CN101426168B - Sounding body abnormal sound detection method and system - Google Patents

Abstract

The present invention provides a method and system for synthesized applying of a series of psychological techniques such as high-level weak signal extraction and human auditory mental model in conventional audio measurement unit and measuring the abnormal sound of the abnormal sound of a sound producer. The present invention adopts a special signal to excite the sound producer to be tested; obtains an abnormal sound curve by measuring current signals and sound pressure signals at both ends of the sound producer and performing time-domain tracking high-pass filtering or EMD process to the two kinds of signals; performing post-processing to the audio response abnormal sound curve by considering the human auditory mental model, to obtain a final audio response abnormal sound curve; and then combining the electrical response abnormal sound. This method can fast and accurately detect whether the abnormal sound exists, which is suitable for production line use.

Description

A kind of sounding body abnormal sound detection method and system

Technical field

The present invention relates to method and test macro that sounding body abnormal sound detects.After obtaining sounding body electroresponse and acoustic response for tested sounding body excitation distinctive signal, by being done, two kinds of signals obtain abnormal sound curve after time domain is followed the tracks of high-pass filtering or EMD processing, consider that people's psychoacoustic model carries out reprocessing to the abnormal sound curve of acoustic response and obtains the abnormal sound curve of last acoustic response, just can judge comprehensively in conjunction with the abnormal sound curve of electroresponse harmony response abnormality sound curve whether sounding body has abnormal sound thereafter.

Background technology

Measure the meaning of sounding body abnormal sound:

The abnormal sound of sounding body can appear at research and development and batch two stages of product of sounding body usually, owing to there is no correct design sounding body structure or selecting suitable material, causes producing abnormal sound when the sounding body vibration sounding when development.Batch product during the stage due to produce in enormous quantities at sounding body and make in have some unpredictable situations, for example: sneak into impurity in magnetic loop, the coil installation site is incorrect, the situation such as individual material is off quality, part position bonding is bad.The capital makes the sounding body of producing send abnormal sound, and these situations generally can not be avoided when producing line production in enormous quantities sounding body fully, and these sounding bodies can be sent out the unacceptable ear-piercing sound of seller when the excitation normal audio.As defect ware, sounding body manufacturing enterprise can not have these product of abnormal sound to consign to their client.At this moment traditional way is that the place one's entire reliance upon tonequality of enterprise's production line is listened and surveyed the workman and pass judgment on by ear, because workman's qualification is different, each workman's ear is different to the subjective feeling of tonequality, some subjective factors of workman in addition, make to be difficult to unify the workman to the judge of the pure tone tonequality of sounding body with an objective standard, cause enterprises production efficiency and product quality to be difficult to improve.Simultaneously, because industry does not have unified standard, the dispute about abnormal sound between manufacturer and client happens occasionally, and the objective judgement of abnormal sound is the insoluble problem of industry always.

Existing technology and method:

A) human auditory system structural model method is mentioned a kind of method that detects sounding body abnormal sound based on the human auditory system structural model in United States Patent (USP) 5884260.The method is with the bank of filters with a plurality of parallel branchs of signal input that measures, and each branch comprises a band pass filter of series connection link, a rectifier and a low pass filter.The passband of each band pass filter and the time constant of each low pass filter are corresponding to every characteristic of human auditory system structural system.Each band pass filter has enough decay beyond their passband, in order to isolate fundametal compoment from harmonic wave.The time constant of the amplitude of each band pass filter and phase response and each low pass filter changes the waveform of this analyzed signal, and is restricted, in order to detect short but distortion that amplitude is very large of duration.This method provides a kind of detection method relevant to the human auditory system structural model, but can not compare with other method of measurement, and is difficult to explain from the objective viewpoint of physics.

B) discrete frequency excitation and detection method, the people such as Pascal Brunet have proposed use discrete frequency excitation sounding body in the paper " detection of loud speaker loose structure " (Loose Particle Detection in Loudspeaker) that the 115th [world] Audio Engineering Society conference delivered, obtain using common high pass filter to carry out filtering after acoustic response, whether signal after filtering is detected tested sounding body by certain algorithm calculating RMS value abnormal sound.The deficiency of the method is 1, adopts the discrete frequency pumping signal just to illustrate is not to produce pumping signal in whole setting frequency range, and the abnormal sound that some only just are triggered in characteristic frequency is not triggered under such excitation.2, a simple high pass, rather than the high-pass filtering of tracking pitch variation makes the method helpless when distinguishing normal harmonic distortion and the abnormal sound of high order.3, the method has very large contradiction on accuracy of detection and test speed.

C) Time-frequency Analysis: the method that the open book 200610011612.6,200610014967.0 of Chinese patent application, 200710178411.X have adopted the signal that detection is obtained to do time frequency analysis is processed, and do again some reprocessings, as Random data processing, pattern learning and pattern classification etc.Their employings of having that different is short time discrete Fourier transform, some employings Wavelet Transform.

