CN106323452B - Detection method and detection device for abnormal sound of equipment - Google Patents
Detection method and detection device for abnormal sound of equipment Download PDFInfo
- Publication number
- CN106323452B CN106323452B CN201510389956.XA CN201510389956A CN106323452B CN 106323452 B CN106323452 B CN 106323452B CN 201510389956 A CN201510389956 A CN 201510389956A CN 106323452 B CN106323452 B CN 106323452B
- Authority
- CN
- China
- Prior art keywords
- sample
- sound
- characteristic parameter
- abnormal sound
- clustering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 230000002159 abnormal effect Effects 0.000 title claims abstract description 95
- 238000001514 detection method Methods 0.000 title claims description 43
- 238000000034 method Methods 0.000 claims abstract description 23
- 238000012545 processing Methods 0.000 claims abstract description 14
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000008569 process Effects 0.000 claims description 7
- 230000035945 sensitivity Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 abstract description 9
- 230000005236 sound signal Effects 0.000 abstract description 5
- 230000008901 benefit Effects 0.000 abstract description 3
- 238000007621 cluster analysis Methods 0.000 abstract 3
- 238000012706 support-vector machine Methods 0.000 description 26
- 238000001228 spectrum Methods 0.000 description 16
- 238000012360 testing method Methods 0.000 description 10
- 238000004422 calculation algorithm Methods 0.000 description 7
- 238000010586 diagram Methods 0.000 description 7
- 230000002950 deficient Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000004519 manufacturing process Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000007635 classification algorithm Methods 0.000 description 3
- 238000000691 measurement method Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000010224 classification analysis Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000003780 insertion Methods 0.000 description 1
- 230000037431 insertion Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000002045 lasting effect Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000013102 re-test Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Landscapes
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The invention discloses a method and a device for detecting abnormal sound of equipment, wherein the method comprises the following steps: collecting a sound signal when the equipment runs; preprocessing the collected sound signal to obtain a sound processing signal; extracting a plurality of feature parameters from the sound processing signal; performing cluster analysis and SVM linear classification on the extracted characteristic parameters and samples in a database; and predicting whether the sound signal is abnormal sound according to the clustering analysis and the SVM linear classification result. The invention detects the abnormal sound of the equipment by combining the cluster analysis and the SVM linear classification, so that the problem of the large-scale nonlinearity of the sound sample in the feature space can be eliminated by the cluster analysis, and the classification operation is carried out on the small-scale sound sample meeting the linear hypothesis by the SVM linear classification. Compared with the prior art, the method has the advantages of simple operation, high prediction precision and the like.
Description
Technical field
The present invention relates to a kind of detection method and device of machine abnormal sound, espespecially a kind of to be based on SVM linear classification and cluster
The abnormal sound detection method and device of analysis.
Background technique
The automatic test of abnormal sound always is the problem of factory testing in the past few years.What product issued in normal work
Noise is to belong to normal noise or abnormal noise (abnormal sound), and past production test is all with consummate operation people
Member is judged that the sound including carrying out to equipment such as fan, transformer, power supply product, projector, motors is tested.Sound
Test must rely on the judgement of operating personnel, and anthropic factor is element most rambunctious in a detection system, is constrained
The realization of product quality ZD Zero Defect target.Necessary personnel's operation simultaneously also limits the realization of production automation.
Existing sound test, there is following methods:
Use sound power level.This mode can measure the general power or total volume of volume, if total volume is higher
, it is bad for can detecting.But cacophonia, typical from the exception of some component, power meter can not tell sound
Constituent.If the component of the exception is not main power component, can not usually be detected.
It is measured using frequency spectrograph.Frequency spectrograph can check the component value of each frequency, be particularly suitable for the inspection of narrow band signal
Out.Such as the single frequency tone that buzzer issues, it can be easy to detect through frequency spectrograph.Spectrum analysis is also that current engineering unit makes
For doing the main tool of abnormal sound judgement.However most abnormal sound mode is BROADBAND NOISE, point on each single-frequency
Amount is not significant, therefore frequency spectrograph is difficult to tell the form of BROADBAND NOISE, and there are significant False Rate, nothings for this decision procedure
Method provides high detection validity.
Use power sound spectrum.The measurement method incorporates the measurement of acoustical power and frequency spectrum, uses various octave power spectrum
Weighted mode measure.This measurement method is closer to response of the human ear to sound, but the basic of it limits and first two
It is identical.Normal noise and abnormal sound difference in the performance of power spectrum are small, and the mobility scale of normal noise may just be more than
This small difference, so that Detection capability is very limited.
It is measured using acoustic mass.Including with loudness (Loudness), roughness (roughness), sharpness
(sharpness) etc. the acoustic mass factor measures.The weighted of these acoustic mass simulates people because of response, and physical quantity is switched to people
The impression of ear, closer to people to loudness, roughness, the feeling of sharpness.However these acoustic mass factors actually can not be effective
Whether symbolizing the exception of noise.It is substantially that weighted calculation obtains on the basis of frequency spectrum or power spectrum, detection
Restriction factor is identical as frequency spectrum or power spectrum.
