measurement techniques in structural acoustics
Transcription
measurement techniques in structural acoustics
MEASUREMENT TECHNIQUES IN STRUCTURAL ACOUSTICS SESSIONS Vibration Damping, Energy and Power Flow G. Pavic LVA, INSA de Lyon, 69621 Villeurbanne, France, pavic@lva.insa-lyon.fr The quadratic quantities of interest in vibration analysis are kinetic and potential energy and power flow density (intensity). All of these quantities can be measured. The classical procedures for the measurement of energy in vibrating linear structures are based on static representation of the Hooke’s law. As a rule, structural damping is not taken into account in a correct manner which can lead to large underestimation or overestimation of true energy levels. The influence of structural damping is demonstrated for some characteristic cases of plates and beams. P ∝ A+2 − A−2 ENERGY, POWER AND DAMPING The input power P, overall energy E, frequency ω and damping η are related by an approximate formula: P ≈ 2 E kinη ω ≈ Eη ω (1) Expressed in terms of modal parameters – natural frequencies υ, modal shape φ, modal force f and mode amplification factor Ω – the overall kinetic energy and total input power are given by: 2 Ekin 2 fq ω2 fq ω 2 , = P= ∑ ∑η q υ q 2 4m q Ω 2 2m q Ωq q f q = ∑ Fn (ω )φ q (z n ) Ω q = υ (1 + jηq ) − ω 2 q (2) (3) where A denotes the wave velocity amplitude. If damping is admitted, (1) becomes more complex; the difference of amplitude squares is changed into: A+2 e − kxη 2 − A−2 e kxη 2 + ηA+ A− sin( ∆ϕ − 2kx) (4) where η - loss factor, k - wavenumber, ∆ϕ - wave phase difference and x - distance along the rod’s axis. The power variation is both exponential and sinusoidal. The power flow in flexural vibration is of much more complex nature as not only the travelling waves but also the decaying ones carry energy. Fig. (2) shows the distribution of power flow in a clamped-clamped 1m long steel beam, excited at 0.2m, loss factor 1%. 2 n The simple formula (1) holds for kinetic energy only at resonances. Its extension to the total energy is valid in cases where the kinetic and potential energies are similar, i.e. again close to resonances, Fig. 1. Fig. 1. Energy in a multi-DOF system. Dotted – resonances. Eq. (1) is valid for either the whole structure or its subassemblies exhibiting resonant behaviour. It cannot be applied at the local level, i.e. for analysis of energy distribution within structure / subassemblies. The role of damping on energy distribution has to be examined. Fig. 2. Power flow in a clamped-clamped beam normalized to unit input power. Loss factor 1%. Negative and positive values indicate divergence in power flow at the excitation. Fig. 3 gives the energies along the beam at 1kHz. SPATIAL DISTRIBUTION OF ENERGY AND POWER The power flow in lossless rods expressed in terms of amplitudes of propagating vibration waves reads: Fig. 3. Spatial distribution of energy in a clamped-clamped beam. Frequency 1kHz. SESSIONS Away from the excitation region and boundaries the two energies are almost identical, while near the singularities the difference between the two increases. Fig. 4 shows the frequency distribution of power flow, normalised to the input power, at 53 equidistant positions covering the entire beam span. Fig. 7. Intensity in a plate shown by the intensity divergence. Plate loss factor: left 0,1% (enhanced 10 times), right 10%. The presence of structural damping couples in power travelling and evanescent waves. This applies not only to plates and beams but to elastic structures in general. DAMPING vs MEASUREMENTS Fig. 4. Normalized power flow in a clamped-clamped beam at equidistant positions along the beam span. Fig. 5 shows the total kinetic and potential energies in a 1.5m×1.25m×20mm steel plate excited at 4 points. Energy, energy density and vibration intensity can all be measured in certain types of structures1-6. When measuring these quantities structural damping must not be disregarded. Fig. 8 shows the measurement error for the beam as defined in the previous section, caused by neglecting structural damping. The curves concern two classical techniques, the 4-point and the 2-point one. Fig. 5. Ratio of kinetic and potential energies in a plate. The damping is seen to little affect the ratio of the two energies which stays close to unity except in the region of very low relative resonance density. The damping however does affect to a great deal the distribution of energy density, Fig. 6 and of power flow, Fig. 7. These two figures refer to the frequency of 500 Hz where the overall kinetic and potential energies are of close level. Fig. 6. Lagrangian energy density in a plate at 500 Hz. Loss factor: left 0,1%, right 10%. n excitation points. Fig. 8. Error of power measurement in a clamped-clamped beam caused by neglecting damping. Spacing: 4 cm. REFERENCES 1. D.U. Noiseux, J. Acoust Soc. Am. 47, 238-247 (1970). 2. G. Pavic, Proc. 3 Intl. Congress on intensity techniques, Senlis, France, 1990, pp 21-28. 3. E.G. Willimas, H.D. Dardy and R.G. Fink, J. Ac. Soc. Am., 78, 2061-2068 (1985). 4. J.C. Pascal, T. Loyau and X. Carniel, J. Sound Vib.,161, 527-531 (1993). 5. J.R.F. Arruda, J.P. Campos and J.I. Piva, Proc. Intl. Conf. Noise Vib. Eng., Leuven, Belgium, 1996 Vol. 1, 641652. 6. Yu.I. Bobrovnitskii, Acoustical Physics, 45, 260-271 (1999). SESSIONS Energy source localization using the reactive structural intensity P. S. L. Alves and J. R. F. Arruda Department of Computational Mechanics, State University of Campinas, C.P. 6122, 13083-970 Campinas, SP, Brazil [psl,arruda]@fem.unicamp Structural Intensity analysis can be a powerful tool in noise and vibration control problems. The active part of the structural intensity is usually measured. It corresponds to the period-averaged value of the force times the velocity, and is related to energy propagation in a waveguide. The reactive part of the structural intensity, related with the energy of standing waves within the waveguide, is usually not analyzed. The authors have recently shown that if properly separated into its wave components, reactive intensity maps agree well with operational modes, clearly indicating the nodal lines in thin flat plates. It is well established that active intensity plots show the energy flow paths and the divergence of the active intensity can indicate the location of energy sources and sinks. For highly reverberant structures, however, measuring active intensity becomes awkward, and intensity plots fail to indicate the region where the energy is injected or dissipated in the structure. In this paper, the possibility of localizing energy sources based on the divergence of the reactive structural intensity and on the distribution of the potential and kinetic energy densities within the structure is investigated. Numerical and experimental results are presented for beams with point excitation. The structural intensity formulation is based on the Euler-Bernoulli theory. A finite element model is used in the numerical simulation. The conditions under which it is possible to effectively detect the location of point energy sources using the proposed method are addressed. POWER FLOW DEFINITIONS In the frequency domain, power flow is a complex vector defined as the product of a force (or moment) and the complex conjugate of the corresponding velocity. The expression for its active component is given by: 1 I = ℜfF (ω)V (ω) g 2 (1) This component corresponds to the period-averaged value. It provides information about the main energy flow paths within the structure, thus enabling the identification of the energy sources and sinks. Active power flow per unit area is also referred to as ‘structural intensity’. Structural intensity is defined as the product of the stress tensor and the velocity. For lightly damped structures, analogously to acoustic intensity, it is awkward to measure the structure intensity [1]. In this case, the component of the energy injected in the structure is mainly related with the reverberation of energy and is usually called reactive power, which is given by the following expression: 1 PR = ℑfF (ω)V (ω) g 2 (2) This component originates the standing waves or modes shapes [2,3]. Since the active and reactive structural intensity components are vectorial quantities, the corresponding energy flow across a closed contour around a point may be obtained. This quantity corresponds to the energy flow balance and may be computed by the divergence of the corresponding map. The active energy continuity equation for the active component is given by following expression [4]: (3) ∇ a = I hΠi where I represents the active power due to the body forces and hΠi the time average of the power dissipated per unit volume. Analogously, the energy continuity equation for the reactive compenent is given by [4]: ∇ r = PR 2ω(hU i hT i) (4) where PR is the reactive power and hU i and hT i are the time averages of the strain energy density and the kinetic energy density, respectively. Active and reactive components of the power per unit width for Bernoulli-Euler beams are obtained by substituting the expressions of moment and shear force: Ix (x; ω) = EI ∂2 w ∂2 w ℜf 2 ∂x2 ∂x∂t + ∂3 w ∂w g ∂x3 ∂t (5) EI ∂2 w ∂2 w ∂3 w ∂w ℑf + g (6) 2 ∂x2 ∂x∂t ∂x3 ∂t Period-averaged values for the strain and kinetic energy are given by: PRx (x; ω) = 2 hU i = 14 EI ∂∂xw2 ∂2 w ∂x2 hT i= 14 ρAω2w w (7) NUMERICAL RESULTS A numerical simulation was performed using a 0.0319x0.00315x1.5 m aluminum beam (E=70 10 9 SESSIONS N/m2 , ρ=2780 kg/m 3 ) modeled with 60 two-node Bernoulli-Euler elements (Fig. 1). In order to simulate a semi-anechoic termination, localized stiffness and damping coefficients were used: KT =1000N/m; CT =100Ns/m (translational); KR =100Nm/rad, CR =10Nms/rad (rotational). The divergence of the reactive power and the term related with the distribution of the strain and kinetic energies were obtained at 105Hz (Fig. 2a). The difference between the two curves shows the location where the energy is injected in the structure (Fig. 2b). 0.6 0.7 0.5 0.6 0.4 0.5 0.3 0.4 0.2 0.3 0.1 0.2 0 0.1 −0.1 0 −0.2 0 0.5 1 1.5 −0.1 0 0.5 1 [m] 1.5 [m] (a) (b) FIGURE 2. (a) Divergence of the reactive power ‘:’ and the distribution of strain and kinetic energies ‘-’; (b) difference between the two curves. EXPERIMENTAL RESULTS An experiment was conducted using an aluminum beam plunged into a sand box to simulate a semianechoic termination (Fig. 3). The point where the force is applied and the beam dimensions are the same considered in the numerical simulation. FRF’s were measured with white noise excitation in the frequency range 0-800 Hz and responses were measured with a laser Doppler vibrometer. The experimental results shown in Figs. 4a and b indicate that it was possible to localize the energy source using the reactive power divergence. However, this was only possible in higher frequencies and with very accurate measurements. Using the reactive power implies computing third order spatial derivatives from measured data. It is known that differentiation amplifies noise and, thus, computing the third order derivative is critical. Very accurate results and careful curve fitting of the spatial data are absolutely essential, as well as high density spatial data. The authors are currently investigating the practical limitations of the method and extending it to plate structures. FIGURE 3. Experimental setup. 0.04 0.08 0.02 0.06 0 0.04 −0.02 0.02 −0.04 −0.06 0 0 0.5 1 −0.02 1.5 0 (a) 0.5 1 1.5 (b) FIGURE 4. (a) Divergence of the reactive power ‘:’ and the distribution of strain and kinetic energies ‘-’; (b) difference between the two curves at 317Hz. REFERENCES ACKNOWLEDGMENTS The authors are grateful to the Fundação de Pesquisa do Estado de São Paulo (FAPESP) for the financial support. 1. F. J. Fahy, Sound Intensity, Elsevier Applied Science, London, GB. 2. P. S. L. Alves, J. R. F. Arruda, L. Gaul and S. Hurlebaus, Power Flow Estimation Using Pulse ESPI, In: Proc. of the 4th Int. Conf. on Vibration Measurements by Laser Techniques: Advances and Applications, pp.362-373, (2000), Ancona, Italy. 3. P. S. L. Alves and J. R. F. Arruda, Active and Reactive Power Flow Estimation Using Mindlin Plate Theory, In: Proc. of the 9th Int. Symposium on Dynamic Problems of Mechanics, pp.459-464, (2000), Florianópolis, Brazil. 4. K. S. Alfredsson, Active and Reactive Structural Energy Flow, Trans. ASME, Volume 119, pp. 77-79. FIGURE 1. FEM model. SESSIONS Measuring the Potential Energy of Structure Vibrations Yu.I.Bobrovnitskii Laboratory of Structural Acoustics, Mechanical Engineering Research Institute of Russian Academy of Sciences, 101990 Moscow, Russia, E-mail: bobrovni@orc.ru While the kinetic energy of a forced vibrating structure can be estimated by measuring the vibration velocity at a number of points, direct measurement of the potential energy of vibrations is a problem due to difficulties in mounting sensors and measuring the stresses all over the structure. In this paper, a method is reported that allows one to estimate the potential energy of vibrations, as well as other energy characteristics, in a rather economic way - through the measurement of the input impedance of the structure. It is shown in computer simulation and experimentally that the method gives reliable and accurate estimates of the potential energy in the low and middle frequency range. INTRODUCTION of the driving point velocity response be known (or The potential energy of a forced vibrating structure characterizes the stresses averaged over the structure and therefore can be useful in many practical applications. At present, the potential energy can not be measured directly by the existing methods because of the necessity to measure six strain components over the entire structure. Recently, a very efficient method has been proposed [1,2] that allows one to estimate the energy characteristics of linear vibrating structures using minimum data measured with usual facilities. To obtain the potential energy, loss factor or other energy characteristics, one does not need, in this method, to measure or compute the vibration response over the entire structure; neither does one need to possess a vibrational model of the structure. The only quantities needed are the complex amplitudes of the external forces (or the input impedances of the structure with respect to these forces) and the complex amplitudes of the velocity response at the driving points. In the simplest case, when the structure is driven by a single external force it is sufficient to measure only two quantities – the amplitude of the force and the amplitude of the driving point velocity. The method is mathematically well based and verified in computer simulations and experimentally. Here, the method is further developed, the accent being made on estimation of the potential energy. measured). The parameters of the system as well as the system response at other points are supposed unavailable. According to the proposed method, one can, using the given quantities, f and v, to estimate the potential energy U as MAIN RELATIONS Consider a linear elastic system with continuous or lumped parameters and with any type of damping, performing harmonic vibrations under the action of the external point force fexp(-iωt). Let the complex amplitude f of the force and the complex amplitude v 1 ∂z (ω ) z (ω ) − | v | 2 Im α ∂ω ω 8 (1) 1 ∂y (ω ) y (ω ) − U ≅ − | f | 2 Im β . ∂ω ω 8 (2) U ≅− or Here, the input impedance z(ω) and input mobility y(ω) are the functions of the measured quantities: z(ω) = f / v, y(ω) = v / f . α and β are closed to unity correction coefficients introduced in paper [2]. It has been proved [1] that when the elastic system under study has no damping, relations (1) and (2) are mathematically correct, i.e. they give the exact values of the potential energy at all frequencies, the correction coefficients α and β being equal to unity . When the elastic system has damping, these relations give good results at the frequencies where the system is mass or rigidity controlled. In the vicinity of the natural frequencies where the system is damping controlled the true value of the potential energy are restored with the help of the correction coefficients. The frequency range of validity of the estimates and of the whole method at its present stage comprises the frequencies where the resonance peaks of the system vibration SESSIONS response are distinctly separated from each other, and the difference between two adjacent resonance frequencies is at least three times greater than the width of the resonance peaks. For not heavily damped structures (with the material loss factor less than 0.1) these are low and middle frequencies. COMPUTER SIMULATION AND EXPERIMENT To illustrate the accuracy of the proposed estimates for the potential energy consider a straight uniform rod of length l free of stresses at one end which executes longitudinal vibrations under the action of an external harmonic load applied to the other end. The vibrations are assumed to be governed by the classical equation of Bernoulli with complex Young’s modulus, Ec=E(1-L0) 0 being the material loss factor. Figure 1 shows the numerically in a computer. The main problem encountered during the experiment was differentiation of the imaginary part of the input impedance and mobility with respect to frequency. Differentiation is a poor conditioned operation extremely sensitive to errors in the data: small deviations in the data (measured with inevitable noise) may lead to large errors in the derivatives. This problem has been overcome with the help of the Pade approximation as a smoothing procedure for the input impedance and mobility. The Pade approximation, i.e. the representation by a ratio of two polinomials, is a “natural” descriptor for impedances and mobilities much more efficient than usually used polinomials. When the total order of the two polinomials was twice the number of resonance (antiresonance) frequencies in the frequency range under study, their ratio provided very accurate approximation to the input impedance and mobility (<1.5%) as well as to their derivatives with respect to frequency (<5%). Such obtained estimates of the potential energy of the experimental beam is shown in Figure 2. There is a good FIGURE 1. Potential energy of the rod vs frequency: exact – solid line, estimated – crosses. k is the wavenumber; material loss factor is equal to 0.05. potential energy of the kinematically excited rod: the exact values computed analytically are presented by the solid line and the estimates via the input impedance are shown by crosses. The estimated potential energy is practically equal to the exact one at low frequencies (kl<8) and does not differ from it more than 10% at middle frequencies (kl<20). At higher frequencies, where the elastic wavelength is less than one third of the rod length, the difference increases. The estimates (1) and (2) have been verified in a laboratory experiment with a flexurally vibrating beam (4x5x150 cm) driven by a shaker. Measured were the amplitudes and phases of the driving force and acceleration at the driven point. Based on them, the velocity, the input impedance, its reciprocal – the input mobility, their derivatives with respect to frequency and the estimates of the potential energy were obtained FIGURE 2. Potential energy of the flexurally vibrating beam: measured by the proposed method (solid line) and by two independent methods (dashed line and stars). agreement between the estimated values and the values obtained by other methods. The experiment confirmed that the proposed method gives reliable estimates of the potential energy at least at low and middle frequencies. REFERENCES 1.Yu.I.Bobrovnitskii, J.Sound Vibr. 217, 351-386 (1998). 2.Yu.I.Bobrovnitskii, M.P.Korotkov, Acoust. Physics, 46, 655-662 (2000). SESSIONS Techniques for Characterising Vibration Inputs to Structures in Multiple Source Situations P.R. Wagstaff, R. Dib and J-C. Henrio Department G.S.M., University of Compiègne, BP 60319, 60206 Compiègne Cedex, France. (peter.wagstaff@utc.fr) Practical techniques of characterising unknown dynamic inputs to structures rely on accurate measurements of the response of the system and the application of inverse techniques to identify the different forces and moments applied at the interfaces between the sources and the receiving structure. The sources may be the result of several generating mechanisms contained within a single mechanical unit linked to the structure or the combination of several different components, that are connected independently. This paper discusses some aspects of methods that may be used to separate and evaluate the characteristic force inputs generated by a specific source at the structure interface in a multiple source situation. Inverse methods of force identification are linked with the signals from reference transducers to reduce the effects of interference from secondary sources. It is shown that different types of estimator produce different types of result if signal conditioning is applied to the inverse method. The application of these techniques is illustrated by experiments conducted on a test bench intended for electric motors. The aim of the experiments was to test different ways of reducing errors in identifying the force inputs generated by the motor due to the vibrations of the generator used as a brake on the test bench. INTRODUCTION APPLICATION The characterisation of the structure borne excitation resulting from primary or secondary vibration sources of industrial machinery and other problems linked to noise and vibration has been a popular topic of research in recent years. The resolution of an inverse problem is often necessary to obtain a satisfactory estimate of the excitation under realistic operating conditions. With larger systems and structures it may be possible to place force transducers and accelerometers between the primary source and the receiving structure without markedly changing the characteristics of the interactions, but in most cases it is preferable to retain the same conditions of liaison between the source and the structure as those of normal operation. Some of the physical limitations and the methods of obtaining the types of measurement required using the classical direct inverse problem approach are described and discussed as well as methods of improving the quality of the results in multiple source situations. The particular application that has been the object of our investigations is the characterisation of the structural excitation induced by an asynchronous electric motor fixed to a test bench. The problems of qualifying the structural excitation of the motor in this situation are similar to those faced by many applications dealing with characterising noise and vibration in industrial machinery. In this case the motor is the primary source which interests us and the brake (II in figure 1). used to provide the load on the motor and thus simulating normal operation is a secondary source of vibration, which for certain frequencies is completely independent of the motor excitation and for other frequencies is completely correlated with it. The coupling between the brake and the motor and the structure is relatively strong because both are mounted on flanges bolted rigidly to the structure, so any reference signals designed to differentiate between the sources are liable to be affected by both.. VI V II 2 III 3 1 5 IV 4 P o s itio n s o f acc e lero m e te rs I C h arg e a m p lifie rs Pow er A m p lifiers i D ata a cq u is itio n r1 1 r2 2 P o s itio n s o f ex c itatio n s an d refe ren c e p o in ts FIGURE 1. Set-up used for characterising the transfer between each point of excitation and the structure SESSIONS BASIC PRINCIPLES The measurement of the frequency response functions between each point of excitation on the flange and the accelerometers characterising the response is carried out using a shaker and a force transducer in the absence of the primary source. The motor excitation is assimilated to 4 force inputs normal to the flange, each force centred on the positions of the four bolts fixing the motor to the flange. If the motor is the only source of excitation of the structure, these forces may be identified by measuring the cross spectral response matrix of the accelerometer responses [G xx ] during normal operation and solving the pseudo inverse problem to find the unknown force input matrix G ff in expression (1) with the aid of the measured processing this data are available, either the principal component or partial coherence techniques and the pseudo-inversion technique is then applied to the conditioned accelerometer response matrix data instead of the directly measured response using the same FRF matrix as before. EXPERIMENTAL RESULTS 0 -1 0 -2 0 -3 0 dB -4 0 -5 0 -6 0 [ ] -7 0 -8 0 FRF (frequency response function) matrix [H]. [Gxx] = [H ]H [Gff ][H ] M ×M M ×N -9 0 -1 0 0 (1) N × N N ×M The pseudo inversion of the FRF matrix is achieved with the aid of SVD (singular value decomposition). The principles of this kind of identification are well known, but some points require a reminder. Firstly the inertia elements upstream of the points of force application during FRF measurements should be removed. In the case of rotating machinery this means removing the shaft if one is trying to identify the excitation applied by the shaft to the bearings. The drawback is that the structure is no longer under the same static gravity loading which may result in measured FRF values being different from those associated with true running conditions. This static load may be simulated during the FRF measurement with the aid of elastic tension elements. The problems associated with the effects of multiple sources may be illustrated by the MIMO (Multiple Input Multiple Output) model presented below. 0 1000 2000 3000 4000 F re q u e n cy (Hz ) 5000 6000 7000 FIGURE 3. Measured force (black), direct inverse (blue) conditioned inverse (green). The above results were obtained with one reference and one shaker replacing the motor excitation in order to investigate the different errors possible and compare directly with the known input measured with a force transducer. The conditioned inverse estimation is completely coincident with the measured force spectrum whereas the direct inverse underestimates the force in certain zones due to the effects of noise. For two independent sources and shakers the results for one of the identified force spectra is presented below using different estimates for the maximum and minimum force contributions coherent with the references on the force transducers. 0 -20 L12 Y1 ( f ) Σ H 21 Amplitude dB -40 H 11 -60 -80 X 1( f ) H 12 L12 X 2( f ) Y2 ( f ) Σ H 22 H 1n L12 -120 . . . . . . . Σ -100 0 1000 2000 3000 4000 Fréquence (Hz) 5000 6000 7000 FIGURE 4. Measured force (black), maximax (green), max (blue), mini (red). Yn ( f ) H 2n FIGURE 2. MIMO model for two inputs and N outputs. The input signals representing the characteristics of the two different sources are used to condition the measured output response spectra and to distinguish between the forces due to each source. Two methods of In this case the maximum estimator is collinear with the measured force and seems the better choice, but the reference signals are normally accelerometers which are more contaminated by the secondary sources at the modal frequencies. In the fuller version of this paper, extensions of these methods are presented to try and identify the real force contribution of each source, even in the presence of inter-source reference contamination. SESSIONS Vibrational Power Transmission in Structures Built by Dynamically Mismatched Substructures J. Lianga and B.A.T. Peterssonb a Department of Aeronautical and Automotive Engineering, Loughborough University, U.K. b Institute of Technical Acoustics, Technical University of Berlin, Germany The vibrational power transmission in structures made of mismatched substructures is discussed. It is revealed that power supplied by a force into such structures is localised in the directly driven substructure. INTRODUCTION In the field of solid-state physics, a phenomenon called Anderson localisation [1] was discovered that, in the presence of local defects, renders the normal modes of a nominally periodic lattice to be confined in the region close to driving point and therefore the magnitude of response will decay exponentially away from this region. In a series of studies, the authors of references [2,3] attempted to bring this case to structural acoustics, and proved that for periodic structures with extended disorder such a behaviour appears, under the condition that the structure is large. Because of the imposed condition, the results obtained are less interesting to vibration engineers wishing to see that vibration can be confined in a small region. This leads to the present work through which a more effective way is sought to localise structural vibration in the vicinity of the driving point. The configuration of the structure to be considered is such that the substructures are cascaded and the dynamics of each substructure are mismatched to those of its neighbours and strictly, therefore, no periodicity of the structure is required. ANALYSIS YJJ(1) and YJJ( 2 ) , the coupling between the two substructures can be described by the ratio [4] cp21 = YJJ(1) / YJJ( 2 ) (1) The interface force at the joint is found to be ìcp - cp 2 + L cp 21 (2) FJblock = FJblock í 21 -1 21 - 2 FJ = 1 + cp 21 î1 - cp 21 + cp 21 - L where FJblock is the reaction from the joint J when it is blocked. If the dynamics of the two substructures are mismatched, i.e. -1 cp21 << 1 , FJ cp21 << 1 or is approximated to zero order as FJ » 0 for cp21 << 1 , -1 or FJ » FJblock for cp21 << 1 . Having determined the point mobility of the structure from (1) (1) Ypp = Ypp - YpJ FJ / Fp , one obtains the zero order approximation of input power as ìï Pin(1), free for cp21 << 1 2 1 Pin = Re Y pp Fp » í (1),block (3) -1 << 1 for cp21 2 ïî Pin where the superscript free denotes the situation where the joint J is dynamically free. Equation (3) states that the power input into such a structure is approximately equal to that for the case where the second substructure is removed, leaving the joint either free or blocked. ( ) The power transmitted into the second substructure is Fp p (1) (2) J Figure 1 A two-element structure coupled at J Commence with a simple case where the structure is made of two substructures coupled at a joint J , as depicted in Figure 1. By means of the uncoupled point mobilities of the two substructures at the contact point ( ) 2 determined by Pt = 12 Re YJJ( 2) FJ . It is noted that the zero order approximation of FJ vanishes for cp21 << 1 and hence for this case it is necessary to approximate FJ to the first order such that block 21 J FJ » cp F . Thus, the transmitted power is found to be of the order SESSIONS ±1 Pt ~ cp21 Pin (4) where the positive sign is used for cp21 << 1 whereas the negative sign is associated with -1 cp21 << 1 . Equation (4) indicates that the power transmitted into the second substructure is smaller than the input power ±1 by the factor cp21 . By carrying out such analysis recurrently for the structures made of n mismatched substructures arranged on a chain, it can be demonstrated that the transmitted power obeys Pt J ( i +1 ) ~ cp(±i1+1)i Pt J i , i = 1,2, L n (5) This means that a significant reduction of vibrational power can be achieved in the area a few spans of substructure away from the vicinity of the excitation more than one point or even a continuous line or surface. If the multiple points or the continuous line or surface is scattered or distributed over an area with typical dimensions substantially smaller than the wavelength of the governing wave, the multiple-point or continuous connection will present negligible effect and can be treated as a single point connection [5]. If, on the other hand, the typical dimensions of the connecting area are comparable with or larger than the wavelength, the effects of interactions between coupling points and components of motion and excitation must be taken into account. Upon focussing on the overall vibration transmission, extensions to encompass the multi-point and component cases have been proposed [6] Further investigations, however, are required to gain more insight into the influence of the interaction and to be able to circumvent restrictive assumptions imposed. CONCLUDING REMARKS DISCUSSIONS The present analysis is subject to certain conditions. First, it is apparent that the results given in equations (3-5) may fail in cases where the dissipation in one or more substructures vanishes. Such a failure is caused by the truncation of the series of the interface force because the power transmissions associated with the higher order terms of this force are neglected. With a decrease of the losses, the neglected components of the power become increasingly important. Therefore, the application of the present analysis should be restricted to situations where the damping in each substructure is sufficiently large such that the power dissipated in a substructure is higher than the power transmitted into its downwards neighbour c.f. the relaxed concept of weak coupling [5] Second, for the situation of broadband excitation, judging the dynamic mismatch by means of cp(±i1+1)i << 1 , i = 1,2, L is not adequate because cp(±i1+1)i is an oscillatory function of frequency. This difficulty may be overcome by assessing the overall Despite some unresolved practical problems, the results obtained have some potentially useful implications. First, in companion with damping treatment for each substructure, mismatching can localise power in the vicinity of the excitation and hence significantly reduce power transmission in structures. Second, to estimate the power fed to a structure constituted by a set of mismatched substructures, it is sufficient to consider the driving substructure only. Third, an efficient way to apply damping treatment for such built-up structures is to add damping to the directly driven substructure, or in the vicinity of excitation. REFERENCES 1. 