Summary of the invention

The objective of the invention is to propose a kind of quick on-line testing sounding body abnormal sound method, the method more reliably and accurately detects sounding body abnormal sound by doing respectively detection after obtaining sounding body electroresponse and acoustic response for tested sounding body excitation distinctive signal.at first detection method can be selected to adopt time domain to follow the tracks of the high pass filter method harmonic wave more than certain exponent number is extracted, perhaps select to adopt the HHT fast algorithm of improvement to do empirical mode decomposition (EMD) to response signal, extract the above decomposed signal of certain exponent number, the high order small-signal from electric response extracted directly done Threshold detection and judged whether abnormal sound thereafter, and the high order small-signal that extracts in acoustical signal is first extracted in conjunction with the abnormal tone signal that the human auditory system mental model really can responsive be heard people's ear, do again Threshold detection and judged whether abnormal sound.Process in conjunction with the signal of two passages like this, consider again simultaneously the detection of people's psychoacoustic model and determination methods can be fast, stablize and judge accurately tested sounding body whether abnormal sound is arranged.

How online, fast, reliably what the present invention will solve is exactly to judge tested sounding body whether abnormal sound is arranged.Original employing discrete frequency will certainly be ignored some potential sounding body abnormal sounds as the detection method of pumping signal, and accurately and on testing efficiency there is contradiction in the method in test result, being not suitable for producing line uses, the processing method in more original proportion territory is used because its Computational Complexity also is not suitable for producing line as Time-frequency Analysis etc.

The object of the present invention is achieved like this:

sounding body abnormal sound detection method of the present invention, be specially: set up test macro as shown in Figure 1, (the time domain waveform figure of test signal as shown in Figure 2 with special test signal, time-frequency figure is as shown in Figure 3) encourage tested sounding body, obtaining tested sounding body both end voltage, the flow through electric current of sounding body, after the sound pressure signal that sounding body sends, doing signal according to Fig. 4 processes, at first use signal to noise ratio detection module (19) to do detection to three signals of gained, secondly extract the signal of telecommunication of tested sounding body and the high-order small-signal in acoustical signal according to the signal that obtains after detecting by high-order weak signal extraction module (20), signal of telecommunication Threshold detection module (21) utilizes the thresholding that signal of telecommunication thresholding generation module (25) generates to do to the signal of telecommunication high order small-signal of extracting the judgement that whether comprises abnormal sound again, and after the high order small-signal of acoustical signal is introduced into the abnormal tone signal that according to the human auditory system mental model, people's ear really can responsive be heard in psychoacoustic model processing module (22) and extracts, entering acoustical signal Threshold detection module (23) judges whether comprising abnormal sound in its acoustic response in conjunction with the thresholding that acoustical signal thresholding generation module (26) generates again, at last the testing result of the testing result harmony signalc threshold detection module (23) of signal of telecommunication Threshold detection module (21) is input in the comprehensive judge module of abnormal sound (24) and comprehensively judges, whether provide tested sounding body has last judgement and the output judged result of abnormal sound.

the test macro of sounding body abnormal sound detection method of the present invention comprises: the constant voltage power amplifier (4) of PC or exclusive equipment (6), data collecting card (5), belt current measuring ability, Sound Attenuator or artificial ear (3), measurement microphone or microphone array (2), PC or exclusive equipment (6) are by program control data capture card (5) emission exciting test signal (7), be carried in tested sounding body (1) two ends after constant voltage power amplifier (4) power amplification of this test signal (7) through the belt current measuring ability, current signal (9) with this amplification voltage signal (8) and the sounding body of flowing through is input in data collecting card (5) simultaneously, use measurement microphone or microphone array (2) that the sound pressure signal (10) that tested sounding body (1) sends under pumping signal is gathered and be input to data collecting card (5) in Sound Attenuator or artificial ear (3), exclusive equipment (6) comprises having the industrial computer of controlling function and Presentation Function, independent embedded device, data collecting card (5) is used for completing mould/number and D/A switch.

Because the present invention has adopted above-mentioned technical scheme, therefore have the following advantages:

1) test macro is built conveniently, and cost is low, and test macro can not only be used for the measurement of sounding body abnormal sound, can also be used for the measurement of other parameters.Such as: frequency sound test, total harmonic distortion test, phase test, testing impedance and the test of sounding body linear dimensions etc.

2) selection of test signal, determined its guarantee to motivate might the prerequisite of sounding body abnormal state under, the testing time is short as much as possible.

3) adopt that two kinds of different high-order weak signal extractions can guarantee not omit with some Weak Anomaly signal extractions out, guaranteed abnormal sound detection accuracy.

