Noise diagnostic system

0Associated
Cases 
0Associated
Defendants 
0Accused
Products 
21Forward
Citations 
0
Petitions 
2
Assignments
First Claim
1. A method for diagnosing noise sources including the steps of:
 measuring sound at a plurality of measurement points around said noise sources;
determining the amplitude at a plurality of frequencies of said sound at said measurement points; and
reconstructing the acoustic field at points other than said measurement points based upon said amplitude at said plurality of frequencies of said sound at said points by solving the Helmholtz equation.
2 Assignments
0 Petitions
Accused Products
Abstract
A method for reconstructing the acoustic field on the surface of a vibrating object based on measurements of the radiated acoustic pressure includes solving the Helmholtz equation directly using the expansion of a set of independent functions that are generated by the GramSchmidt orthonormalization with respect to the particular solutions to the Helmholtz equation. The coefficients associated with these independent functions are determined by requiring the assumed form of solution to satisfy the pressure boundary condition at the measurement points. The errors involved in these coefficients are minimized by the least squares method. Once these coefficients are specified, the acoustic pressure at any point, including the source surface, is completely determined.
33 Citations
View as Search Results
VEHICLE DIAGNOSIS BASED ON VEHICLE SOUNDS AND VIBRATIONS  
Patent #
US 20170076514A1
Filed 09/11/2015

Current Assignee
GM Global Technology Operations LLC

Sponsoring Entity
GM Global Technology Operations LLC

Vehicle diagnosis based on vehicle sounds and vibrations  
Patent #
US 9,824,511 B2
Filed 09/11/2015

Current Assignee
GM Global Technology Operations LLC

Sponsoring Entity
GM Global Technology Operations LLC

ACOUSTIC ANALYSIS APPARATUS FOR VEHICLE  
Patent #
US 20100299107A1
Filed 03/11/2010

Current Assignee
Mazda Motor Corporation

Sponsoring Entity
Mazda Motor Corporation

METHOD FOR RECONSTRUCTING AN ACOUSTIC FIELD  
Patent #
US 20110172936A1
Filed 06/26/2009

Current Assignee
BRUEL KJAER SOUND VIBRATION MEASUREMENT AS

Sponsoring Entity
BRUEL KJAER SOUND VIBRATION MEASUREMENT AS

Method for reconstructing an acoustic field  
Patent #
US 8,731,851 B2
Filed 06/26/2009

Current Assignee
BRUEL KJAER SOUND VIBRATION MEASUREMENT AS

Sponsoring Entity
BRUEL KJAER SOUND VIBRATION MEASUREMENT AS

Material wear indication system  
Patent #
US 20090107795A1
Filed 10/26/2007

Current Assignee
RollsRoyce Corporation

Sponsoring Entity
RollsRoyce Corporation

Material wear indication system  
Patent #
US 7,847,679 B2
Filed 10/26/2007

Current Assignee
RollsRoyce Corporation

Sponsoring Entity
RollsRoyce Corporation

System and method for visualizing sound source energy distribution  
Patent #
US 20070223711A1
Filed 05/24/2006

Current Assignee
Sanyang Industry Company Limited, National Chiao Tung University

Sponsoring Entity
Sanyang Industry Company Limited, National Chiao Tung University

Panel acoustic contributions examination  
Patent #
US 20070189550A1
Filed 02/14/2006

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Wear indicating friction disc  
Patent #
US 20070095626A1
Filed 10/28/2005

Current Assignee
Copeland Andrew, Burkholder Phillip

Sponsoring Entity
Copeland Andrew, Burkholder Phillip

Wear indicating friction disc  
Patent #
US 7,469,777 B2
Filed 10/28/2005

Current Assignee
RollsRoyce Corporation

Sponsoring Entity
RollsRoyce Corporation

Farfield analysis of noise sources  
Patent #
US 20060080418A1
Filed 10/13/2004

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Farfield analysis of noise sources  
Patent #
US 7,330,396 B2
Filed 10/13/2004

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Noise, vibration and harshness analyzer  
Patent #
US 20040243351A1
Filed 06/30/2004

Current Assignee
VETRONIX CORPORATION

Sponsoring Entity


Reconstruction of transient acoustic radiation from a finite object subject to arbitrarily timedependent excitation  
Patent #
US 20050150299A1
Filed 01/08/2004

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Reconstruction of transient acoustic radiation from a finite object subject to arbitrarily timedependent excitation  
Patent #
US 6,996,481 B2
Filed 01/08/2004

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Method and apparatus for reconstructing and acoustic field  
Patent #
US 6,615,143 B2
Filed 03/01/2001

Current Assignee
Wayne State University

Sponsoring Entity
Wayne State University

Method for determining the steady state behavior of a circuit using an iterative technique  
Patent #
US 6,493,849 B1
Filed 09/05/2000

Current Assignee
Cadence Design Systems Incorporated

Sponsoring Entity
Cadence Design Systems Incorporated

Method for determining the steady state behavior of a circuit using an iterative technique  
Patent #
US 6,636,839 B1
Filed 09/05/2000

Current Assignee
Cadence Design Systems Incorporated

Sponsoring Entity
Cadence Design Systems Incorporated

Portable acoustic impedance measurement system  
Patent #
US 6,134,968 A
Filed 07/19/1999

Current Assignee
Spirit Aerosystems Incorporated

Sponsoring Entity
The Boeing Co.

Method for determining the steady state behavior of a circuit using an iterative technique  
Patent #
US 6,151,698 A
Filed 04/28/1997

Current Assignee
Cadence Design Systems Incorporated

Sponsoring Entity
Cadence Design Systems Incorporated

Far field acoustic radiation reduction  
Patent #
US 5,381,381 A
Filed 09/30/1993

Current Assignee
United States Of America As Represented By The Secretary Of The Navy

Sponsoring Entity
United States Of America As Represented By The Secretary Of The Navy

Wavenumberadaptive control of sound radiation from structures using a `virtual` microphone array method  
Patent #
US 5,519,637 A
Filed 08/20/1993

Current Assignee
Mcdonnell Douglas Corporation

Sponsoring Entity
Mcdonnell Douglas Corporation

Mirror with photonic band structure  
Patent #
US 5,365,541 A
Filed 01/27/1993

Current Assignee
Northrop Grumman Systems Corporation

Sponsoring Entity
TRW Limited

Acoustic fluid flow monitoring  
Patent #
US 5,285,675 A
Filed 06/05/1992

Current Assignee
University of Florida Research Foundation Incorporated

Sponsoring Entity
University of Florida Research Foundation Incorporated

Adaptive system for controlling noise generated by or emanating from a primary noise source  
Patent #
US 5,347,586 A
Filed 04/28/1992

Current Assignee
Siemens Energy Incorporated

Sponsoring Entity
Westinghouse Electric Company LLC

Active vibration control  
Patent #
US 5,170,433 A
Filed 12/11/1989

Current Assignee
ADAPTIVE AUDIO LIMITED

Sponsoring Entity
ADAPTIVE CONTROL LIMITED

Diffraction tomography systems and methods with arbitrary source and detector array surfaces  
Patent #
US 4,598,366 A
Filed 01/25/1984