2. 3. 4. value of cp (±i1+1)i , i = 1,2, L in the frequency range of 5. interest. As a consequence, the analysis can only be regarded applicable for the overall vibration transmission. 6. P.W. Anderson, Physical Review 109, 1492-1505(1958) C. H. Hodges, JSV 82(3), 411-424(1982) C. H. Hodges and J. Woodhouse, Report on Progress in Physics 49, 107-170(1986). J. M. Mondot and B.A.T. Petersson, JSV 114(3), 507518(1987). L. Cremer and M. Heckl and E. Ungar, Structure-borne sound. Berlin: Springer-Verlag, 2nd Edition, 1988 J. Liang and B.A.T. Petersson. Dominant dynamic characteristics of built-up structures. (to appear in JSV) Third, herein point connections are assumed between substructures, which differs in some aspects from practical situations where the connections involve SESSIONS The Method of Local-Global Homogenization (LGH) for Structural Acoustics D. B. Bliss and L. P. Franzoni Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC, USA A homogenization method for complex structures, valid at all frequencies, is being developed with emphasis on structural acoustics. Applications include naval and aerospace structures. The approach differs from classical homogenization and utilizes a local-global decomposition facilitated by adding and subtracting canceling smooth forces. The smooth global problem, which can be solved independently, has an infinite-order structural operator; and periodic discontinuities are replaced by equivalent distributed suspension terms. The global problem can be solved efficiently since all rapidly varying scales have been removed. The local problems provide transfer function information, and are only solved afterward if needed. Several problems have been homogenized: a beam with periodic impedance discontinuities, a fluid-loaded membrane with discontinuities; and a cylindrical shell with periodic ribs. An interesting feature is that the effects of fluid radiation can be transferred entirely to the smooth global problem, while evanescent modes are absorbed into the global operator. [Sponsored by the U.S. Office of Naval Research]. INTRODUCTION Many important structural systems having discontinuities, braces, and/or attachments at regular intervals are spatially periodic, or very nearly so. Prime examples are aircraft fuselages, truss structures, and ribbed hulls. Even when forced at a single frequency, the response occurs in a broad spectrum of spatial wavenumbers. Structures such as fuselages and hulls are fluid loaded, and their response is altered accordingly. Not only the structural motion, but also the acoustic radiation, scattering, or interior sound field, may be of interest. Calculating the motion of such structures is a complex and computationally expensive task. The disparity of scales requires high numerical resolution. The forcing may be a series of locally applied forces or continuously distributed. If the structure is spatially periodic, the response will exhibit stop- and pass-bands and have a discrete wavenumber spectrum. If the structure is not strictly periodic, there will be a distributed wavenumber spectrum and the structural response may exhibit localization. Often the low wavenumber (long wavelength) portion of the response is of primary interest, since it models the gross vibratory response. For fluid loaded structures, the low wavenumber part of the response is most efficiently coupled to the acoustic field, since low wavenumbers give supersonic phase speeds. The goal is to isolate the low wavenumber problem from the overall problem. Because the low wavenumber problem is smooth and contains transfer function information from the high wavenumber part of the problem, the approach is a homogenization method. It differs from classical homogenization, and it is valid for the full frequency range. The low wavenumber problem is smooth and global; namely it spans the structure. The high wavenumber problem can be thought of as a series of contiguous local solutions between the discontinuities. Therefore the method is called Local-Global Homogenization. Since the global problem has a known degree of smoothness, there are potential advantages in accuracy and efficiency, if the approach can be extended to numerical methods. Philosophically, the approach is an analytical reformulation, prior to solution, to allow the direct and efficient calculation of the most important aspects of complex problems. FORMULATION For illustration, Fig. 1 shows a 1-D primary structure, such as a beam, string, or membrane, with impedance discontinuities and distributed forcing. L L L L L L discrete impedances FIGURE 1. Forced structure with discontinuities. The equation of motion for the structure, assuming harmonic time dependence, is of the form: L[hG + åh L ] = å d (x - xm )iwhZ m + Fe -iax + f s e -igx - f s e -igx Here L is the structural operator and the displacement h has been written as a sum of the global displacement and sets of local displacements between the discontinuities. On the right-hand-side a smooth force has been both added and subtracted, so the SESSIONS original equation is unchanged. The wavenumber of the applied forcing is a ; for the smooth (slowly varying) force (fs) it is g = a - 2np/L, where n is chosen to minimize g. The wavenumber g corresponds roughly to the smallest wavenumber (longest wavelength) that passes through the discontinuities. By superposition, the problem can be separated into a global equation and local equations that apply on open intervals between discontinuities, L[hG ] = f s e -igx and L[h L ] = Fe -iax - f s e -igx . To solve the local problem, a system of equations is constructed. The number of unknowns is equal to the number of constants in the homogeneous solution plus the number of smooth forces introduced. The number of conditions that must be satisfied due to geometric and natural boundary conditions equals the number of homogeneous constants; therefore one additional constraint can be added per smooth force. For example, it may be desirable to remove the lowest wavenumber from the contiguous local solutions. This constraint adds another equation to the linear system and allows for solution of the unknown smooth force in terms of the applied forcing and the global motion. In addition to this condition, the system of equations is comprised of matching conditions between sub-intervals at the discontinuities and a phase-shifting condition. Returning to the global equation, the smooth force on the right-hand-side can then be replaced by the expression for fs found from the local system of equations. Re-arranging so that all of the terms containing the global variable are on the left side and terms containing the forcing are on the right side, the resulting global equation is then inverse transformed. In spatial variables, the equation governing the global motion consists of a new structural operator that contains an infinite number of even spatial derivatives plus a suspension term. For example, the global equation for a membrane is given by: FLUID-LOADING EFFECTS The local and global solutions can be structured so that the lowest wavenumbers are contained only in the global solution and the higher wavenumbers are contained entirely in the contiguous local solutions. Because the lowest wavenumbers have the highest phase speed, these are the modes that radiate into the fluid. Thus fluid radiation can be confined entirely to the global problem. Furthermore, through an accurate approximation, the evanescent fluid modes can have their loading effect absorbed into the local solutions and thereby used to modify the transfer function effect appearing in the global operator. Thus the effect of evanescent fluid modes can be moved to the global operator, and only radiating modes need to be considered in the global problem. RESULTS AND CONCLUSIONS Example problems based on membranes, beams, and cylindrical shells with periodic discontinuities have been homogenized. For the membrane, unequal mass discontinuities were homogenized; in all other cases thus far, the discontinuities have been identical. For these problems, the dispersion relation for the global problem will exhibit the expected stop- and pass-band behavior of Bloch waves; see Fig. 3. Re[gL] 3 2 1 5 10 15 20 -1 b bL -2 -3 Im[gL] 0.4 0.2 5 10 15 20 bL -0.2 æ iwz L2 n ¶ 2 n ö iwz (cos gL - cos kL ) ÷h = ç1 - cos kL sin kL + å Fe - igx 2n ÷ ç 2 k ( L k2 - g 2 k2 - a 2 n 2n )! ¶x ø è ( )( ) On the right-hand-side, the original forcing is operated on by a filtering function so that the forcing is now slowly varying. Schematically, the homogenized original problem is shown in Fig. 2. L L L L L L New operator includes suspension FIGURE 2. Global problem after homogenization. -0.4 FIGURE 3. Dispersion relation for the beam with periodically spaced mass discontinuities. Local-Global Homogenization is a promising new approach for separating an original problem that contains fast and slow spatial variations into components with different wavenumber content. The smoothly varying global problem often contains all of the information that is desired and it is easily converged, thus providing computational savings. SESSIONS Diagnostics of structure vibrations in acoustic frequency range with the aid of self-organizing feature maps S. N. Baranov, L. S. Kuravsky Problem Laboratory of Mathematical Modeling attached to the Computer Center of Russian Academy of Sciences, c/o “Rusavia”, 6 Leningradskoye Shosse, 125299 Moscow, Russia. E-mail: rusav@aha.ru Failure diagnostics for the structures suffered vibrations in acoustic frequency range is presented. Normalized spectral characteristics of structure response measured in checkpoints are used as indicators to be analyzed. Self-organizing feature maps (Kohonen networks), for which output variables are not required, detect faults. Simultaneous application of different networks duplicating each other makes it possible to improve the quality of recognition. Principal component analysis is employed to reduce the number of variables under study. An aircraft panel with different combinations of attached defective dynamic suppressors is considered to demonstrate features of the approach. Tests have demonstrated high effectiveness of the presented way of recognition and showed the advantages of neural networks over cluster analysis in recognition problems. Technical diagnostics is one of the most typical spheres where neural networks are used. Under consideration here is failure diagnostics of the structures suffered vibrations in acoustic frequency range. This diagnostics is carried out on the base of spectral characteristics measured in structure checkpoints. It is supposed that neither all possible types of damages nor corresponding changes induced in the spectral characteristics may be predicted beforehand. Because of multiplicity of structure types and their applications, it is impossible to generalize considerably the problem solutions. Therefore employed for method demonstration is a specific system including a simply supported steel sandwich rectangular panel and two attached 1-degree-offreedom elastic vibration suppressors with fluid friction. Its dynamic behavior was simulated on the basis of models and methods described in paper [1]. Positions of suppressors were optimized. Wide-band random processes represented test acoustic loads. System conditions were estimated via standardized power spectral densities of accelerations in a checkpoint. (In general case, some checkpoints may be used.) Initial data to estimate such characteristics may be obtained with the aid of accelerometers. Standardizing spectral densities makes it possible to analyze only qualitative shape of structure response spectra and not to take into account the response level. The following system conditions were simulated: OK – both suppressors work properly, Only1 – suppressor 2 is defective, Only2 – suppressor 1 is defective, Panel – both suppressors are defective, Nonlin – non-linear suppressor response (shock interaction of the moving element and stopper). The first variant corresponds to normal operating mode, and the following four ones represents system damages. Since all the damages are not assumed to be known before diagnostics, it is impossible to apply ordinary neural networks with supervised learning for their detection. Self-organizing feature maps (Kohonen networks) [2-3], for which output variables are not required, may be useful in this case. Self-organizing feature maps have an output layer of radial units [4]. This layer is also called a Topological Map and, as a rule, is laid out in a 2- or 1dimension space. Starting from an initially random set of centers, the Kohonen algorithm successively tests each training case and selects the nearest (winning) radial unit center. This center and the centers of neighboring units are then updated to be more like the training case. As a result of a consequence of such corrections, some network parts are attracted to the training cases, and similar input situations activate the groups of units lying closely on the Topological Map. A self-organizing feature map is taught to “understand” input data structure in such a way and to solve the classification problem. The idea, on which this network is based, was originated by analogy with some known features of the human brain. If clustering of input data is completely or partially ascertained, semantic labels might be attached to certain units of the Topological Map. When a classification problem is solved, so called accept threshold is set. It determines the greatest distance on which recognition occurs. If the distance from the winning element to an input case is greater than this threshold, it is supposed that the network has not made any resolve. When units are labeled and SESSIONS thresholds are determined properly, the self-organizing feature map may be used as a detector of new events: it informs about input case rejection only if this case differs from all labeled radial units significantly. The given approach supports diagnostics of both known in advance and unknown damages. Simultaneous application of different networks duplicating each other makes it possible to improve the recognition quality. Frequency ranges are used as variables, and the values of normalized power spectral densities at the centers of these ranges – as cases. Thus, each complete case represents a separate power spectral density. In the test example, initial variants of neural networks with 3×3, 4×4 and 7×7 output layer dimensions were trained to recognize the states OK and Only1. Later on, the conditions Only2, Panel and Nonlin arose successively. After detection of new damage types, network training was carried out again, with corresponding labels being assigned to units of the Topological Maps1. It is convenient to estimate the recognition quality via the percentage of correctly identified situations. Two sorts of errors may occur: errors of the 1st type, when some unknown system state is identified as known one, and errors of the 2nd type, when some system state that has been known before is identified as unknown one or incorrectly. Application of networks duplicating each other2 made it possible to avoid errors of the 1st type in 99-100% of analyzed cases and errors of the 2nd type – in 98-99% of such cases. When all variants of system damages are known before, the problem is essentially simpler. One can employ traditional neural networks with supervised learning to solve it. Perceptrons were the best for the test problem: networks of 100%-recognition were revealed. Radial basis function networks turned out to be less accurate. Neural networks are, of course, not the only way to solve recognition problems. The same purposes may be achieved by means of other procedures – for example, cluster analysis that is intended for partition of an initial object set into classes following a given criterion. Comparison of both techniques makes it possible to draw the following conclusions: ♦ cluster analysis does not yield distinct criteria for classification: one cannot always distinguish qualitatively new and old-type damages – the result depends on critical distance selection; ♦ cluster analysis is less reliable than neural networks; ♦ cluster analysis needs more computer resources than neural networks. Principal components analysis and factor analysis are employed to reduce the number of input variables under study if the number of frequency ranges to be taken into account is too great and worsens network characteristics. These methods extract few latent hypothetical variables that explain approximately all the set of observed ones. As for the test problem, input data capacity might be reduced up to 2 latent variables, with the errors being avoided in 93-100% of cases. Nonlinear transforms on the base of autoassociative neural networks [5] are used in more complicated situations. As a rule, reduction of problem dimension prunes the number of neurons and, therefore, improves characteristics of network training. CONCLUSIONS 1. 2. 3. 4. REFERENCES 1. 2. 3. 4. 1 Working with a real structure, examination to reveal the failure nature must be fulfilled before new training and label assigning. Otherwise, the network will only be able to inform of an appearance of some new, unknown earlier, damage type. 2 Recognition results were selected “by a majority”. Self-organizing feature maps, whose training data do not contain output variables, make it possible to diagnose conditions of vibroacoustic systems in situations where neither all possible damage types nor corresponding changes induced in observed characteristics are not predictable beforehand. If all types of system damages are known beforehand, ordinary neural networks with supervised learning (perceptrons, radial basis function networks) may be used for diagnostics. Test results showed that neural networks were more efficient recognition tools than cluster analysis. Reduction of problem dimension (with the aid of principal components analysis, etc.) improves characteristics of network training. 5. Kuravsky, L. S., and Baranov S. N., “Selection of optimal parameters for acoustic vibration suppressors”, in Proceedings of the 7th International Conference on Recent Advances in Structural Dynamics, Southampton, United Kingdom, 2000. Kohonen, T., Biological Cybernetics 43, 59-69 (1982). Kohonen, T., “Improved versions of learning vector quantization”, in Proceedings of the International Joint Conference on Neural Networks, San Diego, USA, 1990. Haykin, S., Neural networks: a comprehensive foundation, Macmillan Publishing, New York 1994. Kramer, M. A., AIChe Journal, 37, 233-243 (1991). SESSIONS On the Sound Transmission through a Truncated Conical Shell and its Coupling to a Cylindrical Shell P. Neplea, C. Lesueurb a EADS AIRBUS SA, 316 Route de Bayonne, 31060 Toulouse Cedex 03, France LRMA ISAT, 49 rue Mademoiselle Bourgeois BP 31, 58027 Nevers Cedex, France b The present paper deals with the vibroacoustic behaviour of a truncated isotropic conical shell either rigidly backed or coupled to a stiffened cylindrical shell for the purpose of understanding airborne sound transmission through aircraft cockpit. To understand sound transmission mechanisms through a truncated cone and quantify the influence of its coupling to a cylinder on them, experimentation has been carried out on three cases : (a) a rigidly backed truncated cone, (b) an assembly of a truncated cone and a stiffened cylinder, (c) a stiffened cylinder. The lower and upper ring frequencies of the cone are fR2=1800Hz and fR1=4500Hz and its critical frequency is fc=12000Hz. fR2 and fR1 correspond to the ring frequencies of cylindrical shells with radius equal to the larger and the smaller radii of the cone. The Noise Reduction (NR) of the truncated cone for cases a and b, and the NR of the cylinder for case c, have been measured under diffuse sound field conditions in the 100Hz-10kHz range. The two major results are the superposition of the NR curves of the truncated cone for cases a and b above 0.56*fR2 and the similitude of the transmission behaviour of the truncated cone to that of a cylinder. INTRODUCTION In the context of understanding airborne sound transmission through aircraft cockpit, the present work deals with the vibroacoustic behaviour of a truncated isotropic conical shell, either rigidly backed or coupled to a stiffened cylindrical shell. The aircraft cockpit is modeled by a truncated isotropic conical shell of constant thickness and the aircraft structure (cockpit + cabin) by an assembly of a truncated cone and a stiffened cylinder. In the literature, no information could be found concerning sound transmission through such structures. So we set up a data bank in order to understand transmission mechanisms through a truncated cone and quantify the influence of its coupling to a cylinder on them. Experimentation (Noise Reduction (NR) + cavity modes) has been carried out on three cases : (a) a rigidly backed isotropic truncated cone, (b) an assembly of an isotropic truncated cone and a stiffened cylindrical shell, (c) a stiffened cylindrical shell. DESCRIPTION OF THE EXPERIMENTS Figure 1 shows the geometry and the dimensions of the shells. The NR of the truncated cone for cases a and b, and the NR of the cylinder for case c, have been measured under diffuse sound field conditions in the 100Hz10kHz range. For each case, the excitation was a white continuous noise generated by a loudspeaker, and four microphones (two external, two internal) were used. For each microphone, the measurement has been made for 12 circumferential different positions, with the same radial and longitudinal positions (cf. figure 2). The Noise Reduction (NR) has been obtained by : NR (f ) = 10 log10 2 pext 2 pint S (1) V <pext2>S and <pint2>V are respectively mean external and internal quadratic sound pressures with S and V being the cone surface and volume for cases a and b, and being the cylinder surface and volume for case c. . INTERPRETATION OF THE EXPERIMENTAL NR CURVES To refer to the transmission mechanisms of a cylinder, let’s consider the lower and upper ring frequencies of the truncated cone, fR2=1800Hz and fR1=4500Hz, and its critical frequency fc=12000Hz. fR2 and fR1 correspond to the ring frequencies of cylindrical shells with radius equal to the larger and the smaller radii of the cone. Figure 3 shows the Noise Reduction (NR) in the SESSIONS 100Hz-10kHz range for the three cases. The two major results are the superposition of the NR curves of the truncated cone for cases a and b above 0.56*fR2 (fig.3, curves 1, 1’) and the similitude of the transmission behaviour of the truncated cone for cases a and b to that of a cylinder (fig. 3, curves 1, 1’, 2). Note however that the plateau ends with the ring frequency for the cylinder (fig. 3, curve 2), while it ends at 1.56*fR2 (or 0.62*fR1) for the cone (fig. 3, curves 1, 1’, zone 2), that is between the lower and upper ring frequencies. Moreover, the NR increases h D1 hc Scone h1 L Hemispheric nose PERSPECTIVES An analytical approach and full scale experimentation on an airplane cockpit at EADS AIRBUS SA France will soon complete our study. 24 stringers Vcone D2 hn Scylinder with a 12 dB/octave (fig.3, curves 1, 1’, zone 3), which is not the classical mass law tendency behaviour (6dB/octave) observed for a cylinder (fig.3, curve 2). Vcylinder End plate h2 End plates h3 Cone : h=1mm D1(external)=382mm D2(external)=942mm L=950mm h1=25mm h2=30mm hn³10mm Cylinder : hc=1mm D2(external)=942mm Lc=1800mm h3=30mm The 24 stringers are equally spaced. Both shells and stiffeners are made of aluminium Lc FIGURE 1. Thin isotropic conical shell and stiffened cylindrical shell – Geometry and dimensions Case a Case b Fixation bar for microphones l l=150mm l l l l l Case c l l=150mm l External microphones (distance from the skin»10mm) l l=150mm Internal microphones (distance from the skin »50mm) FIGURE 2. Microphones positions for Noise Reduction measurements (1) Zone 1 Zone 2 Me mbra ne be ha viour with minima due to c a vity mode s P la te a u 10 dB Zone 3 Zone 4 Inc re a se De c re a se with a be low fc 12 dB/ oc t slope (1) 10 dB (3 ) (3 ) (4 ) (4 ) (1') fR2 100 1000 F re que nc y ( H z) (2 ) fR1 10000 100 fR2 1000 fR1 10000 F re que nc y ( H z) FIGURE 3. Measured Noise Reduction (NR) under diffuse sound field conditions. (1) NR of the truncated cone case a, (1’) NR of the truncated cone case b, (2) NR of the stiffened cylindrical shell, case c, (3) Mass law under diffuse sound field conditions (6dB/octave slope), (4) 12 dB/octave slope. (1), (1’), (2) are analysed in narrow band Df=4Hz. SESSIONS Sound Radiation from Visco-elastically Damped Plates Excited by a Random Point Force K Akamatsua and T Yamaguchib a Machinery Acoustics, 1-1-2-314 Obanoyama Shinohara, Nada-ku, Kobe 657-0015, Japan b Gunma University, 1-5-1 Tenjin-cho, Kiryu, Gunma 376-8515, Japan The acoustic radiation from baffled finite plates with free viscoelastic damping layers excited by a random point force is studied in order to evaluate damping treatment performances under the condition of neglecting fluid loading. The vibration response of the plates with damping layer is obtained by the finite element method. Modal damping ratios are estimated from undamped normal mode results by means of the modal strain energy method. Expressions for the surface acoustic intensity and the radiated sound power are derived in the transform formulation and evaluated numerically. An experimental study is carried out to measure the surface intensity distributions and to compare them with the analytical results. ANALYTICAL METHOD The surface acoustic intensity distribution on a baffled finite plate of arbitrary configurations excited by a stationary random force is derived. The surface acoustic pressure radiated from the plate is described in the frequency domain as [1] ωρ P(x,y,ω ) = 2 (2π ) =ωρF −1 ∞ V˜ (ς x ,ς y ,ω ) ∫∫ k −ς −ς 2 −∞ 2 x 2 y e ( j ς x x +ς y y ) dς dς x y [ V˜ (ς ,ς ,ω )G˜ (ς ,ς )] x y x y (1) where F −1[ ] denotes the inverse Fourier transform, G˜(ς x ,ς y ) is the Green function and V˜(ς x ,ς y ,ω ) is the Fourier transform of the plate vibration velocity V(x,y,ω ) in the spatial co-ordinate. V(x,y,ω ) is given by V(x,y,ω ) = jωF(ω )∑ H r (ω )φ r (x0 ,y 0 )φ r (x,y) (2) r H r(ω ) = 1 m rω (1 −ω ω 2r + jηω ω r ) 2 r (3) 2 where ω r , m r , φ r (x,y) , η r are the radian natural frequency, modal mass, mode shape function and modal loss factor for r th mode, and F(ω ) is the applied force onto point (x 0 ,y0 ) . The acoustic intensity at the position (x,y) on the plate is given by 1 ∗ (4) I(x,y,ω ) = P(x,y,ω )V (x,y,ω ) 2 Numerical Results In the numerical analysis, an aluminum plate with free layer damping treatment with dimensions of 480 × 360 mm and thickness of 2 mm clamped in a frame with thickness of 10 mm was used. A square baffle of length 2.5 m was placed around the plate. The undamped mode shapes and modal parameters were computed for the composite plate with the viscoelastic material treated as it were purely elastic, then the modal loss factors were obtained by the modal strain energy method [2]. The FFT algorithm and the averaged Green function developed by Williams [3] are used to compute the acoustic pressure on the source plane. The baffle plate area is divided into a lattice of 128 ×128 points with the lattice spacing of 20 mm. Figure 1 shows the acoustic intensity distributions with F(ω ) =1 N for the octave band centered at 250 Hz. The total acoustic power radiated from the plate is obtained by a summation of the acoustic intensity distribution over the plate surface. Two normalized values of the power, the radiation efficiency and the power conversion efficiency, are shown in Figures 2 and 3. The power conversion efficiency is defined as the ratio of the acoustic power Wa to the vibratory power Wk supplied to the plate, ς = Wa Wk Wk = Wa + Wk0 (5) Wk0 is the input power to the plate neglecting the back reaction of the radiated acoustic pressure. It is shown that the radiation efficiency is independent of damping, while the power conversion efficiency depends on damping. EXPERIMENT In order to evaluate the surface intensity patterns and the acoustic power for consistency of predicted trends, an experiment measuring the acoustic intensity field near a vibrating plate driven by a point force with white spectrum was carried out. SESSIONS without damping layer (250 Hz) 20 10 0 19 -10 10 1.5 with damping layer (250 Hz) y 1 0.5 the driving point accelerance. The nearfield estimate of the surface acoustic intensity of the baffled plate was measured using the two-microphone technique. The vibratory input power is given by 2 1 F A ∗ (6) Wk = Re{FV } = sinφ 2ω F 2 where A F is the magnitude of and φ is the phase of the driving point accelerance. Figure 4 shows the experimentally measured surface intensity patterns. The overall agreement between the predicted and measured intensity distributions is good. Table 1 compares the predicted and measured power conversion efficiency. The trend of effects of damping is consistent except 125Hz band. without damping layer (250 Hz) 1.5E-04 0 1.0E-04 -0.5 5.0E-05 Power conversion efficiency [-] Sound radiation efficiency [-] -1 17 FIGURE 1. Surface intensity patterns, octave band at 250 Hz, upper; undamped, lower; damped. 0.0E+00 -5.0E-05 0 10 6 5.0E-05 -1 with damping layer (250 Hz) 10 2.5E-05 -2 10 0.0E+00 -3 10 without damping layer with damping layer -4 10 60 80 100 300 Frequency [Hz] FIGURE 4. Measured surface intensity patterns, octave band at 250 Hz, upper; undamped, lower; damped. FIGURE 2. Sound radiation efficiency. 0 10 Table 1. Power conversion efficiency. 125 Hz band 250 Hz band 500 Hz band undamp damp undamp damp undamp damp Predicted 0.596 0.022 0.089 0.0059 0.124 0.0126 Measured 0.050 0.009 0.130 0.0068 0.200 0.0169 -1 10 -2 10 -3 10 without damping layer with damping layer -4 10 -2.5E-05 500 60 80 100 300 Frequency [Hz] 500 FIGURE 3. Power conversion efficiency. The plate assembly was hung vertically at the center of a baffle with dimensions of 1.5 ×1.8 m . A minishaker was attached to the plate via an impedance head and driven by a white noise source with frequency range up to 1000 Hz . The output signals from the impedance head are fed to the FFT analyzer to obtain CONCLUSIONS The acoustic radiation from plates with free viscoelastic damping has been studied analytically and experimentally. The overall agreement between the predicted and measured is good. REFERENCES 1. H. Peng and R. F. Keltie, J. Acoust.Soc.Am. 85 (1), 57–67 (1989). 2. J. Kanazawa, T. Yamaguchi and K. Akamatsu, J. Acoust. Soc. Am. 100(4 Pt. 2) p 2754 (1996). 