CN101846594A - Fault detection device based on beam forming acoustic-image mode recognition and detection method thereof - Google Patents

Abstract

The invention relates to a fault detection device based on a beam forming acoustic-image mode recognition and a detection method thereof, belonging to the industrial detection field. When the noise of a noise source is positioned through acoustic image, a sensitive characteristic is extracted by using the characteristic extraction technology of image processing, so as to represent the acoustic images in normal and failed conditions, and then the characteristic vectors are subjected to training classification by using a support vector machine in mode recognition, and finally fault diagnosis and online monitoring are performed. The invention gives consideration to the characteristics of sound source recognition and frequency domain analysis, develops a fault detection device based on beam forming acoustic-image mode recognition and a method thereof by referring the application of the image diagnosis technology in other fields, perfects the traditional acoustic fault diagnosis technology and widens the application range of the beam forming sound source recognition technology.

Description

Form the failure detector and the detection method thereof of acoustic image pattern-recognition based on wave beam

Technical field

What the present invention relates to is a kind of device and method of industrial detection technical field, specifically is a kind of failure detector and detection method thereof that forms the acoustic image pattern-recognition based on wave beam.

Background technology

Mechanical fault diagnosis technology in the industrial and mining enterprises mainly is based on vibration signal measurement and analysis, and formed complete theoretical system and the method for a cover, and as: spectrum analysis technique, the analysis of amplitude parameter index, shock pulse technology, resonance demodulation technique etc.Method for diagnosing faults based on vibration signal must be the surface of transducer arrangements at vibrating machine, yet arrange relatively difficulty for rugged surroundings lower sensors such as complex component vibration surface, moving component, high temperature or greases, only can analyze the vibration signal of some isolated measuring points of vibration surface, therefore vibration information that can only the consersion unit part, and can not get under many circumstances be concerned about that the vibration information of parts, equipment integral vibration situation are difficult to reproduce; The occasion that vibration transducer is installed need be shut down for some simultaneously, bigger economic loss will be brought because shut down to install; In addition, because the diversity of equipment failure, fault signature also has nothing in common with each other, vibration performance and not obvious under some failure condition, and other features (as acoustic feature) are apparent in view, therefore are necessary to seek a kind of effective contactless monitoring and analysis means.

Mechanical noise is containing the important information of machine state, can carry out monitoring and fault diagnosis to the mechanical movement state by noise signal analysis.Based on the fault diagnosis of acoustical signal is the new developing direction of fault diagnosis field in recent years, this acoustics diagnose technological merit is outstanding: non-contact measurement, equipment are simple, speed soon, does not influence equipment operate as normal and on-line monitoring etc., especially can be applied to vibration signal and be difficult for the occasion measured.Perfect acoustics diagnose technical requirement can either can be carried out spectrum analysis simultaneously by the effective recognition sound source position.It is several that existing identification of sound source method mainly contains subjective estimate method, branch's running method, surface strength method, near field measurement method etc., but be subjected to the restriction of physical construction and working environment easily.

Find through retrieval prior art, Chinese patent literature CN1584433, open day 2005-2-23, put down in writing a kind of " based on the noise from air conditioner source discrimination of neural network ", the principle of this technology is to introduce algorithm for pattern recognition to judge sound mainly is by which noise source to be produced, and its shortcoming is to know the main sound source of machinery in advance after known several main sound source spectrum distribution.

The microphone array measuring technique is a kind of important method of research noise source, be based on the directive property principle of microphone array, sound source distribution to body surface is measured, find the overriding noise source position, and obtain the principal character of radiated sound field and the physical mechanism that sound source produces, taken into account the characteristics of identification of sound source and frequency-domain analysis.Wave beam forms (Beamforming) as a kind of identification of sound source algorithm based on microphone array, be widely used with aeronautical and space technology, automobile industry, industrial non-destructive testing technology and military target identification, location and follow the tracks of, and the bibliographical information that is directly used in fault diagnosis seldom.

Further retrieval is found, Patricio Ravetta etc. mentions in inter-noise2009 " Noise Source Identification onRotating Machinery:A Novel Health Monitoring Approach Using Acoustic Phased Arrays " literary composition the method for vibration signal diagnostic analysis is applied in the acoustic signal, mainly is based on sound the shake correlativity and the spectrum analysis of signal.The array acoustical signal can obtain mechanical surface sound radiation pressure relative size distribution pattern after forming through the frequency domain wave beam, can obtain the position of noise source under a certain frequency from figure.