Use the resonance demodulation technique of accelerometer.This is a kind of method with vibration substitution abnormal sound test, in accelerometer
On resonant frequency, if depositing the resonance that will inspire accelerometer plus lasting excitation.The source of this Persistent Excitation is exactly
Abnormal sound.This method is proved in the mechanical abnormal sound measurement of rotation be effective, especially Bearing exception person detection.Bearing is different
Chang Shihui inspires a succession of BROADBAND NOISE, and the frequency component of these noises has covered the resonant frequency of accelerometer and excited altogether
Vibration.Resonance frequency signal is demodulated into available driving source characteristic frequency.This method regards effectively, still in certain tests
Wide usage is not high.Most of abnormal patterns all can not be detected effectively.
With standard, anechoic chamber is measured.This measurement method needs a standard anechoic chamber and the very high sound of a set of grade
Sound measuring system, it is very low that advantage is that the ambient noise in anechoic chamber is pressed into, while with the voice acquisition system of low noise,
The ambient noise of measuring signal can be preferably minimized, maximum can be mentioned to the distinguishing ability of the broadband signal of abnormal sound in this way.
This method is due to equipment manufacturing cost valuableness, use when being suitable only for researching and analysing, and is difficult to be applied to and manufacture in link.
However, no matter existing test mode will be pre- on obtained measured value using above any test mode
Determine standard value.For example, the response of the available each frequency of spectrum measurement, general concept be research and analyse it is normal with it is abnormal
Product, specification value when trying to find out sound measurement on frequency spectrum.However the essence of abnormal sound is not single frequency tone, but simultaneously
The component in various broadbands is generated, so that the component of different frequency does not have independence, each frequency can not be just set from scientific principle
The specification value of component.This characteristic explains why pass by attempt to measure and establish specification in various manners, but always
Effective decision procedure can not be found.
Based on above-mentioned, due to the dependence between frequency component, reasonable specification value is established, it cannot be for each frequency point
Amount goes to analyze, it should at least analyze the linear combination of each component.Therefore it is existing reference statistics linear classification algorithm into
Row classification analysis.The algorithm of linear classifier needs first to establish non-defective unit sample and defective products sample.With known non-defective unit feature and
Know defective products feature, a classifying face is calculated, then classification judgement is carried out to sample to be tested with the classifying face.It can use
Statistical classification algorithm has very much, including linear regression, logistic regression, generation algorithm, bayesian algorithm, svm classifier algorithm etc..
SVM (support vector machines) sorting algorithm is used in this project, SVM thinks there is biggish robustness in many papers.
Statistical classification algorithm does not go to calculate specification to independent each feature, carries out classification judgement to the linear combination of feature, this
Sample is more conform with the basic principle that abnormal sound is multiple spectra essence.By testing preliminary analysis, SVM really can be effectively to given
Database sample sound determined.
Reality finds that some significant abnormal sound can be misjudged in sorting algorithm when using.A kind of possible reason
It is SVM is a kind of introduction learning algorithm, it needs to learn from the introduction of database the construction of classifying face, so these are misjudged
Sound, which is added in database, perhaps can solve problem.However actually these significant abnormal sound are added in sample data library,
Classifying face has but been obscured instead, and reduces the validity of svm classifier.This phenomenon is inquired into, the possible reason is using statistics
Linear classifier, is based on the assumption that the linear separability of object of classification, however linear separability 100% can not actually meet,
In the condition for being unsatisfactory for linear separability, SVM will be unable to correctly classify to determine.Significant abnormal sound because significantly from
Group, in feature space farther out with the distribution of main sound classification, the hypothesis of linear separability is no longer set up, therefore adds it number
According to the construction of unfavorable classifier instead in library.
Summary of the invention
In view of the problems of the existing technology, the purpose of the present invention is to provide a kind of methods of automatic checkout equipment abnormal sound
And device.
The detection method of equipment abnormal sound of the invention, includes the following steps:
Acquire voice signal when equipment operation；
The voice signal of acquisition is pre-processed, a sound processing signal is obtained；
Multiple characteristic parameters are extracted from the sound processing signal；
Clustering and SVM linear classification are carried out to the sample in extracted the multiple characteristic parameter and database；
It whether is abnormal sound according to voice signal described in the clustering and the SVM linear classification prediction of result.
Further, carrying out pretreatment to the voice signal of acquisition includes:
The voice signal of acquisition is converted to the standard sound pressure value of frequency domain according to the sensitivity of microphone；
Calibration process is carried out to the standard sound pressure value using weighted factor.
Further, the clustering includes:
According to the Euclidean between the sample in extracted the multiple characteristic parameter and the database in feature space
Distance carries out clustering；
The corresponding voice signal of the multiple characteristic parameter and the database are judged according to the result of clustering
In sample between whether peel off.
Further, the clustering further comprises:
Calculate the Euclidean distance between the sample in the multiple characteristic parameter and the database in feature space；
When the Euclidean distance calculated is greater than or equal to a threshold value, then the result of the clustering is to peel off；
When institute's Euclidean distance calculated is less than the threshold value, then the result of the clustering is not peel off.
Further, the sample in the database includes normal sound sample and abnormal sound sample.
Further, the detection method further include:
Establish equipment normal sound sample and abnormal sound sample；
According to the characteristic parameter in the characteristic parameter and the abnormal sound sample in the normal sound sample, calculate
To the classifying face in feature space.
Further, the SVM linear classification includes:
SVM is carried out according to the positional relationship between extracted the multiple characteristic parameter and the classifying face linearly to divide
Class；
Judge that the corresponding voice signal of the multiple characteristic parameter is located at described point according to the result of SVM linear classification
The abnormal sound sample side in class face or the normal sound sample side of the classifying face.
Further, the SVM linear classification further comprises:
Calculate the relative positional relationship of the multiple characteristic parameter Yu the classifying face；
When the multiple characteristic parameter is located at the abnormal sound sample side of the classifying face, then the multiple feature is joined
The corresponding voice signal of number is classified as abnormal sound.
Further, when the result of the clustering is to peel off, then the sound according to the prediction of result of the clustering
Sound signal is abnormal sound；
When the result of the clustering is not peel off, then the sound according to the prediction of result of the SVM linear classification
Whether signal is abnormal sound.
The detection device of equipment abnormal sound of the invention, comprising:
Acquisition unit, for acquiring voice signal when equipment operation；
Pretreatment unit obtains a sound processing signal for pre-processing to the voice signal of acquisition；
Extraction unit extracts multiple characteristic parameters from the sound processing signal；
Analytical unit, for carrying out clustering to the sample in extracted the multiple characteristic parameter and database；
Taxon is linearly divided for carrying out SVM to the sample in extracted the multiple characteristic parameter and database
Class；
Predicting unit is for the voice signal according to the clustering and the SVM linear classification prediction of result
No is abnormal sound.
Further, the pretreatment unit includes:
The voice signal of acquisition is converted to the standard sound pressure of frequency domain by converting unit according to the sensitivity of microphone
Value；
Calibration unit carries out calibration process to the standard sound pressure value using weighted factor.
Further, the analytical unit includes:
First computing unit, for calculating between the sample in the multiple characteristic parameter and the database in feature sky
Between Euclidean distance；
First result output unit, when the Euclidean distance calculated is greater than or equal to a threshold value, then the cluster
The result of analysis is to peel off.
Further, the sample in the database includes normal sound sample and abnormal sound sample.
Further, further includes:
Database sharing unit establishes the normal sound sample and the abnormal sound sample of equipment；
Second computing unit, according to the spy in the characteristic parameter and the abnormal sound sample in the normal sound sample
Parameter is levied, the classifying face in feature space is calculated.
Further, the taxon includes:
Second computing unit, for calculating the relative positional relationship of the multiple characteristic parameter Yu the classifying face；
Second result output unit, when the multiple characteristic parameter is located at the abnormal sound sample side of the classifying face,
The corresponding voice signal of the multiple characteristic parameter is then classified as abnormal sound.
The present invention detects equipment abnormal sound in such a way that clustering and SVM linear classification combine, so can be with
Exclude sample sound large-scale nonlinear problem in feature space with clustering, it is small-scale to meet line
Property assume sample sound, carry out sort operation with SVM linear classification.Compared with prior art, there is easy to operate, prediction essence
Spend the advantages that high.
Detailed description of the invention
Fig. 1 is the flow diagram of the detection method of the equipment abnormal sound of one embodiment of the invention；
Fig. 2 is the flow diagram of the detection method of the equipment abnormal sound of another embodiment of the present invention；
Fig. 3 is to carry out pretreated process to collected sound signal in the equipment abnormal sound detection method of one embodiment of the invention
Schematic diagram；
Fig. 4 is the flow diagram of clustering in the equipment abnormal sound detection method of one embodiment of the invention；
Fig. 5 is the flow diagram of SVM linear classification in the equipment abnormal sound detection method of one embodiment of the invention；
Fig. 6 is the structural schematic diagram of the detection device of the equipment abnormal sound of one embodiment of the invention；
Fig. 7 is the structural schematic diagram of the detection device of the equipment abnormal sound of another embodiment of the present invention.
Specific embodiment
As shown in Figure 1, the detection method of the equipment abnormal sound of one embodiment of the invention, includes the following steps:
Step S21: voice signal when acquisition equipment operation.Wherein, the voice signal when operation of acquisition equipment, can be with
It is obtained by the sonic transducer installed in equipment.
Step S22: the voice signal of acquisition is pre-processed, a sound processing signal is obtained；
Step S23: multiple characteristic parameters are extracted from the sound processing signal；
Step S24: clustering and SVM line are carried out to the sample in extracted the multiple characteristic parameter and database
Property classification, wherein the sample in database is divided into normal sound sample and abnormal sound sample, and normal sound sample and different
It include multiple characteristic parameters in normal sample sound.
Whether step S25: being different according to voice signal described in the clustering and the SVM linear classification prediction of result
Sound.
It is specifically included as shown in figure 3, carrying out pretreatment to the voice signal of acquisition in step S22:
Step S221: the voice signal of acquisition is converted to the standard sound pressure of frequency domain according to the sensitivity of microphone
Value, merely for time domain sound pressure level, the consistency of retest is bad, and repetition can be improved with frequency spectrum or spectra calculation
Property, the frequency spectrum of expansion can be from 0Hz-22kHz, however the present invention is not limited thereto；
Step S222: calibration process is carried out to the standard sound pressure value using weighted factor, for the sum for allowing people to experience
High level on frequency spectrum can correspond to, such as A weighting can filter out the indistinguishable frequency of human ear, provide side to screen out garbage signal
It helps.Option there are four methods of weighting in the present invention, A weights (A-weighting), directly mapping (Direct map), insertion adds
Power (Interpolate weighting) does not weight (No weighting).A-weighting is to be A- to waveform
weighting.This function major function is not intended to do A weighting weighting, but is used to lead after doing Microphone calibration
Enter microphone receptance function use.Direct map is directly multiplied by defeated after map array by waveform after Fourier transform
Waveform after weighting out needs the duration and sample frequency according to input waveform using direct map, confirms the length of its frequency spectrum
Degree and frequency interval, and the completely corresponding map array (Map array) of frequency point is generated, after being directly multiplied by after output weighting
Waveform.
As shown in figure 4, clustering described in step S25 specifically includes:
Step S251: the Euclidean between the sample in the multiple characteristic parameter and the database in feature space is calculated
Distance；For example, two n-dimensional vector a (X11,X12,…X1n) and b (X21,X22…X2n) between Euclidean distance d12:
Step S252: judge whether Euclidean distance calculated is greater than a threshold value；
Step S253: when the Euclidean distance calculated is greater than or equal to a threshold value, then the knot of the clustering
Fruit is to peel off；
Step S254: when institute's Euclidean distance calculated is less than the threshold value, then the result of the clustering is not
It peels off.
As shown in Fig. 2, the detection method of the equipment abnormal sound for another embodiment of the present invention.Itself and equipment shown in Fig. 1
The detection method of abnormal sound is compared further include:
Step S11: equipment normal sound sample and abnormal sound sample are established.Wherein, acquisition equipment first operates normally
Voice signal in certain time obtains normal sound sample after above-mentioned pretreatment and feature extraction.Then, equipment is acquired
Voice signal in misoperation certain time obtains abnormal sound sample after above-mentioned pretreatment and feature extraction.However,
Normal sound sample and abnormal sound sample might not be as set forth in the present embodiment by subsequent foundation, normal sound sample and
Abnormal sound sample may be to have preset before this.
Step S12: joined according to the feature in the characteristic parameter and the abnormal sound sample in the normal sound sample
Number, is calculated the classifying face in feature space.The calculating of classifying face is calculated using existing linear discriminant function.
As shown in figure 5, SVM linear classification described in step S25 specifically includes::
Step S255: calculating the relative positional relationship of the multiple characteristic parameter Yu the classifying face, that is, calculates multiple spies
Levy the Euclidean distance of the normal sound sample of parameter and classifying face and the abnormal sound sample of multiple characteristic parameters and classifying face
Euclidean distance, judged according to the length of multiple characteristic parameter Euclidean distances be located at normal sound sample side be still located at it is different
Normal sample sound side；
Step S256: Euclidean distance and characteristic parameter and abnormal sound between comparative feature parameter and normal sound sample
Euclidean distance size between sound sample；
Step S257: when the Euclidean distance between characteristic parameter and abnormal sound sample is less than between normal sound sample
Euclidean distance when, the abnormal sound sample in the multiple characteristic parameter distance classification face is closer, positioned at the different of the classifying face
Normal sample sound side, then be classified as abnormal sound for the corresponding voice signal of the multiple characteristic parameter；
Step S258: when the Euclidean distance between characteristic parameter and normal sound sample is less than between abnormal sound sample
Euclidean distance when, the normal sound sample in the multiple characteristic parameter distance classification face is closer, just positioned at the classifying face
Normal sample sound side, then be classified as normal sound for the corresponding voice signal of the multiple characteristic parameter.
It may also be linearly non-thread between the sample in the multiple characteristic parameter and database of acquisition due to that may be
Property.And SVM linear classification method can not carry out Accurate Prediction to non-linear partial.Therefore, non-linear in multiple characteristic parameters
Part is to be excluded according to the result of clustering, that is, divide when clustering in above-mentioned steps S25 in the embodiment of the present invention
The result of analysis is to peel off, then directly predicts that the voice signal is abnormal sound according to the result of the clustering.When the cluster
The result of analysis is not peel off, then prove its with the sample in database be it is linear, so can be according to the SVM linear classification
Prediction of result described in voice signal whether be abnormal sound, i.e., when the multiple characteristic parameter is located at the abnormal sound of the classifying face
The corresponding voice signal of the multiple characteristic parameter is then classified as abnormal sound, when the multiple feature by sample side
Parameter is located at the normal sound sample side of the classifying face, then returns the corresponding voice signal of the multiple characteristic parameter
Class is normal sound.
As shown in fig. 6, the detection device 4 of the equipment abnormal sound for one embodiment of the invention, comprising:
Acquisition unit 41, for acquiring voice signal when equipment operation；
Pretreatment unit 42 obtains a sound processing signal for pre-processing to the voice signal of acquisition；
Extraction unit 43 extracts multiple characteristic parameters from the sound processing signal；
Analytical unit 44, for carrying out cluster point to the sample in extracted the multiple characteristic parameter and database
Analysis；
Taxon 45, it is linear for carrying out SVM to the sample in extracted the multiple characteristic parameter and database
Classification；
Predicting unit 46 is used for the voice signal according to the clustering and the SVM linear classification prediction of result
It whether is abnormal sound.
Pretreatment unit 42 in the present embodiment specifically includes:
The voice signal of acquisition is converted to the standard sound pressure of frequency domain by converting unit according to the sensitivity of microphone
Value；
Calibration unit carries out calibration process to the standard sound pressure value using weighted factor.Wherein, converting unit and calibration
The processing method of unit has been explained in aforementioned abnormal sound detection method, and details are not described herein.In addition, pretreatment unit of the invention
Structure be not limited thereto, can be omitted converting unit or calibration unit.
Analytical unit 44 in the present embodiment includes:
First computing unit, for calculating between the sample in the multiple characteristic parameter and the database in feature sky
Between Euclidean distance；
First result output unit, when the Euclidean distance calculated is greater than or equal to a threshold value, then the cluster
The result of analysis is to peel off.
As shown in fig. 7, the detection device of the equipment abnormal sound for another embodiment of the present invention.Itself and equipment shown in Fig. 3
The detection device of abnormal sound is compared further include:
Database sharing unit 31 establishes the normal sound sample and the abnormal sound sample of equipment；
Computing unit 32, according to the feature in the characteristic parameter and the abnormal sound sample in the normal sound sample
The classifying face in feature space is calculated in parameter.
Taxon 45 includes: in the present embodiment
Second computing unit, for calculating the normal sound sample and exception of the multiple characteristic parameter Yu the classifying face
The relative positional relationship of sample sound；
Second result output unit, when the multiple characteristic parameter is located at the abnormal sound sample side of the classifying face,
The corresponding voice signal of the multiple characteristic parameter is then classified as abnormal sound.
It is particularly shown and described exemplary embodiments of the present invention above.It should be understood that the present invention is not limited to
Disclosed embodiment, on the contrary, it is intended to cover comprising various modifications within the scope of the appended claims and equivalent
Displacement.
Claims (14)
1. a kind of detection method of equipment abnormal sound, includes the following steps:
Acquire voice signal when equipment operation；
The voice signal of acquisition is pre-processed, a sound processing signal is obtained；
Multiple characteristic parameters are extracted from the sound processing signal；
Clustering and SVM linear classification are carried out to the sample in extracted the multiple characteristic parameter and database；
It whether is abnormal sound according to voice signal described in the clustering and the SVM linear classification prediction of result；
Wherein the clustering includes:
According to the Euclidean distance between the sample in extracted the multiple characteristic parameter and the database in feature space
Carry out clustering；
Judged in the corresponding voice signal of the multiple characteristic parameter and the database according to the result of clustering
Whether peel off between sample.
2. detection method as described in claim 1, which is characterized in that carry out pretreatment packet to the voice signal of acquisition
It includes:
The voice signal of acquisition is converted to the standard sound pressure value of frequency domain according to the sensitivity of microphone；
Calibration process is carried out to the standard sound pressure value using weighted factor.
3. detection method as described in claim 1, which is characterized in that the clustering further comprises:
Calculate the Euclidean distance between the sample in the multiple characteristic parameter and the database in feature space；
When the Euclidean distance calculated is greater than or equal to a threshold value, then the result of the clustering is to peel off；
When institute's Euclidean distance calculated is less than the threshold value, then the result of the clustering is not peel off.