3. E G. Williams and J. D. Maynard, J. Acoust.Soc.Am. 72 (6), 2020-2030 (1982). SESSIONS Nearfield effects on acoustic radiation modes of a structure O. Schevina , P. Herzogb and M. Rossia a Laboratory of Electromagnetics and Acoustics, Swiss Federal Institute of Technology, 1015 Lausanne, Switzerland b Laboratoire de Mécanique et d’Acoustique, CNRS, 31 chemin Joseph Aiguier, 13402 Marseille Cedex 20, France The use of acoustic radiation modes to characterize the behaviour of a structure has received increasing attention since the beginning of the 90’s, especially for active control applications (ANC). The main advantage of this expansion is that it involves independant radiation contributions, so that a truncated series leads to a well-defined accuracy for the sum of its terms. However, these modes are defined from the active power radiated by the vibrating surface; they are thus related to the far-field pressure, or to the real part of the radiation impedance on the surface itself. As many ANC systems minimize a weigthed sum of squared inputs, their performances can at first glance be guaranteed only if they involve specific structural sensors, or microphones located in the far field. However, none of these solution is as cost-effective as could be microphones located close to the structure. Conversely, near-field microphones pick up the reactive components of the pressure field, and the transfer matrix estimated is in that case is no more characteristic of the radiated power. This paper proposes to discuss how radiation modes are related to the near field, and the consequences on the radiation control of a structure from bringing the microphones close to it. INTRODUCTION ANC systems are usually aimed to control the acoustic pressure in the far field, and thus they should use microphones located far away from the noise source, in such a way that they pick up a signal representative of the noise to be cancelled. If microphones are located in the vicinity of the source, they pick up local acoustic phenomena which are not propagated further. These phenomena, frequently grouped under the term "nearfield", tend to modify the behaviour of the control system, owing to the fact that the controller seeks to reduce an acoustic pressure which is not representative of the acoustic power radiated to the far field. Because cheap microphones may be used reliably at low frequencies, and that reducing the necessary amount of wiring has a significant effect on the cost of the system, it was found interesting to study the possibility of bringing such sensors closer to the source, and to analyze the consequences of this design. Equation (1) is a good estimate only if the microphones are placed in the far field and spaced homogeneously around the noise source. By expressing transfer functions Z between the pressures p at each microphone and velocities vs at each point of a discrete model of the source surface, equation (1) can be written in the following matrix form [1] : Ŵa vH s H vs (2) 1 H where H ρc Z ∆S Z is the radiation operator, and ∆S is a diagonal matrix containing areas associated with each microphone. The notation H denotes the hermitian operator. By introducing the eigenvalue/eigenvector decomposition of the operator : H QH ΛQ (3) in which Q is an orthogonal matrix of eigenvectors and Λ a diagonal matrix of eigenvalues λi , equation (2) can be written as Ŵa cH Λ c N ∑ λ2i ci 2 (4) i 1 RADIATED POWER Controllers available on the market are usually designed to minimize a sum of squared pressures picked up by a certain number of microphones. Weighted by the medium characteristics and the area implicitely associated with each measurement position, this sum is an estimate of the active power radiated by the source : N Ŵa p2 ∑ ρc ∆Si i 1 (1) where c Q vs is a vector of coupling coefficients between the surface velocity vs and each eigenvector. The radiated power estimate is therefore formulated as a sum of radiation modes which have the property of radiating power independently of each other. In order to study the influence of these modes in the nearfield, we consider the trace of the radiation modes on the source surface, and compute their effects, on a surface conformal [2] to the source at a given distance, as the quadratic estimate defined by equation (1). This estimate is then compared to the true active power radiated, computed in the far field. SESSIONS Source 0 0.1 1 10 distance ( m) FIGURE 1. Evolution of the acoustic field with distance. Left : vibration pattern of the source surface. right : acoustic field radiated by the source on 3 different conformal surfaces. NEARFIELD EFFECTS We consider here a planar rectangular surface 1.7 x 3 m vibrating and radiating an acoustic field at 100 Hz. Figure 1 illustrates the evolution of the pattern of the acoustic field with increasing distance, showing the complexity of the local components in the nearfield. When bringing the microphones closer, this reactive component of the acoustic field becomes non-negligible in comparison with the active one, and this causes the controller to get a bad estimate of the radiated power if the microphones are positionned closer than about 1 m. The pressure picked up in the nearfield, for each radiation mode pattern, is the sum of the pressure related to its contribution to radiated power, and the reactive pressure corresponding to local disturbances. These two components are out of phase, and thus the pressure magnitude in the nearfield is systematically overestimated and so is the related power contribution. Table 1 gives in the case of that planar surface, the power attenuation that would be achieved by the controller if microphones were in the nearfield or in the far field, for a given number of modes included in the ANC design. Table 1. ANC performances with microphones in the nearfield or far field. Number of modes 1 2 5 10 Attenuation (dB) d 0 5 m d 10 m 1.55 3 7 13 3 4 12 20 Conversely, for a given mode, distance and frequency, the ratio between the active and reactive components of the pressure is fixed by the trace on the surface of the radiation mode considered. Assuming that the reactive components generated by different modes do not interact significantly, the overestimation can be fully determined and taken into account in the controller as an additional weighting factor. In practice, such a weighting factor has been found necessary only if microphones have to be placed in the very nearfield. For intermediate distances and reasonable microphone locations, the weighting factors mentionned in equation (1) seem sufficient. This has been confirmed in the case of more complex non-planar sources, which behaviour revealed no major differences with planar ones. CONCLUSION We have shown that active power radiated by a radiation mode tends to be overestimated by an ANC controller using close microphones. Conversely, a method is proposed to take this into account by a suitable correction of the controller input weights. ACKNOWLEDGMENTS Authors wish to thank the swiss institutions CTI and PSEL for their financial support, as well as the ABB Sécheron company for their active participation. REFERENCES 1. S. J. Elliott and M. E. Nelson, J. Acoust. Soc. Am. 94, 2194-2204 (1993). 2. C. I. Holmer, J. Acoust. Soc. Am. 61, 465-475 (1977). SESSIONS Sound reflection characteristics of suspended panel array T. Yokotaa, S. Sakamotoa, and H. Tachibanaa a Institute of Industrial Science, Univ. of Tokyo, Komaba 4-6-1, Meguro-ku, Tokyo 153-0041, JAPAN The frequency characteristics of sound reflection by suspended panel arrays often equipped in halls are investigated by numerical analysis based on the Fresnel-Kirchhoff diffraction theory. Three types of panel arrays with different shapes and arrangements are set and their reflection characteristics are compared. INTRODUCTION In order to reinforce the sound reflection, suspended panel arrays are often equipped in halls. However, it is known that the sound reflection frequency characteristics of suspended panel arrays are apt to be uneven due to the interference between discrete reflections from each panel [1,2]. In this study, this problem is investigated by numerical analysis based on the Fresnel-Kirchhoff diffraction theory by setting three kinds of suspended panel arrays with regular and irregular spatial distributions. From the results, the way of making the reflection characteristic flat is considered. Type-1 Type-2 Type-3 CALCULATION CONDITION As the typical suspended panel arrays, three kinds of variations (Type-1: circle, Type-2: triangle and Type-3: random) shown in Fig. 1 were set in this study. The reflection frequency characteristic for each panel array was calculated by the diffraction theory based on the FresnelKirchhoff approximation according to Babinet’s principle. Figure 2 shows the position of the sound source, calculation area and the projection area of the panel arrays. In the calculation area, 2809 (53x53) observation points were assumed at an interval of 0.5 m. At each observation point, the velocity potential of the reflected sound was calculated for 84 frequencies from 90 Hz to 2240 Hz which were chosen at an equal interval on logarithmic scale. After the calculation, the results were averaged in every 1/3 octave band and they were normalized by the value calculated for the direct sound from the mirror image of the source (diffraction factor: DF). (open area ratio : 50%) FIGURE 1. Arragements of the reflectors Section Panel Array R1 Source Plan(half area of the sound field) Calculation Area (observation points : 0.5m intervals) C.L. RESULTS AND DISCUSSIONS Figure 3 shows the spatial distribution of DF in the calculation area in the case of Type-1, in which different R1 Source 01 2 5 10 [m] FIGURE 2. Calculation set-up SESSIONS 500 Hz (1/3 oct. band) 0 2k Hz (1/3 oct. band) DF[dB] 125 Hz (1/3 oct. band) Source Source -30 FIGURE 3. Distributions of reflected sound level 10 0 DF[dB] -10 Type-1 Type-2 Type-3 -20 -30 125 250 500 Frequency [Hz] 1k FIGURE 4. Reflection frequency characteristics at R1 Type-1 10 Small Type-2 CONCLUSIONS From the results of this numerical study, it has been confirmed that the suspended panel arrays regularly distributed in space are apt to produce uneven reflection frequency characteristic due to the interference between discrete reflections. This problem can be mitigated to some extent by designing the panels in irregular shape and distributing them randomly in space. 2k SD [dB] interference patterns are clearly seen in each frequency band. These patterns are caused by the interference between the reflections from the regularly distributed discrete panels. Figure 4 shows the reflection frequency characteristics at the specular reflection point of the center reflector (R1 in Fig.2) for the three types of panel arrays. In the cases of Type-1 and Type-2, remarkable dips are seen at around 250 Hz, whereas the frequency characteristic is relatively even in the case of Type-3. In order to evaluate the evenness of the reflection frequency characteristic at each observation point, standard deviation (SD) of the value of DF in each 1/3 octave band in the frequency region was calculated. As a result, Fig.5 shows the spatial distributions of SD for each type of panel array. In the results for Type-1 and Type-2, it is seen that a lot of areas colored dark (SD is large) are regularly distributed. It means that there are many areas where the reflection frequency characteristic is much uneven. On the other hand, in the case of Type-3, the distribution pattern seems to be vague and areas colored light (SD is small) are large. It means that the area where the reflection frequency characteristic is relatively flat is expanded. 0 Large Type-3 Evenness of frequency characteristics Source REFERENCES [1] R.W.Leonard, L.P.Delsasso, and V.O. Knudsen, J.Acoust. Soc. Am. 36 (12), 2328 - 2333 (1964) [2] B. G. Watters, L. L. Beranek, F. R. Johnson, and I. Dyer, Sound 2, 26 - 30 (1963) FIGURE 5. Distributions of SD SESSIONS Absorption and transmissibility of coupled microperforated plates T. DUPONT, G. PAVIC and B. LAULAGNET Laboratoire Vibrations et Acoustique, INSA de Lyon, France The microperforated plate, backed up by a cavity and a rigid wall, is usually used for sound absorption. Based on this configuration, the well-known mathematical model by Maa [1, 2] has shown a good agreement with measurements. However, this model fails if a microperforated plate is coupled to another flexible plate. The latter case is the subject of the present paper. At first a one-dimensional case is studied, that of a microperforated plate coupled with a rigid but movable spring-supported piston. A comparison done between modelling and Kundt tube experiment has shown a good agreement. The main effect which enables absorption of a microperforated plate is the viscosity of air flowing through perforations. Two methods have been suggested to increase this effect. A two-dimensional model in oblique incidence has been then developed considering infinite plates. A wave approach using plate impedance has been applied to express the principal acoustic indicators: transmission, absorption and reflection. A parallel impedance scheme has been used to take in account the vibration of the microperforated plate. Different cases have been analysed: the isolated microperforated plate, coupled with a thin plate, a thick plate and a rigid wall. 1. ONE DIMENSIONAL SYSTEM A mathematical model of absorption and transparency of a micro-perforated plate (MPP) system was produced. The model uses the MPP’s impedance as given by Maa [1,2]. In order to understand the acoustics of the system and to validate the developed model, an industrial MP made by a Swedish manufacturer Sontech was tested in a Kundt’s tube. The MPP was backed by a cavity and a rigid cap. A very good match was found between the experiment and the model, Fig. 1. Fig 1. Simulation (---) and measurement (-o-) of absorption in the Kundt tube. Sontech MPP: slit perforations and louvre structure coupled to a rigid wall via 10cm deep cavity. Dural, thickness 1,25 mm, estimated equivalent diameter of circular perforation ∅ 0,25mm, estimated perforation rate 3,5 %. implies that it is difficult increasing absorption without modifying the MPP’s parameters. Secondly, the absorption model had been extended by replacing the rigid wall by a piston (mass with spring and damping). The piston is used here to represent the action of a second plate in the 1D model. The mass, the spring and the piston damping represent a resonant system equivalent to one of the modes of the second plate, i.e. one additional degree of freedom. The piston was found to affect absorption only near its resonance frequency. In this region, the absorption is completely modified w/r the previous case; the piston’s resonance and anti resonance are clearly identifiable (Fig. 2). Fig 2. Computed absorption coefficient of a cavity-backed MPP: (dural, circular holes, thickness 1,25 mm, perforation rate 3 %. Cavity termination: --- rigid wall; -•- 1cm thick steel piston loaded by 2.42e6 N/m spring. Structural damping 0,01. Cavity depth: 10 cm. The perforation’s parameters (perforation rate and hole size) were found to be the key factors governing the 2. TWO DIMENSIONAL SYSTEM: system’s absorption. As the viscosity of the fluid inside INFINITE PLATES the perforations is the principal absorption effect here, two methods were attempted to increase it: a) heating the plate and b) replacing the air in the perforations To account for possible variations in the angle of sound with a more viscous fluid (oil). No major improvement incidence, a further study was made of an infinite MP was found either by experiment or by simulation. This plate in tandem with an ordinary infinite plate. The transmission loss, the reflection and the absorption SESSIONS factors were computed. The classical formula for the MPP impedance does not take into account the plate vibration. The motion of a MPP can modify the part of velocity responsible for the viscous effect (global viscosity). If this motion is of comparable level and phase with the acoustic velocity, the viscous effect will be reduced, while the transparency will be affected too. In order to include this effect in the model and to keep the classical MPP impedance model valid, a parallel impedance branch was added to the existing one: frequencies a screen effect takes place producing high reflection of the composed system. However, the higher the frequency the narrower the absorption band, thus at high frequencies the reflection dominate the absorption implying an increase of the transmission loss with frequency. The coincidence occurs around 1200 Hz, Fig. 3 – top. At this frequency and at certain angles the transmission loss becomes very small. It can be seen that the MPP used reduces the coincidence effect without requiring 1 = 1 + 1 , where Zvib is the plate vibration any absorption material: in the example shown the TL difference MPP system – simple plate is around 15 dB. Ztotal Z vib Z MPP A reversed system has been also tested: source – air – impedance under sound wave excitation and ZMPP is simple plate – air - MPP. Although reflection and MPP the MPP impedance of a single MPP based on the absorption reciprocity do not apply, one can show that Maa’s model. the transmission loss is here perfectly reciprocal. For the reversed system the reflection is very high: it tends very quickly towards unity with frequency, exactly like the single plate reflection, which in turn induces very low pressure in the air gap and consequently a very low MPP absorption. Because of very high reflection, the inverse configuration has lost all of the MPP advantages. Thus when using a MPP, it has to be mounted facing the side of the acoustic source, or inside a room which needs sound isolation. 3. CONCLUSION A MP used in a lightweight composite panel is a good solution of noise control. Its presence leads to non negligible system absorption without the need to add any classical absorption material. In addition, it gives rise to transmission loss, reduces the reflection effect and weakens the coincidence effect. A MPP application could be a viable solution to the low frequency sound transmission in thin plates. Moreover, contrary to porous materials commonly used for absorption, a MPP is inflammable and shows very good hygienic features. This system could be used in transport industry, civil engineering, and in acoustic panel design. REFERENCES [1] Fig 3. Transmission loss (top) and reflection coefficient (bottom) at 45° incidence. -•- simple plate (thickness: 2mm, [2] damping: 1%); --- the same simple plate preceded by a 15 cm deep cavity and a MPP (steel, circular perforation, thickness [3] 1.5mm, ∅ 0.5mm, perforation rate 0.75). One can see on Fig.3 up to 15 dB rise of transmission [4] loss in the presence of the MP. In contrast to the simple plate case, the MPP system does not produce a [5] reflection coefficient which sharply jumps to unity because of the absorption effect, Fig. 3 - bottom. [6] As the frequency rises further, the difference in the transmission loss globally increases (at 4000 Hz the difference exceeds 20 dB). The reflection increases with frequency owing to the MPP alone. At very high MAA, Da You. Theory and design of microperforated panel sound-absorbing constructions. Beijing, China : Scientia Sinica, 1975, vol. 18, n°1, p 55-71. MAA, Da You. Wide-band sound absorber based on microperforated panels. Beijing, China : Chinese journal of acoustics, 1985, vol. 4, n°3, p 197-108. MAA, Da You. Potential of microperforated panel absorber. JASA, 1998, vol. 104, n°5, p 2861. NILSSON, A., NILSSON, E. Sound transmission through honeycomb panels. Proceedings of Congress: Modern practice in stress and vibration analysis Dublin, 1997. KANG, J., FUCHS, H. V., Predicting the absorption of open weave textiles and microperforated membranes backed by a air space. Stuttgart: JSV, 1999, vol. 220, n°5, p 905-920. FUCHS, H.V., ZHA, X. Acrylic-glass sound absorbers in the plenum of Deutscher Bundestag. Stuttgart : Applied Acoustics, 1997,vol 51, n°2, p 211-217. SESSIONS Unsteady dynamics using local and global energy models M. N. Ichchoua , F. S. Suia and L. Jezequela a Department of Mechanical Engineering, Ecole Centrale de Lyon, 69130 Ecully, France A new energy method, called as transient local energy approach (TLEA) is proposed to predict the structural transient dynamic response in the time domain. For the purpose of comparative studies, the energy expressions got from three methods (TLEA, TSEA and exact results) are given, respectively. These studies indicate that the difference between TLEA and TSEA depends on the different description of time-varying energy flow transferred between two coupled subsystems. The comparisons show that TLEA is much more accurate than TSEA. INTRODUCTION - LITTERATURE POSITION OF THIS WORK SECOND ORDER TRANSIENT LOCAL AND GLOBAL EQUATIONS The transient behavior, such as shock and impact, can cause a broad frequency structural vibration and noise, especially in high and mid frequency band. In the early development of the well known statistical energy analysis (SEA), Manning and Lee (1968) proposed a method based on steady-state power balance equation to deal with the mechanical shock transmission [1]. However, as was pointed out by Manning and Lee [1], TSEA was not developed formally, because the definition of coupling loss factor in transient condition was just "transplanted" from steady-state SEA, it seems not to be always appropriate and reliable. More recently, Pinnington and Lednik [2, 3], published additional TSEA study by comparing with exact results for two-degree-freedom model, the conclusion they got were not always satisfactory to predict the transmitted energy precisely. This paper will present a new Transient Local Energy Approach (TLEA) and its discretized format. The two oscillators system and the corresponding two coupling subsystems are studied to illustrate the reliability of TLEA by comparing the solution with the exact results and TSEA. Second order local energy equation FIRST ORDER TRANSIENT SEA EQUATIONS dE1 + η1 ωE1 + η12 ωE1 dt η21 ωE2 I s t) = 1 ∂~I (~s; t ) ηω ∂t c2 ∇W (~s; t ) ηω (3) The energy equation is obtained from the energy balance as ∂2W (~s; t ) ∂t 2 ∂W (~s; t ) 2 + (ηω) W (~s; t ) = 0 ∂t (4) then, formula (4) is called TLEA equations. This equation is different when comparing it to Nefske’s equation [7]. Precisely, the later and the classic TSEA do not consider the time-varying part of the energy flow term, and this ignorance results some inevitable errors. The detailed discussions can be seen in the reference [4, 5]. c2 ∇2W (~s; t ) + 2ηω The TLEA equation will be discretized so that it can be used in the interconnected subsystems or multi-DOF (1) dE2 + η2 ωE2 + η21 ωE2 η12 ωE1 (2) dt The TSEA energy expressions and the parameters definition are almost the same as that got by Pinnington and Lednik [2], therefore, the comparison can be made easily between TLEA, TSEA and exact solution. Πin2 = ~(~; Second order global energy equation position with TSEA The model in Figure 1 is used for TSEA study. The energy balance associated to this model is: Πin1 = Let us define W (~s; t ) as the total energy density associated and ~I (~s; t ) as the active energy flow. c is the energy velocity, the same as the group velocity of waves in the slight damping media. After some mathematics it can be shown [4]: 1 0 0 1 0 1 0 1 0 1 0 1 F2(t) F1(t) k1 k2 k m1 m2 c1 c2 x1(t) x2(t) 1 0 0 1 0 1 0 1 0 1 0 1 0 1 FIGURE 1. Two-degree-of-freedom model. SESSIONS Table 1. Parameters of two different oscillators Test Oscillators Mass (kg) Inherent loss A 1 2 2.5 2 0.08 0.093 Coupling stiff. (N/m) Block freq. (rad/s) CLF ηi j ; η ji Coupling rate † η =η ij i 1000 1072.4 0.26 0.243 3.25 2.60 5105 † : i; j = 1; 2; i 6= j 50 0.25 (a) 40 Transmitted energy power(W) Energy(J) 0.2 0.15 0.1 0.05 0 0 0.02 30 20 10 0.04 0 0.08 −10 (b) 0 0.005 0.01 0.015 0.02 0.025 0.03 Time(s) Energy(J) 0.06 FIGURE 3. Comparison of transmitted energy flow (from element 1 to 2) with the coupling ratio r = 2. —, TLEA solutions; , TSEA solutions; , exact results. 0.04 0.02 0 −0.02 0 0.02 0.04 Time(s) REFERENCES FIGURE 2. Comparison of energy results of two different oscillators. Test A. (a): Input energy, (b): Transmitted energy. —, , TLEA solutions. exact results; , TSEA solutions; oscillator. Precisely, the distributed structure can be divided into some discreted element. Supposed the nodal value is zero order in the finite elements and the concept of total energy rather than energy density turns the TLEA equation into the form: Πin1 = 1 d 2 E1 dE1 +2 + η1 ωE1 + η12 ωE1 η1 ω1 dt 2 dt η21 ωE2 Πin2 = 1 d 2 E2 dE2 +2 + η2 ωE2 + η21 ωE2 2 η2 ω1 dt dt η12 ωE1 (5) (6) The coupling loss factor used in TLEA and TSEA is the same definition in steady state condition. NUMERICAL SIMULATIONS - FIRST VERSUS SECOND ORDER MODELS 1. J. E. Manning and K. Lee, Shock and Vibration bulletin. 37 (4):65–70, 1968. Predicting mechanical shock transmission. 2. R. J. Pinnington and D. Lednik. Journal of Sound and Vibration, 189(2):249–264, 1996. Transient statistical energy analysis of an impulsively excited two oscillator system. 3. R. J. Pinnington and D. Lednik. Journal of Sound and Vibration, 189(2):265–287, 1996. Transient energy flow between two coupled beams. 4. M. N. Ichchou, F. Sh. Sui, and L. Jezequel. CAA’2000 Congress, Sherbrooke, September, 2000. Transient local energy: theory and application. 5. M. N. Ichchou. Oral presentation at SEANET meeting, KTH, June, 2001. Unsteady SEA: Formulations and numerical examples. 6. F. Sh. Sui, M. N. Ichchou and L. Jezequel. Journal of Sound and Vibration, In press, 2001. Prediction of vibroacoustics energy using a discretized transient local energy approach and comparison with TSEA. 7. D. J. Nefske and S. H. Sung. NCA Publication, 3, 1987. Power flow finite element analysis of dynamic systems: Basic theory and application to beams. The time-varying energy results of three methods are compared numerically (see Table 1). The comparisons show that TLEA is much more accurate than TSEA (see Figure 2 and Figure 3). SESSIONS Optimal Design of Stockbridge Dynamic Vibration Neutralizer: Comparation Between BFGS and Genetical Algorithms S.E. Floodya and J.J. de Espíndolab a Acoustics Department, Universidad Tecnológica V. Pérez Rosales, Brown Norte 290, Ñuñoa. Santiago, Chile b Vibration and Acoustics Laboratory, Universidade Federal de Santa Catarina, Campus Trinidade. Florianópolis, SC Brasil The present paper deals with an extension in the study of vibration dynamic vibration neutralizers applied to complex structural systems, that are independent of geometry, mass, stiffness and damping distribution, introducing the concept of Equivalent Generalized Quantities. The basic idea is to transform the mechanical impedance that the dynamic vibration neutralizer transfers in the primary system’s point of connection, in generalized quantities of mass and damping, depending of the frequency. This methodology was applied to design a viscoelastically modified Stockbridge dynamic vibration neutralizer, applying the Finite Element Method, extending the proposed theory of generalized equivalent quantities for several degrees of freedom that can be presented due to the characteristics of this dynamic vibration neutralizer. The cost function was made using the maximum absolute values of the principal coordinates of the joint system, over a frequency range, this cost function has a lot of locals minimums in the viable region. Comparations between a minimization algorithm using derivates, the BFGS and Genetic Algorithm were made, to obtain the optimal dimensions of this mechanical device. INTRODUCTION The dynamic vibration neutralizers, also called vibration absorbers, are devices or structures (secondary systems), that are fixed to another structure (primary system), to reduce vibration levels. They act over the primary system applying reaction forces and dissipating vibration energy. The classic theory of vibration neutralizers introduced by Den Hartog [1] for viscous neutralizers, called MCK is difficult to apply; therefore is inadequate for complex mechanical systems, where many modes can contribute in the response of the primary system. Espíndola and Silva [3] introduced the concept of Generalized Equivalent Quantities, in the study of vibration neutralizers applied to complex structural systems, that are independent of geometry, mass and stiffness distribution. The basic idea is to transform the mechanical impedance, of the neutralizer’s coupling point to the primary system, in generalized quantities of mass and damping that depend of the frequency. With the generalized quantities, is possible to formulate the compound equations of motion in terms of the generalized coordinates of the primary system only. After the equations are written in the principal coordinates, retaining them that correspond to the frequency band of interest, where the problem of high response resides. Then, the computations are made in a modal subspace, with a minimum number of equations. This method will be used to project a viscoelastically modified Stockbridge vibration neutralizer, the viscoelastic material will be added to increase de dissipation of the vibration energy. The effects of the frequency dependence in this type of materials will be studied FEM MODEL OF A STOCKBRIDGE VIBRATION NEUTRALIZER Basically the neutralizer is composed by a central mass, two sandwich (metal – elastomer) beams and two tuning masses, as been showed in Fig. 1. The finite element method has been used to model in the most general way possible the neutralizer’s behavior. The complex and frequency dependent stiffness matrix is a result of the presence of the viscoelastic material in the structure. This lead to an associated eigenvalue problem, which has frequency dependence and will be solved using the technique presented in Espíndola and Floody [4] paper. The motion equation of the secondary system is: Mq DD + K (ω )q = f (t ) (1) SESSIONS Central Mass Elastomer Tuning Mass Steel FIGURE 1 Stockbridge vibration neutralizer Finite Element model FIGURE 2 FRF Compound System Genetic Algorithm. The dynamical stiffness at the root of the viscoelastically modified Stockbridge neutralizer isb given by Espíndola, Floody [4]. PROJECT VARIABLES AND COST FUNCTION The project variables to optimize are the physical dimensions of the neutralizer; this can be represented as a vector: x = [l1 , l 2 , l 3 , h1 , h2 , h3 , h4 , t ] T (2) Where l ´s represent the lengths, h ´s correspond to the heights of the metal and elastomer layers and t is the width of the neutralizer. The cost function to minimize, proposed by Espíndola and Bavastri [3], is the modulus of a vector formed by maximum values of the principal coordinates of the compound system. This can be expressed in the equation (3). The minimization method used was the genetic algorithm to avoid the great number of local minimum of the cost function. Comparations between a minimization algorithm using derivates, like the BFGS were made. The results are shown in the figure 2 and 3. min β (x , Ω ) 2 Ω1 ≤ Ω ≤ Ω 2 0 < l i ≤ Li 0 < hj ≤ H j 0<t ≤T i = 1, l ,3 j = 1, l ,4 β i (x, Ω ) = max Pi , s (x, Ω ) (3) FIGURE 3 Algorithm FRF Compound System BFGS REFERENCES 1 Den Hartog, J.P., 1956, “Mechanical Vibrations, McGraw-Hill, New York. 2 Espíndola, J.J, Silva, H.P., 1992, “Modal Reduction of Vibrations by Dynamic Neutralizers: A General Approach”, 10th International Modal Analysis Conference, San Diego, California, pp. 1367-1373. 3 Espíndola, J.J., and Bavastri, C.A., 1997, “Reduction of Vibration in Complex Structures with Viscoelastic Neutralizers - A Generalized Approach and a Physical Realization”, Proceeding of DETC’97, 1997 ASME Design Engineering Technical Conferences, September 14 - 17, 1997, Sacramento, California. 