex optimization problem. There are a large number of optimization methods that can be employed to nd a minimum of (K, ). e However, any such minimum could be just a local minimum of . e Moreover, to use any optimization routine, we have to rst truncate the innite series term in 8 . This can be done by truncating the series at a point which corresponds to a mode that is located at a frequency which is considerably higher than c . This will not cause a major difculty since as i , then i too. It can be shown that in this case the very high order terms of the series will approach zero. At this stage, we point out that if is xed, then will be a e convex function of K. Therefore, for a xed , the optimum K can be found to be: Fi
i N 1

where K is a matrix of pure gains which has similar dimensions as H i s. To be consistent with the SISC case, we will determine K and such that this following cost function is minimized E(K, ) W(s)(G(s) G (s)) 2 . 2 For a multivariable transfer function F(s), F(s) 2 2 1/2 tr F * ( j )F( j ) d . Moreover, tr M is the trace of the square matrix M. We choose W(s) to be a diagonal matrix, whose diagonal elements are ideal low-pass lters, i.e., W diag(w1 ,w2 , . . . ,w N ) and w i is dened by w i ( j ) 1 for c w c and zero elsewhere. Now, the cost function E(K, ) can be re-written as: W s
i N 1

1 s2

2 Hi i

1 s2

2 2K 2

(12)

This is equivalent to: E K, 1 2

c

1
2 c

2 i

ln 1
2 c

c c

1
i

ln
c c

i i

c c

(9)
i N 1 j N 1

1
2 i 2 2 j 2

ln

tr H i H j

Hence, to avoid the constrained optimization problem 7 , one could x at a frequency higher that c , and then, determine K from 9 . However, this approach may not lead to a low error. Therefore, in general, it is recommended that the optimization problem 7 be solved directly using the available numerical techniques.

1
2 2 2

tr K K

2
2 2

1
i N 1 2 i 2

tr H i K

d .

(13)

Extension to Multivariable Systems

In many cases, it may not be possible to achieve the necessary performance by a single actuator and sensor. Should a multiple number of sensors and actuators be necessary in control of a distributed system, it is essential that the effect of higher order modes that are truncated is captured in an optimal way. In this section, we extend the procedure that was developed in the previous section to the multivariable transfer functions of reverberant plants. In the multivariable case, the transfer function matrix of the system is given by: G s
i 1

The rst part of the cost function is independent of K and . Therefore, the optimization problem can be reduced to that of nding the matrix K and the resonant frequency c such that the following cost function is minimized. K, E 1 2 2
i N 1 c 2 2 2 c

1 2 1
2 3

ln
c c

c c

tr K K

1
2 i i i c c

ln

1
i

1 s2

ln

tr H i K

(14)

2 Hi i

(10)

where H i is a matrix whose dimensions are determined by the number of actuators and sensors, i.e., inputs and outputs of the system. If the system has n outputs and m inputs, then H i s are m n matrices. Transfer function matrix G(s) has an interesting property. All of the individual transfer functions of G(s) share similar poles. Moreover, if the actuator and sensors are collocated, G(s) will be a square transfer function matrix whose diagonal transfer functions possess minimum-phase zeros only. However, the off-diagonal transfer functions may have nonminimum-phase zeros since they correspond to noncollocated actuators and sensors. Following 5 , it can be argued that truncating this model, could seriously disturb the zeros of the diagonal transfer functions of 10 . However, the effect on the off-diagonal transfer functions may be less severe. This section is aimed at extending the model correction technique of the previous section to the case of multivariable systems. Here, we approximate G(s) by a nite number of modes. It is our intention to approximate the effect of the truncated higher order modes on the low-frequency dynamics of G(s) by a second order transfer function matrix as follows. Journal of Vibration and Acoustics

It should be clear now that the problem is reduced to solving (K, ). the optimization problem minK Rm n , E c This optimization can be solved numerically. However, it does not necessarily possess a global minimum. As a result, any solution found may only be a local minimum of . To this end, we E point out that if is xed, the optimal K can be found by setting the derivative of with respect to K to zero see page 592 of 7 . E The optimal K is found to be, K
c 2 2 2 c

1 1 2
c c 3

1 ln 1
i i i c c i N 1 2 i 2

ln

ln

c c

Hi .