4. detection method as described in claim 1, which is characterized in that the sample in the database includes normal sound sample
With abnormal sound sample.
5. detection method as claimed in claim 4, which is characterized in that the detection method further include:
Establish equipment normal sound sample and abnormal sound sample；
According to the characteristic parameter in the characteristic parameter and the abnormal sound sample in the normal sound sample, it is calculated
One classifying face of feature space.
6. detection method as claimed in claim 5, which is characterized in that the SVM linear classification includes:
SVM linear classification is carried out according to the positional relationship between extracted the multiple characteristic parameter and the classifying face；
Judge that the corresponding voice signal of the multiple characteristic parameter is located at the classifying face according to the result of SVM linear classification
Abnormal sound sample side or the classifying face normal sound sample side.
7. detection method as claimed in claim 6, which is characterized in that the SVM linear classification further comprises:
Calculate the relative positional relationship of the multiple characteristic parameter Yu the classifying face；
When the multiple characteristic parameter is located at the abnormal sound sample side of the classifying face, then by the multiple characteristic parameter pair
The voice signal answered is classified as abnormal sound.
8. detection method as described in claim 1, which is characterized in that
When the result of the clustering is to peel off, then the voice signal according to the prediction of result of the clustering is different
Sound；
When the result of the clustering is not peel off, then the voice signal according to the prediction of result of the SVM linear classification
It whether is abnormal sound.
9. a kind of detection device of equipment abnormal sound, comprising:
Acquisition unit, for acquiring voice signal when equipment operation；
Pretreatment unit obtains a sound processing signal for pre-processing to the voice signal of acquisition；
Extraction unit extracts multiple characteristic parameters from the sound processing signal；
Analytical unit, for carrying out clustering to the sample in extracted the multiple characteristic parameter and database；
Taxon, for carrying out SVM linear classification to the sample in extracted the multiple characteristic parameter and database；
Predicting unit, for the voice signal according to the clustering and the SVM linear classification prediction of result whether be
Abnormal sound；
Wherein the clustering includes:
According to the Euclidean distance between the sample in extracted the multiple characteristic parameter and the database in feature space
Carry out clustering；
Judged in the corresponding voice signal of the multiple characteristic parameter and the database according to the result of clustering
Whether peel off between sample.
10. detection device as claimed in claim 9, which is characterized in that the pretreatment unit includes:
The voice signal of acquisition is converted to the standard sound pressure value of frequency domain by converting unit according to the sensitivity of microphone；
Calibration unit carries out calibration process to the standard sound pressure value using weighted factor.
11. detection device as claimed in claim 9, which is characterized in that the analytical unit includes:
First computing unit, for calculating between the sample in the multiple characteristic parameter and the database in feature space
Euclidean distance；
First result output unit, when the Euclidean distance calculated is greater than or equal to a threshold value, then the clustering
Result be peel off.
12. detection device as claimed in claim 9, which is characterized in that the sample in the database includes normal sound sample
Sheet and abnormal sound sample.
13. detection device as claimed in claim 12, which is characterized in that further include:
Database sharing unit establishes the normal sound sample and the abnormal sound sample of equipment；
Computing unit, according to the characteristic parameter in the characteristic parameter and the abnormal sound sample in the normal sound sample,
The classifying face in feature space is calculated.
14. detection device as claimed in claim 13, which is characterized in that the taxon includes:
Second computing unit, for calculating the normal sound sample and abnormal sound of the multiple characteristic parameter Yu the classifying face
The relative positional relationship of sample；
Second result output unit then will when the multiple characteristic parameter is located at the abnormal sound sample side of the classifying face
The corresponding voice signal of the multiple characteristic parameter is classified as abnormal sound.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510389956.XA CN106323452B (en) | 2015-07-06 | 2015-07-06 | Detection method and detection device for abnormal sound of equipment |