4) the high-order small-signal of acoustical signal is carried out the processing of relevant human auditory system mental model, can simulate the processing procedure of people's ear in the audition stage, the abnormal tone signal that people's ear really can responsive be heard extracts, the result that obtains like this is consistent with the result that the actual user work is recognized, and further improves the accuracy that detects.

5) whether the abnormal sound of composite electrical signal and acoustical signal two paths of signals detects and can have abnormal sound detect and judge to sounding body more comprehensive, reliable, careful and accurately, makes result more credible.

Description of drawings

Fig. 1 measurement system diagram.

Fig. 2 test and excitation time domain plethysmographic signal figure.

Fig. 3 test and excitation signal time-frequency figure.

Fig. 4 signal processing flow figure.

Fig. 5 improves HHT fast algorithm flow chart.

The current signal time domain waveform figure that Fig. 6 records.

The sound pressure signal time domain waveform figure that Fig. 7 records.

The different sound curve of Fig. 8 electroresponse and thresholding figure.

Fig. 9 acoustic response high-order small-signal figure.

The different sound curve of Figure 10 acoustic response and thresholding figure.

Figure 11 does not have the acoustic response EMD exploded view of abnormal sound sample.

Figure 12 has the acoustic response EMD exploded view of abnormal sound sample.

Figure 13 does not have the acoustic response instantaneous frequency figure of abnormal sound sample.

Figure 14 has the acoustic response instantaneous frequency figure of abnormal sound sample.

Embodiment

The present invention is further illustrated below in conjunction with drawings and Examples.

According to Fig. 1 test system building, this test macro comprises: the constant voltage power amplifier (4) of PC or exclusive equipment (6), data collecting card (5), belt current measuring ability, Sound Attenuator or artificial ear (3), measurement microphone or microphone array (2), exclusive equipment (6) comprises having controls function and the industrial computer of Presentation Function, independent embedded device, and data collecting card (5) is used for completing mould/number and D/A switch.in this test macro, PC or exclusive equipment (6) are by program control data capture card (5) emission exciting test signal (7), be carried in tested sounding body (1) two ends after constant voltage power amplifier (4) power amplification of this test signal (7) through the belt current measuring ability, current signal (9) with this amplification voltage signal (8) and the sounding body of flowing through is input in data collecting card (5) simultaneously, use measurement microphone or microphone array (2) that the sound pressure signal (10) that tested sounding body (1) sends under pumping signal is gathered and be input to data collecting card (5) in Sound Attenuator or artificial ear (3).Here all signal processing flows carry out in PC or exclusive equipment (6), utilize the program in PC or exclusive equipment (6) to complete.

Be provided with signal to noise ratio detection module (19), high-order weak signal extraction module (20), signal of telecommunication Threshold detection module (21), psychoacoustic model processing module (22), acoustical signal Threshold detection module (23), the abnormal comprehensive judge module of sound (24), signal of telecommunication thresholding generation module (25) harmony signalc threshold generation module (26) in PC or exclusive equipment (6).

Described signal to noise ratio detection module (19) is done the state of signal-to-noise of voltage signal (8), the current signal (9) of input, sound pressure signal (10) and is detected judgement, to provide warning in signal to noise ratio during less than 10dB (this value is different according to the different tested sounding body type of selecting), expression collection signal this moment is not suitable for carrying out the computing of back, if signal to noise ratio more than or equal to 10dB, can further be calculated;

The extraction that described high-order weak signal extraction module (20) will adopt current signal that optional two kinds of algorithms of different obtain detection and sound pressure signal to do the high-order small-signal, a kind of algorithm adopts time domain to follow the tracks of high pass filter, according to the exponent number of setting, the signal of telecommunication or the acoustical signal of input are done the time domain high-pass filtering of following the tracks of fundamental frequency, only keep high-order small-signal part; Another kind of algorithm adopts the HHT fast algorithm of improvement, and the signal of telecommunication and the acoustical signal of input are done empirical mode decomposition (EMD), isolate relatively and the various empirical modals of exciting test signal, and the above empirical modal summation of exponent number is specified in output.

It is as follows that the HHT fast algorithm of improvement obtains the algorithm flow of signal empirical mode decomposition (EMD) in described high-order weak signal extraction module (20), flow chart as shown in Figure 5:

(1) initialization: r ₀=x (t), i=1;

(2) if i＜=N (N represents the exponent number that needs are analyzed) carries out (3) and processes, otherwise current result of calculation is exported in the algorithm end.