(15)

Illustrative Example

In this section we apply the model correction method that has been developed in this paper to a simply supported beam with JULY 2000, Vol. 122 333

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

homogeneous material properties. Dimensions and physical properties of the beam are explained in 6,8 . The beam is assumed to be 1.3 m long and it is assumed that a point force is applied to the beam at a distance of 0.075 m from one end of the beam. Here, we are concerned with the transfer function from the applied point force to the displacement at the very same point, i.e., a collocated transfer function. In Fig. 1, we compare the truncated two mode model of the beam with its thirty mode model. Here we are allowing for a 0.3 percent modal damping for each mode. It can be observed that the error caused by truncation is considerably high. In the same gure, we have plotted the corrected two mode model which includes the correcting out-of-bandwidth mode. This term is determined using the optimization procedure developed above. The optimization is done using the constrained optimization routine of

the Matlab Optimization Toolbox. The initial conditions were set at K 0 6 and 0 110.8 rad/sec. Moreover, c was chosen to be c ( 2 3 )/2, i.e., 110.71 rad/sec. A minimum was found at 176.98 rad/sec. K 2.52 and In Fig. 2, we plot the frequency responses of the error systems for the truncated model plus corrected models that are obtained by adding feed-through terms to the truncated model as suggested in 5 and 8,6 . We also plot the error system corresponding to the corrected model that is proposed in this paper. It can be observed that all of the corrected models have lower errors than the truncated one. Moreover, it can be observed that the corrected model with the out-of-bandwidth mode results in a very small error, particularly at frequencies closer to the highest frequency of interest, i.e., c .

Fig. 1 Comparison of the frequency responses of the thirty mode model of the beam with its two mode model and the corrected model with an out-of-bandwidth mode

Fig. 2 Comparison of the in-bandwidth error of the truncated model with three other corrected models

334 Vol. 122, JULY 2000

Transactions of the ASME

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Conclusions

In this paper we presented a method of minimizing the inbandwidth error that arises in truncated assumed modes models of structures. This was done by adding a second order resonant term to the truncated model of the structure. The resonant frequency of this term was chosen to lie out of the bandwidth of interest. The resonant frequency and the gain of the correction term were determined by optimizing the H2 norm of the error system over the in-bandwidth frequency range. The optimization was shown to be numerically tractable. It was shown that if the resonant frequency of the correction term was xed, an analytic expression for the optimal gain could be found.

References
1 Meirovitch, L., 1986, Elements of Vibration Analysis, 2nd edition, McGrawHill, Sydney.

2 Fraser, A. R., and Daniel R. W., 1991, Perturbation Techniques for exible Manipulators, Kluwer Academic, Massachusetts. 3 Alberts, T. E., and Colvin, J. A., 1991, Observations on the nature of transfer functions for control of piezoelectric laminates, J. Intell. Mater. Syst. Struct., 8, No. 5, pp. 605611. 4 Hong, J., Akers, J. C., Venugopal, R., Lee, M., Sparks, A. G., Washabaugh, P. D., and Bernstein, D., 1996, Modeling, identication, and feedback control of noise in an acoustic duct, IEEE Trans. Control Syst. Technol., 4, No. 3, pp. 283291. 5 Clark, R. L., 1997, Accounting for out-of-bandwidth modes in the assumed modes approach: Implications on colocated output feedback control, ASME J. Dyn. Syst., Meas., Control, 119, pp. 390395. 6 Moheimani, S. O. R., 2000, Minimizing the effect of out of bandwidth modes in truncated structure models, to appear in ASME J. Dyn. Syst., Meas., Control. 7 Lewis, F. L., 1992, Applied Optimal Control and Estimation, Prentice Hall, New York. 8 Moheimani, S. O. R., 1999, Minimizing the effect of out of bandwidth modes in the truncated assumed modes models of structures, in Proc. American Control Conference, pp. 27182722, San Diego, CA.

Journal of Vibration and Acoustics

JULY 2000, Vol. 122 335

Downloaded 12 Mar 2012 to 203.200.35.12. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm