WO2003012778A2

WO2003012778A2 - Method and system for active reduction of sound emissions from drive mechanisms

Info

Publication number: WO2003012778A2
Application number: PCT/DE2002/002459
Authority: WO
Inventors: Jörgen ZILLMAN; Rudolf Maier
Original assignee: Eads Deutschland Gmbh
Priority date: 2001-07-20
Filing date: 2002-07-05
Publication date: 2003-02-13
Also published as: DE10135566A1; DE10135566B4; WO2003012778A3

Abstract

The invention relates to a method for active reduction of sound emissions from drive mechanisms. A sonic field is measured by a plurality of sensors (20;220), said field consisting of a primary sonic field and a superposed secondary sonic field for the reduction of sound. Actuators (230) for producing the secondary sonic field are controlled by a regulating device (215) which minimises the measured sonic field on the basis of a known transfer function between the actuators (230) and the sensors (20;220). Several oscillation modes are determined from the sonic field, whereby the amplitudes thereof are used as an input variable for the regulating device (215).

Description

Process and system for actively reducing engine noise emissions

9. The invention relates to a method and a system for actively reducing the sound radiation from engines according to the preamble of claims 1 and 9, respectively.

In the known methods for actively reducing the sound radiation, the sound field, which is generated, for example, in an engine during operation, is superimposed on an anti-sound field. Actuators are used to generate the anti-noise field. The sound field is measured with sensors or microphones and the anti-sound field is generated with actuators or loudspeakers.

In the known methods, a large number of actuators (N) and sensors (M) are required to reduce the sound radiation. The actuation signals of the actuators are determined from the sensor signals in accordance with the known LMS algorithm. This known control method minimizes the quadratic error. When using this method on aircraft engines, however, there is a large discrepancy in the rate of convergence of individual sound components, which considerably limits the achievable sound reduction. The application of the Newton method therefore tries to adjust the convergence, but leads to problems with the stability. One measure to increase stability is to use an LMS controller with leakage parameters. However, this parameter also limits the achievable noise reduction.

In the document EP0654901 A1, for example, a system for generating a secondary signal is shown, with which a primary signal is extinguished. An error signal is determined by means of sensors and the control signals for the actuators are determined via a correlation matrix. The correlation matrix has an eigenvalue distribution with a value that is essentially equal to 1. In the article "Flight Test of ASAC Aircraft Interior Noise Control System", Dan Palumbo et al in American Institute of Aeronautics & Astronautics, A99-27885, AIAA-99-1933, pp. 852 to 862, a method is described in which To reduce turboprop noise, a sound field is recorded with a large number of microphones, while an anti-sound field is generated with a large number of actuators The actuators are determined from the sensor signals using the LMS algorithm, and the principal component LMS algorithm is used to improve the convergence properties and to limit the actuator performance.

The problem with the known methods and systems is that adaptation must be as quick and efficient as possible, but the computing power available is limited. In engines, especially in aircraft engines, there are therefore usually still very disturbing sound components which cannot be reduced or cannot be reduced sufficiently. The result is noise that is emitted by the engine or aircraft engine and can result in nuisance or damage to health for residents.

The object of the invention is to reduce the radiated sound of an engine, in particular an aircraft engine, even more effectively.

This object is achieved by the method according to patent claim 1 and by the system according to patent claim 9.

Further advantageous features, aspects and details of the invention result from the dependent claims, the description and the drawings.

In the method according to the invention for actively reducing the sound radiation from engines, a plurality of sensors are used to measure a sound field which results from a primary sound field and a secondary sound field superimposed on it for sound reduction, actuators for generating the secondary sound field being controlled by means of a regulating device, which minimizes the measured sound field due to a known transfer function between the actuators and the sensors ocnaiireld several vibration modes are determined, the amplitudes of which are used as an input variable for the control device.

A diagonal matrix is advantageously determined from the transfer function and serves in the control device as a transfer function between the sensors and the actuators.

In particular, the control device can operate according to the LMS method.

The oscillation modes can preferably comprise radial oscillation modes.

