CN105637331B - Abnormal detector, method for detecting abnormality - Google Patents
Abnormal detector, method for detecting abnormality Download PDFInfo
- Publication number
- CN105637331B CN105637331B CN201480056176.7A CN201480056176A CN105637331B CN 105637331 B CN105637331 B CN 105637331B CN 201480056176 A CN201480056176 A CN 201480056176A CN 105637331 B CN105637331 B CN 105637331B
- Authority
- CN
- China
- Prior art keywords
- signal
- operating portion
- periodic component
- monitored object
- ingredient
- 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
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H3/00—Measuring characteristics of vibrations by using a detector in a fluid
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
- Testing And Monitoring For Control Systems (AREA)
Abstract
In the case where the signal source of noise changes in very short time, noise is effectively inhibited, only efficiently extracts the periodic component of the signal from monitored object operating portion.In being controlled to the measuring device that the constant cycle repeats same action, using the volume of each frequency of operation noise to control the property that the period of the constant times in period repeats model identical, from sensor signal separation cycle ingredient and aperiodic ingredient, thus, the accurately period sound in separation and Extraction monitored object operating portion, and execute abnormality detection processing (referring to Fig. 7).
Description
Technical field
The present invention relates to abnormal detector, method for detecting abnormality and computer-readable storage mediums, such as are related to
From the abnormal abnormality detection technology carried out using A/D converter in the signal detection signal after digital conversion.
Background technique
From the viewpoint of the device maintenance, it is strongly desired the abrasion of the gear in the operating portion in detection device, driving portion
Act undesirable abnormality detection technology.Particularly, operating portion physical quantity as sound, vibration, deformation in device is led
Cause is prone to anomalous variation, therefore, is detected using sensor as microphone, vibrating sensor, strain gauge transducer
It is abnormal.But in such as these sensors, even if being equipped with sensor near the operating portion of the object carried out abnormality detection,
Signal other than the operating portion of object mixes, and is difficult to carry out with signal of the enough S-N ratios (S/N ratio) to object mostly
Detection.
Therefore, it is necessary to the Signal separator skill of the signal in the operating portion of object is only extracted from obtained sensor signal
Art.There are the operating portion of multiple objects, need to extract each operating portion respectively from obtained sensor signal
Signal.
So far, as signal separation techniques, the isolation technics using multiple sensor signals is mainly had studied.These technologies
Be using between sensor time difference or difference of vibration according to each signal source the different technologies to separate signal.As representativeness
Source separation technology, there are the sef-adapting filters methods such as minimum variance beamformer method.In the sef-adapting filter method
In, it can be used only by the signal in the operating portion of object without separating sound by multiple input mode filters of other signals
Sound.The signal from sensor element quantity -1 signal source is able to suppress in sef-adapting filter method principle.
Therefore, as signal source, there is only sensor elements other than the operating portion in addition to the monitored object of abnormality detection
In the case where -1 signal source of quantity, the signal in the operating portion from monitored object can be accurately extracted in principle.
On the other hand, it was known that in the case where existing more than -1 signal source of sensor element quantity, come from monitored object
Operating portion signal extraction accuracy deterioration.But even seen for a long time in the presence of more than sensor element quantity -1
In the case where signal source, signal also and not always is issued from whole signal sources.For example, in the case where signal source is operating portion,
It accordingly acts or stops with the action mode of device, therefore, only issue signal from the signal source in the case where movement.
It is therefore contemplated that even in the case where seen for a long time in the presence of -1 signal source of sensor element quantity is more than, such as
Fruit is seen with the short time, is that there is only -1 signal sources below of sensor element quantity mostly.Utilizing the existing of the situation
It in sef-adapting filter method, has the following structure, for the purpose of the signal source for efficiently inhibiting each time, and at every moment obtains
Sensor signal so that the shapes of multiple input mode filters is changed (referring to non-patent literature 1).In this way, making
For the technology for extracting monitored object signal, generally, using the continuous value for changing filter so as to the variation phase with sound field environment
Adaptively efficiently inhibit the noise suppressed method of the sef-adapting filter of the noise at each moment.