(3) screening i intrinsic mode functions (IMF) function, concrete steps are:

(a) initialization: h ₀(t)=r _i(t), k=1;

(b) extract h _k-1(t) local maximum and minimum;

(c) local maximum and local minimum are carried out respectively spline interpolation 3 times, obtain h _k-1(t) envelope up and down;

(d) calculate h _k-1The average m of envelope up and down (t) _k-1(t);

(e) calculate h _k(t)=h _k-1(t)-m _k-1(t);

(f) if satisfy stop condition, i.e. the standard deviation of double the selection result $\mathrm{SD}=\underset{t=0}{\overset{T}{\mathrm{\Σ}}}\frac{{|h_{k-1}\left(t\right)-h_k\left(t\right)|}^2}{h_{k-1}{\left(t\right)}^2}$ In the time of between representative value 0.2 to 0.3, make C _i(t)=h _k(t), otherwise make k=k+1, continue (b) step.

(4) definition r _i(t)=r _i-1(t)-C _i(t).

(5) if r _i(t) still have plural extreme point, i=i+1 is arranged, went to for (3) step.Otherwise decomposable process is completed, r _i(t) be the high-order harmonic wave component that obtains after required N decomposition.

Primary signal is by these IMF components that decomposite and average or trend term (high-order harmonic wave component) expression the most at last for such decomposable process, and expression is as follows:

$X\left(t\right)=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{C}_j\left(t\right)+R_n\left(t\right)$

Check this detection method whether effective method is as follows:

Can adopt IMF component is wherein first done the HIilbert conversion:

$\widetilde{C_j\left(t\right)}=\frac{1}{\mathrm{\π}}\∫\frac{C_j\left(\mathrm{\τ}\right)}{t-\mathrm{\τ}}\mathrm{d\τ}$

Try to achieve again the phase function of analytic signal:

${\mathrm{\θ}}_j={\tan }^{-1}\frac{\widetilde{C_j\left(t\right)}}{C_j\left(t\right)}$

Obtain at last its instantaneous frequency value:

$W_j\left(t\right)=\frac{d{\mathrm{\θ}}_j}{\mathrm{dt}}$

To W _j(t) do detection and just can find out clearly whether testing result is accurate.

Described signal of telecommunication Threshold detection module (21) compares the signal of telecommunication high-order small-signal (being the abnormal sound curve of the signal of telecommunication) that has calculated and the thresholding of setting, there is the situation that exceeds thresholding just to be output as 0, be illustrated in the tested sounding body signal of telecommunication abnormal sound is arranged, do not surpass thresholding and just export 1, represent in the tested sounding body signal of telecommunication without abnormal sound.

Described signal of telecommunication thresholding generation module (25) can select to set an absolute value according to input message, also can will manually listen survey certainly do not have the sample of abnormal sound first to carry out one or many test of the present invention and the electroresponse high-order small-signal result that it obtains is done a statistical average, then with reference to suitably setting out corresponding thresholding after surplus.Computing formula is as follows:

${\mathrm{Limit}}_c=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}C\left(i\right)+{m\arg \mathrm{in}}_c$ Or Limit _c=Max{C (i) }+margin _c

Limit wherein _cExpression calculates the electroresponse thresholding, and N represents that people's ear listens survey certainly there is no the number of abnormal sound sample, the electroresponse high-order small-signal result that C (i) expression people ear listens survey certainly not have abnormal sound sample to obtain through the present invention's test, margin _cExpression considers that product detects redundancy and the suitable threshold margin that increases.Two formula can be selected arbitrarily as required, but should be noted that the sample requirement of setting thresholding when setting strict, this with regard to needs have one reject mechanism with test result obviously and the inconsistent sample test result of other test results weed out and do not add up the generation of thresholding.Chauvenet standard or Peirce standard have been adopted in the present invention.

Described psychoacoustic model processing module (22) is done reprocessing according to the human auditory system mental model to the acoustical signal high-order small-signal of input, the real abnormal tone signal that can responsive hear of people's ear is extracted obtain the abnormal sound curve of acoustic response.Here the human auditory system mental model comprises: temporal masking, frequency domain masking effect, human auditory system Scale Model of Loudness etc.Temporal masking is mainly simulated the effect of mutually sheltering that occurs in not two or more different tone signal in the same time specifically, the frequency domain masking effect is mainly simulated at synchronization fundamental frequency and the low order harmonics masking effect to different tone signal, and the human auditory system Scale Model of Loudness is mainly simulated people's ear weighting characteristic to different frequency loudness under different condition.

Described acoustical signal Threshold detection module (23) compares the thresholding of the abnormal sound curve of acoustic response and setting, there is the situation that exceeds thresholding just to be output as 0, be illustrated in tested sounding body acoustical signal abnormal sound is arranged, do not surpass thresholding and just export 1, represent in tested sounding body acoustical signal without abnormal sound.