For example, the sensors are arranged in the form of rings aligned parallel to one another, with at least one modal amplitude preferably being used as input variable for the control device for each ring.

A temporal and / or a spatial Fourier transformation for individual ring arrangements of sensors with the greatest amplitude is advantageously carried out in order to determine a dominant mode. The vibration modes are preferably dominant vibration modes, e.g. B. can be determined by discrete Fourier transformation.

The weighting coefficients are preferably used to concentrate the control on dominant azimuth and / or radial modes, in which case a multidimensional adaptation algorithm can finally be split into a plurality of one-dimensional control units.

The system according to the invention for actively reducing the sound radiation from engines comprises a large number of sensors for measuring a sound field, a large number of actuators for generating a secondary sound field which is superimposed on a primary sound field of the engine, and a control device for controlling the actuators, the measured one Sound field is minimized, and wherein the control device one or more dominant vibration modes are supplied as input variables, with z. B. a multi-dimensional adaptation algorithm is split into several one-dimensional control units.

The control device is advantageously equipped to carry out the method according to the invention. During operation, the actuators and / or sensors are preferably arranged in the form of parallel rings, the common axis of which coincides with the axis of an engine duct of a sound-emitting engine.

Multi-level signal conditioning, for example, means that the control effort is concentrated on dominant azimuth or radial modes using weighting coefficients. For this purpose, model assumptions about the mode propagation in the channel in particular are integrated into the algorithm. Finally, the multidimensional adaptation algorithm is split up into several one-dimensional control units.

The weighting coefficients used are assigned to physical processes in the engine duct. It is thus possible to specifically influence the radiation characteristics from the end of the channel. A separate step size parameter can be defined for each partial control, which optimally adapts the adaptation speed and enables a manipulated variable weighting of the actuator signals. The effort for the calculation of the modal intermediate quantities is integrated into the calculation of the independent control units and therefore does not require any additional computing time during the control process. Partial regulations with a small contribution to the overall error can be omitted. This significantly reduces the computing time. It is also possible to reduce the computational effort by only considering modal quantities that are dominant in the primary sound field.

The method offers the possibility of reducing a sound field, for the reduction of which a large number of sensors and actuators is required, with minimal computational effort and, at the same time, of specifically influencing the properties of the algorithm by varying control parameters.

In particular, the sound radiation from the intake of aircraft engines, which is essentially caused by the rotor-stator interaction, is reduced or canceled out by the superimposition with an anti-noise field. The spectrum of the radiated sound power has clear tonal components with multiples of the leaf repetition frequencies. These harmonic sound components are canceled out by a suitable regulation.

The invention is described below by way of example using preferred embodiments with associated figures, in which: Fig. 1 shows schematically a channel, in particular engine channel with an annular sensor

Shows order; 2 shows a block diagram of a regulation; 3 shows a block diagram of a further preferred regulation; 4 shows a further block diagram of a regulation; and FIG. 5 shows a system according to the invention on an engine in a schematic representation as

Longitudinal section shows.

FIG. 1 shows as a preferred embodiment of the invention a system for active noise reduction on a duct 10 in the form of an engine duct. A rotor 11 is located within the channel 10. Annular arrangements of sensors 20 are arranged on a wall 12 of the channel 10, each ring 21, 22, 23 extending in the circumferential direction of the channel 10, so that each ring 21, 22, 23 encloses a specific channel cross section at a specific position x1, x2, x3 in the longitudinal direction x of the channel 10. At an inlet opening 13 of the channel 10, a medium flows into the interior of the channel at a speed v, while it exits at an outlet opening 14.

The sound field in channel 10 can be described with the following equation.

in which

x describes the position in the longitudinal direction of the channel 10, r describes the position in the radial direction, φ is position in the azimuthal direction, t is the time, p describes the sound field, σ is the eigenvalue of the vibration mode of order mn. the function f _nm {r „ _m r) d \ e describes the sound pressure distribution of a mode of order mn in the radial direction, m is the index for the order of the mode, which is also evident from the phase dependence of the sound pressure in the circumferential direction φ by the expression e ^{~ Jω>} becomes clear, and n is the index for the radial order of fashion.