4 Espíndola, J.J, Floody, S.E., 2001, “On the Modeling of Metal – Elastomer Composites Strutures : A Finite Element Method Approach”, PACAM IV – DINAME VI, Pan American Congress in Applied Mechanics, Rio de Janeiro, Brasil. SESSIONS Some Aspects concerning a Method for SEA Loss Factor Estimation K-O. Lundberg Department of Engineering Acoustics, LTH, Lund University P.O.B. 118 SE-22100 Lund Sweden, Karl-Ola .lundberg@skane.se A method for experimental estimation of SEA loss factors based on CMTF:s (complex modulation transfer functions), has been earlier reported [1,2]. In the method the low frequency part of the CMTF curve is fitted to a SEA model. The used SEA model has minimum-phase transfer functions. The CMTF curve, however, is not minimum-phase due to a propagation delay. To yield better SEA parameter estimations, the delay should be estimated and removed by shifting the origin of the squared impulse response. Alternatively, using the Hilbert transform, the minimum-phase version of the CMTF can be computed from its magnitude function. The SEA loss factors are estimated for the non-modified, the time-shifted and the Hilbert manipulated response. In the fit procedure the CMTF curve up to the "knee" has been used, rather arbitrary. The damping can be found in the very low frequency part of the CMTF curve, which is obvious from the moment theorem. It means that the centre of gravity time of the squared impulse response is equal to minus the slope of the phase function of CMTF at zero frequency. The centre of gravity time, the slope of the phase function and the decay constant from decay rates of the reverberation curve are computed. INTRODUCTION A wavefront emanating from a source of sound travels at the velocity of sound and reaches a receiver after some time. Thus due to the finite wave velocity there is a propagation delay between the source and a receiver. If the system under test is finite, the response to a pulse of short duration is the direct sound followed by a series of reflections. From this response, or rather the squared impulse response, the damping in the system can be found. -9 7 x 10 first time of arrival 1 0 0 0.005 0.01 0.015 time (s) 0.02 0.025 0.03 FIGURE 1. Initial part of a squared sound pressurevolume velocity impulse response. Measured in a reverberation chamber at a source-receiver distance of 3.15 m in a 1/3-octave band centred at 1250 Hz. The first time of arrival is at 0.0102 s. The used SEA model consists of lumped boxes, while the system under test is a wave propagation system. When estimating SEA parameters, therefore, the propagation delay should be removed. SHIFTING THE ORIGIN OF TIME One possibility is to visually inspect the squared impulse response and then shift the origin of its time record to the moment at which the response starts to build up. This is usually done when damping is derived from decay rates of the reverberation curve; the origin is then shifted to the first time of arrival. An impulse response of a physically realizable system is a causal function. The squared impulse response is causal as well. Then some relations in signal theory can be used. Let g (t ) be a causal if (w ) function and G (w ) = A(w )e its Fourier transform. If G is minimum-phase-shift, then the phase function can be uniquely determined from the natural logarithm of the magnitude function. They are related by the Hilbert transform, i.e. f = Hilbert{ln A} . Thus the minimum-phase version of the phase function can be computed from a given magnitude function. The CMTF is defined as the Fourier transform of the squared impulse response, low-pass filtered. SESSIONS -9 14 x 10 THE MOMENT THEOREM The centre time t s is the centre of gravity along the time axis of the squared impulse response. For a single-degree-of-freedom system with a decay constant d, the inverse of the centre time is 2d. The moment theorem [3] relates the first moment of a time function to the frequency derivative of its Fourier transform at zero frequency: The centre time of the squared impulse response is equal to the slope of the phase function of CMTF at zero frequency. 2 0 -2 0 0.005 0.01 0.015 time (s) 0.02 0.025 0.03 FIGURE 2. The Hilbert-manipulated squared impulse response. ½CMTF½ 1 1_ Ö2 Table 1. Estimates of 2d (=wh factor) from nonmodified squared impulseresponse 5.80 0.170 from leftshifted 0.0102 s squared impulseresponse 6.22 0.160 1 / t s (1/s) 5.89 6.26 Slope of ÐCMTF(0) (s) -1/slope (1/s) -0.170 -0.160 -0.158 6.26 6.09 6.38 6.31 a0 (1/s) t s (s) 0.99 Hz=W0mag/2p 0 0 0 1 5.89 W 0 ph (1/s). See 5.47 2 3 modulation frequency (Hz) 4 5 = -p/4 -p/2 0 0.87 Hz = W 0ph/2p 1 2 3 modulation frequency (Hz) 4 5 FIGURE 3. The CMTF. The solid line corresponds to the non-shifted, the dashed to the Hilbert-manipulated and the dotted to the left-shifted 0.0102 s, squared impulse response. Upper: Magnitude function (All three curves coincide) Lower: Phase function. The dasheddotted line is the used SEA model 6.24 CMTF (0) 2 See Fig.3,upper Reverberation time T [-5,-25] dB (s) 13.82/T (1/s) T [-5,-20] dB (s) 13.82/T (1/s) T [-5,-15] dB (s) 13.82/T (1/s) EDT [0,-10] dB (s) 13.82/EDT (1/s) slope= –0.160 b0 jW + a 0 where the polynomial coef. are derived from a least square curve fit of the Hilbert-manipulated CMTF using the interval from zero to the frequency where ÐCMTF is about -40°. From Hilbertmanipulated squared impulseresponse 6.35 Fig. 3, lower CMTF (W0 mag ) = Ð CMTF where h is the loss 2.21 6.25 2.12 6.52 2.16 6.39 2.18 6.34 ACKNOWLEDGEMENTS The author is grateful to Prof. S. Lindblad for fruitful discussions. REFERENCES 1. Lundberg K-O, Acustica united with acta acustica, 83, 1-9 (1997). 2. Lundberg K-O, Building Acoustics 8 57-74 (2001). 3.Skudrzyk E. The Foundations of Acoustics. SESSIONS Identification and quantification of damping mechanisms of active constrained layer damping Hélène Illaire, Nicolas Poulain, Wolfgang Kropp Department of Applied Acoustics, Chalmers University of Technology, S-41296, Göteborg, Sweden To optimise ACLD (Active Constrained Layer Damping) treatments, it is essential to identify and quantify the different mechanisms leading to the reduction of vibrational energy in the structure. Shen [1] showed that the power input by an external excitation in a beam treated with ACLD is dissipated by the shearing of the viscoelastic layer and by the absorption of mechanical energy by the actuator. In this work, it is shown that the actuator also change the structure impedance, and thus the power input by external disturbances. The motion of a fully treated beam is calculated with a wave approach model. Then, the power balance of the structure is derived, and ACLD treatments’ mechanisms of action are identified and quantified. INTRODUCTION The principle of ACLD is to replace the constraining layer of conventional constrained layer treatment by an actuator (Fig. 1). The actuator will increase the shear in the viscoelastic layer; it will also apply a force on the beam, so that energy is also dissipated through active damping. - The viscoelastic layer is only shearing, with a constant shear angle across its depth. The equation of motion for the transversal deflection is derived by writing the equilibrium of forces and moments on each layer. It is a differential equation of order 6. Solving the characteristic equation gives 6 wave numbers associated with 6 waves propagating on the structure. Six boundary conditions are necessary to determine the amplitudes of these waves. In this work, each end of the beam is resting on a spring. These boundary conditions are chosen because they are easy to implement experimentally. FIGURE 1. Principle of ACLD. Experimental validation of the model Understanding and quantification of ACLD treatments’ mechanisms of action is essential for their optimisation. These mechanisms consist in dissipation of energy by the viscoelastic layer, in absorption of energy by the actuator, and as shown here, in reduction of external input power due to the change of structure impedance induced by the actuator. The motion of a fully treated beam is calculated with a wave approach model. Then, the power balance of the structure is derived, and ACLD treatments’ mechanisms of action are studied. DESCRIPTION OF THE MODEL The model is based on the work of Baz [2]. It uses a wave approach to calculate the motion of a beam fully treated with ACLD. The model makes use of several assumptions: - The transversal deflection is the same for all layers, i.e. there is no compression with respect to the thickness. - Only bending waves propagate in the base and cover beam. Classical Euler-Bernouilli assumptions apply for these layers. FIGURE 2. Acceleration at the driving point of the test beam: model (plain) and measurements (dotted). The model is validated for a passive configuration. An aluminium beam of dimensions 0.4 x 0.03 x 0.003 m was treated on its whole length with a 3M damping tape. The beam was resting on springs made of foam and was excited at one end with a shaker. The displacement at the excitation point was measured with an accelerometer. The mass-loading effect of the SESSIONS measurement equipment was included in the model. A first measurement with the untreated beam was performed in the range 0-2 kHz, to determine the stiffness of the springs, and to tune the Young’s modulus of the aluminium. Then the measurement was repeated with the treated beam. Fig. 2 shows a good agreement between the results of the measurement and of the model. Experimental validation of the model for an active configuration is part of ongoing work. input power is strongly reduced by the active control in this case. IDENTIFICATION AND QUANTIFICATION OF DAMPING MECHANISMS In this work, a quantification of the different damping mechanisms based on an energy approach is proposed. The power balance in the structure has to be fullfiled: Wext + Wa = Wshear , FIGURE 3. Displacement at the driving point of the test beam: without control (dotted) and with control (plain). (1) where Wext is the power input of the external force applied by the shaker, Wa is the power input of the actuator, Wshear is the power dissipated in the viscoelastic layer, and <⋅> denotes the time averaging. Wa can be positive or negative, depending on the amplitude and phase applied to the actuator. The force applied by the actuator is changing the impedance of the structure and thus the power input of the external force. This change of the power input can be quantified as ∆Wext = Wext p − Wext , (2) where Wext p is the power input of the external force FIGURE 4. Power balance of the beam at the 1st mode when the active control is turned off. Simulations Simulations are calculated on a test beam to illustrate the damping mechanisms. The test beam is the same as the beam used to validate the model, except that freefree boundary conditions are now assumed for the sake of simplicity. The actuator is driven with a voltage of amplitude 50 Volts. The phase shift between the control voltage and the external force is π. The displacement of the driving point with and without control is shown in Fig. 3. The amplitude of the 1st mode is decreased of 9 dB, and this of the 2nd mode of 6 dB. The even modes are not affected by the control, since the actuator covers the whole length of the beam. The power balance is then calculated at the 1st mode. The quantities Wext , Wext p , Wshear and Wa are plotted in Fig. 4. The results show that the external CONCLUSION This study shows that the reduction of input power induced by the actuator can substantially contribute to the reduction of vibrational energy in the structure. Therefore, taking this mechanism into account when optimising ACLD treatments could lead to a substantial improvement of the efficiency of such treatments. ACKNOWLEDGEMENTS This research is sponsored by Volvo Research Foundation, Volvo Educational Foundation and Dr Pehr G Gyllenhammar Research Foundation. REFERENCES 1. Shen, I.Y., Journal of Vibration and Acoustics 119, 192199 (1997) 2. Baz, A., Active constrained layer damping, in Proceedings of Damping’93, San Francisco, 1993, pp. IBB 1-23 SESSIONS A Wavenumber Approach for the Response of Aircraft Sidewalls to Random Pressure Fluctuations C. Maury, P. Gardonio and S.J. Elliott Institute of Sound and Vibration Research, University of Southampton, Southampton SO17 1BJ, U. K. Under cruise conditions, aircraft fuselages are exposed to high-level fluctuating pressures mainly due to aerodynamic turbulence and jet noise. In order to analyse the resulting sound transmission through aircraft sidewalls, simulations have been carried out to predict the response of a fuselage panel for two types of random pressure field, namely a diffuse acoustic field and a turbulent boundary layer excitation. An analytical model is first presented for determining the vibro-acoustic response of an aircraft panel to a large class of random excitations in the wavenumber domain. From the simulations, several trends have been pointed out: (i) both resonant and non resonant modes contribute to the panel response over a broad frequency range, below the acoustic coincidence frequency for a diffuse field excitation, and over the hydrodynamic coincidence area for a turbulent excitation; (ii) the panel structural modes are more efficiently excited with a diffuse acoustic field than with turbulent wall-pressure fluctuations. INTRODUCTION This paper is concerned with the use of a wavenumber-frequency approach to determine the vibro-acoustic response of a fuselage panel to wall-pressure fluctuations either due to a fully developed turbulent boundary layer (TBL) or to an acoustic diffuse field. The first configuration is encountered either during flight conditions or in a wind-tunnel facility. The second case is usually achieved in a transmission suite facility. The objective of this paper is to show that, in order to simulate the response of a TBL-excited panel in a laboratory, it is not sufficient to generate a reverberant field of similar level than the wall-pressure fluctuations due to a TBL. THEORY Measurements performed by Wilby et al. on the forward part of an aircraft fuselage under cruise conditions have shown that at high subsonic Mach number ( U ∞ = 225 m s ) and for frequencies above 400 Hz, the vibrations induced by the TBL wall-pressure fluctuations imparted on the fuselage shell are only correlated over a fuselage panel, i.e. the area between two adjacent frames and stringers [1]. Therefore, a simplified, but still representative model is to consider an array of uncorrelated fuselage panels [2, 3]. Each fuselage panel is modelled as a flat aluminium panel set in an infinite rigid baffle, simply-supported along its boundaries and tensioned because of the cabin pressurisation effect. The results presented here have been calculated for a panel with Young’s modulus E = 71GPa , mass density ρ s = 2700 kg / m 3 , Poisson ratio ν = 0.33 and a structural damping of 1%. The panel dimensions are l x = 200 mm , l y = 170 mm . The panel thickness is h = 2 mm . The panel is excited either by a TBL or an acoustic diffuse field. The Efimtsov model has been used for the spectrum of the TBL excitation [4]. The excitation spectrum for an acoustic diffuse field is given by spatially Fourier transforming a sinc correlation function and so, is represented in the wavenumber domain by a function which is constant within the disk k x2 + k y2 ≤ ω co and zero-valued elsewhere. Using a modal formulation, the spectrum for the sound power radiated can be written as: Φ Wr (ω ) = ρ 0 c0 ω 2 Tr Y H JYR , 2 [ ] (1) where Y stands for the modal resonance matrix, R is the radiation matrix and J is the matrix of the modal joint-acceptances an element of which is given by: 2 j mn (ω ) = ∫∫ Φ p (k;ω ) Smn (k) d 2 k l x2 l y2 Φ 0 (ω ) (2) ∞ where Φ p (k; ω ) is the excitation spectrum and th Smn (k) is the spatial Fourier transform of the mn panel mode. jmn (ω ) represents a measure of the SESSIONS coupling between the main wall-pressure fluctuations and the major sensitivity region of the mnth structural mode. 0 −5 Modal Jointfflacceptances −10 RESULTS Figure 1 represents the sound power inwardly radiated by the panel and normalised by the power spectrum of the excitation Φ 0 (ω ) for both types of excitations. From the differences observed, it is clear that, in order to simulate the sound transmission through a TBL-excited panel, it is not sufficient to use a reverberant acoustic field of similar mean-square pressure level. The sound power radiated by the fuselage panel (SPR) is 10 to 20 dB higher for an acoustic diffuse field excitation than for a TBL excitation, and the difference is accentuated when the frequency increases. −30 10Log10(ΦW /Φ0(ω)) (dB rel. 1 W/Pa2) −40 −50 −15 −20 −25 −30 −35 −40 −45 −50 0 1000 2000 3000 4000 5000 6000 Frequency (Hz) FIGURE 2. The modal joint-acceptances for the modes (1,1), (1,2), (4,3) and (7,2) of the panel; Bold curves: acoustic diffuse excitation; thin curves: TBL excitation. All these properties are summarised in Figure 3 which represents the eigenfrequencies of the panel modes against streamwise mode number, along with the acoustic and hydrodynamic coincidence lines. The proximity of the eigenfrequencies to the coincidence lines confirms the frequency ranges over which highly excited modes mainly contribute to the panel response for both kind of excitations. Diffuse field −60 r 6000 −70 5000 TBL −80 Acoustic matching lines 4000 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 Frequency (Hz) −90 6000 Frequency (Hz) 3000 2000 Hydrodynamic matching lines FIGURE 1. The sound power inwardly radiated by a fuselage panel normalised by the excitation spectrum. Bold curve: acoustic diffuse excitation; thin curve: TBL excitation. An interpretation can be given in terms of the wavenumber sensitivity of the panel structural modes to a given form of excitation, also represented by the modal joint-acceptances. They have been represented in Figure 2 for some panel modes for which the resonant frequencies have been located by a tip. For each mode, the maximum occurs at a frequency for which the spacewise variations of the mode are mostly coincident with those of the forcing field. For a TBL excitation, it can be seen that, below 2 kHz, both resonant and non-resonant highly excited modes mainly contribute to the panel response whereas, above 2 kHz, the panel modes are inefficiently excited. For a diffuse field excitation and up to 10 kHz, both resonant and nonresonant modes are efficiently excited by the reverberant field and so the SPR levels do not decrease with frequency as in the TBL case. 1000 0 0 1 2 3 4 5 6 7 8 Streamwise mode number FIGURE 3. Location of the panel eigenfrequencies in the frequency-streamwise wavenumber domain: -o-, m=1; -|-, m=2; -◊-, m=3; -∇-, m=4. REFERENCES 1. J. F. Wilby and F. L. Gloyna, J. Sound and Vib. 23(4), 443-466 (1972). 2. C. Maury, P. Gardonio and S. J. Elliott, Modelling of the flow-induced noise transmitted through a panel ISVR Technical Report no 287, 2000. 3. C. Maury, P. Gardonio and S. J. Elliott, Active control of the flow-induced noise transmitted through a panel AIAA Journal, 2001 (In press). 4. B. M. Efimtsov, Soviet Physics Acoustics 28(4), 289-292 (1982). SESSIONS Experimental characterization of the coupling between extensional and bending waves in a beam with discontinuity M.-H. Moulet, F. Gautier and J.-C. Pascal Laboratoire d’Acoustique de l’Université du Maine, UMR CNRS 6613, Avenue Olivier Messiaen, 72085 Le Mans Cedex 9, France In order to predict vibratory behavior of complex assembled structures, accurate models of the junctions between sub-structures are required. Because of the junctions, out-of-plane and in-plane vibrations are coupled. With a wave approach, this coupling can be described with a scattering matrix, whose terms are interpreted as reflection, transmission and coupling coefficients. The objective of this paper is to present a specific measurement technique allowing us to determine the scattering matrix terms. This technique is developed and validated in the simple case of a beam with a discontinuity, which induces a coupling between bending and extensional waves. INTRODUCTION In aeronautical or automotive sectors, complex assembled structures are widely used. Junctions between sub-structures couple out-of-plane and inplane vibratory displacements. In the case of beams, junctions couple bending and extensional waves. A simple analytical model of such coupling is developed and applied to periodic structures in [1]. In [2], power flow is measured in T-beams, where bending and extensional waves are coupled, without the junction being characterized. Characteristics of a joint, which constitutes a junction between two beams, are measured for bending waves only, in [3]. The aim of this paper is to characterize experimentally the coupling between bending and extensional waves in a beam with a discontinuity. First, the scattering matrix of a junction is presented in the case of a beam with a mass, which constitutes the studied discontinuity. Then, an experimental technique provides us the scattering matrix terms. Finally, experimental terms are compared with theoretical predictions. SCATTERING MATRIX OF A JUNCTION We study the case of a prismatic beam with an added mass (see figure 1), whose centre of gravity, G, of the mass is not on the neutral axis of the beam. Such a discontinuity, which is a simple junction separating two parts of the beam, couples extensional and bending waves. - F [A- C- E-] w (x) y . O. G u-(x) [B+ D+ F+] w+(x) yG u+(x) x xG = 0 [B- D- F-] [A+ C+ E+] FIGURE 1. Beam with an added mass . On both sides of the discontinuity, the longitudinal displacement u and the transverse flexural displacement w are expressed as: u ± ( x) = A ± e jke x + B ± e − jke x , (1) w± ( x) = C ±e jkb x + D±e− jkb x + E ±e−kb x + F ±ekb x , (2) where ke and kb are the wave numbers associated to longitudinal and flexural wave motion. The superscripts + and – refer to front (x>0) and back (x<0) of the discontinuity respectively. The discontinuity can be intrinsically characterized by a scattering matrix defined by: V out = SV in . (3) The amplitude vectors Vin = [A-;B+; C-; D+; E-; F+] t and t Vout = [A+;B-; C+; D-; E+; F-] represent waves coming into and coming out of the junction. In the case of a widened junction, when far field approximation is valid, the discontinuity can be ~ characterized by a reduced scattering matrix S defined ~ by [ A+ ; B − ; C + ; D − ]t = S [ A− ; B + ; C − ; D + ]t . Making use of the motion equations of the added mass, it can ~ be shown that the scattering matrix S has the theoretical following form: SESSIONS Te R ~ S = e Ceb − Ceb Re Cbe Te − Cbe − Ceb Tb Ceb Rb − Cbe Cbe Rb Tb |Te| |Re| (4) kb l |Tb| phase (rad) phase (rad) |Ceb| kf l kb l u out-of-plane vibrometer beam w(ω) FIGURE 2. Experimental set-up The beam is excited by a shaker at one end and is free at the other end. The induced vibratory field is scanned by two laser vibrometers: the first one measures flexural velocity w; the second one measures longitudinal velocity u. For each vibrometer and for each part of the beam (x<0 or x>0), two measurement points are needed in order to determine the amplitude vectors Vin and Vout. Because such a measurement provides us only 4 relations between the 6 scattering matrix terms defined in (4), two configurations of the excitation source have to be used. For the first configuration, a harmonic force is applied in the longitudinal direction (Ox). For the second configuration, the force is applied in the transverse direction (Oy). The amplitude vectors Vout and Vin being determined in the two configurations, the six scattering matrix terms can be calculated with an inversion procedure. EXPERIMENTAL RESULTS kf l kb l FIGURE 3. Magnitude of the scattering matrix terms associated to the discontinuity (continuous line: experimental, dotted line: theoretical) phase (rad) in-plane vibrometer kf l kb l |Cbe| phase (rad) u(ω) (1) w shaker e(ω) kf l kb l phase (rad) mirror added mass analyzer N 1= 1 MEASUREMENT TECHNIQUE In order to measure the scattering matrix terms, a specific experimental set-up is defined on figure 2. kb l |Rb| f ( x ) = ∑ α n sin( nπl x ) phase (rad) where Ri are reflection coefficients, Ti are transmission coefficients and Cij are coupling coefficients. Subscripts b and e are related to bending and extensional waves respectively. As shown in figure 3, the magnitudes of the experimental kscattering matrix terms have kf l the same fl frequency evolution and the same order of magnitude as the magnitudes of the theoretical terms. However, for some frequencies, discrepancies remain which can be explained by ill conditioning of the matrix to be inverted and by position error of in-plane vibrometer; because of this error, longitudinal motion induced by extensional waves can not perfectly be separated from longitudinal motion induced by bending waves. CONCLUSION AND PROSPECTS A specific experimental set-up is developed and tested in order to measure the scattering matrix terms of an added mass on a beam. The qualitative agreement between measures and theoretical prediction could be improved with a spatial filtering to separate the different waves and regularization methods to avoid the ill conditioning of the matrix to be inverted. REFERENCES -4 The studied beam is 2m long (l), has a 1.6 10 m cross-section, and is made with aluminum whose Young’s modulus is E=67.5 GPa and whose density is ρ=2700kg.m-3. The position of the added mass (M=74.10-3 kg) is defined by the coordinates yG=3.45.10-2 m and xG=0 (see figure 1). 1. D. J. Mead and Š. Markuš, J. Sound Vib, 90(1), 1-24 (1983). 2. R. P. Szwerc, C. B. Burroughs, S. A. Hambric and T. E. McDevitt, J. Acoust. Soc. Am., 107(6), 3186-3195 (2000) 3. G. Pavic, Proceedings of International Conference NOVEM 2000, 31Aug.-2Sept.2000, Lyon, France SESSIONS Diagnostics of structure vibrations in acoustic frequency range with the aid of self-organizing feature maps S. N. Baranov, L. S. Kuravsky Problem Laboratory of Mathematical Modeling attached to the Computer Center of Russian Academy of Sciences, c/o “Rusavia”, 6 Leningradskoye Shosse, 125299 Moscow, Russia. E-mail: rusav@aha.ru Failure diagnostics for the structures suffered vibrations in acoustic frequency range is presented. Normalized spectral characteristics of structure response measured in checkpoints are used as indicators to be analyzed. Self-organizing feature maps (Kohonen networks), for which output variables are not required, detect faults. Simultaneous application of different networks duplicating each other makes it possible to improve the quality of recognition. Principal component analysis is employed to reduce the number of variables under study. An aircraft panel with different combinations of attached defective dynamic suppressors is considered to demonstrate features of the approach. Tests have demonstrated high effectiveness of the presented way of recognition and showed the advantages of neural networks over cluster analysis in recognition problems. Technical diagnostics is one of the most typical spheres where neural networks are used. Under consideration here is failure diagnostics of the structures suffered vibrations in acoustic frequency range. This diagnostics is carried out on the base of spectral characteristics measured in structure checkpoints. It is supposed that neither all possible types of damages nor corresponding changes induced in the spectral characteristics may be predicted beforehand. Because of multiplicity of structure types and their applications, it is impossible to generalize considerably the problem solutions. Therefore employed for method demonstration is a specific system including a simply supported steel sandwich rectangular panel and two attached 1-degree-offreedom elastic vibration suppressors with fluid friction. Its dynamic behavior was simulated on the basis of models and methods described in paper [1]. Positions of suppressors were optimized. Wide-band random processes represented test acoustic loads. System conditions were estimated via standardized power spectral densities of accelerations in a checkpoint. (In general case, some checkpoints may be used.) Initial data to estimate such characteristics may be obtained with the aid of accelerometers. Standardizing spectral densities makes it possible to analyze only qualitative shape of structure response spectra and not to take into account the response level. The following system conditions were simulated: OK – both suppressors work properly, Only1 – suppressor 2 is defective, Only2 – suppressor 1 is defective, Panel – both suppressors are defective, Nonlin – non-linear suppressor response (shock interaction of the moving element and stopper). The first variant corresponds to normal operating mode, and the following four ones represents system damages. Since all the damages are not assumed to be known before diagnostics, it is impossible to apply ordinary neural networks with supervised learning for their detection. Self-organizing feature maps (Kohonen networks) [2-3], for which output variables are not required, may be useful in this case. Self-organizing feature maps have an output layer of radial units [4]. This layer is also called a Topological Map and, as a rule, is laid out in a 2- or 1dimension space. Starting from an initially random set of centers, the Kohonen algorithm successively tests each training case and selects the nearest (winning) radial unit center. This center and the centers of neighboring units are then updated to be more like the training case. As a result of a consequence of such corrections, some network parts are attracted to the training cases, and similar input situations activate the groups of units lying closely on the Topological Map. A self-organizing feature map is taught to “understand” input data structure in such a way and to solve the classification problem. The idea, on which this network is based, was originated by analogy with some known features of the human brain. If clustering of input data is completely or partially ascertained, semantic labels might be attached to certain units of the Topological Map. When a classification problem is solved, so called accept threshold is set. It determines the greatest distance on which recognition occurs. If the distance from the winning element to an input case is greater than this threshold, it is supposed that the network has not made any resolve. When units are labeled and SESSIONS thresholds are determined properly, the self-organizing feature map may be used as a detector of new events: it informs about input case rejection only if this case differs from all labeled radial units significantly. The given approach supports diagnostics of both known in advance and unknown damages. Simultaneous application of different networks duplicating each other makes it possible to improve the recognition quality. Frequency ranges are used as variables, and the values of normalized power spectral densities at the centers of these ranges – as cases. Thus, each complete case represents a separate power spectral density. In the test example, initial variants of neural networks with 3×3, 4×4 and 7×7 output layer dimensions were trained to recognize the states OK and Only1. Later on, the conditions Only2, Panel and Nonlin arose successively. After detection of new damage types, network training was carried out again, with corresponding labels being assigned to units of the Topological Maps1. It is convenient to estimate the recognition quality via the percentage of correctly identified situations. Two sorts of errors may occur: errors of the 1st type, when some unknown system state is identified as known one, and errors of the 2nd type, when some system state that has been known before is identified as unknown one or incorrectly. Application of networks duplicating each other2 made it possible to avoid errors of the 1st type in 99-100% of analyzed cases and errors of the 2nd type – in 98-99% of such cases. When all variants of system damages are known before, the problem is essentially simpler. One can employ traditional neural networks with supervised learning to solve it. Perceptrons were the best for the test problem: networks of 100%-recognition were revealed. Radial basis function networks turned out to be less accurate. Neural networks are, of course, not the only way to solve recognition problems. The same purposes may be achieved by means of other procedures – for example, cluster analysis that is intended for partition of an initial object set into classes following a given criterion. Comparison of both techniques makes it possible to draw the following conclusions: ♦ cluster analysis does not yield distinct criteria for classification: one cannot always distinguish qualitatively new and old-type damages – the result depends on critical distance selection; ♦ cluster analysis is less reliable than neural networks; ♦ cluster analysis needs more computer resources than neural networks. Principal components analysis and factor analysis are employed to reduce the number of input variables under study if the number of frequency ranges to be taken into account is too great and worsens network characteristics. These methods extract few latent hypothetical variables that explain approximately all the set of observed ones. As for the test problem, input data capacity might be reduced up to 2 latent variables, with the errors being avoided in 93-100% of cases. Nonlinear transforms on the base of autoassociative neural networks [5] are used in more complicated situations. As a rule, reduction of problem dimension prunes the number of neurons and, therefore, improves characteristics of network training. CONCLUSIONS 1. 2. 3. 4. REFERENCES 1. 2. 3. 4. 