Existing technical literature
Non-patent literature
Non-patent literature 1:L.J.Griffith and C.W.Jim, " An alternative approach to
linearly constrained adaptive beamforming、”IEEE Trans.Anntenas Propagation、
Vol.30, i.1, pp.27-34, Jan.~1982.
Summary of the invention
Subject to be solved by the invention
However, it is more bad that performance is more sharply then followed in the variation of shape in such structure disclosed in non-patent literature 1
Change.Even if because sound, vibration are not the sampling rates of 10kHz or so, if the type of signal source does not change in at least several seconds
It can be followed, however in the case where the time interval of very short time within the type of signal source was with 1 second is changed, then
Follow difficulty.In sef-adapting filter, in order to learn the time that the information for the noise that should be disappeared there will be the several seconds, therefore, come
Not as good as the operation noise for the measuring device for following sound field environment and being changed with the short time within 1 second, can not successfully inhibit to make an uproar
Sound.This with will be controlled as repeating the measuring device of same action as the case where object using the constant cycle being identical.
The present invention is made in view of such situation, provide it is a kind of for be controlled as repeating with the constant cycle into
In the measuring device of row same action, efficiently inhibit to be changed relative to sound field environment with the short time (for example, within 1 second)
Measuring device operation noise noise, and extract the technology of monitored object signal.
Means for solving the problems
In order to solve the above problems, inventor is conceived to, and is being controlled as repeating same action with the constant cycle
In measuring device, the volume of every frequency of operation noise is to control the period that the period of the constant times in period repeats model identical
Sound;And other than operation noise around noise be and control period independently existing aperiodic sound.Also, in the present invention,
Assuming that being blended in sensor signal to control the signal after the duplicate period sound of the constant doubling time in period and aperiodic discord
In, each period sound, aperiodic sound are separated.Also, it extracts from the period sound after separation via the fortune with monitored object
Turn the identical transmitting of sound and handle and reach the ingredient of sensor, and is carried out abnormality detection for the ingredient extracted.
That is, abnormal detector of the invention executes study processing and abnormality detection processing.Study processing is needle
The periodic component changed to volume with the constant times in the control period of measuring device and the non-week independently changed with the control period
The sensor signal that phase ingredient mixes executes maximal possibility estimation to sensor signal using the information in control period, divides as a result,
From periodic component and the aperiodic ingredient, and calculate the change that the characteristic quantity of the periodic component isolated is indicated with probabilistic manner
The processing of dynamic probability distribution.In addition, abnormality detection processing is the biography that used in will be handled in study other than sensor signal
Sensor signal uses the abnormal processing of the result detection signal of study processing as check object.
Further feature related to the present invention becomes clear according to the description of this specification, attached drawing.In addition, passing through
The form of the range of the combination of element and a variety of elements and detailed description below and appended claims and reach and real
Existing form of the invention.
It is illustrated it is to be understood that the description of this specification is only typical, without the model of any pair of claim of the invention
It encloses or the meaning that application examples is defined.
Invention effect
In accordance with the invention it is possible to accurately from the signal component from various operating portions and the noise outside device
The mixed sensor signal of ingredient extracts the signal in monitored object operating portion.
Detailed description of the invention
Fig. 1 is the figure for indicating to be arranged the hardware configuration example of the measuring device of abnormal detector of the invention.
Fig. 2 is the block diagram for indicating the outline structure of abnormal detector of embodiment of the present invention.
Fig. 3 is to schematically illustrate each temporal frequency for the signal included by the measurement sensor 104 of measuring device
The figure of ingredient.
Fig. 4 is the figure for indicating the structural example of display picture 1400 of embodiment of the present invention.
Fig. 5 is the figure for indicating the structural example of initial setting screen.
Fig. 6 is the flow chart for illustrating all processing summaries of the abnormal detector of embodiment of the present invention.
Fig. 7 be indicate embodiment of the present invention abnormal detector performed by learning program 400 processing structure
Figure.
Fig. 8 is the figure of the presumption example for the volume for indicating each periodic component and aperiodic ingredient.
Fig. 9 is the figure for indicating the data configuration of periodic component statistic DB404.
Figure 10 is the data configuration for indicating the transmission function in each monitored object operating portion kept in registration ingredient DB407
Figure.