Described acoustical signal thresholding generation module (26) can select to set an absolute value according to input message, also can will manually listen survey certainly do not have the sample of abnormal sound first to carry out one or many test of the present invention and the acoustic response high-order small-signal result that it obtains is done a statistical average, then with reference to suitably setting out corresponding thresholding after surplus.Computing formula is as follows:

${\mathrm{Limit}}_A=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}A\left(i\right)+{m\arg \mathrm{in}}_A$ Or Limit _A=Max{A (i) }+margin _A

Limit wherein _AExpression calculates the electroresponse thresholding, and N represents that people's ear listens survey certainly there is no the number of abnormal sound sample, the acoustic response high-order small-signal result that A (i) expression people ear listens survey certainly not have abnormal sound sample to obtain through the present invention's test, margin _AExpression considers that product detects redundancy and the suitable threshold margin that increases.Two formula can be selected arbitrarily as required, but should be noted that the sample requirement of setting thresholding when setting strict, this with regard to needs have one reject mechanism with test result obviously and the inconsistent sample test result of other test results weed out and do not add up the generation of thresholding.Chauvenet standard or Peirce standard have been adopted in the present invention.

The comprehensive judge module of described abnormal sound (24) to the signal of telecommunication Threshold detection of input as a result harmony signalc threshold testing result do a comprehensive judgement, only have when two detections all by the time just provide sounding body without the detection judgement of abnormal sound, when there being any one to detect the obstructed out-of-date detection judgement that tested sounding body has abnormal sound that provides, can be more flexibly in actual the use logical combination of testing result be judged, combination table is as follows:

Signal of telecommunication Threshold detection result	Acoustical signal Threshold detection result	The comprehensive detection result
			Abnormal sound is arranged	Abnormal sound is arranged	Abnormal sound is arranged
Without abnormal sound	Abnormal sound is arranged	Abnormal sound or indefinite is arranged Annotate
			Without abnormal sound	Without abnormal sound	Without abnormal sound
Abnormal sound is arranged	Without abnormal sound	Abnormal sound is arranged

Annotate: may be that acoustic environment is influential to acoustical signal Threshold detection result on every side this moment, can consider again to detect once, with the result that obtains determining again.

Whether signal flow graph recited above can be considered to reconfigure in actual use, have abnormal sound to make a decision such as only detecting the signal of telecommunication or acoustical channel to tested sounding body.

The detailed process that the present invention measures the sounding body linear dimensions is:

Test macro is set according to the user by PC or exclusive equipment (6) and is produced an exciting test signal (7), and this exciting test signal (7) is the continuous logarithmic swept-frequency signal, is defined as follows formula:

$\mathrm{Stim}\left(t\right)=U\sin \left[\frac{{\mathrm{\ω}}_{1}T}{\ln \left(\frac{{\mathrm{\ω}}_2}{{\mathrm{\ω}}_1}\right)}(e^{\frac{t}{T}\ln \left(\frac{{\mathrm{\ω}}_2}{{\mathrm{\ω}}_1}\right)}-1)\right]$

Wherein: U is the test signal amplitude, and T is the test signal time, ω ₁The initial frequency of test signal, ω ₂The termination frequency of test signal (7), time domain waveform as shown in Figure 2, time-frequency figure is as shown in Figure 3.This test signal (7), have the characteristics such as frequency changes continuously, the testing time is controlled, select the benefit of this pumping signal to be: 1, this frequency sweep form will encourage all frequencies in the setpoint frequency section rather than some discrete frequencies of selection of logarithm saltus step, so just can detect in whole frequency range whether tested sounding body has abnormal sound and can not cause some potential abnormal sounds not to be energized out because of the problem of the signal of excitation.2, can under the condition that guarantees measuring accuracy and accuracy, accelerate to greatest extent test speed.

by data collecting card (5), test signal (7) is input in power amplifier (4) and with the amplification voltage signal (8) of exporting and is input to simultaneously tested sounding body (1) and data collecting card (5), and the current signal (9) of tested sounding body (1) of flowing through is input in data collecting card (5), the acoustical signal that tested sounding body (1) is launched in Sound Attenuator or artificial ear (3), gather acoustical signal (10) and be input in data collecting card (5) with measuring microphone or microphone array (2).

recording desired signal (8), (9), (10) after, the signal that these signals carry out is as shown in Figure 4 processed, detect the abnormal sound that whether has of tested sounding body, specific as follows: at first with the voltage signal (8) that records, whether the signal to noise ratio that current signal (9) and sound pressure signal (10) are input to the signal of the gained of judgement test in signal to noise ratio detection module (19) meets the demands, if less than 10dB (this value is different according to the different tested sounding body type of selecting), output warning signal (17), prompting test signal signal to noise ratio is not enough, need to take to increase the test signal amplitude this moment, reduce the measures such as noise, if detection signal-to-noise ratio greater than 10dB, is input to voltage signal (8), current signal (9) and sound pressure signal (10) in high-order weak signal extraction module (20).The current signal that records in this example (9) as shown in Figure 6, sound pressure signal (10) is as shown in Figure 7

The extraction that high-order weak signal extraction module (20) will select current signal that one of two kinds of algorithms of different obtain detection and sound pressure signal to do the high-order small-signal according to the algorithm control information (18) of input.