Current Assignee
Schlumberger Technology Corporation

Sponsoring Entity
Schlumberger Technology Corporation

Acoustic emission source location on platelike structures using a small array of transducers  
Patent #
US 4,592,034 A
Filed 11/15/1982

Current Assignee
Cornell Research Foundation Incorporated

Sponsoring Entity
Cornell Research Foundation Incorporated

Long wavelength acoustic flowmeter  
Patent #
US 4,445,389 A
Filed 09/10/1981

Current Assignee
UNITED STATES OF AMERICA AS REPRESENTED BY THE SECRETAR OF COMMERCE

Sponsoring Entity
UNITED STATES OF AMERICA AS REPRESENTED BY THE SECRETAR OF COMMERCE

Portable apparatus for measuring acoustic impedance at the surface of curved sound absorber  
Patent #
US 4,305,295 A
Filed 12/26/1979

Current Assignee
The Boeing Co.

Sponsoring Entity
The Boeing Co.

Radio frequency resistivity and dielectric constant well logging utilizing phase shift measurement  
Patent #
US 4,012,689 A
Filed 10/24/1974

Current Assignee
Texaco Incorporated

Sponsoring Entity
Texaco Incorporated

DEVICE FOR MEASURING ACOUSTIC QUANTITIES  
Patent #
US 3,658,147 A
Filed 06/29/1970

Current Assignee
United States Of America As Represented By The Secretary Of The Navy

Sponsoring Entity
United States Of America As Represented By The Secretary Of The Navy