The expression in brackets describes the propagation behavior of mode mn in the positive axis direction with complex amplitude A * _m and in the negative axis direction with the amplitude Λ- mn '

Because the sound pressure is only measured on the duct wall, the further description of the sound pressure field is limited to values of the radius r of r = 1 standardized with the duct radius. In the formula (1) the radial distribution for a mode is reduced to a factor fmn (σmn), which can also be taken into account by the amplitudes A * _m and A ^~ _m . Furthermore, it makes sense to combine all radial modes with the same circumferential order. The following formula then results for the wall sound pressure

The propagation behavior of the channel modes is determined by the wave vectors k _M in direction and speed. The vector is composed of the two orthogonal wave numbers k * _n and k _rφ . The wave number in the r-φ plane is determined from the eigenvalue of the mode and the channel radius R.

K _Φ = ^ - (4) The amount of the wave vector k _M corresponds to the free field wave number for a stationary medium

k = -, the quotient of angular frequency and propagation speed. If a uniform flow velocity c is superimposed on the Mach number M of sound propagation, the wave number results as follows:

kli - kl _m + k] _φ = (kM -k _nm ) ² . Converting the equation to k _mn gives the equation

The direction of propagation of the modal wave fronts in the channel is given by the formula

given. For the case M = 0, this angle also corresponds to the main direction with which the acoustic energy is emitted into the free field.

From equation 2 it can be seen that from the Fourier components of the sensor signals (for the frequency to be controlled, a vector E (jω) with elements for each sensor) by means of a discrete spatial Fourier transformation in the azimuthal direction θ, the modal amplitudes A _m on each microphone ring (vector E _M (j <o)) can be calculated. The operation can be implemented by a matrix MS, which is orthonormal if the sensors are arranged equidistantly on each ring.

E _M = MS * E (7)

In a similar way, the excitation of an azimuth mode is realized in that the actuators, like the sensors, are arranged in rings and are linked to one another by a matrix MA. The matrix MA represents the inverse discrete spatial Fourier transformation.

Q _A = MA * Q _M (8) The calculation of the amplitude for mode m on the ring / is independent of the sensor signals on the rings r ≠ i. Each line of MS therefore contains only Si elements that are not equal to 0. Si is the number of sensors on ring i with the azimuthal angular positions φi to φsj. The line of the matrix for the modal amplitude for m = mi on ring i is:

MS _mi = [e ^JmA e ^j " ^h . E ^Φs> \ (9)

The vector for calculating the control signals of the actuators on the ring i with the mode mi results from:

φι bis are the actuator positions.

Figure 2 shows a block diagram as an example of the inventive control.

With the transfer matrix C, the function block 100 represents the propagation processes in the channel between the actuator signals QA and sensor signals E. The further blocks symbolize functions that are implemented by the controller. The sensor signals result from a superposition of the primary sound P with the secondary sound field S generated by the controller. The blocks with the transformation matrices MS and MA have already been explained. The vector EM is used to represent the error in the form of the amplitudes of azimuthal modes.

The diagonal matrix W enables weighting of the individual azimuthal modes within the control process. If n diagonal elements w (i, i) = 0 are set, these modes are excluded from the control. This procedure allows a significant reduction in the computing time required if the primary sound field can only be reproduced by a few modes (eg buzz-saw noise, Tyler-Sofrin mode in rotor-stator interaction). The error function is as follows:

J _w (W real) (11)

Up to this step, the weighted modal amplitudes EW and the modal amplitudes of the actuator control QM form the input and output signals for the LMS algorithm. The control task is then split into independent sub-regulations. The calculations required for this only relate to the transfer function between these two input and output variables, which is specified as follows.

WE _M = E _w = WC _Mod -Q _M

^ WMod ⁼ "'^ Mod (

Using the singular value decomposition (SVD), the transfer matrix C ^^ is converted into a diagonal matrix D, which contains the roots of the real eigenvalues of C ^^ C ^^, and into the right and left-hand eigenvectors in the matrices V and U split.