1 Working with a real structure, examination to reveal the failure nature must be fulfilled before new training and label assigning. Otherwise, the network will only be able to inform of an appearance of some new, unknown earlier, damage type. 2 Recognition results were selected “by a majority”. Self-organizing feature maps, whose training data do not contain output variables, make it possible to diagnose conditions of vibroacoustic systems in situations where neither all possible damage types nor corresponding changes induced in observed characteristics are not predictable beforehand. If all types of system damages are known beforehand, ordinary neural networks with supervised learning (perceptrons, radial basis function networks) may be used for diagnostics. Test results showed that neural networks were more efficient recognition tools than cluster analysis. Reduction of problem dimension (with the aid of principal components analysis, etc.) improves characteristics of network training. 5. Kuravsky, L. S., and Baranov S. N., “Selection of optimal parameters for acoustic vibration suppressors”, in Proceedings of the 7th International Conference on Recent Advances in Structural Dynamics, Southampton, United Kingdom, 2000. Kohonen, T., Biological Cybernetics 43, 59-69 (1982). Kohonen, T., “Improved versions of learning vector quantization”, in Proceedings of the International Joint Conference on Neural Networks, San Diego, USA, 1990. Haykin, S., Neural networks: a comprehensive foundation, Macmillan Publishing, New York 1994. Kramer, M. A., AIChe Journal, 37, 233-243 (1991). SESSIONS Cluster Control of Total Acoustic Power Radiated from a Planar Structure Using Smart Sensors Nobuo Tanaka Department of Mechanical Engineering, Tokyo Metropolitan Institute of Technology, Tokyo, Japan This paper deals with the minimization of total acoustic power radiated from a vibrating distributed-parameter planar structure. This paper presents a novel control method utilizing both the smart sensors and cluster actuation, thereby enabling one to suppress the total acoustic power without causing observation/control spillover problems. First, the acoustic power matrix of a planar structure is shown to be expressed in a form of a block diagonal matrix by reordering the columns and rows of the matrix, and hence the suppression of the power mode defined in each block matrix leads to the suppression of the total acoustic power. Then, cluster control consisting of cluster filtering and cluster actuation is introduced, which permits one to control each cluster independently. Finally, the experiment is conducted, demonstrating the validity of the proposed method for suppressing the total acoustic power. INTRODUCTION Minimization of the structural kinetic energy of a vibrating structure does not necessarily mean the minimization of the total acoustic power radiated from the structure. Sometimes the sound level increases while the vibration level decreases. Furthermore, the use of point sensors and actuators for controlling the structural vibration causes both the observation and control spillover problems leading to the instability of a control system. To overcome the problems, this paper presents a novel control method utilizing both smart sensors [1] and cluster actuation [2], thereby enabling the suppression of the total acoustic power without causing observation/control spillover problems. The smart sensor implemented with signal processing functions is realized by shaping PVDF film sensors with the aim to extract the acoustic power mode that is the common thread connecting directly the vibration field and acoustic field. Therefore, the suppression of the power mode leads to that of total acoustic power. The cluster actuation is a control strategy to activate the specific targeted cluster without inducing control spillover in the sense of cluster, its conceptual background being based upon cluster filtering; that is, all the structural vibration modes of a rectangular panel, for instance, may be filtered into four clusters - odd/odd modal cluster, odd/even modal cluster, even/odd modal cluster and even/even modal cluster. Among these, the odd/odd modal cluster is the greatest contributor to the total acoustic power. By introducing smart sensing and cluster control, the extraction and suppression of each cluster without causing spillover can be performed. This paper begins by discussing the acoustic power modes of a vibrating panel, presenting a method for suppressing the total acoustic power by suppressing the acoustic power mode. By employing the cluster control comprising both smart sensors and cluster actuation, the smart cluster feedback control system is constructed. Finally, the experiment is carried out, demonstrating the validity of the proposed method for suppressing the total acoustic power of a vibrating panel. ACOUSTIC POWER MODE Consider some generic structure, subject to harmonic excitation by an unspecified primary forcing function. Then the total acoustic power radiated from the structure is writ- ten as Pw = v HAv = v HQΛ ΛQ – 1 v = u HΛu (1) where v is the vector of complex modal velocity amplitudes, H is the matrix Hermitian, A is some real, symmetric, positive-definite acoustic matrix, u is the power modal amplitude vector, Q is the modal matrix and Λ is the diagonal matrix obtained by the orthonormal transformation A = QΛ ΛQ – 1 (2) As can be seen from Eq. (1), u is given by u = Q – 1v = Q Tv (3) Accordingly, the velocity at x of a structure is expressed by v(x) = Ψ T(x)v = Φ T(x)u (4) where Ψ and Φ are the vectors of the eigen function and power mode function, respectively. The relevance between these vectors are given by Φ(x) = Q TΨ(x) (5) The total acoustic power in Eq. (1) is, then, expressed as Nm Pw = Σλ i=1 i ui 2 (6) where λ is the eigenvalue of A, which is always positive and real due to the property of A, and hence Pw is always suppressed when ui, the power modal amplitude, is reduced. Furthermore, the acoustic power modes are combinations of like-index structural modes (odd/odd modes, odd/even modes, even/odd modes and even/even modes) which contribute independently to the acoustic radiation. CLUSTER CONTROL The difficulty in controlling the vibration of a distributed-parameter structure lies in the infinite number of eigenfunctions (structural modes) present in a “real” system. The approach presented in this paper for tackling this problem is “cluster control” that consists of both “cluster filtering” and “cluster actuation”. Cluster filtering places structural modes into a finite number of clusters, each clus- SESSIONS ter possessing some common property, while employing cluster actuation excites targeted clusters independently. Thus, by using both cluster filtering and actuation, cluster control avoids observation/control spillover in the sense of cluster. The end result means that groups of modes can be treated independently, and it becomes possible to preferentially direct control effort to the most bothersome clusters. This is an important result for structural acoustics, as the clusters containing volumetric modes can be preferentially dampened using a DVFB-like approach with guaranteed stability. Unlike a conventional modal control approach using point sensors and point actuators for suppression of structural modes, cluster control aims to suppress the cluster of interest, leading to suppression of all structural modes belonging to the cluster. With a view to giving further insight into the significance of cluster control, it is worthwhile using the specific example of controlling the odd/odd modal cluster. First, in order to construct the cluster control system, 4 sensors for cluster filtering and 4 actuators for cluster actuation are needed. The output signal eo/o of a cluster filtering on the odd/odd modes is described by 4 4 REFERENCES [1] N.Tanaka, S. D. Snyder and C. H. Hansen, “Distributed Parameter Modal Filtering Using Smart Sensors” Transactions of the ASME, Vol.118 pp.630-640, 1996 [2] S. D. Snyder, N. Tanaka and Y. Kikushima, “The Use of Optimally Shaped Piezo-electric Film Sensors in the Active Control of Free Field Structural Radiation, Part 1: Feedforward Control” ASME J. Vib. Acoust. Vol.117, pp.311-322, 1995 ∞ Σ w(r i,t) =iΣ= 1 kΣ= 1 ϕk(r i)η k(t) i=1 ∞ (7) Σ ϕ ok / o(r 1)ηok / o(t) k=1 Phase deg =4 Mobility Then, using Eq. (7) as a feedback control signal weighted with the feedback gain of go/o, the control force is given by ∞ 4 f(r,t) = – 4 g o / o Σ ϕ ok / o(r 1)η k (t) Σ Fiδ(r – r i) o/o (8) Introducing the force polarities, multiplying Eq. (8) by ϕi(r), integrating it over the domain D and substituting the resulting control force into the equation of motion, it is expressed in a form of a modal coordinate system as i=1 η i(t) + ωi2η i(t) ∞ = – 16g o / o Σ ϕ ok / o(r 1)η k (t) ϕ i(r 1) o/o Gain dB Gain dB Observe from Eq. (9) that the output signal from the odd/ odd modal cluster is fed back only to the odd/odd modal cluster, thereby avoiding the control/observation spillover in the sense of cluster. In exactly the same way, control of the other clusters is performed by introducing the polarities of the sensors and actuators . Figure 1 shows the experimental result for suppressing smart sensor outputs extracting odd/odd modal cluster, odd/ even modal cluster, even/odd modal cluster and even/even modal cluster by using the direct feedback control. As is seen from the figure, all the sensor outputs corresponding to the power modes are suppressed significantly without causing instability of the feedback control system. Spectrum 0 0 – 20 (b ) O dd/ o dd mo da l c l us te r – 50 – 70 (c) O d d/ev en m od al c l u ste r – 50 – 70 – 30 (d ) E ve n/od d mo da l cl us te r – 50 – 70 – 30 Gain dB if η i(t) corresponds to the odd/odd modes . –18 0 20 – 30 k=1 if η i(t) does not correspond to the odd/odd modes (9) 0 – 40 – 30 Gain dB k=1 (a) D riv i ng p o int mo b il i ty 1 80 Gain dB e o / o(t) = CONCLUSIONS A new control approach for suppressing the total acoustic power radiated from a planar structure using both distributed-parameter sensors and cluster actuation has been presented. It was found that the smart sensors based upon the shaped PVDF film may extract the targeted power mode. It was also found that the cluster actuation enables the independent excitation on the targeted cluster without causing control spillover in the sense of cluster. Experimental results demonstrate the validity of the proposed method for suppressing the total acoustic power radiated from a vibrating panel. (e) E ve n/ev en mod al cl u ste r – 50 – 70 10 1 00 Frequency 5 00 Hz FIGURE 1 Smart sensor outputs after cluster control SESSIONS Zonal and global control of vibrational structural intensity in an infinite fluid-loaded elastic plate Jungyun Won and Sabih I. Hayek Active Vibration Control laboratory Department of Engineering Science and Mechanics Penn State University, University Park, PA 16802, U.S.A. In this paper, the active control of active vibrational structural intensity (VSI) in an infinite elastic plate in contact with a heavy fluid is modeled by the Mindlin plate theory of bending. The plate is excited by a point force, which generates a vector active VSI field in the plate. Point force actuator(s) at arbitrary location(s) on the plate are utilized to minimize the VSI in a desired zone or in a large global area of the plate. The influence of the number of controller actuators and the location of these controllers relative to the source region for the minimization of the total VSI in a region is explored. INTRODUCTION Vibrational energy flow, in the form of structural intensity, is one of the most useful ways to understand the paths of vibration transmission and propagation. When a structure is coupled to a heavy fluid, the energy flows from a structure, through a coupling medium to the farfield and sometimes flows back to the structure. In this paper, the structural intensity of fluid-loaded Timoshenko-Mindlin plate with harmonic point forces are calculated for an infinite plate. The cost function to be minimized is the magnitude of structural intensity vector at chosen reference point(s). z Acoustic Media y Fc Fo the acoustic pressure on the plate, Fo is the applied force, and F is ∑y/∑r+y/r. The acoustic pressure, p, which acts on the plate is described by the scalar wave equation: where k is the acoustic wavenumber, and p is the acoustic pressure. A boundary condition is needed to couple the acoustic pressure on the surface of the plate to its vibration: ∂p ∂z ψb = FORMULATION ρ sh3 2 ω ) Φ − G ′h ∇ 2 w = 0 (1) 12 δ (r ) − Pa = − ρ s h ω 2 w G ′h ( ∇ 2 w + Φ ) + Fo 2π r where D is the bending stiffness, G´ is the adjusted shear modulus, rs is the density of the plate, h is the thickness of the plate, w is the forcing frequency, Pa is (3) = ρ oω 2 w ρ ⋅ f1 ( ρ ) H 0(1) (rkρ ) w Fb dρ = ε h 2 −∫∞ f ( ρ ) − ρ f 1 ( ) 2 ρ 2 −1 (4) ψ Fb ∞ ρ 2 H1(1) (rkρ ) dρ = ε kh 2 −∫∞ f ( ρ ) − ρ f 1 ( ) 2 ρ 2 −1 (5) ∞ wb = FIGURE 1. Physical geometry of plate ( D ∇ 2 − G ′h + z =0 To solve these equations, it is necessary to apply Hankel transform. After solving the transformed equations simultaneously with the boundary condition and applying inverse Hankel transform, the displacement and shear deformation can be obtained. The normalized solutions are given by: x Two simultaneous time-harmonic TimoshenkoMindlin plate equations of motion describing the motion of the plate in terms of the displacement, w and the shear deformation angle, y are [1]: (2) (∇ 2 + k 2 ) p = 0 where f1 ( x) = 1 + K s 2 Ω 2 x 2 − K s 2 K d 2 Ω 2 2 2 f 2 ( x) = ( x 2 − K s )( x 2 − K d ) − 1 Ω 2 and Ω = ω ω c . wc is a classical coincidence frequency. Time-averaged structural intensity is expressed as: SIb = SI 1 6(1−υ) dwb * = − Re[κ 2 +ψb ) ⋅ iΩwb ] ( 3 2 Dk h ωc 2 (kh)2 dkr (6) ψ dψ 1 * − Re[( b +ν b ) ⋅ iΩψb ] dkr kr 2 SESSIONS 6pi -5pi -4pi 2 0 4 2 pi 2pi 3pi 4pi 5pi 6pi -pi 0 X*kf pi 2pi 3pi -2 -2 2 -2 0 3pi 4pi 5pi 6pi FIGURE 2. VSI and VSI reduction in dB : Controller at (p, 0), Ref. Point at (4p, 0) If one uses two controllers and one reference point, the resultant minimization of VSI is comparable to one controller/reference point. In this case, the value of the optimal control force is not unique, and the control algorithm gives the closest value from the starting point -2pi -2pi -3pi 2 pi 2pi 3pi 4pi 5pi 6pi -4 -10 -8 -12 8 -4 -12 -10 -8 -6 -4 -2 02 10 4 4 126 12412268024 8 8 10 -2 -4 68 4 02 0 2 -2 -8 -6 -5pi 0 X*kf -12 -3 -10 -1-800 -4 -2 -8 -6 -4pi -6pi -5pi -4pi -3pi -2pi -pi 10 -2 0 2 -6 22420 -2 -4 -2 0 -8 10 -pi -6 -4 0 -2 -4 0 Y*kf 0 2 4 0 -pi -4 -8 -2 -4 -2 40 1280 -22 -24 -26 -28 -30 -2 -4 -2 0 -4 2pi 6pi -6pi -6pi -5pi -4pi -3pi -2pi -pi -4 0 X*kf pi -6 pi 5pi -6 0 X*kf 4pi 30 2286 Y*kf pi 0 -6 -4 -2 -4 2pi 2 4 2pi 186 0 -2 3pi -2 0 3pi 10 -6 -4 -6pi -6pi -5pi -4pi -3pi -2pi -pi 3pi 2 Y*kf 4pi -6 Y*kf 5pi 4pi -6 2pi 3pi 4pi 5pi 6pi 6pi 5pi -6 2 2 2pi -4 4 pi -28 -12 -14 -16 -18 -20 -22 -24 -26 -6 6 0 X*kf -8 pi 6pi -6pi 30 2826 24 22 20 18 16 14 12 10 8 4 6 -pi -6 0 X*kf 6pi -5pi 0 -6pi -5pi -4pi -3pi -2pi -pi 5pi 30 281428 12 26 64 12 10 8 -5pi -6pi 4pi pi 2 -4 -4pi -5pi 3pi 2 0 -4pi 2pi 64 -3pi 26 2224 86 2 -2 -2pi -3pi 10 8 64 2 0 0 -2pi 2 4 -pi 0 -60 -8 -4 8 6 4 2 -2 -1 0 -pi pi 10 8 2 0 -4 6 Fig. 4 shows the results for two controllers at (≤p, 0) and two reference points at (≤4p, 0). It shows the minimized VSI field and the significant reductions are attained in the region near the entire x-axis. 68 1012 1416 18 24 60 114 121 0 2 64 8 -4 -2 pi pi 0 X*kf -4 10 12 24 4 FIGURE 4. VSI and VSI reduction in dB : Controllers at (≤p, 0), Ref. Points at (≤4p, 0) 4 2pi 2pi 26 24 22 20 18 16 14 12 -5pi -4pi 0 26 -4 0 2 4 16 14 12 24 0 28 10 -2 30 28 Y*kf 4 10 8 6 -6pi -6pi -5pi -4pi -3pi -2pi 28 30 -4pi -6pi 24 320 26 8 86 -5pi -4 -2 -3pi 108 2 0 -4pi 0 16 14 12 4 -2pi 0 8 10 2 114 16 18 20 22 0 68 1 24 28 26 22 24 20 18 16 14 12 -pi -3pi 2830 26 24 -2 -pi -2pi 10 8 6 26 0 18 20 22 0 0 pi pi 4 18 20 22 2pi 2pi 6 3pi 1412 18 16 20 24 22 6 -4 4 6 3pi 4pi 2 10 8 4pi -6pi -5pi -4pi -3pi -2pi -pi 0 3pi 6pi 0 2 0 5pi 5pi -2 4pi 6 8 3pi 4pi 4 5pi 6pi 6pi -4 5pi 2 4pi When two controllers are used to control the VSI at two reference points, the resulting reduction in VSI is more global than two previous cases. -3pi 5pi 10 4 -6pi -6pi -5pi -4pi -3pi -2pi 2 0 X*kf FIGURE 3. VSI and VSI reduction in dB : Controllers at (≤p, 0), Ref. Point at (4p, 0) 6pi 6pi 2 4 30 12 16 18 20 26 30 6 -4 -6pi -5pi -4pi -3pi -2pi -pi RESULTS The control of VSI depends on the relative location of the source and the controller. By varying the location of the controller relative to the source, it was found that efficient reductions were attained at the locations which are multiples of half fluid-loaded structural wavelengths. The closer the controller is to the source, the more global is the reduction in VSI. Fig. 2 shows the VSI and VSI reduction in dB with the controller at (p, 0) and the reference point at (4p, 0), and it shows VSI reduction in the region near the xaxis. 8 6 1210 8 -2 -5pi 30 26 24 22 20 141618 12 0 -6pi 28 6 81 0 4 -4 4 2 -3pi 142224 -22 -24 -26 -28 -30 -8 -402468 -2-6 -1 -2 -2 121086 0 -4pi 30 28 26 24 22 2018 1614 2 -2pi 1214 16 18 2022 24 26 6 8 10 4 0 -3pi 0 24 26 28 -2 -2 Y*kf 0 -pi -4 164 28 0 -pi -2pi 12108 6 122802 30 pi 24 2pi 0 -2 -2 3pi 8 262 1 164 3pi 18 20 2pi 22 24 pi 4pi 6 8 0 4pi 5pi Y*kf 4 6 5pi 108 12 6pi -4 2 The integration paths and branch cuts for the integrations above have been used by several authors such as in [1]. After using the paths, several integrations occur during the calculation of wb and yb, and can be calculated using numerical techniques. For the horizontal axis of plots, normalized distance, kf r, which is normalized to fluid-loaded structural wavelength, is used. Hence, 2π in the horizontal axis means a full fluid-loaded structural wavelength. For numerical calculation, parameters of a steel plate in water are used. The excitation frequency is W=0.2. One/two point-force controllers are used to control the structural intensity at one/two reference points The structural intensity at the reference point is a second order polynomial in terms of a real and an imaginary part of the control force(s) when one controller is used. To assure that the VSI is minimized, the square of the magnitude of the VSI vector is minimized. The steepest gradient search method is used to find the minimum of the function, which is a numerical optimization technique. which is zero in this paper. Fig. 3 is the result of two controllers at (≤p, 0) and one reference point at (4p, 0). Y*kf NUMERICAL CALCULATION 2pi 3pi 4pi 5pi 6pi FIGURE 5. VSI and VSI reduction in dB : Controllers at (p, 0), (0, p), Ref. Point at (4p, 0) In fig, 5, two controllers of the same forces are located at (p, 0) and (0, p) with two reference points at (4p, 0) and (0, 4p). The results are symmetric with respect to x-y diagonal. The VSI was reduced near the positive x and y-axes. REFERENCES 1. C. Seren and S. I. Hayek, J. Acoustic Soc. Am. 86(1), 195-209 (1989) SESSIONS Multiple-Aspect Acoustic Scattering from Fluid-Filled Cylinders Measured at Sea A. Tesei SACLANT Undersea Research Centre, V.le S. Bartolomeo 400, 19138 La Spezia, Italy. tesei@saclantc.nato.int Multiple-aspect backscattering is studied at low-intermediate frequency from fluid-loaded, fluid-filled, thin-walled cylinders. The dynamics of predicted elastic waves are described in the frequency-aspect domain by extending to liquid-filled shells the thinshell theory originally developed for air-filled shells. The models are validated by at-sea measurements of a water-filled cylinder. INTRODUCTION Multi-aspect elastic backscattering from fluid-filled, thin-walled, cylindrical shells is studied in the ka range (1,20). A canonical water-loaded, flat-ended, thinwalled, steel cylinder is considered. In previous work [1][2] the elastic contribution from a water-filled shell insonified at broadside was predicted and in [3] theoretical considerations were validated on at-sea data. This work extends the analysis to the shell response at oblique incidence. The aspect-dependent dynamics of the expected elastic waves are formalized by extending thin-shell theory to liquid-filled shells. The proposed models have been validated by the analysis of at-sea measurements. THEORETICAL CONSIDERATIONS A number of elastic wave families were proven [1] to be generated by infinite cylindrical shells insonified at broadside. For ka∈(1,20), in the case of a thin-shell steel cylinder with relative thickness h=d/a=0.024 (d being the wall thickness and a the radius), the wave families common to both air-filled and liquid-filled shells are the outer-fluid-borne A Scholte-Stoneley and shell-borne A0 and S0 Lamb-type waves [1]. If the shell is liquid-filled, the filling causes the generation of a number of additional elastic waves (Fig. 1): the inner and outer Scholte-Stoneley S wave, the inner ScholteStoneley A wave and multiple periodical internal bounces rl’ of the first, second, and higher orders [2]. The frequency modes f nl of each wave l (n is the modal order) can be expressed in terms of the properties of target and inner and outer media [3]. FIGURE 1. Wave travel paths at normal incidence. FIGURE 2. (a) Wave number decomposition at oblique incidence. (b) Path of a generic helical wave. According to thin-shell theory [1][4][5] developed for air-filled shells, at oblique incidence the elastic response of the cylinder is mainly characterized by shell-borne waves (Lamb-type S0 and shear S waves). Outer-fluid borne Scholte-Stoneley A wave modes are predicted at low frequency. At oblique incidence (Fig. 2) the wave travel path around the shell from circular is expected to be helical, according to the decomposition of the wave number k along radial (θ) and axial (z) directions [1][4][6]: k 2 = k z2 + kθ2 , k = 2πf nl 2πf nl , kz = sin α c ext c ph (1) where cext is the sound speed of the outer medium and α the incident angle. As phase matching of the helical wave occurs when kθ=2πn/P, where P is the perimeter of the projection of wave travel path on the radial plane [6] (Fig. 1), from Eq. (1) we obtain: ( sin 2 α = c ext / c ph )2 1 − (n c ph )2 / ( f nl P )2 (2) If the wave is (quasi-)nondispersive (e.g., S0, shear and S waves) its phase speed cph (approximately) tends to a constant sound speed that depends on the wave nature and the physical properties of the medium (e.g., for shear waves cph→cs). Hence the mode loci of (quasi-)nondispersive waves are obtained by substituting cph in Eq. (2) with the appropriate speed. Each locus is a quasi-parabolic curve centred at broadside with an asymptote at its critical angle [4]. SESSIONS Thin-shell theory is extended here to liquid-filled shells off broadside. The dynamics of inner-fluid-borne S wave and periodical internal bounces rl’ are expected to be similar to those described in Eq. (2) for air-filled shell waves. However these waves have a travel speed tending to the sound speed of the inner medium only when they travel along the shell walls. Then, under the assumption of thin shell, they propagate through the shell walls and are expected to re-radiate outside the shell with a phase speed tending to either the shear or the compressional speed of the shell itself, i.e., with cph→cp|s. Under this hypothesis the model of mode loci of inner Scholte-Stoneley S and periodical internal bounces is obtained by extending the procedure described above for empty thin-shells: ( ) ( ) 2 2 sin 2 α = cext / c p|s 1 − (n cin )2 / f nl P FIGURE 4. Identification of inner-fluid-borne waves. (3) where P = 2π (a − d ) for the inner-fluid-borne surface wave S and P = 4l ' sin (π /(2l ) )(a − d ) for a periodical internal reflection rl’ (l’=1,2,…). EXPERIMENTAL RESULTS Free-field measurements were performed in a basin from a water-filled, 0.25m radius x 2m, 6mm thinwalled steel cylinder [3]. The object, suspended underwater from a floating frame, could rotate around its minor axis while insonified by a parametric sonar with a Ricker 8kHz pulse. Backscattered response was received by a hydrophone in monostatic configuration. The multiple-aspect spectral representation of the data (Fig. 3) shows a regular, patterned texture generated by a number of elastic wave mode loci, the presence of which is a first indication that the insonified object is a shell with circular cross-section. A more detailed analysis allows the identification of mode loci belonging to S0 Lamb-type, shear and outer-fluid A waves in accordance with thin-shell theory (Fig. 3). FIGURE 5. Model-based interpretation of the response of a water-filled cylinder insonified at broadside. Additional shear and compressional wave mode loci are identified as belonging to inner-fluid-borne surface waves and periodical bounces (Fig. 4) from the resonance mode identification of the response at broadside (Fig. 5). The good agreement between theory and experiment encourages the extension of the analysis to cylinders lying proud on or buried in the sea bottom and to more complex objects. ACKNOLEDGMENTS The author is grateful to A. Maguer, B. Zerr and W. Fox, who was the scientist in charge of the trial, for the fruitful scientific discussions and the continuous support. Many thanks go to J. Fawcett for his fundamental contribution to target scattering modeling. REFERENCES FIGURE 3. Elastic wave identification of water-filled shell response from thin-shell theory. 1. N.D. Veksler, Resonance Acoustic Spectroscopy, Springer Verlag, Berlin, 1993. 2. J.-P. Sessarego, J. Sageloli, C. Gazanhes, H. Überall, J. Acoust. Soc. Am. 101 (1), 135-142 (1997). 3. A. Tesei, W.L.J. Fox, A. Maguer, A. Løvik, J. Acoust. Soc. Am. 103, 2813 (1998). 4. M. L. Rumerman, J. Acoust. Soc. Am. 93 (1), 55-65 (1993). 5. J.A. Fawcett, SACLANTCEN Internal Report 273 (1998). 6. C.N. Corrado, Ph.D. Thesis, MIT, Cambridge (1993). SESSIONS Vibro-Acoustic Wave Transmission and Reflection in a Hose-Pipe System Yun-Fan Hwang Applied Research Laboratory, Penn State University, P.O. Box 30, State College, Pa 16804-0030, U.S.A. An analysis of sound and vibratory transmission and reflection losses in a fluid filled planar piping system which consists of straight pipe segments, flexible hose, elbows and/or U-joints is discussed in this paper. Calculation of the transmission losses for a hose-pipe system has been widely discussed in the literature. The reflection characteristics of a hose-pipe system, however, have not received proper attention. In this paper, we calculate the reflection and transmission coefficients of a piping system simultaneously. The numerical example shows that very large pressure and bending wave transmission losses occurred in a hosepipe system are actually caused by reflection rather by attenuation or dissipation by the system. INTRODUCTION Analyses of sound and vibratory transmission and reflection losses in a fluid filled planar piping system which consists of straight pipe segments, flexible hose, elbows and/or U-joints are concerned of this paper. In the approach defined in Ref. [1,2] for a planar (in y-z plane) piping system, the axial force (Fz), transverse force (Fy), moment (Mx), displacements (Uy, Uz), rotations (ψx), fluid pressure (p) and its particle velocity (v) at both ends of a pipe or hose segment are related by a transfer matrix. Waves in fluid are treated as plane waves, which are dynamically coupled with the pipe or hose wall. The fluid wave speeds are calculated using the modified bulk modulus resulting from the Poisson coupling with the pipe wall. Timoshenko beam theory is use to represent the bending vibration of the pipe. An elbow or U-joint is handled by representing it as several straight segments. The axis for any of the segments is rotated with respect to that of the previous segment with an incremental angle. The sum of all of the incremental angles is equal to the turning angle of the elbow or U-joint (90o or 180o, respectively). Accordingly, the transfer matrix for each of these segments must be multiplied by a point matrix (to correct the rotation of the axis) so that it will be dynamically comparable with all other segments. The relationship between the dynamic parameters at two ends of a piping assembly is represented by the system transfer matrix, T, which is obtained through a successive multiplication of the transfer matrices representing the variouspipe segments, from up to down stream, of the piping system. That is, S d = TSu (1) where S=[Uz, p, v, Fz,Uy,Ψx, Mx, Fy]T, the superscript T denotes the transpose of a matrix, and Su and Sd denote S at the up and down stream terminations of the piping system, respectively. From this, the response of the entire system to an external excitation can be solved by specifying the boundary conditions and excitation. In order to determine the transmission and reflection losses, the system must be connected to a semi-infinite pipe at its up stream where an incident wave is launched toward the piping assembly, and there is a semi-infinite pipe at the down stream where the transmitted waves will propagate away from the system. For the eiωt time harmonic function, the incident, reflected and transmitted waves may be defined as [ Fz , p, Fy ]inc = [ A1 e −iλ1u z [ Fz , p, Fy ]rfl = [ B1 e B6 e , A2 e −iλ1u z − λu3 z −iλu2 z , B2 e , A3e −iλu2 z −iλu4 z , B5e ], −iλu4 z (2) , ], (3) −iλd z −iλd z −iλd z [ Fz , p, Fy ]trn = [ B3 e 1 , B4 e 2 , B7 e 3 , B8e− λ3 z ] , d (4) respectively where λi accompanied with superscripts u and d are the up and down stream propagating wavenumbers for various types of waves defined in Ref. [1]. With some algebraic manipulation, the reflection and transmission constants, B = [B1 B2 B3 B4 B5 B6 B7 B8 ]T (5) can be determined by solving B = ( D − TC ) −1 TE (6) where E = [iλ1u ( ρ p Apω 2 ) −1 A1 , A2 ,−iλu2 ( ρ f ω 2 ) −1 A2 , − A1 , iλu4 (( ρ p Ap + ρ f A f )ω 2 ) −1 A3 , ( ρ p I p ω 2 − ( λu4 ) 2 EI p ) −1 A3 , iλu4 EI p ( ρ p I pω 2 − (λu4 ) 2 EI p ) −1 A3 ,− A3 ]T . (7) C and D are square matrices where SESSIONS C11 = − E1 , C 22 = 1, C 32 = − E 3 , C 41 = −1, C 55 = − E 5 , C56 = −λu3 (( ρ p Ap + ρ f A f )ω 2 ) −1 , C65 = ( ρ p I pω 2 − ( λu4 ) 2 EI p ) −1 , C66 = (( ρ p Ap + ρ f A f )ω 2 + (λ3u ) 2 EI p ) −1 , C75 = −iλu4 EI p C65 , C76 = −λ3u EI p C66 , C85 = −1, C86 = −1 ; D13 = iλ1d ( ρ p Apω 2 ) −1 , D24 = 1, D34 = −iλd2 ( ρ f ω 2 ) −1 , D43 = −1, D57 = iλd4 (( ρ p Ap + ρ f A f )ω 2 ) −1 , calculated transmission and refection coefficients of the pressure and bending waves. It is shown that both pressure and bending waves are effectively transmitted to the other end only at low frequencies. At higher frequencies, the transmission losses are high because most of incident wave energies are reflected. This analysis indicates that the large transmission losses of the hose-pipe system shown in Fig.1 should attributed mainly to the system reflection of the incident waves, rather than to the system dissipation. In other words, the incident wave energy in the upstream (left side of the system) remains in the upstream. In this case a dissipative element installed in the upstream of the hose-pipe system may be necessary to reduce the reflected energy. D58 = λ3d (( ρ p Ap + ρ f A f )ω 2 ) −1 , D67 = ( ρ p I pω 2 − ( λ4d ) 2 EI p ) −1 , Reference D68 = ( ρ p I pω 2 + ( λ3d ) 2 EI p ) −1 , 1. M.W. Lesmez, D.C. Wiggert, and F.J. Hatfield, J. Fluid Eng. 112, 311-318 (1990). 2. M.L. Munjal and P.T. Thaeani, J. Acoust. So. Am. 101(5), Pt.1, 2524-2535 (1997). . D77 = iλd4 EI p D67 , D78 = λ3d EI p D68 , D87 = −1, D78 = −1 , and where all of the Cij and Dij not shown above are zeros and all nomenclatures not defined above are same as that of Ref. [1]. From Eq. (2) to (4), it is obvious that lB4/A2l2 and lB2/A2l2 are the fluid pressure wave transmission and reflection coefficient, respectively. The transmission and reflection coefficients for other types of wave can be defined in the similar manner. A hose-pipe system shown in Fig. 1 consists of six segments. Segments 1, 3 and 6 are steel pipes, each of them is 20 cm long. Segments 2 and 4 are 90o elbows and the radius of the bend for both is 15 cm. Segment 5 is a rubber hose which is 50 cm long. For simplicity, all pipe and hose segments are having the same outside diameter (3 cm) and the same wall thickness (3 mm). The Young’s modulus, Poisson ratio, and specific gravity of the hose are assumed to be 2x108 Pa, 0.48 and 1.5, respectively. Each of the elbows is approximated by four straight segments. Both the left and right terminations of the assembly were assumed to fit with semi-infinite steel pipes of the same diameter and thickness. The fluid in the piping system is water. A unit incident pressure or bending wave inside the left semi-infinite pipe is assumed to propagate toward the assembly. The incident wave will be transmitted to the right semi-infinite pipe in terms of fluid pressure, bending and longitudinal waves. Fig. 2 shows the 4 z 5 6 3 2 1 FIGURE 1. A simple pipe/hose assembly where the arrows indicate the direction of wave propagation. 