A kind of algorithm is to adopt time domain to follow the tracks of high pass filter, according to the exponent number of setting, the signal of telecommunication or the acoustical signal of input are done the time domain high-pass filtering of following the tracks of fundamental frequency, only keep high-order small-signal part, and the result that will handle is input to respectively signal of telecommunication Threshold detection module (21) and psychoacoustic model processing module (22).The signal of telecommunication high-order small-signal that in this example, this algorithm process current signal (9) obtains (being the abnormal sound curve of the signal of telecommunication) (11) as shown in Figure 8, the acoustical signal high-order small-signal (13) that the sound pressure signal of processing (10) obtains is as shown in Figure 9.

Described signal of telecommunication thresholding generation module (25) can select directly to set an absolute value, also can will manually listen survey certainly do not have the sample of abnormal sound first to carry out one or many test of the present invention and the electroresponse high-order small-signal result that it obtains is done a statistical average, then with reference to suitably setting out corresponding thresholding after surplus.Computing formula is as follows:

${\mathrm{Limit}}_c=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}C\left(i\right)+{m\arg \mathrm{in}}_c$ Or Limit _c=Max{C (i) }+margin _c

Limit wherein _cExpression calculates the electroresponse thresholding, and N represents that people's ear listens survey certainly there is no the number of abnormal sound sample, the electroresponse high-order small-signal result that C (i) expression people ear listens survey certainly not have abnormal sound sample to obtain through the present invention's test, margin _cExpression considers that product detects redundancy and the suitable threshold margin that increases.Two formula can be selected arbitrarily as required, but should be noted that the sample requirement of setting thresholding when setting strict, this with regard to needs have one reject mechanism with test result obviously and the inconsistent sample test result of other test results weed out and do not add up the generation of thresholding.Adopt the Chauvenet standard to adopt above formula to generate the different sound curve of signal of telecommunication thresholding (22) (as shown in Figure 8) to the input sample signal of telecommunication (19) in this example, and be entered into signal of telecommunication Threshold detection module (21).

Signal of telecommunication Threshold detection module (21) compares the signal of telecommunication high-order small-signal (being the abnormal sound curve of the signal of telecommunication) (11) that has calculated and the different sound curve of the signal of telecommunication thresholding (22) of setting, and output electrical signals judged result (12) is to the comprehensive judge module of abnormal sound (24).Have the situation that exceeds thresholding just to be output as 0, being illustrated in the tested sounding body signal of telecommunication has abnormal sound, does not surpass thresholding and just exports 1, and representing does not have abnormal sound in the tested sounding body signal of telecommunication.In this example test result as shown in Figure 8, the result of output is 1, there is no abnormal sound in the signal of telecommunication of tested artificial body for generating.

Psychoacoustic model processing module (22) is done reprocessing according to the human auditory system mental model to the acoustical signal high-order small-signal (13) of input according to the model control information (27) of input, the real abnormal tone signal that can responsive hear of people's ear is extracted obtain the abnormal sound curve of acoustic response (14).Here the human auditory system mental model comprises: temporal masking, frequency domain masking effect, human auditory system Scale Model of Loudness etc.Temporal masking is mainly simulated the effect of mutually sheltering that occurs in not two or more different tone signal in the same time specifically, the frequency domain masking effect is mainly simulated at synchronization fundamental frequency and the low order harmonics masking effect to different tone signal, and the human auditory system Scale Model of Loudness is mainly simulated people's ear weighting characteristic to different frequency loudness under different condition.Before this example is processed, acoustical signal high-order small-signal (13) result as shown in Figure 9, obtains the different sound curve of acoustic response (14) result as shown in figure 10 after processing through psychoacoustic model processing module (22).

Acoustical signal thresholding generation module (26) can select directly to set an absolute value, also can will manually listen survey certainly do not have the sample of abnormal sound first to carry out one or many test of the present invention and the acoustic response high-order small-signal result that it obtains is done a statistical average, then with reference to suitably setting out corresponding thresholding after surplus.Computing formula is as follows:

${\mathrm{Limit}}_A=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}A\left(i\right)+{m\arg \mathrm{in}}_A$ Or Limit _A=Max{A (i) }+margin _A

Limit wherein _AExpression calculates the electroresponse thresholding, and N represents that people's ear listens survey certainly there is no the number of abnormal sound sample, the acoustic response high-order small-signal result that A (i) expression people ear listens survey certainly not have abnormal sound sample to obtain through the present invention's test, margin _AExpression considers that product detects redundancy and the suitable threshold margin that increases.Two formula can be selected arbitrarily as required, but should be noted that the sample requirement of setting thresholding when setting strict, this with regard to needs have one reject mechanism with test result obviously and the inconsistent sample test result of other test results weed out and do not add up the generation of thresholding.Adopt the Chauvenet standard to generate the different sound curve of acoustical signal thresholding (22) (as shown in figure 10) according to above formula to input sample acoustical signal (21) in this example, and be entered into acoustical signal Threshold detection module (23).