E _W = V ^H -DU ^H -Q _M (14)

The substitution of VE _w = E _K and U ^H Q _M = Q _K for a diagonal transfer function between S ^ and E _κ ..

E _K = DQ _K (15)

The LMS adaptation rule for minimizing JW is therefore

Because of the diagonal matrix D ", there is a split into partial regulations.

With the vector K, the adaptation steps of each individual partial control are optimized during the control process.

The transfer function CM _od is measured once before the control. Every time the modal weighting factors W are changed, the matrices C _Moώ U, V and D have to be recalculated. Before the regulation begins, the matrices for calculating QA and ER are combined.

Q _A = MA-U- Q _K = TA-Q _K (18)

E _K = V -W-MS-E = TS-E (19)

The error function remains unchanged from the diagonalization.

J _W = E _W -E _W = E _K -V -V -E _K = E _K -E _K (20)

Very small error sizes E) can be excluded from the regulation. If the error function is shown as a function of the actuator signals, it can be seen that partial controls with small diagonal elements D (i, i) can only reduce the error with a very high expenditure of power and are therefore also usefully excluded from the control process.

Figure 3 shows another block diagram as an example of a control. The directional characteristic is influenced.

This embodiment is based on the idea that a further saving in actuator power can be achieved if the control instead of the weighted sum of the amplitude squares of azimuthhalmodes minimizes the sum of the amplitude squares of radial modes Amn. Each radial mode has a dominant skin radiation direction (Equation 6). With w weighting coefficients in the diagonal matrix W, the control can be concentrated on an angular range.

As with the weighting of the azimuthal modes, the radial modes must be determined beforehand. The model of mode propagation in the channel serves this purpose (equation 3). The wavenumbers are given by equations 4 and 5. Equation 3 is now given in matrix notation. A vector with the radial modes Amn, here given for the sake of simplicity only to an azimuth mode order, is linked via a channel matrix Syst with the azimuth modes (vector Am) at the L sensor rings.

A _M = Syst-A " _m

Syst = (22)

₂ ~ ßmQ ^x L

By inverting this channel matrix, the matrix TR is obtained, which enables the radial modes to be calculated from the amplitudes of the azimuth modes.

E _R = TA -E _M (23)

TA = system ^~ (24)

With the expansion of equation 12 on the left and right side with W-TR, the following relationship results for the transfer function between Qmod and EW.

W -TR -C _Mod -Q _M = W -TR -E _M

CψR _ad - W - LR - _Mod

U and V are determined as described above using SVD from matrix C, WRad - The system and method shown reduce, in particular, the sound radiation from the intake of aircraft engines, which is caused by the rotor or the rotor-stator interaction, by superimposition with an anti-noise field. The spectrum of the radiated sound power has clear tonal components with multiples of the leaf repetition frequencies. These harmonic sound components are triggered by a suitable regulation.

In particular, it is possible and often completely sufficient to modify only the radiation characteristics from the entry into the far field. In the case of aircraft engines, for example, only those noise components that radiate to the side of the ground are annoying.

Further possible details of the invention are described below, which can be used in various embodiments:

From the transfer function (system matrix) between actuators and sensors, transformation matrices are calculated, with the help of which the independent coordinates of the control system are formed in a post-processing and preparation stage from the vectors of the actuator and sensor signals. These independent coordinates are used as input and output variables for the LMS controller.

The LMS controller is well known and is described, for example, in Nelson, P.A. and Elliot, S.J .: "Active Control of Sound", Academy Press, London, 1992.

By using the independent coordinates, the system matrix is diagonalized and the complexity of the controller is reduced from the dimension NxM to N. This reduces the computing effort and the stability is increased.

While in general each amplitude of the sensor signals contributes to the error criterion to the same extent, the contributions of the individual coordinates can be very different. The amplitudes of coordinates that only make a small contribution can be considered as noise components and can therefore be excluded from the control. The complexity is further reduced by the factor (NX) / N. The entire control system is split into N or NX independent control clusters with only one input and output variable. To use a Newton algorithm, it is therefore not necessary to invert a fully occupied matrix, but only a diagonal matrix.