10 10 10 COEFFICIENTS NUMERICAL EXAMPLES y 10 10 10 10 10 2 PRESSURE TRANS PRESSURE REFLT BENDING TRANS BENDING REFLT 1 0 -1 -2 -3 -4 -5 0 200 400 600 FREQUENCY, HZ 800 1000 FIGURE 2. Predicted transmission and reflection coefficients of a pipe/hose assembly (Fig. 1). SESSIONS Coherence of Seismoacoustic Waves from Explosive Sources V.S.Averbakha, N.N.Gerdyukovb, I.N.Didenkulova, V.I.Dudinb, V.N.Erunovb, S.A.Lobastovb, A.P.Marysheva, S.A.Novikovb, A.A.Stromkova, V.I.Talanova a Institute of Applied Physics, Nizhny Novgorod, Russia Russian Federal Nuclear Center-VNIIEF, Sarov, Russia b Explosive sources are widely used for excitation of seismic waves. It is usually assumed that higher charge of the source allows one to ensure deeper seismoprospecting. However the main problem in this case is high energy losses in the shock wave resulting in weak dependence of seismic waves intensity measured far from explosive charge source. Coherent methods in seismoprospecting have attracted increasing attention during last years. Coherent signal processing permits an increase in the signal-to-noise ratio. Special electromagnetic sources were mainly used in coherent seismoacoustics so far. The present paper describes a new direction in coherent seismoacoustics that is based on the use of controllable explosive sources. Experimental investigations were performed with the system of synchronized smallcharge explosive sources. The results show high coherence of seismoacoustic waves generated by explosive sources, and the possibility of using it for seismoacoustic prospecting INTRODUCTION EXPERIMENT AND RESULTS Explosive sources have been used for excitation of seismic waves since long for seeking resources and studying subsoil assets [1]. In this case, the use of high-power sources allows us to increase the seismic sounding depth. The main obstacle in this way is weak dependence of the amplitude and generated seismic waves on the charge power Q [1]. It is known also that only from 0.01 to 1% of the total explosion energy is transferred to the seismic wave energy [2]. Moreover, to examine subsoil assets, it is desirable to use directed seismic waves. Coherent methods of seismic diagnostics have been recently developed [3,4]. They are based on the formation of seismic fields with controllable spacetime structure, which allows us to perform coherent processing of signals scattered from soil inhomogeneities. Such a processing allows us to receive rather weak scattered signals and, correspondingly, increase the sounding depth. In this case, it is also possible to substantially increase the working frequency range up to several hundred Hz, which improves the spatial resolution. Until recently, the studies in the field of coherent high-frequency seismoacoustics were conducted with the help of coherent electromagnetic-type sources developed at the Institute of Applied Physics [3,4]. In this paper we consider a new direction in coherent seismoacoustics related to the possibility of using controlled low-power explosive sources developed in VNIIEF. The scheme of the experiments was as follows. Explosive sources (with power from 1 to 50 g) were located at a depth of 30 cm in sandstone-loam soil. The charges were exploded by an electric detonator ensuring time synchronization of about 1 µs. Seismic receivers were located along the radial line with respect to the charge. The interval between the receivers was 0.5 m. Therefore, the receiving system was a linear antenna. The distance from the charge to the first seismic receiver was 1 m. Experimental series with "point" charges 1, 2, and 8 g of explosive were carried out. An explosion is known to generate different-type seismoacoustic waves such as longitudinal, shear, surface, and deeprunning [1]. These waves propagate at different velocities and can also be reflected from the layered soil structure and scattered from various inhomogeneities. All this results in the delay of explosion signal. It was observed that for small time intervals from the explosion time, the spectrum is up to 400 Hz. At later times, the spectrum narrows mainly due to a decrease in the upper frequency boundary, and then a region of dominating frequencies is gradually formed in the interval approximately from 25 to 50 Hz. To investigate the coherence of seismoacoustic signals from different explosions we subsequently exploded three point charges with mass 1 g of explosive under the same conditions. The results are shown in Fig. 1. The correlation factor values between signals are not worse than 0.97 for different receivers. SESSIONS 500 5.5 400 5 300 4.5 Distance (m) Amplitude 200 100 0 -100 4 3.5 3 2.5 2 -200 1.5 -300 1 -400 0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0 Time delay (s) FIGURE 1. Seismic signals from 3 consequent explosions. Such a high correlation is due to the explosion synchronization system used. The obtained data allow us to speak of the possibility of coherent processing of signals from different explosions. Consider this issue in more detail. In accordance with the law of explosion identity [1,2] we have u=k(Qp)n, where u is the particle velocity in the seismic wave field, k is the proportionality factor, n is the exponent of power, Qp=Q1/3/r is the reduced charge mass, Q is the charge value, and r is the distance from the explosion point to the observation point. For body seismic waves we have n=1 [1]. From the identity law it is obvious that the use of a series of small explosions is more efficient than simultaneous blasting of a charge with total power of the whole series. Assuming that we conducted N similar experiments with charges Q let us determine the effective value of the charge Qeff required for generating a signal of the value coinciding with that of the resulting signal obtained from coherent addition of signals in a series of N explosions. From the identity law it follows that after the coherent addition of signals from N explosions the resulting value of u is equivalent to one experiment with the explosive mass Qeff=N3Q. Therefore, coherent addition of signals is by a factor of Qeff/NQ=N2 more efficient than the blasting of one charge with total power NQ that is similar to that in a series of N explosions. For example, adding the signals from ten subsequent experiments with Q =10 g, we obtain the same result as that for one experiment with Q =10 kg of explosive. The possibility of coherent addition of signals from explosion sources allows us also to form controlled seismic fields, in particular, directed seismic beams. Due to this, we can further increase the seismic sounding depth. Such an experiment was conducted with a plane square-shaped distributed explosive source 0.5 0.5 m2⋅ with total charge value 50 g of explosive. In Fig. 2 we show halftone image of signals 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 Time delay (s) FIGURE 2. Halftone image of seismic signals received by ten aligned receivers separated by 0.5 m (the vertical axis) as a function of time delay (the horizontal axis). received by ten receivers as a function of delay time counted off from the charge blasting time. In the figure, one can see the equiphase lines (hodographs) of waves. The first hodograph corresponds to the Rayleigh wave propagating at a velocity of 270 m/sec, and the second hodograph corresponds to a body wave refracted at the boundary between dry and moisturesaturated sand at the 2.5 m depth. CONCLUSION Along with the use of small-power charges that do not significantly interfere with the soil structure, the high degree of the time synchronization of explosion allow us to ensure high coherence of seismic signals. The high coherence of signals allows us to perform accumulation of seismic waves from a subsequent series of similar explosions and form seismic fields with the required space-time structure. All this allows us to substantially increase the efficiency of seismic sounding, reduce the total power of used charges, and improve safety during explosive experiments. ACKNOWLEDGMENTS This work was supported by ISTC (project 1369) and in part by RFBR (99-02-16957, 00-15-96741). REFERENCES 1. 2. 3. 4. Sadovsky M.A. Selected Papers. Geophysics and Physics of Explosion. Nauka, Moscow, 1999, pp.20-80. Kuzmenko A.A., et al. Seismic action of explosions in rocks. Nedra, Moscow, 1990, pp.325-390. Averbakh V.S., et al. Acoustical Physics, 44, 725-732 (1998). Averbakh V.S., et al. Acoustical Physics, 47, 435-441 (2001). SESSIONS Experimental Determination of SEA Parameters for the Prediction of Noise Pressure Level in Cylindrical Cavity A. Botteona, S. Gergesa a Vibrations and Acoustics Lab, Mechanical Engineering Department, UFSC, Florianópolis, Brazil Statistical Energy Analysis – SEA is a noise and vibration prediction method effective at mid and high frequencies. Some acoustic and structural systems have its response to random excitation extended to the high frequency region, for instance the response of aircraft fuselage to the propulsion system and excitation by turbulent boundary layer. The necessary parameters to the application of the method are modal density, damping loss factor, power input and coupling loss factor, they describe the dynamic characteristics of structural components and its connections. They can be measured or calculated using theoretical derivation. This paper presents the experimental determination of modal density and damping loss factor, in a 0.95 diameter, 1.20 height aluminum structure closed in both sides, that simulates one section of a airplane fuselage and its inner cavity. The results are compared to theoretical prediction. The validation of this simplified model is the first step to obtain a reliable complex model for the prediction of SPL inside airplanes. The complex model will be used as a prediction tool at design stage in the development of new aircraft. INTRODUCTION In SEA a system is divided in subsystems, each of these represents a mechanism to store energy. SEA is based on a power balance. In the steady state, the power input in a subsystem is either dissipated internally or transmitted to another subsystem. The dissipated power is proportional to the energy level of the subsystem and its damping factor. The power flux between the subsystems is proportional to the coupling loss factor and to the difference of the averaged modal energy level. Through this power balance its possible to obtain a system of linear equations, in which the level of energy of each system is the unknown variable. In this experiment the model conceived composes of 3 subsystems, cylindrical shell, inner cavity and surrounding air. The test set up, suspended cylinder and detail of impedance head, rod and shaker is shown in figure 1. FIGURE 1 MODAL DENSITY Modal density is defined as n( f ) = c N Df , where fc Df is the central frequency and NDf is the number of modes within Df. When a structure is excited by a broadband white noise, a very large number of modes appear in the high frequency region, making difficult to compute the response of each mode individually, therefore the approach used is based on methods that consider averaged values of the mobility. The concept of point mobility represents the capacity a component has to absorb power. The modal density was determined according to equation (1). n (f ) = n(f) MA 1 f2 4 * MA * Re < Y > Df òf 1 (1) Modes/Hz total mass of the cylinder Re<Y> real part of the drive point mobility The drive point mobility was measured using single point excitation by shaker and a impedance head. Figure 1(right). The results are presented (Figure 2) in constant bands of 100 Hz. These are compared to Szechenyi [1] semi-empiric results, and data given by commercial software. Damping has been added to the structure in order to avoid negative components in the real part of the drive point mobility The added mass existing between the transducer and the structure has to be compensated by according to following SESSIONS equation Yreal = Ymeasured 1 - iwM added Ymeasured (2) [3] loss was not considered because in SEA it is related to the coupling loss factor. Impedance head was used because the phase difference between force transducer and accelerometer can cause errors in the results. The drive point mobility was measured in three different points and the acceleration was measured in six different locations for space averaging. The obtained data, presented in Figure 3, tend to values between 0.0005 and 0.002 for 1 < f < 4.5 kHz. Clarkson obtained a constant value of 0.001 for non stiffened cylinders. FIGURE 2: modal density ( constant bands 100Hz) ****** experimental data ++++ data given by commercial sofwae AutoSEA .…….. results for the equivalent flat plate when f® ¥ -------- results calculated by Szechenyi formula For cylindrical shells, just before the ring frequency, high values of the modal density are observed due the grouping of structural ressonances. DAMPING LOSS FACTOR Damping is defined as being the ratio between the energy dissipated per cycle and the maximum vibratory energy. h= Ediss/ cicle Wdiss Wdiss (3) = = 2 2 2pM < v > 2pf .M < v > w.M < v2 > The damping loss factor has specific values for each mode. However according to Lyon [4] an average value per band is required. The structural damping is responsible for the dissipation for the system vibratory energy. As it varies depending on the material and geometry it is normally determined experimentally. There are two methods to determine structural damping, power injection method and decay method. In this work we have used the first., the former is the subject of the next tests. 2 Win = Frms Re < Yreal > (4) 1 f 2 Win .w.df h(f ) = Df òf 1 MA < a 2 > (5) FIGURE 3: Damping Loss Factor - Win method It can be inferred that indirect methods can be used successfully to determine modal density and damping loss factors. These work continues with the determination of the input power and the measurement of the coupling loss factor. Having the data we are able to calculate de energy levels in each subsystem (solving a system of linear equation), and from that derive the SPL in the in the inner cavity, which is our final objective. ACKNOWLEDGMENTS This work is supported by CNPq – Brazilian Research and Development Agency. REFERENCES 1. 2. h(f) band averaged damping Frms complex force amplitude measured in the impedance head <a> spatial averaged acceleration It was assumed that Win = Wdiss, i.e all the dissipated energy was through structural damping. Radiation 3. 4. 5. E. Szechenyi,. 1971, Modal densities and radiation efficiencies of unstiffened cylinders using statistical methods Journal of Sound and Vibration., 1971, 19, pp 65-81. B.L.Clarkson and R.J. Pope, Experimental determination of modal densities and loss factors of flat plates and cylinder. Journal of Sound and Vibration, 1981 77(4), pp 535-549.. K.T. Brown and M.P. Norton. Some comments on the experimental determination of modal densities and loss factors for Statistical Energy Analysis applications. Journal of Sound and Vibration, 1985, 102(4), pp. 588-594. R.H.Lyon, Statistical energy analysis of dynamical systems: Theory and Application, MIT Press, 1975 P. R Keswick and M. P Norton. A Comparison of Modal Densities Measurements. Applied acoustics, 1987, 20 137-153. SESSIONS Linear and Nonlinear Investigating of Concrete J.C. Lacouture, P. Johnson and F. Cohen Tenoudji Laboratoire Environnement et Développement, Université D. Diderot, Tour 33-43, Case courrier 7087, 2 Place Jussieu, 75251, Paris Cedex, France e-mail : lacoutur@ccr.jussieu.fr. In the present work, we are monitoring simultaneously with the use of sonic waves, the linear and nonlinear viscoelastic behavior of concrete during curing from just after mixing and well into the solid state. The concrete is contained in a cell with a Lucite base. Using the complex reflection coefficient of short ultrasonic pulse between the Lucite base wall and the concrete, the linear compressional and shear wave viscoelastic moduli are determined in the linear part of the experiment; the moduli of concrete are obtained during the whole curing process. For the nonlinear experiment part, a high power, low frequency continuous sine compressional wave is transmitted through the medium to investigate the evolution of its nonlinear properties during the cure. Quasi-continuous-wave harmonic generation at different excitation amplitudes are used to extract classical and hysteretic nonlinear parameters which are put in relation with the moduli calculated after the linear measurements. Concrete is well known as a complex and multiscale material. The evolution of this type of material does not stop with time but its long term properties depend strongly upon the curing period. This is the reason why monitoring the properties of concrete during curing is a major topic of interest [1,4]. The hardening of concrete caused by the reaction of hydration of the cement is now a very well known process. Many papers deal with the linear elastic properties of concrete. Furthermore, it is known that hardened concrete shows nonlinear properties [2,3] mainly caused by the presence of microcracking. However, a crossed analysis of the two type of elastic properties has not yet been performed and more generally, to our knowledge, the consolidation transition of a material has not been studied simultaneously by the two methods. This paper begins with a description of the experimental set-up. The different results are then discussed : the evolution of the temperature in the core of the concrete and the linear and nonlinear properties. EXPERIMENTAL PROCEDURE A box of square section 16*16 cm with a base of Lucite contains the concrete poured in it just after mixing (Fig. 1). The experimental set-up can be divided into a linear and a nonlinear part. For the linear measurements, two compressional and shear wide band piezoelectric transducers are used (500 kHz central frequency). These transducers are operated in pulse echo mode. We are looking at the evolution of the reflection coefficient between the Lucite and the concrete over time for compressional and shear waves. From the reflection coefficient and knowing the density, PC with A/D converter concrete cell PZT transducer Amplifier Wave generator Nonlinear part plexiglas echo ASA Linear part FIGURE 1. Experimental configuration. we calculate the elastic moduli of the concrete. In the nonlinear portion of the experiment, we transmit pure tone compressional waves across the sample at different amplitudes and perform the analysis to extract nonlinear effects as the harmonic generation. The fundamental frequency is 8 kHz. Data collection is done every minute and stored into a computer. The PZT emitter and receiver are in direct contact with the concrete. The temperature is also monitored with a thermocouple embedded in the sample and the ambient temperature is measured with another thermocouple. The concrete is a reactive powder concrete made of Portland cement, thin sand, fume silica and superplasticizer. RESULTS AND ANALYSIS In Fig. 2 are plotted the temperature evolution and the shear wave reflection coefficient versus time. The main chemical reaction starts at approximately 20 hours ; here the concrete begins to structure and SESSIONS establishes gradually the connections between solid particles. 3 2.8 36 35 1 34 0.9 0.8 32 0.7 31 |R| (-) 0.6 T 30 29 0.5 28 0.4 27 Temperature (°C) 33 S - waves temperature 1.4 1.2 1 0.8 2f 0.6 3f 28 29 30 31 32 33 34 35 36 37 38 Time (h) 22 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 FIGURE 4. Harmonic amplitude dependence between 28 and 38 hours. Time (h) FIGURE 2. Shear wave reflection coefficient and the temperature in the sample between 0 and 95 hours. 0.5 2.0E+10 0.4 1.5E+10 0.3 Poisson's ratio 1.0E+10 Young's modulus 0.2 Young's modulus (Pa) The shear and compressional wave reflection coefficients leads to the calculation of the velocities and the elastic moduli (Fig. 3). Poisson's ratio (-) 1.6 23 0 5.0E+09 0.1 0.0E+00 25 2 1.8 0 24 0.1 2.2 0.2 25 0.2 2.4 0.4 26 0.3 Harmonic Amplitude dependence (-) 2.6 1.1 26 27 28 29 30 31 32 33 34 35 36 37 38 Time (h) FIGURE 3. Young's modulus and Poisson's ratio between 25 and 38 hours. As seen in Fig. 3, the material keeps a fluid character for 29 hours. Because of the shrinkage of the cement paste, the concrete debounded from the Lucite after thirty eight hours and did not allow us to follow the evolution after this time. In the nonlinear part of the experiment, transmission of 8 kHz sinusoidal wave is possible only after 23 hours, the amplitude of the second harmonic is reliable after 28 hours. In Fig.4 are plotted the variation with time of the power law coefficient of the second and third harmonic amplitudes dependence versus the amplitude of fundamental after 28 hours. The third harmonic amplitude dependence coefficient of two or nearly two shows that we are in a non classical case of nonlinearity instead of the classical case where the dependence is three. We believe that the observed dependence is not exactly two because of the perturbation of the eigen modes within the cell. The harmonic dependence after 28 hours corresponds to a transition period for the material where the largest particles are already connected and the hydrates develop in the pore space as deduced from the linear results [4]. CONCLUSION With the monitoring of different parameters as temperature, linear and nonlinear elastic properties, we collected information about the evolution of the concrete during curing. We have seen a correlation between nonlinear response and linear response that relates to known microstructure. REFERENCES 1. Boumiz A., Vernet C., Cohen Tenoudji F., Advanced Cement Based Materials, 3, 94-106 (1996). 2. Johnson, P.A., Materials World, the Journal of the Institute of Materials, 7, 544-546 (1999). 3. TenCate J., E. Smith and R. A. Guyer, Physical Review Letters., 85, 1020-1024 (2000). 4. Morin V., Cohen Tenoudji F., Richard P., Feylessoufi A. and Vernet C., Proceedings of Intern. Symp. On HPC and RPC edited by P.C. Aïtcin et al. RILEM publisher, Sherbrooke, 1998, 3, pp. 119-126. SESSIONS Generalized solvable models in fluid loaded structures Ivan V. Andronov Institute in Physics, Univ. of St.Petersburg; Ulianovskaja 1-1, St.Petersburg, 198904, Russia The generalized point models of inhomogeneities in elastic plates are suggested. These models take into account geometrical size of the inhomogeneity and correct in some cases classical point models. INTRODUCTION The problems of acoustic waves diffraction by elastic plates or shells with edges, conjunctions, cracks, inclusions and stiffeners attract wide interest caused by the needs of shipbuilding, noise analysis, acoustics of the ocean and other sciences. Much attention is paid to problems that can be solved in a closed form of integrals or series [1, 2]. Then the analysis of physical effects is the most simple. Such are the problems using point models formulated by means of contact conditions fixed in a midpoint of the obstacle. However real inhomogeneities have finite size. Here we present more precise model of crack and describe a class of generalized point models. These models are formulated as operator extensions in the form of zerorange potentials [3, 4]. two terms vanish at orthogonal direction to the plate, while the correction remains nonzero. Moreover for thin plates asymptotic analysis of D j , A and B involved in (1) shows the last term in (1) to be the principal one by the parameter kh (h is thickness of the plate). Figure 1 presents the effective cross-sections Σ computed for a 0 10 20 10 10 10 5 10 3 0 01 and 0 1m (from lower to upper curves). Difference is seen already for very narrow cracks. 0.1Hz 10Hz 100kHz 1kHz -25 -50 DIFFRACTION BY A CRACK Consider infinite elastic plate bounding an acoustic half-space. Let on a segment a a the plate be absent and the edges of the two half-plates be free. The field is generated by an incident at ϕ0 plane wave. Asymptotic by ka 1 analysis allows the far field amplitude of the scattered field to be found [5] -75 FIGURE 1. Frequency dependencies of Σ (dB) for narrow cracks in 1cm steel plate in water for ϕ0 20 iν k2 ss0 k4 2 2 k6 3 3 c c c c π L ϕ L ϕ0 D4 0 D6 0 π M ϕ Ac2 M ϕ0 Ac20 ν ln ka 4 B Ψ Here s sin ϕ, s0 sin ϕ0 , c cos ϕ, c0 (1) cos ϕ0 , L ϕ ik sin ϕ0 M ϕ ν M ϕ k cos4 ϕ k04 4 The following parameters in (1) describe the system plate–fluid: k and k0 are the wave numbers in acoustic media, and in isolated plate, ν ρ0 ω2 D, D is bending stiffness of the plate, ρ0 is the density of fluid, ω is frequency, D4 , D6 , A and B are contact integrals depending only on k, k0 and ν. The first two terms in (1) coincide with the exact expression for scattering by point model [2]. The correction is logarithmically by ka smaller, but the first GENERALIZED POINT MODELS The following approach is often used in boundaryvalue problems for fluid loaded plates. First the general solution U that satisfies all the equations and conditions except the contact conditions is introduced. It contains parameters that are then found an algebraic system originated from contact conditions. Generally U satisfies the problem ∆U k2U 0 d 4 4 ∂U k0 dx4 ∂y y νU 0 3 dj ∑ c j dx j δ x y 0 j 0 Here δ x is the Dirac delta-function and c j are arbitrary parameters. Physically U is the field of point sources on SESSIONS the plate. To formulate the point model that takes into account the geometry of the obstacle, one can add a passive source in the fluid. That is, the first equation in the problem for U is replaced by ∆U k2U ! ! C ln " r & S # o " 1 #' r% 0 (2) Here C is arbitrary and S is fixed positive. If S ! 0, the generalized model is reduced to a classical one. It should be noted that mathematically the condition (2) fixes some self-adjoint extension of the operator in the form of a zero-range potential [3, 4, 6]. The main question that appears in the formulation of a model of particular obstacle is the value of S in (2). The most rigorous way to find S for narrow crack is by comparing the far field asymptotics of scattering by a generalized point model with the asymptotics (1). One can find that logarithmically small corrections in (1) correspond to S ! a& 2 $ The following generalized models are suggested: 1. Point model of crack (see [2]) C ! 0 ξ( ("*) 0 #+! 0 ' ξ( ( (,"-) 0 #+! 0 $ ' 2. Model of a narrow crack (free edges) U C ln " 2r & a # o " 1 #' . ξ( ( "-) 0 #+! 0 ' ξ( ( ( r% "-) 4 0' 0 #/! 0 $ 0.05 5 5005 4 -20 πCδ " x # δ " y #$ An additional parameter C presented in the new model should be defined by an additional “boundary” condition. Mathematically correct formulation appears by setting the following condition in terms of the asymptotics of the solution U when r % 0 U Σ (dB) 3 4 3 5 2 -30 2 -40 1 5 1 FIGURE 2. Effective cross-sections Σ (dB) as functions of halfwidth a (mm) for the models 1–5 CONCLUSION The generalized point models were considered for two-dimensional problems of diffraction by fluid loaded plates with small obstacles. In these models additionally to classical boundary-contact conditions an additional condition was specified for the logarithmic derivative of the acoustic field in the central point of the obstacle. The parameter S in this condition was noted independent of the general parameters of the boundary value problem. It is defined only by the width of a gap in the plate. This fact allows the parameter S to be chosen from the analysis of a simpler problem for absolutely rigid plate. The method of the zero-range potentials [3, 4] allows three-dimensional defects in fluid loaded plates to be modelled in a similar manner. See e.g. [7], where the model of short crack is introduced. 3. Fixed point (see [1]) C! 0 ' ξ "-) 0 #+! 0 ' ξ(0"-) 0 #+! 0 $ 4. Model of a narrow slit with fixed edges U . C ln " 2r & a # o " 1 #' r % 0 ' ξ "-) 0 #1! 0 ' ξ(0"-) 0 #+! 0 $ 5. Model of a bubble C ln " 2r & a # o " 1 #' r % 0 ' U. ξ 2 C3 " R #$ π Note that for a ! 0 the models 2 and 4 are close to the classical models 1 and 3. However, already for very small ka scattered field are significantly different. The bubble (model 5) can not be described by classical boundarycontact conditions. Figure 2 presents the dependence of effective cross-section on half-width a for all the formulated above models. The incidence of 10kHz plane wave at ϕ0 ! 10 3 on a 5mm steel plate in water is taken as an example. REFERENCES 1. I.P.Konovalyuk and V.N.Krasilnikov Problems of diffraction and waves propagation, Leningrad Univ. Press, 4, 149–165 (1965). 2. D.P.Kouzov Appl.Math.Mech. 27, 1037–1043 (1963). 3. B.S.Pavlov Uspekhi matem. nauk 42, 99–131 (1987). 4. S.Albeverio and P.Kurasov Singular perturbations of differential operators and solvable Schroedinger type operators, Cambridge Univ. Press, London Math. Soc. Lecture Notes 221 (2000). 5. I.V.Andronov, B.P.Belinskiy and J.P.Dauer Wave Motion, 24, 101–115 (1996). 6. I.V.Andronov Appl.Math.Mech. 59, 451–463 (1995). 