Acoustical signal Threshold detection module (23) compares the different sound curve of the acoustical signal thresholding (23) of the different sound curve of acoustic response (14) and input, and output acoustical signal judged result (15) is to the comprehensive judge module of abnormal sound (24).Have the situation that exceeds thresholding just to be output as 0, being illustrated in tested sounding body acoustical signal has abnormal sound, does not surpass thresholding and just exports 1, represents in tested sounding body acoustical signal without abnormal sound.In this example test result as shown in figure 10, the result of output is 0, and abnormal sound is arranged in the acoustical signal of tested artificial body for generating.

The comprehensive judge module of described abnormal sound (24) is done a comprehensive judgement to signal of telecommunication Threshold detection result (12) the harmony signalc threshold testing result (15) of input, only have when two detections all by the time just provide the detection judgement that sounding body has abnormal sound, when there being any one to detect not by being to provide the detection judgement that tested sounding body has abnormal sound, can be more flexibly in actual the use logical combination of testing result be judged, combination table is as follows:

Signal of telecommunication Threshold detection result	Acoustical signal Threshold detection result	The comprehensive detection result
			Abnormal sound is arranged	Abnormal sound is arranged	Abnormal sound is arranged
Without abnormal sound	Abnormal sound is arranged	Abnormal sound or indefinite is arranged Annotate
			Without abnormal sound	Without abnormal sound	Without abnormal sound
Abnormal sound is arranged	Without abnormal sound	Abnormal sound is arranged

Annotate: may be that acoustic environment is influential to acoustical signal Threshold detection result on every side this moment, can consider again to detect once, with the result that obtains determining again.

In this example, measurement situation again still, illustrates that tested sounding body has abnormal sound really.

Another kind of algorithm in high-order weak signal extraction module (20) adopts the HHT fast algorithm of improvement, the signal of telecommunication and acoustical signal to input are done empirical mode decomposition (EMD), isolate the various empirical modals of relative and exciting test signal, and the above empirical modal summation of exponent number is specified in output.

It is as follows that the HHT fast algorithm of improvement obtains empirical mode decomposition (EMD) algorithm flow of signal in described high-order weak signal extraction module (20), flow chart as shown in Figure 5:

(1) initialization: r ₀=x (t), i=1;

(2) if i＜=N (N represents the exponent number that needs are analyzed) carries out (3) and processes, otherwise current result of calculation is exported in the algorithm end.

(3) screening i intrinsic mode functions (IMF) function, concrete steps are:

(a) initialization: h _O(t)=r _i(t), k=1;

(b) extract h _k-1(t) local maximum and minimum;

(c) local maximum and local minimum are carried out respectively spline interpolation 3 times, obtain h _k-1(t) envelope up and down;

(d) calculate h _k-1The average m of envelope up and down (t) _k-1(t);

(e) calculate h _k(t)=h _k-1(t)-m _k-1(t);

(f) if satisfy stop condition, i.e. the standard deviation of double the selection result $\mathrm{SD}=\underset{t=0}{\overset{T}{\mathrm{\Σ}}}\frac{{|h_{k-1}\left(t\right)-h_k\left(t\right)|}^2}{h_{k-1}{\left(t\right)}^2}$ In the time of between representative value 0.2 to 0.3, make C _i(t)=h _k(t), otherwise make k=k+1, continue (b) step.

(4) definition r _i(t)=r _i-1(t)-C _i(t).

(5) if r _i(t) still have plural extreme point, i=i+1 is arranged, went to for (3) step.Otherwise decomposable process is completed, r _i(t) be the high-order harmonic wave component that obtains after required N decomposition.

Primary signal is by these IMF components that decomposite and average or trend term (high-order harmonic wave component) expression the most at last for such decomposable process, and expression is as follows:

$X\left(t\right)=\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{C}_j\left(t\right)+R_n\left(t\right)$

Select one there is no the sample of abnormal sound and have the sample of abnormal sound to do 5 IMF decomposition of extraction by this algorithm respectively in this example, result is respectively as Figure 11 and shown in Figure 12.

Check this detection method whether effective method is as follows:

Can adopt IMF component is wherein first done the HIilbert conversion;

$\widetilde{C_j\left(t\right)}=\frac{1}{\mathrm{\π}}\∫\frac{C_j\left(\mathrm{\τ}\right)}{t-\mathrm{\τ}}\mathrm{d\τ}$

Try to achieve again the phase function of analytic signal:

${\mathrm{\θ}}_j={\tan }^{-1}\frac{\widetilde{C_j\left(t\right)}}{C_j\left(t\right)}$

Obtain at last its instantaneous frequency value:

$W_j\left(t\right)=\frac{d{\mathrm{\θ}}_j}{\mathrm{dt}}$

To W _j(t) do detection and just can find out clearly whether testing result is accurate.