23 Claims
 1. A method for diagnosing noise sources including the steps of:
 measuring sound at a plurality of measurement points around said noise sources;
determining the amplitude at a plurality of frequencies of said sound at said measurement points; and
reconstructing the acoustic field at points other than said measurement points based upon said amplitude at said plurality of frequencies of said sound at said points by solving the Helmholtz equation.  View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10)
 measuring sound at a plurality of measurement points around said noise sources;
 11. A method for diagnosing noise sources including the steps of:
 (a) measuring the field acoustic pressure p at a plurality of measurement points around said noise sources;
(b) assuming the acoustic pressure anywhere has the form;
##EQU41## where p and c are the density and speed of sound of the fluid medium, respectively, N is the total number of expansion terms, C.sub.i are coefficients associated with independent functions .psi..sub.i *;
(c) generating said independent functions .psi..sub.i * by using the GramSchmidt orthonormalization with respect to .psi. over the source surface; and
(d) setting the acoustic pressure p* equal to said measured field acoustic pressure p at said measurement points;
(e) solving said coefficients C.sub.i using the least squares method; and
(f) reconstructing the acoustic field at points other than said measurement points by calculating;
##EQU42##  View Dependent Claims (12, 13, 14, 15)
 (a) measuring the field acoustic pressure p at a plurality of measurement points around said noise sources;
 16. A system for diagnosing noise sources comprising:
 a plurality of transducers, said transducers generating a signal indicative of the amplitude and frequency of sound at a plurality of measurement points;
a signal analyzer receiving said signal from said transducers, said signal analyzer generating frequencyamplitude data based upon said signal from said transducers;
means for reconstructing the acoustic field at points other than said measurement points based upon said frequencyamplitude data.  View Dependent Claims (17, 18, 19, 20, 21, 22, 23)
 a plurality of transducers, said transducers generating a signal indicative of the amplitude and frequency of sound at a plurality of measurement points;
1 Specification
This application claims the benefit of U.S. Provisional Application No. 60/006,223, filed Nov. 3, 1995.
BACKGROUND OF THE INVENTIONThe present invention relates generally to a noise diagnostic system and more particularly to a noise diagnostic system utilizing an inventive method, referred to here as the "Helmholtz Equation Least Squares (HELS) Method."
In engineering practice, it is often desired to diagnose noise sources and quantify their strengths in order to reduce resulting noise radiation. Problems of such are categorized by Turchin et al. (1971) as the "inverse problem" which deals with problems of finding unknown sources based on the known consequences. In an inverse acoustic radiation problem, the acoustic quantities on the source surface are determined based on the measured acoustic pressure in the field. One particular application in the automotive industry is to determine the causes of certain amplitudes and frequencies of noises caused by the components of the automobile engine. Generally, a plurality of transducers or microphones are placed near the components of the engine to gather data regarding the frequency and amplitude of the noise caused by the multitude of various components. The data is then analyzed in order to determine which components are causing which frequencies and at which amplitude.
As in all inverse problems, one major difficulty in the inverse acoustic radiation problem is associated with the illposedness. By definition, a problem is wellposed if the solution exists, is unique, and depends continuously on the data; otherwise, it is illposed. Under most circumstances, an illposed problem is very sensitive to the formulations used, and solutions thus obtained do not depend continuously on the auxiliary data. As a result, a slight error in the measured data may lead to an enormous error in the reconstruction of the acoustic quantities on the surface.
The inverse problems have been the subject of extensive studies for the past few decades (Landweber, 1951; Twomay, 1963; Franklin, 1970) and have been documented in detail by many people. The present invention is mainly concerned with an inverse acoustic radiation problem, namely, the reconstruction of the acoustic field on the surface of a vibrating structure from the measurements of the radiated acoustic pressure in the field.
An early approach to the reconstruction of the acoustic field on the surface of a planar source is through near field acoustic holography (NAH) together with a twodimensional fast Fourier transformation (FFT) algorithm (Williams et al., 1980; Maynard et al., 1985; Veronesi and Maynard, 1987). However, this technique requires hundreds or thousands of transducers placed at extremely close range to the components. Further, this technique only works for very simplegeometry components.
This approach was extended, called the generalized near field acoustic holography (GENAH), to cylindrical sources by Williams and Dardy (1987), and further to nonseparable geometries by Borgiotti et al. (1990). In order to take into account of the evanescent field, the hologram surface on which measurements were taken must be very close to the source surface, within onehalf wavelength (Loyau and Pascal, 1988). Sarkissian (1990, 1991, 1992) developed an algorithm based on an expansion of the surface field in terms of a set of real functions for the farfield acoustic holography. All these algorithms were limited to source surfaces with simple geometric shapes. Further, these techniques all required hundreds or thousands of transducers placed at very close range to the sources.
In the early 1990's, the boundary element method (BEM)based Kirchhoff integral theory were used to generate a transformation matrix that correlated the field acoustic quantities to the surface ones in order to reconstruct the acoustic field on an irregularlyshaped surface (Veronesi and Maynard, 1989; Kim and Lee, 1990; Bai, 1992). However, the matrix thus obtained was singular. Hence singular value decomposition (SVD) was used to filter out the evanescent waves and to regularize the matrix. Chao (1987) used an implicit leastsquare method to approximate the reconstruction of the surface acoustic field by minimizing the errors associated with the integral equation approach. Gardner and Bernhard (1988) tried this integral equation approach for an inverse acoustic radiation problem inside a cavity. Numerical examples indicated that the reconstruction error in the interior problem had the same order of magnitude as that in the exterior problem.
Generally speaking, the BEMbased Kirchhoff integral formulation is a natural approach to reconstruct the acoustic field on the surface of an irregularlyshaped object. The main advantage of this approach is the reduction of the dimensions of the problem by one, thus significantly improving the efficiency of numerical computations. However, this approach has two inherent drawbacks resulting from transforming the wave equation into the Fredholm integral equation of the first and second kinds, respectively. The first drawback is associated with the wellknown nonuniqueness of the surface Kirchhoff integral equation, and the second is with the illconditioning of the transformation matrix which makes the inverse acoustic radiation problem an illposed one. The first drawback can be overcome by using the CHIEF method, provided that the overdetermined points are properly selected. The second drawback is less straightforward to deal with than the first one.
To show the existence of the illconditioning difficulty, consider a general Fredholm integral equation of the first kind ##EQU1## where K(x,y) is an arbitrary, integrable kernel, g(x) is given, and f(y) is the sought function. Phillips (1962) has shown that there is no successful way of solving f(y) for an arbitrary kernel K(x,y) when g(x) is specified with only modest accuracy. The reason for that is quite simple: while Eq. (1) can yield g(x) for a given K(x,y) and f(y), its inverse may not be bounded. This can be seen as follows. Let f(y) be the solution to Eq. (1) and add to it a fluctuation f.sub.m =sin (my). Substituting f(y) into Eq. (1) then yields ##EQU2## Since for any integrable kernel, the function ##EQU3## as m.fwdarw..infin.. Hence an infinitesimal change g.sub.m (x).noteq.0 in g(x) will cause a finite change f.sub.m in f(y). Also, we expect that g.sub.m (x).fwdarw.0 as m.fwdarw..infin. faster for a flat smooth kernel than for a sharply peaked one. Hence the success in solving f(y) for a given g(x) depends to a large extend on the accuracy of g(x) and the shape of K(x,y).