| TW104130498A TWI587294B (en) | 2015-07-06 | 2015-09-15 | Detection method of abnormal sound of apparatus and detection device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510389956.XA CN106323452B (en) | 2015-07-06 | 2015-07-06 | Detection method and detection device for abnormal sound of equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106323452A CN106323452A (en) | 2017-01-11 |
| CN106323452B true CN106323452B (en) | 2019-03-29 |
Family
ID=57727303
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510389956.XA Active CN106323452B (en) | 2015-07-06 | 2015-07-06 | Detection method and detection device for abnormal sound of equipment |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN106323452B (en) |
| TW (1) | TWI587294B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105841797B (en) * | 2016-03-15 | 2019-07-02 | 中南大学 | A method and device for detecting abnormal noise of window motor based on MFCC and SVM |
| TWI654599B (en) | 2017-07-07 | 2019-03-21 | 圓展科技股份有限公司 | Electronic apparatus with audio compensating mechanism and audio compensating method thereof |
| TWI680370B (en) * | 2018-03-30 | 2019-12-21 | 逸奇科技股份有限公司 | Time series signal analysis method |
| JP7188597B2 (en) * | 2019-07-30 | 2022-12-13 | 日本電信電話株式会社 | Feature quantity extraction device, anomaly degree estimation device, methods thereof, and program |
| US11835430B2 (en) * | 2019-07-30 | 2023-12-05 | Nippon Telegraph And Telephone Corporation | Anomaly score estimation apparatus, anomaly score estimation method, and program |
| CN110536215B (en) * | 2019-09-09 | 2021-06-29 | 普联技术有限公司 | Method and apparatus for audio signal processing, calculation device, and storage medium |
| TWI761715B (en) | 2019-10-21 | 2022-04-21 | 緯創資通股份有限公司 | Method and system for vision-based defect detection |
| US11326935B2 (en) | 2019-10-21 | 2022-05-10 | Wistron Corporation | Method and system for vision-based defect detection |
| CN110992985A (en) * | 2019-12-02 | 2020-04-10 | 中国科学院声学研究所东海研究站 | Identification model determining method, identification method and identification system for identifying abnormal sounds of treadmill |
| CN113095340B (en) * | 2019-12-23 | 2024-04-16 | 神讯电脑(昆山)有限公司 | Abnormality early warning method of production machine and mass production method of objects |
| TWI751642B (en) * | 2020-08-10 | 2022-01-01 | 騰擎科研創設股份有限公司 | Detection system for abnormal sound detection and cause determination |
| CN113611327A (en) * | 2020-10-23 | 2021-11-05 | 深圳市冠旭电子股份有限公司 | Abnormal sound detection and analysis method and device, terminal equipment and readable storage medium |
| CN112700790A (en) * | 2020-12-11 | 2021-04-23 | 广州市申迪计算机系统有限公司 | IDC machine room sound processing method, system, equipment and computer storage medium |
| TW202229822A (en) * | 2021-01-22 | 2022-08-01 | 建準電機工業股份有限公司 | Method and system for abnormal sound testing of a fan |
| US11735017B2 (en) | 2021-06-23 | 2023-08-22 | Bank Of America Corporation | Artificial intelligence (AI)-based security systems for monitoring and securing physical locations |
| US12387570B2 (en) | 2021-06-23 | 2025-08-12 | Bank Of America Corporation | Artificial intelligence (AI)-based security systems for monitoring and securing physical locations |
| US11954990B2 (en) * | 2021-06-23 | 2024-04-09 | Bank Of America Corporation | Artificial intelligence (AI)-based security systems for monitoring and securing physical locations |
| TWI774472B (en) | 2021-07-13 | 2022-08-11 | 緯創資通股份有限公司 | Abnormal sound detection method and apparatus |
| CN113984191B (en) * | 2021-10-27 | 2024-11-08 | 苏州微著设备诊断技术有限公司 | An intelligent needle-listening system for abnormal sound of power equipment and its diagnosis method |
| TWI889194B (en) * | 2024-02-06 | 2025-07-01 | 三和技研股份有限公司 | Abnormal sound analyzing system and industrial automation system including the same |
| CN119091910A (en) * | 2024-11-08 | 2024-12-06 | 中科声玄(苏州)科技有限公司 | Abnormal noise detection method and system for new energy vehicle motor based on decision-level fusion |
| CN119889359A (en) * | 2025-01-21 | 2025-04-25 | 珠海精实测控技术股份有限公司 | Fan abnormal sound intelligent distinguishing and detecting method and system based on support vector machine |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101846594A (en) * | 2010-06-22 | 2010-09-29 | 上海交通大学 | Fault detection device based on beam forming acoustic-image mode recognition and detection method thereof |
| CN101968379A (en) * | 2010-09-30 | 2011-02-09 | 南昌航空大学 | Method for extracting characteristic information of operating condition vibration signal of aircraft engine rotor system |
| CN103398843A (en) * | 2013-07-01 | 2013-11-20 | 西安交通大学 | Classification method of planetary gear box sun gear faults based on multi-class relevance vector machine |
| CN103645249A (en) * | 2013-11-27 | 2014-03-19 | 国网黑龙江省电力有限公司 | Online fault detection method for reduced set-based downsampling unbalance SVM (Support Vector Machine) transformer |
| CN103810374A (en) * | 2013-12-09 | 2014-05-21 | 中国矿业大学 | Machine fault prediction method based on MFCC feature extraction |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8463719B2 (en) * | 2009-03-11 | 2013-06-11 | Google Inc. | Audio classification for information retrieval using sparse features |