The 1 rank IMF that in this example, decomposition is obtained does the instantaneous frequency value and processes, result is respectively as Figure 13 and shown in Figure 14, really can find that the testing result medium and low frequency that abnormal sound sample is arranged in Figure 14 partly has a large amount of anomalous audio rate compositions, and in Figure 13 just there there is no the testing result of good sample.

It should be noted that at last: above embodiment is only in order to illustrate the present invention and unrestricted technical scheme described in the invention; Therefore, although this specification has been described in detail the present invention with reference to each above-mentioned embodiment,, those of ordinary skill in the art should be appreciated that still and can modify or be equal to replacement the present invention; And all do not break away from technical scheme and the improvement thereof of the spirit and scope of invention, and it all should be encompassed in the middle of claim scope of the present invention.

Claims (4)

1. sounding body abnormal sound detection method, it is characterized in that: set up test macro, with the special tested sounding body of test signal excitation, obtaining tested sounding body both end voltage, the flow through electric current of sounding body, after the sound pressure signal that sounding body sends, at first use signal to noise ratio detection module (19) to do detection to three signals of gained, secondly extract the signal of telecommunication of tested sounding body and the high-order small-signal in acoustical signal according to the signal that obtains after detecting by high-order weak signal extraction module (20), signal of telecommunication Threshold detection module (21) utilizes the thresholding that signal of telecommunication thresholding generation module (25) generates to do to the signal of telecommunication high order small-signal of extracting the judgement whether abnormal sound is arranged again, and after the high order small-signal of acoustical signal is introduced into the abnormal tone signal that according to the human auditory system mental model, people's ear really can responsive be heard in psychoacoustic model processing module (22) and extracts, entering acoustical signal Threshold detection module (23) judges whether comprising abnormal sound in its acoustic response in conjunction with the thresholding that acoustical signal thresholding generation module (26) generates again, at last the testing result of signal of telecommunication Threshold detection module (21) testing result harmony signalc threshold detection module (23) is input in the comprehensive judge module of abnormal sound (24) and comprehensively judges, whether provide tested sounding body has last judgement and the output judged result of abnormal sound,

Selecting special test signal is the continuous logarithmic swept-frequency signal, is defined as follows formula:

$\mathrm{Stim}\left(t\right)=U\sin \left[\frac{{\mathrm{\ω}}_{1}T}{\ln \left(\frac{{\mathrm{\ω}}_2}{{\mathrm{\ω}}_1}\right)}(e^{\frac{t}{T}\ln \left(\frac{{\mathrm{\ω}}_2}{{\mathrm{\ω}}_1}\right)}-1)\right]$

Wherein: U is the test signal amplitude, and T is the test signal time, ω ₁The initial frequency of test signal, ω ₂It is the termination frequency of test signal.

2. sounding body abnormal sound detection method according to claim 1, it is characterized in that: described tested sounding body comprises at least: moving-coil sounding body unit, piezoelectric sounding body unit, dull and stereotyped sounding body unit, sounding body close case system, sounding body phase inverting box system, receiver, microphone system.

3. sounding body abnormal sound detection method according to claim 1 is characterized in that: all signal processing flows in test macro PC or exclusive equipment (6) in carry out, utilize the program in PC or exclusive equipment (6) to complete.

4. the test macro of sounding body abnormal sound detection method according to claim 1, it is characterized in that: the test macro of this foundation comprises: the constant voltage power amplifier (4) of PC or exclusive equipment (6), data collecting card (5), belt current measuring ability, Sound Attenuator or artificial ear (3), measure microphone or microphone array (2), PC or exclusive equipment (6) are by program control data capture card (5) emission exciting test signal (7), be carried in tested sounding body (1) two ends after constant voltage power amplifier (4) power amplification of this test signal (7) through the belt current measuring ability, current signal (9) with this amplification voltage signal (8) and the sounding body of flowing through is input in data collecting card (5) simultaneously, use measurement microphone or microphone array (2) that the sound pressure signal (10) that tested sounding body (1) sends under pumping signal is gathered and be input to data collecting card (5) in Sound Attenuator or artificial ear (3), exclusive equipment (6) comprises having the industrial computer of controlling function and Presentation Function, independent embedded device, data collecting card (5) is used for completing mould/number and D/A switch,

Be provided with signal to noise ratio detection module (19), high-order weak signal extraction module (20), signal of telecommunication Threshold detection module (21), psychoacoustic model processing module (22), acoustical signal Threshold detection module (23), the abnormal comprehensive judge module of sound (24), signal of telecommunication thresholding generation module (25) harmony signalc threshold generation module (26) in PC or exclusive equipment (6).

Images