In image processing field, diagnostic techniques based on image is used very extensive, for example Chinese patent application numbers 200710069113.7, denomination of invention is " method and apparatus of differentiating different variety green tea based on the texture analysis of multispectral image ", its principle is the image that obtains different variety green tea by the 3CCD camera, calculate various green tea image texture characteristics and put into sorter and train classification, and then unknown green tea is carried out diagnostic classification.Reference forms a kind of method for diagnosing faults based on the pattern-recognition of wave beam formation acoustic image based on the diagnostic techniques of image, can expand the application of wave beam formation and the development of promotion acoustics diagnose technology.

Summary of the invention

The present invention is directed to the prior art above shortcomings, a kind of failure detector and detection method thereof that forms the acoustic image pattern-recognition based on wave beam is provided, acoustic pressure distribution or sound source position based on machinery a certain frequency under normal and malfunction can change, thereby in acoustic image, produce corresponding the variation, by acoustic image in noise source position, location, use the Feature Extraction Technology extraction sensitive features of Flame Image Process to characterize normally and the acoustic image under the failure condition, by the support vector machine in the pattern-recognition these proper vectors are trained classification and then carried out fault diagnosis and on-line monitoring then.

The present invention is achieved by the following technical solutions:

The present invention relates to a kind of failure detector that forms the acoustic image pattern-recognition based on wave beam, comprise: microphone, array bracket, data acquisition system (DAS), wherein: array bracket is positioned at sound source one side, several microphones are fixedly set on the array bracket and with data acquisition system (DAS) with array way and are connected, storage of collected to time-domain signal and calculate testing result.

The present invention relates to above-mentionedly form the detection method of the failure detector of acoustic image pattern-recognition, may further comprise the steps based on wave beam:

The first step, the machinery with a plurality of noise sources is carried out normal dividing and setting with the fail operation state, the working background noise that machinery is set is consistent, microphone is arranged on the array bracket, then log-on data acquisition system and each time-domain signal passage that microphone is collected is carried out synchronous acquisition obtain acoustical signal p (t).

Second goes on foot, sets up the sample storehouse S of corresponding acoustical signal p (t), carries out spectrum analysis and obtains sensitive frequency F, adopts the frequency domain wave beam based on spherical wave to form to each sample of sample storehouse S:

$B(\stackrel{\→}{\mathrm{\κ}},\mathrm{\ω})=\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{P}_m\left(\mathrm{\ω}\right)e^{-\mathrm{j\ω}{\mathrm{\Δ}}_m\left(\stackrel{\→}{\mathrm{\κ}}\right)}---\left(1\right)$

Wherein: ω represents responsive circular frequency, and M is the microphone number, P _m(ω) be the spectral magnitude of m microphone sound pressure signal at the frequencies omega place,

When focusing on

During direction, m microphone is with respect to the time delay of reference point, and B is the frequency domain output that wave beam forms focus point.

The 3rd step, output is combined into source image restructuring matrix A at the frequency domain of a plurality of focus points, and source image restructuring matrix A is visualized as acoustic pressure distributed image B and locatees the overriding noise source position, realizes the fault detect location.

The 4th step, wave beam is formed the result carry out feature extraction, after obtaining proper vector sample storehouse, put into support vector machine and train classification, contrast the discrimination under the various conditions, classifier parameters after being optimized, promptly punish parameters C and kernel function coefficient gamma, the array signal that collects is in real time carried out feature extraction and adopts the sorter after optimizing to discern, realize fault diagnosis.

Described feature extraction is to handle by in the following dual mode any one:

A) the reconstruct matrix A is carried out svd and obtains singular value features:

$A=\mathrm{U\Λ}{V}^T=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i{u}_i{v}_i^T=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i{A}_i---\left(2\right)$

Wherein: k is the order of matrix A, α _iBe i singular value of matrix A, make α ₁〉=α ₂〉=... 〉=α _n, m singular value constitutive characteristic vector before getting is used for the training classification.