Because of the presence of this inherent illposedness in the Kirchhoff integral formulation, any slight inaccuracy in the measurements may lead to an erroneous result in the reconstruction. Since all measurements will inevitably involve certain levels of uncertainties either due to random fluctuations or due to the effect of the evanescent waves that decay rapidly as they propagate away from the surface, this illposedness can be a real threat to the success of the inverse acoustic radiation problem. Even slight errors in measurements will lead to erroneous reconstruction. Generally, measurement errors are magnified hundreds of times or thousands of times by this technique. One partial solution is to take numerous measurements at close range. SVD can be used to truncate the order of the matrix and regularize it, thus reducing but not eliminating the reconstruction error. The procedures involved in SVD, however, are timeconsuming, especially at high frequencies. Even with the regularization, the accuracy of reconstruction is still limited to the near field (Kim and Lee, 1990). Moreover, how to handle small singular values in SVD is still a problem open to investigation (Veronesi and Maynard, 1989).
SUMMARY OF THE INVENTIONIn the present invention, a plurality of transducers measure the amplitude and frequency of sound at a plurality of points in the field surrounding the noise sources. The transducers are connected to a signal analyzer which generates frequency versus amplitude data. The frequency/amplitude data is input to a computer which reconstructs the acoustic field at the noise source surface to diagnose the noise sources.
The reconstruction of the acoustic field is tackled by directly solving the Helmholtz equation. The acoustic pressure on the surface of an object is expressed in terms of a set of independent functions which are generated by the GramSchmidt orthonormalization with respect to the particular solutions to the Helmholtz equation. The coefficients associated with the independent functions are determined by requiring the assumed form of solution to satisfy the pressure boundary condition at the measurement points. The errors involved in the coefficients are then minimized by the least squares method. Once these coefficients are specified, the acoustic pressure at any point, including the source surface, is completely determined. Hence this method is not only applicable to reconstruction of the acoustic field on the source surface, but also to the forward acoustic radiation problem. Moreover, it can be applied to sources with both separable and nonseparable coordinates.
BRIEF DESCRIPTION OF THE DRAWINGSFIG. 1 is a schematic of a noise diagnostic system in accordance with the present invention.
FIG. 2 is a flowchart for the computer of the noise diagnostic system of FIG. 1.
FIG. 3 is a schematic showing the locations of measurements with respect to the short cylinder with two spherical end caps.
FIG. 4 shows the comparison of the reconstructed acoustic pressure distribution and the initially calculated one on the surface of the cylinder.
FIG. 5 shows the comparison of the calculated acoustic pressure in the approximated one over a circle of radius r=4.
FIG. 6 shows the comparison of the initially calculated surface acoustic pressure and the reconstructed one.
FIG. 7 shows a comparison of the calculated field acoustic pressure in the approximated one using an 8term expansion over a circle of radius 4.
FIG. 8 shows the comparison of the calculated dimensionless surface acoustic pressure and the reconstructed one along the source generator.
FIG. 9 shows the comparison of the approximated field acoustic pressures using an 18term expansion with the calculated ones over a circle of radius r=4.
FIG. 10 shows the relative errors in reconstruction of surface acoustic pressure for a dilating sphere.
FIG. 11 shows relative errors in reconstruction of surfaced acoustic pressures on an oscillating sphere.
FIG. 12 shows the relative errors in reconstruction of surfaced acoustic pressure on a vibrating cylinder with two spherical end caps.
FIG. 13 shows the relative errors in reconstruction of surface acoustic pressures on an oscillating cylinder with two spherical end caps.
FIG. 14 shows the relative errors in reconstruction of surface acoustic pressures on a cylinder with a vibrating piston set in one endcap.
FIG. 15 shows the relative errors in reconstruction of surface acoustic pressures on a vibrating cylinder with two spherical endcaps.
FIG. 16 shows the relative errors in reconstruction of surface acoustic pressures on an oscillating cylinder with two spherical endcaps.
FIG. 17 shows the relative errors in reconstruction of surface acoustic pressures on a cylinder with a vibrating piston set in one endcap.
DETAILED DESCRIPTION OF A PREFERRED EMBODIMENTA noise diagnostic system 20, free of both nonuniqueness and illposedness difficulties, according to the present invention is shown in FIG. 1. The noise diagnostic system 20 generally comprises a plurality of transducers 22, or microphones connected to a signal analyzer 24. Alternatively, a digital sound processing computer board could be used as a signal analyzer 24. The noise diagnostic system 20 further includes a computer 26 receiving and analyzing the data from the signal analyzer 24.
The noise diagnostic system 20 can be used to diagnose noise sources and quantify their strengths in order to facilitate efforts to reduce noise levels. For illustrative purposes only, and not by way of limitation, the present invention will be shown and described for use in diagnosing the sources of noise from automobile engine 30.
The automobile engine 30 is preferably connected to a dynamometer 36 inside an anechoic chamber 38. At least two transducers 22 are placed adjacent the engine components 42 inside the anechoic chamber 38.
In operation, the transducers 22 measure the frequency and amplitude of noise while the engine 30 is running. The gathered data is sent to the signal analyzer 24 which indicates amplitude as a function of frequency. This frequency/amplitude data is sent to the computer 26. The computer 26 determines the amplitudes of each frequency caused by each engine component 42 by reconstructing the acoustic field on the surface of the components 42 based upon the frequency/amplitude data from the signal analyzer 24. The computer 26 utilizes a method which will be referred to here as the "Helmholtz EquationLeast Squares, or HELS, Method," more fully explained below, to obtain the noise source distribution. This noise source distribution facilitates the reduction of noise by identifying the engine components 42 which are generating certain amplitudes and frequencies.
FIG. 2 is a flowchart overview of the HELS method. In step 50, the solution for the field acoustic pressure is assumed to take the form ##EQU4## In step 52, a particular solution to the Helmholtz equation can be selected in spherical coordinates ##EQU5## In step 54, the independent functions .psi..sub.i * are generated by using the GramSchmidt orthonormalization with respect to .psi. over the source surface. In step 56, the field acoustic pressure p is measured at a plurality of points, using the transducers 22. In step 58, the field acoustic pressure p* is set equal to the measured field acoustic pressure p. In step 60, the coefficients C.sub.i are solved using the least squares method. The acoustic pressure at any point, including the surfaces of the components 42 is then given by ##EQU6##
The Fundamental Theory of the Helmholtz EquationLeast Squares Method utilized by the noise diagnostic system of the present invention is discussed below. Then Numerical Examples of Reconstructions of Acoustic Fields on Source Surfaces with both separable and nonseparable coordinates are presented.
The radiated acoustic pressure in the field satisfies the wave equation, whose Fourier transformation is the socalled reduced wave equation or the Helmholtz equation.
(p)=v.sup.2 p+k.sup.2 p=0, (3)
where a hat for the complex pressure .sub.p is dropped for brevity and k is the wave number.
Equation (3) may be subject to one of the following three types of boundary conditions ##EQU7## where .sub.x .beta. .epsilon..differential..beta., a, b, and g are specified, and .differential./.differential.n represents the normal derivative on the boundary .differential..beta..
Solutions to Eq. (3) subject to boundary conditions (4) can be approximated by a linear combination of the independent functions .psi.* ##EQU8## where p and c are the density and speed of sound of the fluid medium, respectively. The independent functions .psi..sub.i *, can be selected in such a way that they satisfy any one of the following three conditions (Birkhoff and Lynch, 1984):
(i) .psi..sub.i * satisfy the differential equation, but not the boundary condition;
(ii) .psi..sub.i * do not satisfy the differential equation, but .psi..sub.i * satisfies the boundary conditions and .psi..sub.2 *, .psi..sub.3 *, . . . satisfy the homogeneous boundary conditions; or
(iii) .psi..sub.i * satisfy neither the differential equation nor the boundary conditions.
For a differential equation in multiple dimensions