| FR2944903B1 (en) * | 2009-04-24 | 2016-08-26 | Thales Sa | SYSTEM AND METHOD FOR DETECTING ABNORMAL AUDIO EVENTS |
| JP5363213B2 (en) * | 2009-06-30 | 2013-12-11 | 東京エレクトロン株式会社 | Abnormality detection system, abnormality detection method, storage medium, and substrate processing apparatus |
| TW201117190A (en) * | 2009-11-12 | 2011-05-16 | Nat Cheng Kong University | Sound recognition for frog vocalizations |
| WO2012031614A1 (en) * | 2010-09-09 | 2012-03-15 | Autoliv Development Ab | System for determining vibration characteristics of a motor vehicle |
| TWI448978B (en) * | 2011-06-24 | 2014-08-11 | Univ Nat Taiwan Normal | Training method for fault detection system |
| FR2994495B1 (en) * | 2012-08-10 | 2015-08-21 | Thales Sa | METHOD AND SYSTEM FOR DETECTING SOUND EVENTS IN A GIVEN ENVIRONMENT |
-
2015
- 2015-07-06 CN CN201510389956.XA patent/CN106323452B/en active Active
- 2015-09-15 TW TW104130498A patent/TWI587294B/en active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101846594A (en) * | 2010-06-22 | 2010-09-29 | 上海交通大学 | Fault detection device based on beam forming acoustic-image mode recognition and detection method thereof |
| CN101968379A (en) * | 2010-09-30 | 2011-02-09 | 南昌航空大学 | Method for extracting characteristic information of operating condition vibration signal of aircraft engine rotor system |
| CN103398843A (en) * | 2013-07-01 | 2013-11-20 | 西安交通大学 | Classification method of planetary gear box sun gear faults based on multi-class relevance vector machine |
| CN103645249A (en) * | 2013-11-27 | 2014-03-19 | 国网黑龙江省电力有限公司 | Online fault detection method for reduced set-based downsampling unbalance SVM (Support Vector Machine) transformer |
| CN103810374A (en) * | 2013-12-09 | 2014-05-21 | 中国矿业大学 | Machine fault prediction method based on MFCC feature extraction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106323452A (en) | 2017-01-11 |
| TW201703028A (en) | 2017-01-16 |
| TWI587294B (en) | 2017-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106323452B (en) | Detection method and detection device for abnormal sound of equipment | |
| RU2704073C2 (en) | Method and system for training acoustic or vibration analysis of machine | |
| CN114004306A (en) | A device fault assessment system and method based on multi-dimensional data of the Internet of Things | |
| US20220397500A1 (en) | Apparatus and method for detecting microcrack using orthogonality analysis of mode shape vector and principal plane in resonance point | |
| JP2019067197A (en) | Method for detecting trouble sign | |
| CN110398647A (en) | Transformer state monitoring method | |
| CN118897171B (en) | Detection method, device, medium and system for partial discharge capability of switchgear | |
| CN107748209A (en) | A kind of Structure Damage Identification | |
| WO2024116637A1 (en) | Classification device, learning-model generation device, classification method, and learning-model generation method | |
| JP3966500B2 (en) | Defect inspection system due to abnormal noise | |
| CN114186586A (en) | Damage identification method and equipment based on two-dimensional convolutional neural network | |
| JP3594477B2 (en) | Abnormal noise inspection device | |
| CN111122085B (en) | Structure assembly quality evaluation method based on power distribution characteristics | |
| KR101290928B1 (en) | Equipment and method for diagnosing malfunction using sound quality parameter | |
| KR101886210B1 (en) | Measuring instrument reliability evaluation apparatus and operating method thereof | |
| CN118836968A (en) | A method for detecting vehicle noise offline | |
| CN113763679A (en) | Method and device for monitoring abnormal sound of gas-insulated enclosed combined electrical appliance | |
| CN113588069A (en) | Piezoelectric sensor and microphone | |
| Koshekov et al. | Modernization of vibration analyzers based on identification measurements | |
| CN117457026A (en) | Noise detection system and method for riding equipment | |
| Lin et al. | Condition-based shaft fault diagnosis with the empirical mode decomposition method | |
| Tsiapoki et al. | Combining a vibration-based SHM scheme and an airborne sound approach for damage detection on wind turbine rotor blades | |
| CN113302467B (en) | Method for inspecting a component, in particular an injector | |
| JP3132388B2 (en) | Abnormal noise inspection method and apparatus for rotating machine | |
| KR100456841B1 (en) | Method for processing detection signal in duration test of vehicle |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210115 Address after: 215200 no.1688, Jiangxing East Road, Wujiang Economic and Technological Development Zone, Wujiang City, Suzhou City, Jiangsu Province Patentee after: Delta Electronics (Jiangsu) Ltd. Address before: 215200 no.1688, Jiangxing East Road, Wujiang Economic Development Zone, Jiangsu Province Patentee before: DELTA ELECTRONICS COMPONENTS (WUJIANG) Ltd. |
|
| TR01 | Transfer of patent right |