Figure 11 is the tables of data for indicating signal in normal components DB405 to be written to, from monitored object operating portion
The figure of structure.
Figure 12 is the figure for indicating periodic component/aperiodic ingredient study portion 403 details of embodiment of the present invention.
Figure 13 is the flow chart for illustrating every frequency cycle ingredient/aperiodic ingredient study portion 802 processing details.
Figure 14 be indicate embodiment of the present invention abnormal detector performed by abnormality detecting program 600 processing knot
The figure of structure.
Figure 15 be indicate embodiment of the present invention abnormal detector performed by abnormality detection processing 501 processing knot
The figure of structure.
Figure 16 is the figure for indicating periodic component/aperiodic ingredient separation unit 502 details of embodiment of the present invention.
Figure 17 is the figure for indicating periodic component/aperiodic ingredient separation unit 502 variation of embodiment of the present invention.
Figure 18 is the flow chart for illustrating the periodic component of every frequency/aperiodic ingredient separation unit 902 processing details.
Figure 19 is the flow chart of the processing details for specification exception determination unit 503.
Specific embodiment
Hereinafter, being described with reference to embodiments of the present invention.In attached drawing, the identical element of function is sometimes with identical volume
Number indicate.In addition, attached drawing illustrates specific embodiment and embodiment based on the principle of the present invention, however these be in order to
Understanding of the invention is in no way intended to the explanation of limiting the invention property.
Present embodiment is sufficiently illustrated in detail in order to make those skilled in the art implement the present invention, however answers
When being interpreted as that other installation/modes can also be used, carry out while technical idea scope and spirit of the invention can not departed from
The change of construction/configuration or the replacement of a variety of elements.Therefore, description below should not be defined in this explanation.
In addition, as be described hereinafter, embodiments of the present invention, can also be with using the software installation that runs on a general-purpose computer
It is installed using the combination of dedicated hardware or software and hardware.
In addition, in aftermentioned explanation, information of the invention is illustrated by " table " form, however these information can not also be with
The data configuration of table shows, and can also be showed with the data configurations such as list, DB, queue or form other than it.Therefore, in order to
It indicates to be directed to " table ", " list ", " DB ", " queue " etc. sometimes, referred to as " information " independent of data configuration.
Hereinafter, " central processing unit " (also referred to as computer or processor) is illustrated as subject (action subject)
It manages, however can also illustrate everywhere in embodiment of the present invention with various " programs " are action subject.In addition, the one of program
It partly or entirely can use specialized hardware realization, it furthermore can also be with modularization.Program distribution server, storage can also be passed through
Medium installs various programs in each computer.
<structure of measuring device>
Fig. 1 is the figure for indicating the hardware configuration example of the measuring device of abnormal detector of setting embodiment of the present invention.
Measuring device 100 includes turntable 101, the operating portion 102 being made of multiple operating portions, monitored object operating portion
103, measurement sensor 104.
The control that turntable 101 corresponds to measuring device is acted periodically, and with the constant in the control period of device
Constant cycle again repeats same action.Similarly, operating portion 102 is also acted with the constant cycle.For example, to control the period
Constant times constant cycle carry out piston motion.Monitored object operating portion 103 similarly controls the normal of period with device
The constant cycle of several times acts.
In the present invention, monitored object operating portion 103 according to using measurement with the sensor signal that sensor 104 is measured come
Detect due to it is certain will thus carry out the state of movement different from usual.
<structure of abnormal detector>
Fig. 2 is the outline structure for indicating measuring device and abnormality detection system (abnormal detector+monitoring server)
Hardware block diagram.
Measurement is arranged inside measuring device 100 with sensor 104, is passed by microphone, vibrating sensor and/or strain-type
Sensor is constituted.The analog signal being taken by the measurement sensor 104 is converted by A/D converter 201 from analog signal
For digital signal.
Transformed digital signal is sent to abnormal detector 200.Abnormal detector 200 includes central operation
Device 201, nonvolatile memory 202, volatile memory 203.Transformed digital signal is transported to central operation dress
201 are set, and executes aftermentioned learning program and abnormality detecting program.This program is stored in nonvolatile memory 202, and
It is read when program executes.Ensure required working storage when program executes in volatile memory 203.