p(x)!=0, x.epsilon..beta. (6)
subject to the boundary condition with an operator A
Ap(x.sub..beta.)!=G(x.sub..beta.), x.sub..beta. .epsilon..differential..beta. (7)
we can select the independent functions .psi..sub.i * that satisfy any one of the following three integrals (Lieberstein, 1960) and then seek the values of the coefficients C.sub.i ##EQU9## which render the integrals in Eq. (8) minimum. Here W.sub.i,, i=1, to 3, are the weighting functions, and dV and dS are integration elements in the region .beta. and on the boundary .differential..beta., respectively. Once the coefficients C.sub.i, are determined, the acoustic pressure anywhere can be approximated by Eq. (5).
In the present invention, attention is focused on the Helmholtz equation (3) subject to the boundary condition (4), and the independent functions .psi..sub.i * that satisfy the condition (i) are selected.
IIAn important step in the present method is to orthonormalize the independent functions .psi..sub.i * on the source surface with respect to the particular solutions .psi..sub.i * to the Helmholtz equation (3). In terms of the spherical coordinates the Helmholtz equation can be written as ##EQU10## subject to the Dirichlet boundary condition, Eq. (4), and the Sommerfeld's radiation condition, ##EQU11##
The approximate solution to Eq. (9) can be written in the form of Eq. (5), with .psi..sub.i * being generated by a linear combination of the particular solutions to the Helmholtz equation .psi..sub.i * (Vekua, 1953) ##EQU12## where h.sub.m and P.sub.n,m denote the spherical Hankel function and the Legendre function, respectively. The former corresponds to an outgoing wave, appropriate for situations where the acoustic energy is radiating outward into an unbounded medium. The amplitude of such a wave tends to infinity at r=0. However, this does not preclude using it for the present purpose since no physical source has zero radius, and the point r=0 is excluded in the function that is valid only outside the source. For a wave in the interior of an enclosure, where r may be zero, we define the spherical Bessel function .Fourier..sub.m (z) as the real part of h.sub.m (z), and the spherical Neumann function N.sub.m (z) as the imaginary part of h.sub.m (z) (Morse and Ingard, 1968),
h.sub.m (z)=.Fourier..sub.m (z)+iN.sub.m (z), .Fourier..sub.m and N.sub.m real (12)
The imaginary part N.sub.m are subsequently discarded because they have no physical meaning as r.fwdarw.0. The real part .Fourier..sub.m are retained as the particular solutions to the Helmholtz equation for an interior problem, such as for diagnosing noise sources in a passenger compartment of a vehicle.
The functions .psi..sub.i given by Eq. (11) are readily applicable to a spherical source, but may not be ideal for an irregularlyshaped source geometry, especially for those that contain sharp edges, for then a large number of terms in the expansion may be required. However, in engineering applications true sharp edges are rare. They are often rounded. Also, the radiated acoustic pressure from a finite, irregularlyshaped source obeys the spherical spreading law in the far field. Hence Eq. (11) may still be used as an appropriate approximation.
The independent function .psi..sub.i * can now be generated by the GramSchmidt orthonormalization with respect to .psi..sub.i (Pinsky, 1991), ##EQU13## where the inner products are taken over the source boundary .differential..beta., ##EQU14##
The independent functions .psi..sub.i * thus obtained are orthonormal for any source surface
(.PSI..sub.i *,.PSI..sub.j *)=.delta..sub.ij (15)
where .delta..sub.ij is the Dirac delta function. Also, they are uniformly convergent because .psi..sub.i consist of a uniformly convergent series of Legendre functions (Bergman, 1960).
IIITo demonstrate the orthonormalization process discussed above, let us consider the following Laplace equation (Kantorovich and Krylov, 1958) subject to boundary conditions ##EQU15##
The harmonic function u is sought in a rectangular region 1,1; 1,1! with the specified Dirichlet boundary condition (17). The approximate solution to Eq. (16) can be written as ##EQU16##
The first step in solving this problem is to choose a complete set of particular solutions .psi..sub.i to the Laplace equation (16), from which the independent functions .psi..sub.i * can be generated. Although trigonometric functions are a good choice for the particular solutions to a rectangular region, we still use (r.sup.n cos n.theta., r.sup.n sin n.theta.) to form the particular solutions for the purpose of demonstration. The first few terms of the particular solutions can be written as 1, (x,y), (x.sup.2 y.sup.2,2xy), (x.sup.3 3xy.sup.2, 3x.sup.2 yy.sup.3), (x.sup.4 6x.sup.2 y.sup.2 +y.sup.4,4x.sup.3 y4xy.sup.3), . . . .
Next, we apply the GramSchmidt orthonormalization to the particular solutions .psi..sub.i to generate the independent functions .psi..sub.i *. Because of the symmetry of the boundary condition (17) on .psi..sub.i, we only need consider the even functions, namely, 1, (x.sup.2 y.sup.2), (x.sup.4 6x.sup.2 y.sup.2 +y.sup.4), . . . .
The first particular solution is .psi..sub.1 =1, whose normalization with respect to the boundary .differential..beta. is ##EQU17## Hence the first independent function is .psi..sub.1* =1/.sqroot.8.
For the second function .psi..sub.2 =(x.sup.2 y.sup.2), we first calculate the inner product ##EQU18## which means that .psi..sub.1 * and .psi..sub.2 are orthogonal. Thus, we can set x.sub.2 =.psi..sub.2. The normalization of x.sub.2 yields ##EQU19## Therefore the second independent function .psi..sub.2 *=.sqroot.15/64(x.sup.2 y.sup.2).
The third independent function .psi..sub.3 * can be obtained in a similar manner. The inner products of .psi..sub.3 with respect to .psi..sub.1 * and .psi..sub.2 * are ##EQU20## From Eq. (13), x.sub.3 is given by ##EQU21## After the normalization we obtain ##EQU22##
Once .psi..sub.i * are specified, the coefficients D.sub.i in Eq. (18) can be determined by requiring u* to satisfy the prescribed boundary conditions (17) ##EQU23## where (x.sub..beta. y.sub..beta.) .epsilon..beta..
Since .psi..sub.i * are orthonormal with respect to the boundary .differential..beta., we can solve D.sub.i by multiplying both sides of Eq. (19) by .psi..sub.j * and then integrating over .differential..beta. ##EQU24##
Substituting Eqs. (20) into (18) yields the approximate solution u* ##EQU25## In this case, the approximate result at origin is u(0,0)=0.814 and the exact solution is 0.816. More accurate results can be obtained by taking more expansion terms in Eq. (18).
IVNow we derive the formulation for reconstructing the acoustic field on the surface of a vibrating object based on measurements of the radiated acoustic pressure in the field. Such a process is referred to here as the Helmholtz EquationLeast Squares (HELS) method, because it essentially solves the Helmholtz equation using the least squares method.
Consider a closed, smooth, and impermeable surface .differential..beta. immersed in an unbounded medium .beta.. Assume that the surface is vibrating at a constant angular frequency .omega., and that there are no other sources in the medium except the vibrating surface. The acoustic pressure at an arbitrary field point p then satisfies the Helmholtz equation (9) and the Sommerfeld radiation condition (10).
Now, given the source location, geometric shape, and boundary conditions at the measurement points in the field
p(x.sub.s)=p.sub.0 (x.sub.s), (22)
we wish to reconstruct the acoustic field on the source surface.
The first step in solving this problem is to express p* as a sum of independent functions .psi..sub.i *, Eq. (5). Here the independent functions .psi..sub.i * are generated from a complete set of particular solutions .psi..sub.i, given by Eq. (11), through the GramSchmidt orthonormalization, Eq. (13). The inner products in the orthonormalization process must be taken over the entire source boundary .differential..beta. using the Gaussian quadrature technique (Wang, 1995).
Next, the coefficients C.sub.i associated with the independent functions .psi..sub.i * are determined by requiring the assumed form of solution, Eq. (5), to satisfy the boundary conditions at the measurement points, Eq. (22). Suppose that an N term expansion in Eq. (5) is used, and that M measurements are taken, where M.gtoreq.N. Then we can form Msimultaneous algebraic equations for N unknowns ##EQU26##
If the measured data p.sub.0 are exact, then the approximate solution p* converges to the true value as N.fwdarw..infin. (Davis and Rabinowitz, 1961). However, in reality the measured data always contain certain amount of uncertainties either due to random fluctuations or due to the effect of evanescent waves that decay rapidly as they propagate away from the source. Hence, the approximate solution will not converge to the true one. Nevertheless, the error involved in the approximate solution can be minimized by the least squares method ##EQU27##
Substituting the left side of Eq. (23) into (24) and taking the derivatives of the resultant equation with respect to each individual coefficient C.sub.i, we obtain ##EQU28##