7. I.V.Andronov J.Math.Sci. 73, 304–307 (1990). SESSIONS Vibroacoustical Identification of a Double Wall Panel S.J. Pietrzko EMPA-Swiss Federal Laboratory for Materials Testing and Research Ueberlandstrasse 129, CH 8600 Dübendorf This paper focuses on the identification of the structural modal behavior of double wall panels, as well as the acoustic mode shapes in the cavity between the walls. Additionally the model of a whole panel including dynamics of sensors and actuators used for active control was identified. This general state space model was used to design a control system to improve sound transmission of a double wall panel around mass-air-mass resonance. TEST PANELS UNDER STUDY The first panel under study consisted of two monolithic glass panels (717x1091 mm) with a nominal thickness of 6 mm, and a critical frequency of approximately 2106 Hz. The interpanel spacing of 80 mm was filled with air. The panel was mounted in wooden sashes with dimensions x = 1480 mm and y = 1230 mm. For the experiments, the wooden sashes were installed in a concrete wall of a semianechonic 3 chamber with a volume of 78.1 m . The double wall resonance frequency of the pane was calculated to be at 66 Hz, which corresponds well to the observed transmission loss drop in the range around 71 Hz, see Fig.1, where three measured frequency response functions (FRF) are given. FRF's (sound pressure in the cavity/point force). The second panel under study was a double glazed window consisting of a monolithic glass pane with nominal thickness of 10 mm, critical frequency » 1210 Hz), and a monolithic glass pane of 4 mm thickness, critical frequency » 3160 Hz. The interpane spacing of 16 mm was filled with Argon gas. The double wall resonance frequency of the window was estimated at 180 Hz, which corresponds well with the observed drop of the transmission loss in the 200 Hz one-third octave band. PLANT DESCRIPTION The plant used in the experiment has three inputs i.e. a noise disturbance exciting the system from the sending room and two control inputs generated by loudspeakers inside the cavity. The output from the panel is measured in the cavity by two microphones an in the receiving room by a microphone-array, Fig.2. FIGURE 1. Structural (solid) and acoustical frequency response functions of the panel due to a single point force acting on a pane. For this measurement, the sending pane was excited in a corner by a point force, and the driving point acceleration as well as the microphone signals inside the cavity were measured. The driving point response shows clearly the dominating vibratory behavior of the panel around the mass-air-mass frequency. This was also observed in simultaneously measured acoustical FIGURE 2. A double wall plant. Input loudspeakers inside the cavity placed in the corners. Output microphones in the cavity and a microphone-array placed on a hemisphere. SESSIONS MODAL PARAMETERS NOMINAL PLANT IDENTIFICATION For the second panel, 16 pole values (damped natural frequencies, damping ratios) and modal participation factors were calculated using the Least Squares Complex Exponential Method. In the frequency range from 180 to 230 Hz, i.e. around the double wall resonance frequency, there are 6 modes which contribute significantly to the dynamic response. Estimates of the damped natural frequencies stabilized within 3 % and estimates of damping ratios within 5%. In the second stage of parameter estimation, modal vectors were estimated using the Least Squares Frequency Domain Technique. They stabilized within an accuracy of 5%. The modal parameters of the window close to 200 Hz compared with the fundamental mode at 38.2 Hz are presented in the Table 1; the corresponding complex mode shapes are given in Fig. 3. The whole plant (double wall structure including 2 loudspeakers and 3 microphones in the cavity) was identified to build a suitable state space model for controller design. Identification of the system without reconstruction filters established a for this structure typical model of order 50. This model was not suitable to implement for control, because of computational time overflow. After reduction of sampling frequency and using reconstruction filters it was possible to get a good quality low order balanced state space model. The subspace identification methodology was preferred [1]. This model was used to design a robust controller with good performance. Table 1: Panel modal parameters around 200 Hz Freq. Damping [Hz] (%) 182.0 197.8 201.0 219.5 1.7 1.3 1.3 2.3 Modal mass [kg] 1.0 1.0 1.0 1.0 Modal damping [kg/s] 37.80 32.68 33.13 64.04 Modal stiffness [N/m] e+06 1.308 1.545 1.595 1.902 FIGURE 4. Example of an identified FRF (solid) superimposed with a calculated one for a collocated loudspeaker and microphone in a corner. CONCLUSIONS FIGURE 3. Identified modal shapes of the window around the double wall resonance frequency, in the 200 Hz one-third octave band. Identified modal vectors (shapes of the panel walls) were used to verify and update lumped parameter and FEM models. For the identification of complex vibroacoustical structures like double wall panels, the subspace identification method gives reasonable state space models which can be directly used for robust control design. Additionally these models can be verified and updated with experimental modal analysis. Combination of these two methods is recommended. REFERENCES 1. P. van Overschee, B.D. Moor, Subspace Identification for Linear Systems: Theory, Implementation, Applications, Kluver Academic Publishers, Norwell, MA, 1996. SESSIONS Experimental Investigation on Depth Evaluation of Perforation Wenxiao QIAO, Xiaodong JU, Guangsheng DU and Jun FANG Department of Geo-information, University of Petroleum, Dongying, Shandong, 257061 , P.R. China As an important procedure of oil-well completion, perforation is a key technique in petroleum engineering. To date there is no in situ technique to evaluate the quality of perforation. In this paper, a method to measure the depth of perforation is investigated experimentally by using Ultrasonic Pulse Echo Techniques (UPET). The prototype equipment, whose measurement scope is between 0 and 1000mm, is developed successfully to measure the depth of the perforation in steel and sandstone targets. The transducers and the related generators needed, as well as parameters of the data acquisition system, are determined by experiment. Techniques to recognize the perforation and to recognize scattered acoustic wave from bottom of perforation are bring forward for the first time. Furthermore, two criteria are used to search for the scattered wave corresponding to bottom of perforation: Firstly the energy of the scattered wave must be strong enough. Secondly the dominant frequency of the scattered wave must be in the frequency domain of the incident acoustic pulse. All these techniques are testified successfully by experiments. INTRODUCTION As one of the important procedures of oil-well completion, perforation is a key technique in petroleum exploration and production. Of extremely complicated geometrical shapes, perforation tunnels usually have wedgelike outlines with a diameter of about 10 mm at the mouth and a depth of up to 700 mm. A comprehensive inspection of perforation quality, especially the depth evaluation of perforation, is always an interesting problem. BHTV and the new generation of imaging logging tools, such as USI and CBIL, are capable of evaluating the inside wall and circularity of casing and of evaluating the perforation positions, but they can not be used to determine the depth of perforations. There does not exist a device that can directly measure the downhole perforation depth[1~3]. Therefore, it is of great economical and social significance to study methods of downhole perforations depth evaluation and develop a corresponding device so as to help objectively evaluate the perforation effectiveness, optimize perforations techniques and improve oil-gas recovery efficiency. Ultrasonic pulse echo techniques could be a very promising method to provide a downhole perforations depth evaluation. In this article, an experimental research method and the corresponding results are introduced on the evaluation of perforations depth using ultrasonic pulse echo techniques. MEASUREMENT SYSTEM The system shown in figure 1 is mainly composed of a signal generator, an amplifier, an acquisition system, a processing system and a positioning system, the operations of which are all controlled by a computer. The signal source can generate the signal with frequency between 100 kHz to 1 MHz and an amplitude range of 50 to 460V. The resolution and the highest sampling rate of the data acquisition system are 12 bits and 10 MHz respectively. The positioning system is responsible for the translating and rotating the transducer. Computer Controler Driver Step Motor Signal Generator Data acquisition Casing Steel Target Tank Rock Target Perforation Ultrasonic Transducer Perforation FIGURE 1. The experimental measurement system The experimental casing has a length of 352 mm, an external diameter of 160.24 mm and a wall thickness of 10.20 mm. Two holes (each is 20 mm in diameter) are drilled in the casing, which could be connected in alignment with cylindrical steel and rock perforation targets, for the purpose of simulating two perforations of different depths. 4 Steel targets and 4 rock targets are used in this experiment. The steel targets are 150 mm long and their external diameters are 100 mm. In each target there are 3-4 holes, the depths range and the opening diameters of which are 10 to 150 mm and about 10 mm respectively. As for the rock targets, the external diameters are 150 mm and the lengths are 500 mm. In each target there is only one hole, which has a diameter of about 10 mm and a depth of between 130 and 500 mm (penetrated). SESSIONS All the operations such as the rising and rotating of the transducer holder and data acquisition and display are digitally controlled by means of a computer-controlled positioning system and a data acquisition system. When the transducer is rotated for 360 degrees, it is raised 5 mm and 100 times measurements of the pulse echo are conducted. Therefore the casing’s inside wall is scanned. Fi gure 2 shows the reflected ultrasonic waveforms in two different situations: when the transducer’s radiation direction is in alignment with and is not in alignment with the perforations. In the figure, the waveform in the middle is the reflected ultrasonic waveform when the transducer radiation direction is in alignment with the perforation. The waveforms on the top and at the bottom are the reflected ultrasonic waveforms when the transducer radiation direction is not in alignment with the perforations. It is shown in the figure that there are evidently reflected waves at 40 µs if the transducer is not in alignment with the perforation and there is no such reflection when the transducer is in alignment with perforation. 1.5 Amplitude 1.0 0.5 Reflected from Casing 0.0 Scattered from Perforation -0.5 -1.0 0 10 20 30 40 50 60 70 80 9 0 100 110 120 B in figure 3) and their dominant frequencies are close to that of the incidence waves. The depth of perforation could be calculated according to the time when point B is reached and the acoustic speed in the liquid inside the perforation. Each ultrasonic echo waveform is processed so as to determine whether there exist perforations. If there exits a perforation, the corresponding perforation depth is calculated out. All the data such as measurement depth and position, whether there is a perforation and the perforation depth are saved. 0.8 Amplitude(V) EXPERIMENTAL RESULTS 0.4 B 0.0 -0.4 -0.8 0 30 60 90 120 150 Time (µs) FIGURE 3. The scattered waveforms by a perforation in steel target DISCUSSION AND CONCLUSIONS After the experimental study on the downhole perforation depth evaluation, a computer-controlled prototype equipment is developed successfully to measure the depth of the perforations. The key parameters of perforations evaluation device are determined. The digitized signal processing techniques are developed to determine the existence and non-existence of perforations and the depth of a perforation. The technique of tracing the scattered signals (which come from the perforation bottom) backwards from the ending point of the received scattered waveforms is proposed. Time ( µ s ) FIGURE 2. The reflected waveforms of when the radiation direction is in alignment and is not in alignment with the perforations When the transducer’s radiation direction is in alignment with the perforations, the scattered waveforms could be received as shown in figure 3, where point B is the ending point of the scattered wave. Because of the irregularity of the inside of the perforations, there is usually no significant acoustic wave reflection from the bottom of the perforation, but the incidence waves could be scattered all over the inside wall of the perforation. After detailed analysis and comparison, it could be concluded that the scattered waves that are received last are from the bottom of the perforation (corresponding to the point ACKNOWLEDGMENTS Other members of the research group include postgraduate students Cheng Xiangyang, Zhao Huagang, etc. REFERENCES 1. 2. 3. Asheim Harald et al., Determination of Perforation Schemes To Control Production and Injection Profiles Along Horizontal Wells, SPE Drilling & Completion, March 1997, 13~17. Halleck, P. M., Recent Advances in Understanding Perforator Penetration and Flow Performance, SPE Drilling & Completion, March 1997, 19~126. Halleck, P. M., Wesson, D. S., Snider, P. M. and Navarette, M., Prediction of In-Situ Shaped-Charge Penetration Using Acoustic and Density Logs, SPE 22808, 483~490, 1991. SESSIONS Interface mobilities for source characterisation; matched conditions. B.A.T. Petersson1 and A.T. Moorhouse2 1 Institute of Technical Acoustics, Technical University of Berlin, Einsteinufer 25, D-10587 Berlin, Germany. 2 Acoustics Research Unit, Liverpool University, Liverpool L69 3BX, U.K. Interface mobilities have been demonstrated useful quantities in conjunction with the analysis of vibration transmission at multiple point and large area interfaces. The extension of the approach to the related area of source characterisation is logical. Thereby significant conceptual as well as practical simplifications result. The interface mobilities which are the ratios of the spatially Fourier decomposed velocity field to excitation field components, thus retain the formal simplicity and transparency of the single point case. This is at the expense of a slightly more elaborate post-processing of the constituent, computational or experimental data. Herein, the approach is critically examined for a source-receiver configuration, involving multiple contact points. Particular interest is directed to the case where both subsystems have matched or close to matched dynamic characteristics. INTRODUCTION In a suite of recent publications an approach to treat large, continuous, closed contour interfaces has been addressed from various viewpoints, e.g. [1-5]. It is demonstrated that with the introduction of the concept of interface mobility [4], a concept intimately related to the class of Fredholm integral equations, a significant simplification can be obtained for small and intermediate Helmholtz numbers. The main reason being that the transmission problem is transposed into that of an equivalent, single point and single component of motion and excitation, cf. [6]. Its applicability to cases with multiple point connections between the subsystems therefore becomes interesting. Of foremost interest is the application of interface mobilities in conjunction with source characterisation as recognised in [7]. SOME FUNDAMENTALS For continuous velocity fields and force distributions over a closed contour, the interface mobility is defined as, 1 − ik s − ik s Y pq = 2 Y (s | s0 )e p e q 0 dsds0 C C C Consider a multi-point interface between a source and a receiver structure such that a closed contour can be formed, passing all the points. The interface forces can be brought on to a form of a continuous distribution through, F (s) = Fmδ (s − sm ) , ∫∫ ∑ Fq = (∑ F e − ikq sm m ) C, where C is the perimeter of the contour. Similarly, the vibrations along this contour can be expanded as, v p = (1 C ) ∫ v (s)e − ik p s ds . C The complex power transmitted from the source to the receiver is then found to be given by, Q = C2 (∑ Y * pq F p Fq ) 2. From previous studies it is seen that for small Helmholtz numbers, kR , where R is a typical radius or dimension of the interface, the cross-order terms are generally small compared with those of zero and first order. For large Helmholtz numbers, moreover, the cross-order terms asymptotically vanish. With the cross-order terms small compared with the paired terms, the transmission problem and the source characterisation is simply subdivided into a sequence of orders, for instance, C2 2 2 2 (Y F + Y11 F1 L + Y−1−1 F−1 L) 2 00 0 This means that for many engineering applications, the formal simplicity of the single-point, singlecomponent case can be retained and terms included as required. Finally, it opens a viable scheme for source characterisation since all paired orders can be treated as individual source contributions. Q= such that the coefficients of the associated Fourier series become, SESSIONS EXPERIMENTAL EXAMINATION – MATCHED SUBSYSTEM DYNAMICS 10 |Y vF|, [m/Ns] 10 10 10 10 -1 10 10 Power, [W] With particular focus on the implications of closely matched dynamic characteristics of the source and receiver, a simple source was constructed. The mobility of the source system – a finite plate driven by a miniature shaker at an eccentric position – was adjusted by means of point masses to align with that of the receiver – a larger finite plate of the same thickness as that of the source. The source was connected to the receiving plate at four contact points, also positioned eccentrically and the interface contour chosen was a circle passing through these points. The high degree of matching is exemplified in Figure 1 via the ordinary point mobilities of source and receiver at one contact point. Interface mobilities were formed from measured point and transfer mobilities of both substructures. To estimate the transmitted power to the receiver, also the free velocity of the source was registered and Fourier decomposed. As a reference, the power transmitted to the receiver was additionally measured directly at the contact points in the assembled state. 10 10 10 -4 -6 -8 -10 -12 10 1 2 10 Frequency, [Hz] 10 3 Figure 2. Comparison of estimated an measured power transmission. (—) measured in the assembled state, (- - -) estimated using zero and first order interface mobilities and (·······) calculated using the matrix formulation The interface mobility appears to handle the matched condition as well as the full matrix calculation, which means that the estimates are sensitive in regions where the subsystems characteristics are equal in magnitude but phase conjugated. -2 CONCLUDING REMARKS -3 -4 -5 10 10 -2 1 2 10 Frequency, [Hz] 10 3 Figure 1. Ordinary point mobilities of source (——) and receiver (- - -) at a contact point. In Figure 2 are compared the power estimated from zero and first order interface mobilities, and from a full matrix computation with that measured in the assembled state. Although the interface mobility estimate of power, truncated after the first order terms, requires significantly less data it is mostly as accurate as the full matrix mobility calculation and is even more accurate at the higher frequencies. In the upper range moreover, it is more numerically stable than the matrix estimate. By decomposing the source activity, in this case the free velocity, and the structural dynamic characteristics, a set of source orders is established. In a first approximation these can be considered as independent source components to be superimposed. The source components correspond to physically interpretable motion components of the source. It is demonstrated that for common noise and vibration control applications, the few primary orders suffice to handle the structure-borne sound transmission of multi-point installations. REFERENCES 1 2 3 4 5 6 7 Petersson, B.A.T., Proc. Inter-Noise, Honolulu, 1984, pp. 553-558. Hammer, P. and Petersson, B., JSV 129, 1988, pp 119-132 Petersson, B., JSV 176, 1994, pp 625-639. Petersson, B., JSV 202, 1997, pp 511-537. Fulford, R. and Petersson, B.A.T., JSV 232, 1999, pp 877895. Pinnington, R.J. and Pearce, D.C.R., JSV 142, 1990, pp 461479. Petersson, B.A.T., Proc. 6th Congress of Vibr. and Sound, Copenhagen, 1999, pp 5, 2175-218 SESSIONS Development of Sound Radiation Prediction Program using PFFEM Analysis Results Ho-Won Lee, Suk-Yoon Hong and Young-Ho Park Department of Naval Architecture & Ocean Engineering, Seoul National University, Seoul, Korea The sound radiation prediction program is developed using the power flow finite element method(PFFEM) analysis results. PFFEM is a new method used for the prediction of the vibration energy density and intensity of arbitrary shape structures in medium to high frequency ranges. The boundary element method is used for the sound radiation program developed here, and the analysis results calculated by PFFEM is used as the boundary condition to analyze the vibration. Then, the sound radiation analysis can be simultaneously performed. With this program, the vibration and radiation characteristics of complex system structures such as submarine are predicted. 1. INTRODUCTION Commercial analysis programs used in the vibroacoustic field are split into two parts of software such as for the structural vibration analysis and for the sound radiation analysis. Mostly they have been individually developed and used. The purpose of this paper is to structure the system which is able to analyze the vibrational problem and the sound radiation problem at the same time. The power flow finite element method(PFFEM) which is used for the prediction of the vibration energy density and intensity of a complex structure in the medium to high frequency band is used here as a key method of the vibration analysis. The sound field around the structure is studied by solving the surface pressure about the structure by the acoustic boundary element method (BEM). At this time, the vibrational energy density calculated by power flow finite element method is used as boundary condition(e.g. surface normal velocity) of BEM. Then, computations are performed for complex structure such as submarine. where ω is the exciting frequency, η the structural damping loss factor and m represents the wave type. e is the time- and locally space-averaged vibrational energy density and c gm the group velocity. Equation (1) may be written as matrix form with finite element techniques as [K ] {e }= {F }+ {Q } (e) (e) (e) (e) (2) where each term of equation (2) can be written by K F Q (e) mij (e) mij 2 æ c gm ç òD ç ηω ∇ φ i ∇ φ è = = ò (Π m j ö + ηωφ i φ j ÷ dD ÷ ø (3 ) φ i )dD (4) D (e) mij = 2 æ c gm ò çç ηω Γ è ö φ i ( − n ) ⋅ ∇ e ÷d Γ . ÷ ø (5) 2.2 Boundary Element Method If sound wave propagates through a three dimensional domain with small amplitude, wave equation can be expressed as 2. THEORY 2.1 Power Flow Finite Element Method When the vibrational power is input with a unit plate element of structure and in a steady state, governing energy equation can be written as ∇2 p + k 2 p = 0 Equation (6) can be modified for the boundary element method as follows : N − 2 c gm ηω ∇ 2 em + ηω em = Π m (1 ) (6) 9 Aα åå α mj m =1 =1 N ö æ p mα − ç 4π + å C mj ÷ p j = m =1 ø è N 9 Bα vα åå α mj m =1 m (7) =1 SESSIONS where p j is the sound pressure at j-th node, N the number of element and vmα is the surface normal α , α , velocity at m-th element. Amj B mj C mj of equation (7) are defined as follows ; α A mj = ∂ òS ∂ n r m m α B mj = i ωρ C mj = − ikR ö ÷N ÷ ø Figure 3. Vibration energy density for inner plate α (ξ ) J (ξ ) d ξ (8 ) (ξ ) e mj òS R mj (ξ ) N α (ξ ) J (ξ ) d ξ m ∂ ò ∂n Sm æ e − ikR mj ( ξ ) ç ç R (ξ ) mj è rm ö æ 1 ÷ J (ξ ) d ξ ç ç R (ξ ) ÷ ø è mj (9 ) (10 ) Figure 4. Surface pressure distribution where J is the jacobian and N α (ξ ) the shape function. For the energy density of power flow finite element method to be used, it is transformed into velocity. 3. COMPUTATIONAL EXAMPLES Computations are performed for the submarine model which is excited by a harmonic point force located in the middle of the engine room. It is assumed that the structure is in the water. The excitation frequency is f = 100 Hz, and the damping loss factor is η = 0.05 at the inner plate and η = 0.1 at the outer plate. The submarine model for vibration and sound radiation analysis is shown in Figure 1. The vibration energy density obtained from the PFFEM are shown in Figures 2 and 3. Figure 4 shows the surface pressure distribution obtained from the boundary condition of Figure 2. At this time, the sound pressure and intensity distribution around submarine are shown in Figures 5 and 6, respectively. Figure 5. Sound pressure around submarine Figure 6. Sound intensity around submarine 4. CONCLUSIONS Figure 1. Submarine model for the vibration and sound radiation analysis. Figure 2. Vibration energy density for outer plate The program for the sound radiation prediction using PFFEM’s vibration analysis results is developed. The vibration analysis of arbitrary shape structure composed of plates is performed by using the power flow finite element method, and these vibrational results are used as the boundary conditions and the sound pressure field around the structure is calculated by the acoustic boundary element method. This algorism is used for the program of sound radiation prediction. With this program, computations are performed for the case of an underwater submarine model which is excited by a point force located in the engine room, and the sound radiation characteristics of its model are successfully predicted. SESSIONS Order Extraction as Order Waveform Based on the Digital Tracking Filter of Low Orders (II) *) A.A. Petrovskya, A.V. Stankevichb, S.J. Luznevb a Department of Real-time Systems, Bialystok Technical University, Wiejska St. 45A, 15-351 Bialystok, Poland b Department of Computer Engineering, Belarusian State University of Informatics and Raadioelectronics, 6, P. Brovky St., 220027 Minsk, Belarus The order extraction as order waveform in the rotating machine monitoring systems based on the new digital low order filtering are presented in this paper. For increasing of an analyzer dynamic range in the given report is offered to execute resampling amplitude of signal in new designed time moments by the filtering in the order domain, unlike the known approach, where digital filter was used in the time domain. The given digital tracking filters of low orders allows high-performance tracking of harmonic responses of periodic in mechanical acoustical systems. The tracking capabilities are independent of the rate of change of the rotational speed (slew rate). The experiments have shown the possibility to receive a practical dynamic range more than 100 dB. INTRODUCTION The goal of order tracking is to extract selected orders in terms of amplitude and phase, called Phase Assigned Orders, or as waveforms. The order functions are extracted without time delay (no phase distortion), and may hence be used in synthesis applications for sound quality, multiplane balancing, measurements of operational deflection shapes, and dopplerized engine noise tracking. PROCESSING IN ORDER DOMAIN Due to improvements in microprocessor performance, in particularity, digital signal processor, it is feasible to design a digital processor of computed synchronous resampling and order tracking that has negligible internal phase noise. The digital processor can be used to: #1) collect measured data at some fixed rate, digital filtering of the signals using multirate filters; #2) measure and store the arrival times of each synchronizing tachometer pulse simultaneously (it is important to measure the arrival time of each tachometer pulse very accurately to reduce the effects of time jitter); #3) calculate the new digital resampling time points on the base of different rotation models and to store them; #4) interpolate the stored measurement data in some optimum manner to obtain new samples at the desired time points; #5) compute the spectrum in the order domain. THE RESAMPLING AMPLITUDE ALGORITHM The order tracking front-end algorithms in the rotating machine monitoring systems based on the novel digital low order filtering are presented in paper [1]. For increasing of an analyser dynamic range in the given article is offered to execute resampling amplitude of signal in new designed time moments by the filtering in the order domain, unlike the known approach, where digital filter was used in the time domain [2]. As in the case with the frequency domain for a filter in the order domain we shall enter the term “Filter of the Low Orders” (FLO). The proposed approach excludes methodical errors, because in the order domain such filter will have the constant order characteristics and for further processing will be allocated only by part of a spectrum. As for time domain the given filter realizes automatic tracking for the change of the object rotation velocity in the correspondence with an accepted rotation mathematical model. The resampling amplitude algorithm by the interpolation digital filter in order domain has the following steps: #1) to calculate a revolution angle for i resampled point from beginning of a revolution the base of the linear model of object rotation; #2) to calculate a current angle from beginning of j-th revolution for the nearest points stored with SESSIONS constant sampling time period to the resampled point on the base of the linear model of object rotation; #3) to check up the angular distance from the resampled point up to some i-th nearest point stored with constant sampling time period, whether the angular distance is no more than half of length of the impulse response; #4) to calculate the value of the impulse response of FLO with angle defined on the step #2 for the signal points stored with constant sampling frequency; #5) to calculate resampled sample of a signal in i point as a sum input signal points stored with constant sampling frequency weighted with impulse response. FIGURE 1a. Magnitude response of the FLO in the frequency domain TESTING RESULTS The testing results of an asynchronous electric engine in the stage of start-up and some time later in the stage of rated load are shown on the figure 1 (a-d). The range of generated frequencies was closed to vibrations of symmetrical engine with 24 slots of stator. For simulation the classic model of asynchronous engine was used. The results (figure 1) also show that the developed method provides stability of amplitude and rule of spectral lines in order domain do not depend on operations mode of object. The magnitude characteristic of FLO in frequency domain for the given example is shown on the figure 1a. The curve 1 corresponds to the first resampled point, the curve 2 - 32-nd, and the curve 3 64-th resampled points respectively. On the figure 1a the following labels are accepted: w - normalized frequency, mag - amplitude in dB. FIGURE 1b. The dependence of rotation speed CONCLUSION The dynamic range of analyzer with resamling amplitude by the FLO does not depend on the number of experimental points on one complete object revolution, but correct choice of sampling frequency for maximally possible rotation speed of object is important. FIGURE 1c. The tracking contours of the first, tenth and twenty fifth orders REFERENCES 1. Petrovsky A., Stankevich A., Balunowski J. Proc. of the 6th International congress “On sound and vibration”, ICSV’99, 5-8 July 1999, Copenhagen, Denmark. –pp.2985-2992 2. Potter R. Sound and Vibration, September 1990, pp. 30-34. 8) This work was supported by KBN under the grant 7T07B 033 18 FIGURE. 1d. The spectral map in order domain SESSIONS Response of Rectangular Thin Plate to a Point Force and Distributed Excitation P. R. Bonifácio, A. Lenzi Departamento de Engenharia Mecânica, Universidade Federal de Santa Catarina, 88000 Florianópolis, Brasil, arcanjo@emc.ufsc.br, boni@emc.ufsc.br. Several techniques are used to study the frequency response of plate, however, they limit with all side simply supports. In this paper, analytical analyses are presented for various combinations of simply supported (S), clamped (C) and free (F) edges in thin rectangular plate through parametric analysis and is obtained the effect of boundary conditions in the response of finite plates. Although there are several combinations, only three of boundary conditions are shown in this work, as SSSS, SSSC and SSSF, and it can be extended to other cases. Therefore, this parametric analysis concerned with appendage in the form of frequency response of plate by classical Kirchoff’s theory. Numerical calculations are used to check the obtained analytical results. The predicted resonance frequencies curves and mode shape are compared against the Finite Element Method (FEM) and good agreement is found. INTRODUCTION At high frequency in structural analysis, where wavelengths are short, predict the response of structures by analytical methods is often preferable once the solutions by FEM are computationally intensive. In the present paper the curves of frequencies that describe the displacement of the plate are determined for each values of Lx/Ly and for each set of boundary condition in the plate, calculated in relationship with eigenvalue parameters λ = Lx 2ω ρ h / D . The generalized coordinates are expressed through parametric analysis for each set of boundary condition in the plate. Generalized forces of the problem are found to a point force and to a distributed excitation along the plate. Mathematical Procedure ratio for the assumed isotropic material, h, the plate thickness, and (xo,yo) is the excitation position. The displacement at any point on the structure can be represented as [2,3] w( x, y, t ) = m D= Eh 3 (1 + iη ) 12(1 − ν 2 ) (1) (2) where w(x,y) is the flexural displacement, D, the complex bending stiffness, η, the lost factor, ρ, the material density, E, Young’ modulus, ν, is Poisson’s mn ( x , y ) e −iω t (3) n where x = x / Lx and y = y / Ly The case of steady-state harmonic vibration and time dependence of the form e-iwt is considered in the present work. This solution (Eq. 3) must satisfy the boundary conditions by Φ , which represents the base functions that describe the shapes vibrations in the plate. Substituting the strain energy U and kinetic energy T into Lagrange’s equations (Eq. 4), the dynamic response of the plate, excited by point or a distributed force, can be calculated [2]. For the stationary time harmonic response, the kinetic energy T and the strain energy U, can be clearly determined by [3]. The flexural displacement of a thin, transversely vibrating plate, excited by time harmonic point force F(xo,yo), and to a distributed force F(x,y), oscillating with a circular frequency ω, is governed by [1] D∇ 4 w( x, y ) − ρ hω 2 w( x, y ) = Fδ ( x − xo )δ ( x − yo ) ∑∑ qmn (t ) Φ d ∂T ∂T ∂U + = Qmn − dt ∂q!mn ∂ qmn ∂ qmn (4) where qmn represents the mth-nth-generalized coordinate and Qmn is the mth-nth-generalized force in both directions. The generalized force to a point and distributed excitation is expressed respectively as [2] Qmn = 1 1 ∫ ∫ δ ( x − x )δ ( y − y ) Φ o 0 0 Qmn = 1 1 ∫∫ 0 0 o mn ( x , y ) dydx F ( x, y ) Φ mn ( x , y ) dydx (5) (6) SESSIONS 100000 10000 1000 10 w (λ ) Plate SSSF Lx/Ly=1 Lx/Ly=2 Lx/Ly=3 1E-3 1E-8 2 2 Plate SSSS Lx/Ly=1 Lx/Ly=2 Lx/Ly=3 1E-4 1 w (λ ) 2 w (λ ) 1 1E-6 50 100 150 2 200 0 50 100 1/2 λ=Lx ω(ρ h/D) 150 2 Plate SSSC Lx/Ly=1 Lx/Ly=2 Lx/Ly=3 1E-3 1E-7 200 50 100 1/2 λ=Lx ω(ρ h/D) 150 2 200 1/2 λ =Lx ω(ρ h/D) FIGURE 1. Parametric analyses Displacement (m) Displacement (m) 1E-4 1E-5 1E-6 1E-7 Plate SSSF Proposed Model FEM 1E-5 1E-6 Plate SSSC Proposed Model FEM 1E-4 Displacement (m) Plate - SSSS Proposed Model FEM 1E-4 1E-5 1E-6 1E-7 1E-7 0 200 400 600 800 1000 0 200 Frequency (Hz) 400 600 800 0 1000 200 400 600 800 1000 Frequency (Hz) Frequency (Hz) FIGURE 2. Displacement of plates and comparison with FEM being F(x, y) the distributed force acting on the plate. The generalized coordinate on the structure can be obtained substituting from Eq. 3, Eq 5 and Eq. 6 into the Lagrange’s equations (Eq. 4), resulting in q mn (λ ) = Qmn λ 2 1 1 ∂ 2Φ ∂ 2Φ ∂ 2Φ 1 1 ∂ 2Φ 2 ∂ 2Φ 2 − 2(1 − ν ) −( ) dx dy − λ 2 ∫ ∫ Φ 2 dy dx 2 2 0 0 ∂ x∂ y ∂ x ∂ y ∫0 ∫0 ( ∂ x2 + ∂ y 2 ) results are compared with FEM, which show excellent agreements in all cases. Lx S S S S Lx Ly S S F S Lx S S C S Ly Ly FIGURE 3. Boundaries conditions in the plates where λ = Lx 2ω ρ h / D ; Conclusions The parametric analysis consists of finding the mean-square quadratic spatial response of the plate in relationship with λ , as seen in Eq. 7, expressed in (ms2/kg)2. These results can be seen in the Figure 1, which shows results for ν = 0.3 , and several values of Lx/Ly. w2 ( λ ) = 1 1 0 0 ∑∑ ∫ ∫ q m n 2 2 mn (λ )Φ mn ( x , y ) dx dy The phenomenon of the vibration in plates is easily studied through a parametric analysis, thus the main variables of the problem can be analyzed in a better way. Therefore, this work can be used to predict a more complex system as plates coupled in beams. (7) This analysis is considered for three different sets of boundary conditions, as in Figure 3. The results of the displacements are shown in Figure 2, for a plate with Lx=1m, Ly=0.5m, h=0.003m, ρ= 2710 kg/m3, ν=0.3, η=0.01, E=7.2 1010 Pa, xo=0.1m, yo=0.14m. The REFERENCES 1. Graff, K., Wave Motion in Elastic Solids, New York, 1975, pp. 240-250. 