In general, the convergence of the controller is adjusted by a step size parameter. The maximum step size is limited by a stability criterion with regard to the largest eigenvalue of the system matrix. Sound components that are associated with smaller eigenvalues converge more slowly. The method described above ensures that the sound components separated according to their own values are fed to a separate control cluster. For each cluster it is possible to define its own step size parameters that are optimal for this sound component or for fashion.

By comparing the error criteria for each cluster, you can quickly determine in which cluster the parameters are most useful to optimize.

FIG. 4 additionally shows a block diagram of a control in the frequency domain. This is an implementation for a narrowband filtered-x-LMS control in the frequency domain. The signals and transfer functions are complex functions of frequency. The current frequency can be adjusted by suitable sensors, e.g. B. can be determined with the aid of a speedometer signal.

Blocks T1, F and T2 are part of the controller. C is the time variant of the transmission matrix, which describes the acoustic properties of the controlled system and is assumed to be known. This is z. B. experimentally determined in experiments.

To generate the secondary sound field, N actuators are arranged in the engine inlet, e.g. B. flush with the channel wall or integrated in the stator blades. These actuators are excited by the controller with the manipulated variables y in order to trigger the primary sound field p. Downstream from the actuators are M acoustic sensors for measuring the reference or error quantity e, which results from the superposition of the primary and secondary sound fields.

The aim of the control using the filtered-x-LMS algorithm is to minimize the amplitude squares of the error signals. A component of the method is the definition of the matrix T1 and T2, with which the input and output signals of the LMS controller (block F) are conditioned in a suitable manner in order to optimally adapt the convergence properties of the LMS algorithm to the acoustic properties of the transmission path can.

As an alternative to the illustration in FIG. 4, the signals y, e, s and p and the matrix C can relate to modal variables in order to minimize suitable error variables which are related to the channel acoustics of the inlet channel. The method can also be used if the sensor signals e are virtual signals of the sound pressure at positions outside the inlet channel where there are no sensors at the time of the control. The virtual signals are calculated by sensors in the inlet channel using a known transfer function. This is e.g. B. in Roure, A. and Albarrazin, A .: The Remote Microphone Technique for Actice Noise Control, ACTIVE99, Ford Lauderdale, USA, pp1233-1244.

4, the matrices T1 and T2 are calculated from the system matrix C. The system matrix is broken down into a diagonal matrix W, which contains the eigenvalues of C.

C = U -W - V ^H (26)

The right and left matrix contains the right-hand and left-hand orthonormal eigenvectors, so that applies

U "- U = V - V ^H = 7. (27)

/ is the unit matrix.

The matrix T1 is calculated as follows:

Tl = K - W ^H - U "(28)

K is a diagonal matrix with any real diagonal elements that are nonzero. Thus the new error size for the LMS algorithm (block F)

ek = K -W "- U ^H -e (29) The new error function is

The matrix T 2 is the conjugate complex matrix of the right-hand eigenvectors of C.

T2 = V (31)

The transfer function of the open controlled system thus becomes a diagonal matrix with purely real diagonal elements.

ck

= S (32)

The LMS algorithm for adapting the manipulated variables yk at time n results in

yk ^{{n +)} = yk ^{n -a -Ck - ek ⁽ⁿ⁾ or yk = yk -a - ß diag (S) _i ^■ ek?> i = 1 ... _V (33)

The regulation was reduced to N one-dimensional regulations (partial regulations). In addition to the global step size α, a real step size vector ß was defined, with which the convergence speed can be optimized for each partial control. If a = 1 is chosen and is

converts each partial control in an adaptation step to the optimal solution (Newton algorithm. Alternatively, K =

can be selected, which results in ß, = diag (S) _i = 1.

Runtimes in the transmission path and a faulty system matrix C can lead to instabilities of individual sub-regulations. The advantage of the method described here is is that instabilities in a partial regulation i can be suppressed by reducing β without impairing the convergence of the other partial regulations.