Equation (25) can be put in a matrix form
.Fourier.!.sub.M.times.N {C}.sub.N.times.1 ={.beta.}.sub.M.times.1, M.gtoreq.N, (26)
where .Fourier.! represents the transformation matrix that correlates the field measurement to the acoustic field on the source surface, and .beta.! is the matrix that contains the boundary condition. The elements of .Fourier.! and .beta.! are given by ##EQU29##
Note that the transformation matrix .Fourier.! is nonsingular. Hence the coefficients C.sub.i can be solved by inverting the matrix .Fourier.!
{C}.sub.N.times.1 =.Fourier.!.sub.N.times.M.sup..mu. {.beta.}.sub.M.times.1, (28)
where
.Fourier.!.sub.N.times.M.sup..mu. =(.Fourier.!.sup.T .Fourier.!).sup.1 .Fourier.!.sup.T
is called the pseudoinverse, here a superscript T stands for the transposition of a matrix. It can be shown that the condition number of the pseudoinverse is much smaller than that of a direct inverse (Stewart, 1973), hence the system of equations (28) is stable and the accuracy of numerical results for {C} is high.
Note that in deriving Eq. (28) no restrictions have been imposed on the measurement points. They need not be confined to a surface conformal to the source boundary. They can be taken at any point in the field so long as they do not overlap each other. Of course the more measurement points are taken, the more accurate the reconstruction will be. It will be shown, however, that using the present method, a satisfactory reconstruction can be obtained with much less measurement points as compared with the traditional Kirchhoff integral formulation approach. The only thing one needs to keep in mind is that the orthonormalization must be taken with respect to the source surface .differential..beta., but not the measurement surface.
Once the coefficients C.sub.i are solved, the surface acoustic pressure can be determined simply by substituting the surface coordinates into Eq. (5). In fact, the acoustic pressure is now expressed as a function of the measurement coordinates. Hence Eq. (5) can be used to predict the radiated acoustic pressure as well.
We emphasize here that the present method uses the expansion theory to solve the Helmholtz equation directly, and the errors involved in the expansion are minimized by the least squares method. Hence it is free of the nonuniqueness and illconditioning difficulties inherent in the Kirchhoff integral formulation approach. It will be shown that such an approach has a high tolerance in the inaccuracy in the measured data. This characteristic is very appealing because in engineering applications all measurements contain certain levels of uncertainties.
VOne of the advantages that accrues from the least squares method is that it makes the error estimation simple and readily available once the computation is completed. The proof of the least squares error and degree of convergence are wellknown, see for example the papers by Nehari (1956) and Dais and Rabinowitz (1961). Examples of estimating the least squares error can also be found in the paper of Hochstrasser (1958). For completeness, we simply cite a brief account of the least squares error analysis by Davis and Rabinowitz (1961).
Let .beta. be a convex domain and let u(z)(=u(x,y)) be harmonic in .beta.. Designate its values on .differential..beta. by u(s). Let u.sub.1, u.sub.2, . . . , u.sub.n be n harmonic functions which are orthonormal in the sense that ##EQU30##
If a.sub.n are the Fourier coefficients of u with respect to u.sub.n, ##EQU31## Here z.sub.p designates a point on .differential..beta. and Eq. (31) is valid for z .epsilon..beta.. When the system u.sub.1, u.sub.2, . . . , is complete in the space of harmonic functions u(z) with .intg..sub..differential..beta. u.sup.2 (s)ds<.infin., the first bracket in the righthand side approaches zero (Parseval's equation) and Eq. (31) exhibits the pointwise convergence of the Fourier series .SIGMA..sub.m.sup..infin..sub.=1 a.sub.m u.sub.m (z) to the solution of the first boundary value problem.
In the present situation, the sought function is p, which is specified at the measurement point as p.sub.0 (.sub.x s), and the approximated one is p*, which is expressed as a sum of independent functions .psi..sub.i * given by Eq. (5). The mean square of the difference between p(.sub.x ) at any point .sub.x in the field and the approximated one pc.SIGMA..sub.i.sup.N.sub.=1 C.sub.i .psi..sub.i *(.sub.x ) is bounded by ##EQU32## where the integrals on the right side of Eq. (32) are carried over the measurement surface.
Note that the functions .psi..sub.i * are complete and bounded (because .psi..sub.i are complete and bounded), and the radiated acoustic pressure p is harmonic and bounded ##EQU33## Hence if the measured value of p.sub.0 is exact, the first bracket on the right side of Eq. (32) is identically zero and the approximate solution pc.SIGMA..sub.i.sup.N.sub.=1 C.sub.i .psi..sub.i *(.sub.x ) converges to the true solution as N.fwdarw..infin.. In reality, however, the measured acoustic pressure p.sub.0 always contains certain level of uncertainties. Therefore the first bracket on the right side of Eq. (32) is of a finite value.
NUMERICAL RESETSThe HELS method developed above is used here to reconstruct the acoustic fields on various source surfaces. Sources of both separable and nonseparable coordinates are considered. Excellent agreements between the reconstructed acoustic fields and the preselected ones are obtained in all cases. A sensitivity analysis of the HELS method on the level of uncertainties involved in the input data is also carried out. To examine the robustness of the HELS method, some artificial bias and random errors are introduced into the input data. These erroneous input are then used to reconstruct the acoustic field on the source surface, and the magnitudes of the errors are compared to those of the input. Numerical results demonstrate that while the bias errors are transferred to the reconstruction at the same level, the random errors are actually reduced. Under no circumstance will the errors be amplified during the reconstruction. Further, the measurement points need not be limited to the near field or to a conformal surface. They can be anywhere so long as they do not overlap each other.
VIIn this section, numerical examples of the reconstruction of the acoustic fields on various types of source surfaces are shown. Sources with both separable and nonseparable coordinates are considered. In all cases, a surface velocity distribution is preselected. The acoustic pressures on the source surface and in the field are then calculated. The field acoustic pressures are taken as the measured quantities to reconstruct the surface acoustic pressures, which are subsequently compared with the predetermined ones on the surface.
(i) Sources with separable coordinates
The examples concerned with sources with separable coordinates include a dilating sphere, an oscillating sphere, and a vibrating piston set in a sphere. In what follows, we first consider the case of a dilating sphere. The complex amplitude of the radiated acoustic pressure from a dilating sphere is given by (Pierce, 1981) ##EQU34## where p and c are the density and speed of sound of the fluid medium, respectively, V.sub.s is the amplitude of the surface velocity, and a is the radius of the sphere.
The approximated acoustic pressure is given by Eq. (1). The independent functions .psi.i* are generated by the GramSchmidt orthonormalization with respect to the particular solutions .psi..sub.i to the Helmholtz equation
.psi..sub.i (r,.theta.)=h.sub.i (kr)P.sub.i (cos .theta.), (3)
where h.sub.i and P.sub.i denote the spherical Hankel function and the Legendre function, respectively. Note that because of the axisymmetry the dependence of .psi. on the azimuth angle .phi. is omitted.
Since the functions .phi..sub.i are orthogonal on the spherical surface, we can directly set .psi..sub.i *=.psi..sub.i. The coefficients C.sub.i associated with .psi..sub.i * are determined by requiring p* to satisfy the pressure boundary condition at the measurement points. Suppose that an N term expansion in Eq. (1) is used, and that M measurements are taken, where M.gtoreq.N. Then we have Msimultaneous algebraic equations for N unknowns. The errors involved in C.sub.i are minimized by the least squares method and their results are given by
{C}.sub.N.times.1 =.Fourier.!.sub.N.times.M.sup..mu. {.beta.}.sub.M.times.1, (4)
where
.Fourier.!.sub.N.times.M.sup..mu.
represents a pseudoinverse, and the elements of .Fourier.! and {.beta.} are specified in Part I of this paper.
Table 1 lists the calculated values of the first eight coefficients C.sub.i for the dilating sphere at ka=1 and V.sub.s =1 (m/s), which agree very well with the exact solution. In particular, numerical results show that we actually only need one term expansion to approximate the radiated acoustic pressure p. In other words, the acoustic pressure on the surface of a dilating sphere can be reconstructed, theoretically, by taking one measurement in the field only. Of course in practice we would take a few measurements in order to minimize the errors involved in the approximation.