2. Meirovitch, L., Analytical Methods in Vibrations, London, 1967, pp. 30-54. 3. Leissa, A., Vibrations of Plates, Washington, 1969. SESSIONS Numerical Modelling of Sound Transmissions in Buildings at Low Frequencies O. Chielloa, F. C. Sgarda, N. Atallab a Laboratoire des Sciences de l'Habitat, DGCB URA CNRS 1652, Ecole Nationale des Travaux Publics de l'Etat, 69518 Vaulx-en-Velin CEDEX, FRANCE. b Groupe d'Acoustique de l'Université de Sherbrooke, Department of Mechanical Engineering, Univ. de Sherbrooke, Sherbrooke, QC,J1K2R1,Canada This paper presents a general method to investigate the vibroacoustic behavior of complex structures coupled to acoustic cavities at low frequencies. A finite element description is used for structural displacements and fluid pressure fields. The coupled forced system is expanded on the uncoupled structure and rigid cavity modal basis. The structural modal analysis is performed using a free-interface component mode synthesis technique. The method proves to be accurate and well suited to parameter studies on structural complexities. In this paper, the method is applied to investigate the sound transmission paths in buildings at low frequencies. In addition, a general elastic joint between the separating wall and the lateral walls is accounted for in the model. The original free-interface technique has been modified both in the definition of the attachment modes and in the assembly procedure to account for the mass-less joint which is considered as a substructure in the component mode synthesis. An example is presented to validate the approach. Additional applications will be detailed in the poster session. INTRODUCTION THEORY The estimate of the influence of structural complexities (joints, stiffener etc) on the vibroacoustic behavior of elastic structures coupled to acoustic cavities is a common issue encountered by engineers. In building acoustics especially, where standards are more and more strenuous, structural complexities related to the mounting conditions of partitions may have important effects on the acoustic insulation at low frequencies. Deterministic methods such as finite elements coupled to classical modal analysis are appropriate at low frequencies but are often limited by the problem size and not well suited to parametrical studies. Indeed, when the effect of a structural parameter on the vibroacoustic behavior of a system needs to be known, the discretized quantities relative to the whole structure have to be recalculated before to redo the modal analysis accounting for the structural changes. In this paper, component mode synthesis (CMS) is shown to be an adequate tool for solving this kind of problems and deserves to be used whenever possible for the structural analysis. In addition, a free interface method [1,2] is used that allows for a removal of the junction degrees of freedom between the sub-structures in the final system. The coupling of structures with closed acoustic cavities is accounted for by a modal expansion over the uncoupled structural modes and rigid cavity modes. This technique proves to be valid for light fluid and allows for a significant reduction of the system size [3]. The combination of these two techniques enables one to calculate low frequency responses of complex problems with a good accuracy and a reduced computation time. In the following, the two methods are described successively. Then a validation example is presented. During the poster session, a typical configuration encountered in building acoustics will be focused on. Consider the in vacuo free vibrations of a discretized structure, undamped and composed of Nc substructures. In order to calculate the eigenmodes of this structure, displacement vectors u(c) of each substructure c are decomposed in terms of the truncated modal basis of the isolated F(c) completed by a pseudostatic approximation of the unkept modes [2]. u(c) = F(c) q(c)+Gd(c)fjunc(c) (1) where q(c),Gd(c) and fjunc(c) are the modal coordinates vector, the residual flexibility matrix of sub-structure c and the force vector applied on sub-structure c by all the other sub-structures, respectively. Applying continuity relations of forces and displacements at the interface between each sub-structure allows for an elimination of force vectors fjunc(c) in the system of equations together with an expression of the substructure displacements in terms of a global modal coordinates vector qr = {q(1) … q(Nc) }T only, such that u(c)=R(c)qr where xT denotes the transposed of x. Assuming a temporal dependency of the form ejwt, the invocation of the stationnarity of the energy functional of the whole structure expressed in terms of vector qr leads to an eigenvalue problem of the type: (Kr - w2 Mr ) qr = 0 (2) The eigenvectors of the whole structure on the degrees of freedom of each sub-structure c, F(c), are then given by Fs(c) = R(c) Fr, where Fr is the modal basis computed from eigenvalue problem (2). This method is very advantageous compared to a global modal analysis of the whole structure when physical parameters of a single sub-structure are changed since it is only necessary to perform the modal analysis of the substructure of interest and to use the component mode synthesis technique to obtain eq(2). This step requires SESSIONS the resolution of a reduced size eigenvalue problem compared to classical global modal analysis. In the case of high number degrees of freedom problems, the technique allows for an important decrease of the memory used for the computation. Finally, the method is robust and accurate, provided that the number of kept modes of each sub-structures be adequately selected. The forced response of a weakly damped system is now considered. The system comprises a global elastic structure which is coupled to a discretized acoustic cavity. The equations of the system are projected over a dual modal basis composed of the truncated modal basis of the whole structure obtained by CMS and the truncated modal basis of the rigid walled cavity, Ff obtained by a classical modal analysis. They read: ~ 2 éW ù ì Fm ü Cm s -w I ï ê ú ìíq s üý = ïí 1 (3) 1 ~ ~ ý ê CTm W f - I ú îq f þ ï 2 S m ï 2 êë úû îw þ w ~ In (6), W s denotes the generalized complex stiffness whole structure, each plate has been considered as an isolated sub-structure. By retaining the sub-structure modes up to 1500Hz, (1.5 times the upper frequency limit of the excitation spectrum), errors on the eigenfrequencies of the whole structure are less that 1% up to 1000Hz. The calculation of the vibroacoustic indicators has been validated by comparing the present approach to the results of commercial software I-DEAS (SDRC©) for an identical mesh in the case where one wall of the cavity is excited by a decentered point force. Fig.1 exhibits an excellent agreement between both approaches for the mean square velocity of one of the non excited plates. Software I-DEAS results have been obtained from a classical modal analysis of the whole structure, the fluid coupling being accounted for by a BEM. matrix accounting for structural damping such that: Ns Ns ~ ~ T~ Ws = å W(sc ) = å F(sc ) K ( c )Fs( c ) ~ c =1 (4) c =1 where K ( c ) is the complex stiffness matrix of the isolated sub-structure c. I denotes the identity matrix, ~ W f is a diagonal matrix filled with the kept eigenvalues ~ of the cavity, I is the generalized complex compressibility matrix accounting for dissipation within the cavity under the form of a structural damping and Cm is a generalized coupling fluid-structure matrix which can be written as: Ns Ns c =1 c =1 Cm = å C(mc ) = å F(sc ) C( c )Ff T (5) where C(c) is the coupling matrix between sub-structure c and the cavity. Sm is the generalized source vector in the cavity and Fm is the generalized force vector applied on the whole structure. Eq(3) is then solved at each frequency to get the modal coordinates qf and qs associated to the cavity and to the structure respectively. These quantities are used to compute the vibroacoustic indicators. In the following, all the cavity and structure modes up to 1.5 times the upper frequency of the excitation spectrum are kept in the expansions. RESULTS A validation example is now presented. Consider a rectangular cavity with dimensions 0.48´0.40´0.56m and flexible steel walls. The steel characteristics are E=2.1011Pa, n=0.3 and r=7800kgm-3. The cavity is filled with air with density r0=1.2kgm-3 and sound speed c0=340ms-1. The elastic walls are discretized using quadratic 8 nodes shell elements (12´10, 12´14 or 10´14 according to the walls). The cavity mesh is constituted of 6´5´7 quadratic 20 nodes hexaedric elements. The coupling matrices are calculated over the fluid mesh (6´5, 6´7 or 5´7) using quadratic 8 nodes surface elements. The frequency band of interest is [10-1000 Hz]. For the calculation of the modes of the FIGURE 1. Mean square velocity of a non excited wall CONCLUSION This paper has presented a general approach to study the vibroacoustic behavior of complex elastic structures coupled to elastic cavities at low frequency. The main features of the approach lie in the modal analysis of the whole structure using a free surface component mode synthesis together with the use of an uncoupled dual modal basis for the fluid-structure coupling. These two techniques can be combined to speed up the resolution of a complex fluid-structure problem by decreasing the number of degrees of freedom and prove to be appropriate for parameter studies. The convergence of the approach is excellent. Numerical results have been compared successfully to those obtained with a commercial software in the case of a flexible walled cavity filled with air and excited mechanically. During the poster session, a typical building configuration will be considered. REFERENCES 1. Craig, R.R. Jr, and Chang, C.-J. Substructure coupling for dynamic analysis and testing. CR 2781. NASA. 1977. 2. Tournour, M.A. and Atalla, N., Noise Control Engineering Journal, 46, 83-90 (1998). 3. Atalla, N., Tournour, M.A., and Paquay, S., A modal approach for the acoustic and vibration response of an elastic cavity, Proceedings 16th ICA/135th ASA meeting, 1998, pp.189-190. SESSIONS On point excited plates Jonas Brunskog, Per Hammer Division of Engineering Acoustics, Lund University, P.O. Box 118, SE-221 00 Lund, Sweden The present paper presents an expression for the point mobility of infinite plates driven by a perfectly rigid indenter. The problem is of general interest in connection with excitation and transmission of structure-borne sound. The indenter is assumed to be circular, weightless, stiff compared to the stiffness of the plate, and small compared to the wavelength of the bending and quasi-longitudinal waves of the plate. A detailed three-dimensional analysis is used. Traditionally the problem is solved approximately by means of assuming a pressure distribution in the interface between indenter and plate. In the present study the pressure distribution is also assumed, but an optimal choice of the pressure amplitude is found by means of a variational formulation. INTRODUCTION The object of the present paper is to derive expressions for the point mobility of infinite plates when driven by a perfectly rigid indenter. The problem is of general interest in connection with generation and transmission of structure-borne sound. This work, however, have been prompted by a special need; a accurate description of the imaginary part of the point mobility is important e.g. for impact situations where a 'spring' will make the impacting body rebound. The hypothesis is that especially the imaginary part of the mobility is incorrect in the previous derived expressions found in the literacy. The indenter is assumed to be circular, weightless, stiff compared to the stiffness of the plate, and small compared to the wavelength of the bending and quasi-longitudinal waves of the plate. The plate is assumed to be isotropic. The mobility is defined as the complex ration of velocity and force at the intersection between indenter and plate. By means of the classical Kirchhoff thin plate equations, it has been shown [1, 2] that the point mobility of an infinite plate can be written as v 1 (1) Y≡ z = F 8 m′′ B where vz is the vibration velocity, F the driving force, m'' the mass per unit area and B is the bending stiffens. In [3] Ljunggren discusses the accuracy of equation (1) and compare it with Mindlin theory and a threedimensional theory where only the poles corresponding to the bending waves are taking into account. In all analysis describing the motion of 'thin' structures it is assumed that at each point, the two sides of the structure have exactly the same displacement. This is an approximation and when the excitation is concentrated in an area which is comparable with or smaller than the structural thickness, additional weakness effects can occur. Moreover, when the radius of the indenter is comparable or lager than the thickness, additional inertia effects can occur. A more detailed three-dimensional analysis has been used by Paul [4] and Ljunggren [5] for rigid indenter, and Heckl [6] for soft indenter. Heckl and Petersson [7] investigated the influence of different choices of pressure distributions. The boundary value problem for a soft indenter is simpler than that resulting from a rigid indenter. For a soft indenter the pressure distribution under the indenter is constant. Therefore, both outside and under the indenter the boundary condition are Neumann conditions, i.e. prescribed pressures. For a rigid indenter the displacement under the indenter and the pressure distribution outside the indenter is prescribed, i.e. mixed Neumann and Dirichlet conditions. Both Heckl [6], Ljunggren [5] and Petersson and Heckl [6] avoided the problem of solving integral equations by means of assuming a pressure distribution under the indenter. As there is no guarantee that this assumption actually results in a uniform displacement under the indenter, this case will be denoted quasi-rigid. In [5] a comparison between the different results can be found (Figure 9 in the reference). Calculated values of the local reaction for the same numerical values are presented. There is hardly any agreement at all between the results. As Ljunggren points out, this lack of agreement is hardly astonishing in view of the different presumptions used in the different cases. In the present paper the pressure distribution in the interface between indenter and plate is found by means of a variational formulation. SESSIONS FORMULATION The system to be solved is a rigid circular indenter acting on a plate with finite thickness, shown in figure 1. We seek the point mobility Y of the excitation situation, and thereby the force acting in the interface between indenter and plate when the indenter is displaced a distance wieiωt, where the time dependence eiωt is henceforth suppressed. R wi x d z For a stiff indenter, the boundary conditions are w0 (x, y ) = wi , for r ≤ R and z = 0 p(x, y ) = σ z = 0, for r > R and z = 0 (2) where r=(x2+y2)1/2, and a additional condition is p(x,y)=σz=0 for z=d. The tangential shear stresses at the both surfaces are assumed to be zero, i.e. also under the indenter. It should be noticed that the displacement of the indenter wi is a real constant (if the time dependence eiωt is suppressed). The force is found as the integral of the pressure field under the indenter. With the aid of Hankel transforms, the displacement can be written 1 2π ~ p ∫0 iω (AA + AS )J 0 (k r r )k r dk r ∞ (3) where AA = AS = (k (k ) 2 r − iωαk 2 2 µ + β 2 tanh (αd 2) − 4αβkr2 tanh (βd 2) 2 2 r ) R ∫ p( s) s K (r , s)ds = w , r ≤ R . i 0 This is a Fredholms equation of the first kind. VARIATIONAL FORMULATION AND SOLUTION The varational formulation follows the method described in Morse and Ingard [8]. With the new variables q(r)=p(r)r and v=iωw, a variational formulation of the problem is R R R R 0 0 0 0 V = ∫ q(r )v * dr + ∫ q * (r )vdr − ∫ q * (r )∫ q(s )K (r , s )dsdr FIGURE 1. Rigid indenter acting on plate w0 (r ) = 1 iω − iωαk 2 2µ + β 2 coth (αd 2 ) − 4αβkr2 coth (βd 2) 2 Define a kernel K as ∞ K (r , s) ≡ ∫ J 0 (k r s)( A A (k r ) + AS (k r )) J 0 (k r r ) k r dk r 0 The boundary conditions (2) and equation (3) can now be used to rewrite the problem as Assume a pressure distribution that corresponds to the semi-infinite case, used by Ljunggren [5], p(r)=c/(R2r2)1/2. A stationary point is solved for the constant c, which is the best fit for the present pressure distribution. The input mobility can then be determined as Y= 1 2πR 2 r K (r , s )s RR ∫∫ 2 R − r 2 R2 − s2 0 0 dsdr Make use of the relation R ∫ s J (xs) 0 R 2 − s 2 ds = sin (Rx ) x , 0 Thus, the result is Y= 1 sin 2 (kr R ) (AA (kr ) + AS (kr ))dkr . 2π R 2 ∫0 kr ∞ It seems not be possible to calculate this integral analytically, but it is possible to use numerical integration. REFERENCES 1 Boussinesq, J., Applications des Potentiels, GauthierVillars, Paris (1885) 2 Cremer, H. and Cremer, L., Frequenz, 2, 61-84 (1948) 3 Ljunggren, S., Acta Acoustica, 3, 531-538 (1995) 4 Paul, H.S. J. Acoust. Soc. Am., 42 (2), 412-416 (1967) 5 Ljunggren, S. J., Sound Vib., 90 (4) 559-584 (1983) 6 Heckl, M., Acoustica, 49, 183-191 (1981) 7 Petersson, B.A.T. and Heckl, M., J. Sound Vib., 196 (3) 295-321 (1996) 8 Morse, P. and Ingard, U., Theoretical Acoustics, Prinston University press, Prinston, New Jearcy, 1968 SESSIONS Uncontrollable Modes in Double Wall Panels Oliver E. Kaisera , A. Agung Juliusb , Stanislaw J. Pietrzkoc , Manfred Moraria a ETH – Swiss Federal Institute of Technology, Automatic Control Laboratory, CH-8092 Zurich, Switzerland b Department of Applied Mathematics, University of Twente, Enschede, The Netherlands c EMPA - Swiss Federal Laboratories for Materials Testing and Research, CH-8600 Dubendorf, Switzerland Double-glazed windows have a poor transmission loss at low frequency. Since the passive means are more or less exhausted one could think of using an active controller to increase the transmission loss. In the work presented here two speakers in the cavity between the panes are used as actuators. A modal model is derived and validated with data from a laser scanner and measured transfer functions on the structure. From an analysis of this model it is shown that for certain configurations of the double wall panel some modes of the coupled system are uncontrollable and unobservable by speakers and microphones in the cavity, thus limiting the achievable controller performance. This theoretical result is verified by feedforward control experiments on two types of double-glazed windows. For the fully controllable window the transmission loss achieved by the active controller is about twice as large as for the window with uncontrollable modes. INTRODUCTION One way to tackle the control of stochastic noise in three dimensions is to reduce the sound transmission to the zone of interest. In buildings, windows are often the weak link in protecting the interior from outside noise. In particular, double glazed windows have a poor sound insulation at low frequency around the mass-air-mass resonance (double wall resonance). Since the passive means for windows are exhausted, an active controller that increases the transmission loss in the low frequency range is an attractive approach to reduce the noise level in buildings [2]. In the work presented here two speakers in the cavity between the panes are used as actuators similar as in [4] and [3]. The full experimental set-up and its dimensions are given in Fig. 1. Two double panels were investigated. The symmetric configuration consisted of a 6 mm-panel, a cavity of 84 mm, and a second panel of 6 mm. For the asymmetric configuration the second panel was replaced by a 3.2 mmpane. MODELING OF DOUBLE GLAZED WINDOWS For the modeling, a double panel structure can be divided into five subsystems, namely the excitation dynamics, the first panel, the cavity, the second panel, and the radiation. Each subsystem is relatively well understood [1] and can be modeled with a modal approach. The models of the subsystems can then be assembled to a model of the double panel structure as suggested in [3]. In addition, we included models of the speakers and transformed the model into state space form [2]. Validation FIGURE 1. Experimental set-up. The two panes have a size of 717 × 1091 mm and are mounted with wooden sashes in the opening between the sending and the receiving room. The model was validated with a laser vibrometer and by measuring transfer functions. The model not only predicted the mode shapes correctly but also the eigenfrequencies (cf. [2] for details). In Fig. 2 the transfer function from a speaker in the corner to a microphone in the same corner is shown. Apart from a difference in gain which is due to an unknown speaker parameter, the prediction from the model and the measurement agree very well. To make sure that this agreement is not accidental the validation was repeated for different double panel configurations, i.e. the thickness of the panels and the interpanel spacing was varied. In all cases the agreement between the model and the measurement was similar to the one in Fig. 2. SESSIONS magnitude [dB] 20 2 0 1 −20 0 −40 2 10 −1 phase [deg] 0 −100 −2 0.6 −200 0.4 −300 0.2 −400 0 −500 0.4 0.8 0.6 1 FIGURE 4. Poorly controllable mode (symmetric config.) frequency [Hz] FIGURE 2. Transfer function from a speaker in the cavity to a microphone in the cavity. Thin: model. Thick: measurement. contr. symmetric config 0.2 2 10 contr. asymmetric config 0 10 Table 1. Performance comparison of the different controllers. As quality measure the attenuation in dB around the mass-airmass resonance at 80 Hz is used. 0 −10 Symmetric panel Asymmetric panel Feedforward controller with error mics in receiving room 8.5 dB 18 dB Feedforward controller with error mics in cavity 4 dB 7.5 dB Controller −20 −30 50 100 150 200 250 300 350 400 450 500 10 0 −10 −20 −30 50 100 150 200 250 300 350 frequency [Hz] 400 450 500 FIGURE 3. Controllability of the coupled modes for a speaker in the corner (logarithmic scale with base 10). Some of the modes of the symmetric configuration (top) are poorly controllable, whereas all modes of the asymmetric configuration are well controllable. UNCONTROLLABLE MODES IN DOUBLE PANEL STRUCTURES For the optimization of the sensor and actuator locations the controllability and observability grammians calculated from the validated model were used. It was then noticed that some modes of the symmetric configuration have poor controllability for all actuator locations. An analysis of the validated model revealed that the poorly controllable modes correspond to modes where the two panels move in-phase as in Fig. 4. Such modes do not occur in the asymmetric configuration. There, all the modes are well controllable. EXPERIMENTAL RESULTS For both the symmetric and the asymmetric configuration feedforward controllers were implemented with three different actuator locations. These locations were the best three locations found in the actuator optimization [2]. While the performance varied only very little for the different actuator locations a substantial difference between the symmetric and the asymmetric config- uration was noticed. Due to the uncontrollable modes the controller is substantially less efficient for the symmetric panel than for the asymmetric panel (Tab. 1). CONCLUSIONS In [4] and [3] it is pointed out that the positioning of the actuators in the cavity between the panels plays an important role in order to achieve good performance. We showed, that in addition the performance of an active controller for a double glazed window can be substantially improved if the structure is designed for control. For our experimental set-up the performance at the mass-air-mass resonance could be doubled if the panel was designed to have well controllable modes only. REFERENCES 1. C. R. Fuller, S. J. Elliott, and P. A. Nelson. Active Control of Vibration. Academic Press Limited, London, 1996. 2. O. E. Kaiser. Active Control of Sound Transmission through a Double Wall Structure. PhD thesis, Swiss Federal Institute of Technology Zurich, 2001. 3. J. Pan and C. Bao. Analytical study of different approaches for active control of sound transmission through double walls. Journal of the Acoustical Society of America, 103(2):1916–1922, Apr. 1998. 4. P. Sas, C. Bao, F. Augusztinovicz, and W. Desmet. Active control of sound transmission through a double panel partition. Journal of Sound and Vibration, 180(4):609–625, 1995. SESSIONS Testing of the Tyre Vibration States by Speckle Interferometry J. Slabeyciusa, P. Koštiala, M. Držíkb and M. Rypákc a Faculty of Industrial Technologies, Univ. of Trenčín, SK-020 01 Púchov, Slovakia Institute of Construction and Architecture, Slovak Acad. Sci., SK-842 20 Bratislava, Slovakia c Rubber Research Institute, MATADOR, a.s., SK-020 32 Púchov, Slovakia b The frequencies of the basic vibration modes of pneumatic tyres were measured by the method of electronic speckle correlation interferometry. A powerful loudspeaker fed on continually tuned harmonic generator was used for non-contact excitation of the tyre. The setup of electronic speckle correlation interferometer consisting of green light Nd:YAG laser, optical elements and CCD camera has been adjusted. The interference pattern of vibrating tyre surface was obtained by PC processing both the original speckle field image and the image of deformed speckle field caused by vibrations of tyre surface. The obtained results are discussed. INTRODUCTION An automobile tyre should satisfy a number of functional requirements. It must work reliably in a large region of dynamic conditions – from static loadings to high frequency exciting of mechanical vibrations. The pneumatic tyre today is a highly sophisticated engineering structure, which viscoelastic, anisotropic and nonhomogeneous properties have to be taken into account. Consequently, the reliable theoretical and computer analysis of mechanical properties of tyre is extremely difficult [1]. The performance requirements of the system cannot be adequately designed without knowledge of its real dynamic characteristics, where the experimental measurements are necessary. For the measurement of the shape changes of tyre caused by its vibrations, the optical methods are very convenient, because there are non contact and their sensitivity is very high [2]. average method, double exposure technique and realtime visualisation method. The method analogous to holographic interferometry based on the properties of laser speckles and digital processing of image is known as electronic (digital) speckle pattern interferometry [2,5]. In this case, the electronic images of both wave field (corresponding to unloaded and loaded object) are compared in PC by appropriate software. The electronic speckle pattern interferometry tyre testing system, adjusted in our laboratory, consists of Nd:YAG diode pumped laser (50 mW, 532 nm), optical elements, CCD-camera, PC with framegrabber and image processing software. The equipment for tyre clamping and exciting of vibration consists of steady holder, powerful loudspeaker and continually tuned harmonic generator with amplifier. A block scheme of the optical setup is shown in Figure 1. PC CCD Mirror METHOD Mirror Mirror LASER Holographic interferometry is an optical method based on the interference of two optical wave fields – first one is scattered by object in primary state, the second one by object in load state [3,4]. Naturally, these fields cannot really exist in the same time. Hence, at least one of them is reconstructed by holographic way. The interference pattern allow to compute the shape deviation of object in every point. Therefore, the interferogram stores the full information about the deviation of the object surface. The holographic interferometry includes three basic techniques: time- Diffuser Lens Mirror Test Object FIGURE 1. The block scheme of the optical setup for the holographic speckle correlation interferometry. SESSIONS RESULTS AND DISCUSSION The testing object was the tyre MP-12, 175/70 R 13. The tyre fixed on the holder was illuminated by a broadened laser beam. A part of spreaded light beam was separated by using of mirror and directed through another mirror to the small ground screen. Passing through the ground screen the light creates the speckle field directed by the semitransparent mirror into the objective lens of CCD camera. Through this semitransparent mirror the image of the object is also passing and is projected onto the CCD matrix where the interference of both the speckle fields happen. The next image, figure 3, shows the fifth order mode with frequency of 216 Hz and, in the end, figure 4 corresponding to mode of sixth order with frequency 282 Hz. The interference pattern of n-th order vibrational mode consist of 2n separated maxima, symmetrically arranged on the tyre perimeter. FIGURE 4. Interference pattern corresponding to radial vibrational mode of sixth order at frequency 282 Hz. FIGURE 2. Interference pattern corresponding to radial vibrational mode of fourth order at frequency 155 Hz. Any movements of object surface in the line of sight will create the changes in optical paths and thus gives rise to the pattern of interferogram. The tyre was excited by loudspeaker fed on continually tuned harmonic generator. The images corresponding to the motionless tyre and the vibrating one are electronically subtracted and the resulting image with interference pattern is created. As the illustrative examples we present interference patterns for three vibrational modes of tyre investigated: Figure 2 shows a fourth-order radial tyre mode, which frequency was 155 Hz. The contrast of images, obtained by electronic interferometry is not as good as that obtained by classical holographic interferometry. On the other side, the digital electronic interferometry allow avoid the recording media. The process of measurement is convenient and relatively fast. From the irregularities of interference pattern we can judge on non-uniformity of tyre tested. The frequencies of natural vibration modes differ from those of tyre in contact with road. Nevertheless, the reducible experiment, in which the tyre is steady in the massive holder and all the perimeter of tyre is free, can bring useful information about tyre quality. REFERENCES 1. Gardner, I., and Theves, M., Tire Sci. & Technol., 17, 86-99 (1989) 2. Kreis, Th.: Holographic Interferometry. Acad. Verlag, Berlin, 1996 3. Potts, G.R., Tire Sci. & Technol., 1, 255-266 (1973) 4. Keprt, J. and Bartoněk, L., Holographic testing of tyres, in Proc. 35th Int. Conf. on Experimental Stress Analysis. Olomouc 1997, p.162-167 5. Urgela, S., Slabeycius, J., Koštial, P., The Application of Holographic Methods for the Non-destructive Testing of Tyres, in Proc. Slovak Rubber Conf. ’99, Púchov, 1999, p.11-19. FIGURE 3. Interference pattern corresponding to radial vibrational mode of fifth order at frequency 216 Hz. SESSIONS