B) the known image gray level is n, the second order joint probability density function P by asking for image (d θ) obtains gray level co-occurrence matrixes for i, j:

M(d，θ)＝[P(i，j，d，θ)] (3)

Wherein: function P has described image distance on the θ direction has the appearance of gray scale i and j respectively for a pair of pixel of d probability.

Described proper vector comprises: angle second moment, contrast, the degree of correlation, entropy, variance, unfavourable balance distance, mean value and variance, average, the poor variance of difference and difference entropy.

The present invention utilize machinery normal with malfunction under the variation of sound pressure amplitude and distribution, use for reference of the application of diagnostic imaging technology at other field, adopt Flame Image Process, feature extraction and mode identification technology that wave beam formation acoustic image is handled.Experiment shows the validity of method, simultaneous verification cooperate noise source location, the recognition function of acoustic imaging technology, image treatment features extraction and mode identification technology are combined with the acoustic imaging technology, widen acoustic imaging The Application of Technology scope, form the feasibility of an effective Acoustic Based Diagnosis technology of cover and widespread use in engineering.

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention.

Fig. 2 is the embodiment device synoptic diagram.

The acoustic image synoptic diagram that Fig. 3 obtains for embodiment.

Embodiment

Below embodiments of the invention are elaborated, present embodiment is being to implement under the prerequisite with the technical solution of the present invention, provided detailed embodiment and concrete operating process, but protection scope of the present invention is not limited to following embodiment.

As shown in Figure 2, the experimental provision of present embodiment comprises: microphone 1, array bracket 2, data acquisition system (DAS) 3, wherein: array bracket 2 is positioned at sound source 4 one sides, several microphones 1 are fixedly set on the array bracket 2 and with data acquisition system (DAS) 3 with array way and are connected, storage of collected to time-domain signal and have computing machine 5 to calculate testing result.

As shown in Figure 1, present embodiment may further comprise the steps:

1, adopts the overriding noise source of three loudspeaker simulation machineries, distance should be greater than the resolution of identification of sound source system each other for loudspeaker, sound source 1 centre coordinate (0.325m wherein, 0.025m), sound source 2 centre coordinate (0.325m,-0.325m), sound source 3 centre coordinates (0.325m, 0.325m).

It when 2, setpoint frequency is for 2000Hz sensitive frequency, simulate the normal and malfunction of each noise source by the sound pressure level size of adjusting loudspeaker, as: sound source 1 sound pressure level is in normal operating condition when changing between 94dB～100dB, all the other situations are malfunction; Sound source 2 sound pressure levels are in normal operating condition when changing between 80dB～90dB, all the other situations are malfunction; Sound source 3 sound pressure levels are in normal operating condition when changing between 65dB～75dB, all the other situations are malfunction.Set 66 in normal condition sample altogether, 63 of various failure operation state sample.

3, arrange microphone array at distance sound source face 0.5m place, present embodiment adopts square array, array surface 0.5m * 0.5m, 6 * 6 of measurement points, the acoustic field signal of each 36 passages of synchronous acquisition, and the volume coordinate position of record microphone.The layout of loudspeaker and microphone array as shown in Figure 3.

4,128 sound field samples are carried out wave beam formation and calculate source image restructuring matrix and acoustic image, Fig. 3 a and Fig. 3 b are respectively the acoustic image under normal and the malfunction.Restructuring matrix is the complex matrix that comprises phase information, and gray level n get respectively 8,16,32,64 and 128 o'clock the visual acoustic image that changes into of matrix.

5, observe the main sound source position, with actual sound source position contrast, wherein reach the imaging algorithm reason more greatly owing to the sound source sound pressure level differs, main sound source 1,2 can be located, and sound source 3 can't be discerned.Restructuring matrix is directly extracted singular value features, calculates the textural characteristics based on gray level co-occurrence matrixes to determining acoustic image after the gray level, form proper vector sample storehouse.

6, use Libsvm to sample training classification, wherein as shown in table 1 based on the classification results of textural characteristics, as shown in table 2 based on the classification results of singular value features:

Table 1 is based on discrimination contrast (%) under each gray level of statistical nature

Can see that by table 1 discrimination is the highest when gray level is 128,90 ° of directions, reach 86.0%, punish parameters C=128, kernel function coefficient gamma=0.5 this moment.