Next, we consider the case of an oscillating sphere. The complex amplitude of the radiated acoustic pressure in this case is given by (Pierce, 1981) ##EQU35## where
cos .theta.=n.multidot.e.sub.R, here n
is the unit normal on the surface and
e.sub.R
is the unit normal in the direction of wave propagation from the source to the receiver.
To reconstruct the acoustic field on the surface of the oscillating sphere, we follow the same procedures as those outlined in the above. Table 2 shows the comparison of the calculated values of the first eight coefficients C.sub.i with the exact solution at ka=1 and V.sub.s =1 (m/s). As in the first case, we only need one term in the expansion to approximate the radiated acoustic pressure; or equivalently, only one measurement to reconstruct the acoustic field.
The third example is a vibrating piston set in a sphere. The distribution of the surface velocity is given by ##EQU36## where 2.theta..sub.0 is the vertex angle of the piston. The analytic solution for the radiated acoustic pressure is given by Morse and Ingard (1986) in terms of a series expansion. The accuracy of the result in this case depends on the vertex angle 2.theta..sub.0 and the dimensionless wavenumber ka. The larger the values of 2.theta..sub.0 and ka are, the more terms in the expansion are needed.
Table 3 demonstrates the comparison of the calculated values of the coefficients C.sub.i with the analytic ones given by Morse and Ingard for 2.theta..sub.0 =45.degree., ka=1, and V.sub.0 =1 (m/s). An excellent agreement is obtained once again. In particular, the convergence is achieved with five terms in the expansion, which means that the acoustic pressure on the surface of the sphere can be reconstructed with only five measurements in the field.
(ii) Sources with nonseparable coordinates
Spherical sources as discussed above have separable geometries and therefore, analytical solutions for the radiated acoustic pressures can be solved by using the method of separation of variables. In order to examine the feasibility of the HELS method for sources of nonseparable coordinates, we consider a short cylinder 70 with two spherical endcaps 72 for which there exist no closedform solutions, shown in FIG. 3.
To demonstrate the use of the HELS method, we first select a velocity distribution V.sub.s on the source surface and then solve the surface acoustic pressure p.sub.s by using the BEMbased Kirchhoff integral formulation (Wang, 1995),
{p.sub.s }=(A!2.pi.I!).sup.1 B!{V.sub.s }, (7)
where I! is the identity matrix, and the elements of the matrices A! and B! are given by ##EQU37## where J(.xi.) is the Jacobian of the transformation from the global to the local coordinates, S.sub.m is the area of the mth segment divided on the surface, R.sub.j (.xi.) is the distance from the jth node to every point on the mth area segment, and N.sub.a (.xi.) are the secondorder shape functions of the local coordinates (.xi.).ident.(.xi..sub.1,.xi..sub.2) (Wang, 1995).
In the following numerical examples, the radius of the cylinder 70 is set at a=1 (m) and the half length of the cylinder is b=0.5 (m). In solving the surface acoustic pressure using Eq. (7), the surface is divided into 216 quadrilateral segments with 753 nodal points. Once the surface acoustic pressure is determined, the field acoustic pressure can be calculated by the Kirchhoff integral formulation (Wang, 1995)
4.pi.{p}=A!{p.sub.s }B!{V.sub.s }. (9)
In calculating the field acoustic pressure p, the field measurement points 74 are selected to lie along a line parallel to the generator of the cylinder 70 at a distance of 0.1 (m) away from the surface, as shown in FIG. 3. The values of p thus calculated are taken as the measured data and used to reconstruct the surface acoustic pressure p.sub.s, which are subsequently compared with those obtained from Eq. (7).
The first example deals with a vibrating cylinder 70 with a uniform velocity distribution over the entire surface. The surface and field acoustic pressures are calculated by Eqs. (7) and (9), respectively. The approximate solution for the acoustic pressure is expressed as a series expansion, Eq. (1). Note that in this case the closedform particular solution .psi. cannot be found because the finite cylinder has nonseparable coordinates. However, in the far field the radiated acoustic pressure obeys the spherical spreading law. Hence it is appropriate to use the particular solutions .psi..sub.i for a spherical source as an approximation. The independent functions .psi..sub.i * must still be orthonormalized with respect to the cylindrical surface. The inner products involved in Eq. (13) of the FUNDAMENTAL THEORY section must be taken over the entire cylindrical surface. In carrying out these surface integrals, the Gaussian quadrature with double precision are used. As usual, the accuracy of the surface integral increases with the number of segments divided on the surface and the number of internal integration points. Once .psi..sub.i * are specified the coefficients C.sub.i can be determined by requiring p* to match the field acoustic pressure p. The errors involved are minimized by the least squares method.
FIG. 4 shows the comparison of the reconstructed acoustic pressure distribution p* and the initially calculated one p.sub.s on the surface of the cylinder 70 at ka=1. All acoustic pressures are nondimensionalized with respect to pcV.sub.s and plotted along the source generator S. Numerical results demonstrate that the acoustic pressure on the surface can be reconstructed successfully with only a few measurements, or equivalently, a few terms in the expansion in Eq. (1). In FIG. 4, a .diamond. denotes a reconstruction of the surface acoustic pressure with two terms in the expansion or, equivalently, two measurements of the radiated acoustic pressure in the field, a .largecircle. implies a reconstruction with three terms, and a x with five terms in the expansion. Results show that the accuracy of the reconstruction increases with the number of terms taken in the expansion. With N=5, the maximum error in the reconstruction is already less than five percent.
The approximate solution given by Eq. (1) also allows one to predict the radiated acoustic pressure in the field. FIG. 5 depicts the comparison of the calculated acoustic pressure (solid line) using Eq. (9) and the approximated one (dots) over a circle of radius r=4 (m). The agreement between the two results is remarkable. The radiation pattern in this case resembles that of a dilating sphere except with some extra radiation from the side wall of the cylinder.
The second example is concerned with the same cylinder, but oscillating back and forth along the zaxis direction at ka=1. The same procedures as those described in the above are followed. FIG. 6 demonstrates the comparison of the initially calculated the surface acoustic pressure and the reconstructed one. In FIG. 6, a .diamond. depicts the results of using a fourterm, a .largecircle. the results of using a sixterm, and a x the results of using an eightterm expansion. The maximum error in reconstruction with an eightterm expansion is found to be less than five percent.
FIGS. 7 shows the comparison of the calculated field acoustic pressure (solid line) and the approximated one (dots) using an eightterm expansion over a circle of radius 4 (m). A good agreement is achieved. The radiation pattern in this case resembles that of an oscillating sphere. At higher frequencies, more side lobes are expected (Wang and Wu, 1994) because of the effect of reflection from the side wall of the cylinder.
The last example is concerned with a vibrating piston set in one of the spherical endcaps of the cylinder. The piston has a vertex angle of 2.theta..sub.0 =60.degree. and vibrates at an amplitude of V.sub.0 at ka=1. The rest of the surface is stationary. ##EQU38##
The procedures involved in generating the field acoustic pressures and in reconstructing the surface acoustic pressure are exactly the same as those described in the above.
FIG. 8 demonstrates the comparison of the calculated dimensionless surface acoustic pressure (solid line) and the reconstructed one along the source generator S. A .diamond. represents the result of using a tenterm, a .largecircle. the result of using a fourteenterm, and a x the result of using an eighteenterm expansion. Because of the complexity of the problem, more terms are required to ensure the accuracy of the result. The maximum error in the reconstruction with an eighteenterm expansion is found to be less than twelve percent.
FIGS. 9 illustrates the comparison of the approximated field acoustic pressures (dots) using an eighteenterm expansion with the calculated ones (solid line) over a circle of radius r=4 (m). The agreement is excellent. The radiation pattern in this case resembles that of a vibrating piston set in a sphere (Morse and Ingard, 1986).