The output valve exported after being handled using abnormality detecting program is transported to HUB206, is transformed to network transmission number
According to packet.Transformed data packet is transported to HUB212 via network 207.The data received are sent monitoring clothes by HUB212
Business device 211.
Monitoring server 211 includes central operation device 208;Volatile memory 209;Volatile memory 210;It is aobvious
Show device 213;The input unit 214 being made of mouse, keyboard.Central operation device 208 acts monitoring result browser program,
The picture shown in display 213 is generated using the data sent.It is made of in addition, user can be used mouse, keyboard
Input unit 14 come control (instruction) initially setting start, abnormality detection movement beginning.By monitoring result browser program
It is stored on nonvolatile memory 209, and is read when program executes.In addition, ensuring program in volatile memory 210
Required working storage when execution.
<signal (example) included using measurement sensor>
Fig. 3 is schematically illustrate the signal included in measuring device shown in Fig. 1 using measurement sensor 104 every
The figure of the ingredient (example) of a temporal frequency.
Sensor signal is implemented to be transformed to every well known to a person skilled in the art Short Time Fourier Transform, wavelet transformation
The signal of a time (τ) and frequency.In Fig. 3, the volume of each temporal frequency reaches the temporal frequency ingredient of certain threshold value or more
It is showed by black box.It is showed on the contrary, becoming certain threshold value temporal frequency ingredient below by white boxes.In Fig. 3 it is found that
Black lattice and white square become identical patterns with the period identical with the control period.
Consider the sensor measured in the measuring device with multiple operating portions with constant control sampling action letter
Number, the pattern of the similar temporal frequency ingredient in the period of the constant times in control period is indicated as shown in Figure 3.
<the display picture example in monitoring server>
Fig. 4 is the display picture in the display 213 on the monitoring server 211 for indicating to operate using input unit 214
1400 show exemplary figure.
Display picture 1400 is included the multiple buttons operated using input unit 214 as picture structure and (initially sets and press
Button 1401, movement start button 1403, study start button 1404, conclusion button 1405) and each monitored object operating of display
The monitored object operating portion Stateful Inspection picture 1402 of the state in portion.
When pressing pressure initially setting button 1401 using input unit 214, show that picture is used in initial setting.
Fig. 5 be indicate initial setting screen 1900 show exemplary figure.Initial setting screen 1900 shows that user can set
Determine the picture of the sensor intensity (aftermentioned bj (f)) in multiple buttons and monitored object operating portion.Specifically, user is initially setting
Determine to set the index (index) in the monitored object operating portion of setting sensor intensity on picture 1900, it is each setting
After the intensity of sensor, register button 1901 is pressed.Thereby, it is possible to the bj (f) in the monitored object operating portion being consistent registrations
Into aftermentioned registration ingredient DB407.In general, being directed to each monitored object operating portion, 1 measurement sensing is set in its vicinity
Device.It is therefore contemplated that the intensity of the measurement sensor near monitored object operating portion is set as 1, by survey in addition to this
The intensity of quantity sensor is set as 0.After the information in end setup monitored object operating portion, press initial setting terminates to press user
Button 1902.Initial setting screen 1900 terminates as a result, back to display picture 1400.
Referring again to FIGS. 4, pressing learns start button 1404 in display picture 1400 by user, filled in central operation
It sets and executes aftermentioned learning program 400 on 201.In addition, by user's push action start button 1403, to execute abnormality detection
Program 501.In addition, browser program terminates when user presses conclusion button 1405.
Show that each monitored object operating portion is normal or abnormal in monitored object operating portion Stateful Inspection picture 1402.
<processing in abnormal detector is whole>
Fig. 6 is the figure for indicating the process of typical action of each program in abnormality detection system.It is shown in Fig. 4 by user
Display picture 1400 on indicated using input unit 214, start the motion flow.
Firstly, the starting of central operation device 208 shows the browser program of display picture 1400 on monitoring server 211
(S701)。
Then, central operation device 201 presses study start button 1404 to user and responds, and executes learning program
400(S702).By the execution of the learning program, the periodic component of input signal and aperiodic ingredient are separated, only monitored object
The signal (normal components) in operating portion is registered to database (normal components DB405).