Each error quantity of the LMS control contributes to the quadratic error of the sensor signals as follows

Sound components e / c, - that with small amounts of eigenvalues

can only be compensated with comparatively strong excitation amplitudes yk, -. If these parts continue to contribute only slightly to the total error according to equation 35, the regulation of these sound parts (partial control i) can be dispensed with. This is in contrast to the LMS regulation. A limitation of the excitation amplitude is achieved with leakage parameters without influencing the convergence properties of the dominating sound components. The computational effort is considerably reduced due to the limitation to Nx partial regulations. A representation of the errors caused by each element of the error vector ek, according to equation 35, shows which error element ek causes the largest error component and consequently which step size parameter β should be increased. It should be emphasized here that due to the orthogonality of the matrix U in equation 28, the vector elements ek in equation 35 are linked by a real diagonal matrix, so that the calculation of the error components can be implemented with minimal effort.

FIG. 5 shows a further embodiment of an engine 200 with a rotor 21 1 in a longitudinal section. Annular arrangements of sensors 220 in the form of microphones are provided on the outside of the wall 212 of the engine duct 210. In the longitudinal direction of the channel 210 upstream, likewise in the region of the wall 212, there are e.g. B. annular arrangements of actuators 230 or speakers to generate the anti-noise field. A controller 215 is coupled to the sensors 220 and actuators 230 to drive the actuators 230 in response to signals generated by the sensors 220. Further control variables, such as. B. an rpm signal, which indicates the current speed of the rotor, are processed in the controller 215.

The inventive method is advantageously controlled by a processor and can, for. B. can also be implemented as a computer program for controlling the controller 215.

Claims

claims

1. A method for actively reducing the sound radiation from engines, in which a plurality of sensors (20; 220) is used to measure a sound field which results from a primary sound field and a secondary sound field superimposed on it for sound reduction, with actuators (230) for Generation of the secondary sound field can be controlled by means of a control device (215), which minimizes the measured sound field on the basis of a known transfer function between the actuators (230) and the sensors (20; 220), characterized in that several vibration modes are determined from the sound field, the Amplitudes can be used as an input variable for the control device (215).

2. The method according to claim 1, characterized in that a diagonal matrix is determined from the transfer function, which serves in the control device (215) as a transfer function between the sensors and the actuators.

3. The method according to claim 1 or 2, characterized in that the control device (215) operates according to the LMS method.

4. The method according to any one of the preceding claims, characterized in that the vibration modes are or include radial vibration modes.

5. The method according to any one of the preceding claims, characterized in that the sensors (20; 220) are arranged in the form of rings (21, 22, 23) aligned parallel to one another, at least one for each ring (21, 22, 23) modal amplitude is used as an input variable for the control device.

6. The method according to any one of the preceding claims, characterized in that a temporal and a spatial Fourier transformation for individual ring arrangements (21, 22, 23) of sensors is carried out at the greatest amplitude in order to determine a dominant mode.

7. The method according to any one of the preceding claims, characterized in that the vibration modes are dominant vibration modes, which are determined by discrete Fourier transformation.

8. The method according to any one of the preceding claims, characterized in that the control is concentrated on dominant azimuth and / or radial modes by means of weighting coefficients, and finally, a multi-dimensional adaptation algorithm is split up into a plurality of one-dimensional control units.

9. System for active reduction of sound radiation from engines with

a plurality of sensors (20; 220) for measuring a sound field, a plurality of actuators (230) for generating a secondary sound field which is superimposed on a primary sound field of the engine, a control device (215) for controlling the actuator (230), wherein the measured sound field is minimized, characterized in that the control device (215) is supplied with one or more dominant vibration modes as input variables, with a multi-dimensional adaptation algorithm being split up into a plurality of one-dimensional control units within the control device (215).

10. System according to claim 9, characterized in that the control device (215) is equipped for performing the method according to one of claims 1 to 8.

1 1. System according to claim 9 or 10, characterized in that the actuators (230) and / or sensors (20; 220) are arranged in operation in the form of parallel rings (21, 22, 23) whose common axis with the Axis of an engine duct (10; 210) of a sound-emitting engine coincides.