VIIIn the conventional Kirchhoff integral formulation approach, reconstruction of the acoustic pressure on the source surface has been found to be very sensitive to the accuracy of the measurements. Because of the inherent illposedness difficulty in this approach, any slight inaccuracy in the measurements will lead to an enormous error in the reconstruction. Hence measurements must be taken at an extremely close distance to include the evanescent wave effect, and the dynamic range of the measurement device must be very large to ensure a high signal to noise ratio. In many applications, these requirements render this type of approach impractical.
With the HELS method, these requirements are no longer necessary. The measurement points can be selected at any point in the field, so long as they do not overlap each other. To examine the robustness of the HELS method, we here intentionally introduce some bias and random errors into the measurements, and then use these erroneous data to reconstruct surface acoustic pressures.
In practice, the bias error may be due to system errors which may be corrected by calibration. On the other hand, the random error may be due to random fluctuations which may be reduced by taking more averages in the measurements. In this sensitivity analysis, the bias errors are introduced by multiplying the calculated field acoustic pressures by a constant, for example, 1.05 for a fivepercent bias error. Similarly, random errors are introduced by multiplying the field acoustic pressures by a random number sequence scaled to the range of one plus or minus a nominal error. For example, a fivepercent random error is generated by a random sequence of numbers between 0.95 and 1.05. These numbers are then multiplied by the field acoustic pressure values.
The first two examples of the sensitivity analysis are concerned with a dilating sphere and an oscillating sphere. Since in these cases the closedform solutions are available, the bias and random errors are introduced by multiplying the exact values of the field acoustic pressures by some preselected constants or random number sequences.
FIGS. 10 and 11 show the relative errors in reconstruction of surface acoustic pressures for a dilating sphere, an oscillating sphere, and a vibrating piston set in a sphere based on these erroneous input data. In these graphs, the horizonal axes represent the generator of the spherical surface expressed in terms of the polar angle .theta. varying from 0.degree. to 180.degree., and the vertical axes are the relatively errors with respect to the exact surface acoustic pressures. The solid lines denote the relative errors in reconstruction due to a five percent bias error, and the dashed lines represent the relative errors due to a tenpercent bias error in the input. Numerical results show that the same magnitudes of bias errors are transferred to reconstruction without any amplification.
FIGS. 10 and 11 also demonstrate the relative errors in reconstruction of the surface acoustic pressures due to a fivepercent and a tenpercent random errors in the input, respectively. A .largecircle. in these graphs represents the relative errors in the reconstruction cue to a fivepercent random error, and a .diamond. indicates the relative errors due to a tenpercent random error in the input. Numerical results demonstrate that the average levels of relative errors in reconstruction are actually lower than those in the input. This implies that the random errors have been effectively reduced by the least squares method. The largest relative error, about 13.8%, is seen at around .theta.=90.degree. for an oscillating sphere. This is because the amplitude of the acoustic pressure is very small there, hence any roundof error in the computations could lead to a large error in the final results. This roundoff error can be further reduced by increasing the number of segments discretized on the surface and the number of measurements taken in the field.
Next, we perform the sensitivity analysis on sources with nonseparable coordinates. As before, we consider a vibrating cylinder, an oscillating cylinder, and a vibrating piston set in one of the spherical endcaps of a cylinder. Because of the lack of the analytical solutions, the field acoustic pressures are calculated numerically using Eq. (9). Therefore there are inherent errors in the input to begin with. When these erroneous data are used to reconstruct surface acoustic pressures, the results are expected to be inaccurate.
In what follows, we first examine the relative errors in reconstruction due to the inherent errors in the input data. The relative error in this case is defined as the ratio of the absolute value of the difference of the numerical results obtained by using Eqs. (1) and (7) to that of Eq. (7). The solid lines in FIGS. 12 to 14 demonstrate the relative errors of the reconstructed surface acoustic pressures of a vibrating cylinder, an oscillating cylinder, and a vibrating piston set in one of the spherical endcaps of a cylinder, respectively. The average levels of the relative errors are less than two percent for a vibrating cylinder (see FIG. 12), five percent for an oscillating cylinder (see FIG. 13), and seven percent for a vibrating piston in a cylinder (see FIG. 14). The largest error occurs around .theta.=90.degree. for an oscillating cylinder for the same reason as that of an oscillating sphere. These inherent errors can be reduced further by increasing the number of segments discretized on the cylindrical surface and the number of internal points in the Gaussian quadrature.
Now we introduce the bias and random errors to the already inaccurate input data and then reconstruct the surface acoustic pressures. The dashed and dotted lines in FIGS. 12 to 14 represent the relative errors in reconstruction of surface acoustic pressures for a vibrating cylinder, an oscillating cylinder, and a vibrating piston in a cylinder due to a fivepercent and a tenpercent bias errors in the input, respectively. The average levels of the relative errors in reconstruction are six percent for a vibrating cylinder, eight percent for an oscillating cylinder, and ten percent for a vibrating piston set in a cylinder when a fivepercent random errors are introduced into the input. The average levels of the relative errors in reconstruction are 12% for a vibrating cylinder, 13% for an oscillating cylinder, and 13% for a vibrating piston set in a cylinder when 10% random errors are introduced into the input. Note that these relative errors are caused by a combined effect of the inherent errors and the bias errors introduced into the input. It is interesting to note that the relative errors due to bias errors fluctuate in essentially the same patterns as those due to the inherent errors. In the cases of spheres as discussed above, there are no inherent errors. Hence the relative errors due to bias errors in the input remain constant.
FIGS. 1517 demonstrate the relative errors in reconstruction of surface acoustic pressures due to a fivepercent and a tenpercent random errors, respectively. The average levels of the relative errors in reconstruction are two percent for a vibrating cylinder, five percent for an oscillating cylinder, and five percent for a vibrating piston set in a cylinder when five percent random errors are introduced into the input. The average levels of the relative errors in reconstruction are 10% for a vibrating cylinder, 11% for an oscillating cylinder, and 12% for a vibrating piston set in a cylinder when 100 random errors are introduced into the input. Once again, these relative errors are the results of a combined effect of the inherent errors and the random errors introduced into the input data. As in the cases of spheres, the average levels of the relative errors in reconstruction due to random errors are actually lower than those due to bias errors. This indicates that the random errors have been effectively reduced by the least squares method.
The examples exhibited above demonstrate that the HELS method has a much higher tolerance in inaccuracy in the input data than the BEMbased Kirchhoff integral formulation approach. Such a feature can be very appealing because in engineering applications all measurements contain certain levels of uncertainties.
Numerical results demonstrate that the HELS method is robust, and can yield satisfactory results in reconstructing the acoustic pressure on the source surface with only a few measurements in the field. It has no restriction on where to take the measurement point and is applicable to sources with both separable and nonseparable coordinates. The accuracy of reconstruction using the HELS method increases with the number of terms used in the expansion, or equivalently, the number of field measurements.
In accordance with the provisions of the patent statutes, the present invention has been described in what is considered to represent its preferred embodiment. However, it should be noted that the invention can be practiced otherwise than as specifically illustrated and described without departing from its spirit or scope.