Later, central operation device 201 responds user's push action start button 1403, and makes abnormality detecting program 600
It acts (S703).In abnormality detecting program 600, abnormality detection processing 501 is executed whenever obtaining sensor signal (referring to figure
15)。
Then, the central processing unit 208 for monitoring server 211, whenever by the reception abnormality detection of monitoring server 211
When the abnormal determination result of reason 501, it is confirmed whether to have pressed the conclusion button (S704) of browser program.
If having pressed (the case where being "Yes" in S704), the end browser program of central processing unit 208, also, in
Arithmetic unit 201 is entreated to terminate abnormality detecting program 600 (S705).
<learning program>
Fig. 7 is the processing structure block diagram for the learning program 400 for indicating that central operation device 201 executes.
Indexed unit 401 buffers the sensor signal included by measurement sensor 104.Here, the sensing measured
Device signal is indicated by formula (1).
[formula 1]
X (t)=[x1(t),x2(t),…,xM(t)]T……(1)
Here, if M is the quantity of measurement sensor, if the A/D transformation for increasing (increment) when t is every sub-sampling is opened
The sampling number risen after beginning.If T is the transposition operator of matrix or vector.
Frequency decomposition portion 402 is handled by frequency decompositions such as well known Short Time Fourier Transform, wavelet transformations, by each measurement
The time-domain signal of sensor is transformed to temporal frequency domain.Here, when setting index of the m to indicate the number of measurement sensor, x
(t) time-domain signal for m-th of measurement sensor for including in is xm (t).Also, time-domain signal will be converted as after temporal frequency domain
Signal describe be xm (f, τ).Here, the index of frequency is indicated if f is, if τ is to indicate to implement Short Time Fourier Transform every time
When increased frame number variable.If f is the integer value from 1 to k.
Here, the sensor signal measured in temporal frequency domain is indicated by following formula (2).
[formula 2]
X (f, τ)=[x1(f,τ),x2(f,τ),…,xM(f,τ)]T……(2)
The x (f, τ) of multiple frames is stored in buffer by periodic component/aperiodic ingredient study portion 403 first.Here,
If by τ=1,2 ..., the signal until LLt is stored in buffer.If LLt is the natural number in the control period of measuring device
Times.In addition, periodic component/aperiodic ingredient study portion 403 is according to aftermentioned process flow, by x (f, τ) be decomposed into the period at
Divide, aperiodic ingredient.When factoring, if assigning the quantity of periodic component and the quantity of aperiodic ingredient in advance.If after decomposing
Periodic component is ci (f, τ) (index that i is periodic component), if aperiodic ingredient is that rj (f, τ) (j is the rope of non-periodic component
Draw).About the volume of the periodic component ci (f, τ) after decomposition, when the average volume phase that it is equalized on temporal frequency direction
Than being to be evaluated as the period in certain ratio situation below in the average volume that x (f, τ) is equalized on temporal frequency direction
Ingredient is not present.By implementing the processing, even if the quantity of the periodic component of physical presence is less than preset periodic component
Quantity, can also extract the periodic component for being suitable for actual cycle composition quantity.
It is associated with the periodic component ci (f, τ) for being separated and being extracted by periodic component/aperiodic ingredient study portion 403
(accordingly), volume vi (f, τ), the time-invariant covariance matrix of each temporal frequency for the periodic component being consistent are exported
Ri (f) and with control the period remove the periodic component extracted period obtained by value (Pi).In addition, for aperiodic ingredient,
Also the volume of each temporal frequency is exported.
Here, illustrate the presumption of the volume of periodic component and aperiodic ingredient.Fig. 8 shows each periodic components and aperiodic
The presumption example of the volume of ingredient.The ingredient that is changed as volume with controlling the period of the natural several times in period and estimated period at
Point.On the other hand, the volume of aperiodic ingredient is estimated as the ingredient for not having the period for depending on the control period.