Table 2 is based on the discrimination contrast (%) of singular value

By table 2, the singular value number is that 4 o'clock recognition effects preferably reach 89.1%, punishment parameters C=2048, kernel function coefficient gamma=0.125.

Found through experiments, the discrimination that the direct singular value features of extracting restructuring matrix trains classification to obtain is the highest, under the present embodiment condition, it is best to choose the recognition effect that preceding 4 big singular value constitutive characteristic vectors train classification to obtain, reached 89.1%, sample number quantitative limitation and each sound source sound pressure level difference considerable influence recognition effect, but the validity of method has been described near 90% discrimination.

Claims (4)

1. failure detector that forms the acoustic image pattern-recognition based on wave beam, it is characterized in that, comprise: microphone, array bracket, data acquisition system (DAS), wherein: array bracket is positioned at sound source one side, several microphones are fixedly set on the array bracket and with data acquisition system (DAS) with array way and are connected, storage of collected to time-domain signal and calculate testing result.

2. one kind according to claim 1ly forms the detection method of the failure detector of acoustic image pattern-recognition based on wave beam, it is characterized in that, may further comprise the steps:

The first step, the machinery with a plurality of noise sources is carried out normal dividing and setting with the fail operation state, the working background noise that machinery is set is consistent, microphone is arranged on the array bracket, to the log-on data acquisition system and each time-domain signal passage that microphone is collected is carried out synchronous acquisition obtain acoustical signal p (t);

Second goes on foot, sets up the sample storehouse S of corresponding acoustical signal p (t), carries out spectrum analysis and obtains sensitive frequency F, adopts the frequency domain wave beam based on spherical wave to form to each sample storehouse S:

$B(\stackrel{\→}{\mathrm{\κ}},\mathrm{\ω})=\frac{1}{M}\underset{m=1}{\overset{M}{\mathrm{\Σ}}}{P}_m\left(\mathrm{\ω}\right)e^{-\mathrm{j\ω}{\mathrm{\Δ}}_m\left(\stackrel{\→}{\mathrm{\κ}}\right)}$

Wherein: ω represents responsive circular frequency, and M is the microphone number, P _m(ω) be the spectral magnitude of m microphone sound pressure signal at the frequencies omega place,

When focusing on

During direction, m microphone is with respect to the time delay of reference point, and B is the frequency domain output that wave beam forms focus point;

The 3rd step, output is combined into source image restructuring matrix A at the frequency domain of a plurality of focus points, and source image restructuring matrix A is visualized as the acoustic pressure distributed image and locatees the overriding noise source position, realizes the fault detect location;

The 4th step, wave beam is formed the result carry out feature extraction, after obtaining proper vector sample storehouse, put into support vector machine and train classification, contrast the discrimination under the various conditions, classifier parameters after being optimized, promptly punish parameter and kernel function coefficient, the array signal that collects is in real time carried out feature extraction and adopts the sorter after optimizing to discern, realize fault diagnosis.

3. according to claim 2ly form the detection method of the failure detector of acoustic image pattern-recognition based on wave beam, it is characterized in that, described feature extraction is to handle by in the following dual mode any one:

A) the reconstruct matrix A is carried out svd and obtains singular value features:

$A={\mathrm{U\ΛV}}^T=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i{u}_i{v}_i^T=\underset{i=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{\α}}_i{A}_i$

Wherein: k is the order of matrix A, α _iBe i singular value of matrix A, make α ₁〉=α ₂〉=... 〉=α _n, m singular value constitutive characteristic vector before getting is used for the training classification;

B) the known image gray level is n, the second order joint probability density function P by asking for image (d θ) obtains gray level co-occurrence matrixes for i, j:

M(d，θ)＝[P(i，j，d，θ)]

Wherein: function P has described image distance on the θ direction has the appearance of gray scale i and j respectively for a pair of pixel of d probability.

4. the detection method that forms the failure detector of acoustic image pattern-recognition based on wave beam according to claim 3, it is characterized in that described proper vector comprises: angle second moment, contrast, the degree of correlation, entropy, variance, unfavourable balance distance, mean value and variance, average, the poor variance of difference and difference entropy.

Images