Back to Fig. 7, if the quantity of the periodic component extracted is N, then N is saved in periodic component statistic DB404
A periodic component Pi, the volume vi (f, τ) of each T/F and the covariance matrix Ri (f) of each frequency.Fig. 9 is indicated
The data configuration of periodic component statistic DB404.If the control period of device is Lt, then it is directed to each periodic component, by signal
Period/control period, corresponding with signal period (Pi × Lt) of whole frequencies (from 1 to k) and each periodic component number of frames
The covariance matrix of the volume of each T/F and whole frequencies is stored in periodic component statistic DB404.In addition,
The volume vi (f, τ) of each T/F is the information of the intensity changed according to each time, the covariance square of each frequency
Battle array Ri (f) is the information for being transformed to independent of the time and taking the intensity ratio of steady state value.
Return to Fig. 7, registration determination processing 406 differentiate the periodic component that extracts whether be monitored object operation noise, such as
Fruit is the ingredient in monitored object operating portion, then the transmission function and probability distribution of the periodic component are obtained from registration ingredient DB407,
It is written in normal components DB405.Here, according to the monitored object kept (when initial setting) by registration ingredient DB407 in advance
The consistent degree of the time-invariant covariance matrix Ri (f) of the transmission function and periodic component of the ingredient in operating portion, whether presumption
It is the ingredient in monitored object operating portion.
<transmission function in monitored object operating portion>
Figure 10 is indicated by the data configuration of the transmission function in each monitored object operating portion kept registration ingredient DB407.Such as
Shown in Figure 10, in registration ingredient DB407, transmission function bj (f) is registered with for each monitored object operating portion.If setting Lm
For the quantity in monitored object operating portion, then bj (f) is the transmission function of the frequency f in j-th of monitored object operating portion, by sensor
The element of number of elements is constituted.Judged using following formula (3) and the consistent degree of Ri (f).
[formula 3]
Here, trace is the operator of the mark of calculating matrix.H is the operation for taking the Hermite transposition of matrix or vector
Symbol.For each monitored object operating portion (being directed to each j), the periodic component that selecting type (3) becomes maximum, the periodic component is made
Periodic component for j-th of monitored object operating portion is written in normal components DB405.
<normal components DB>
Figure 11 is the tables of data for indicating to be written to signal in normal components DB405, from monitored object operating portion
The figure of structure.In the table, for each monitored object operating portion, it is registered with transmission function aj (f).
Transmission function aj (f) can also be used as bj (f) and register, and can also be used as and is selected as j-th of monitored object fortune
The first characteristic vector of the covariance matrix Ri (f) of the periodic component of transfer part and register.By first as Ri (f) inherently to
Amount is registered, and can be registered comprising in influence of echo, reverberation in measuring device etc., the physical analogy in device design
The transmission function for the influence that can not be learnt in advance, and can expect to improve the judgement precision of abnormal determination processing.
In addition, in normal components DB405 cycle register ingredient probability distribution pj (x).The probability distribution pj (x) is table
The information for showing the normal range (NR) of the signal of the normal components from monitored object operating portion makes to use it to carry out abnormal determination.This
Outside, here, the probability distribution of periodic component refers to the model that the variation of characteristic quantity of periodic component is shown with probabilistic manner, if
X is sound equipment characteristic quantity.As probability distribution, it is able to use Mixed Normal Distribution known to the skilled person in the art etc., arrives mesh
Before until the various distributions that are used as abnormal detection method.In addition, will find out from the separation signal ci (f, τ) of each periodic component
MFCC (Mel frequency cepstral coefficients (Mel Frequency Cepstrum Coefficients)), whole frequencies and whole
Ci (f, τ) or its volume vectorization of frame and used as sonority features amount x.That is, passing through the period from input signal
Ingredient ci (f, τ) extracts the ingredient for becoming sound characteristic, and checks what kind of distribution the ingredient for becoming the sound characteristic becomes,
It can seek probability distribution pj (x).
In addition, in order to seek GMM (gauss hybrid models (Gaussian Mixture Model)) equiprobability according to data
Distribution, for Counting statistics amount, needs multiple measurement.In this case using such as flowering structure: abnormality detection journey is performed a plurality of times
Sequence 501 sees the periodic component detected as the signal in j-th of monitored object operating portion as multiple measurement, according to
Data seek the probability distribution of GMM.
Transmission function bj (f) can be using the structure by initially setting user interface (referring to Fig. 5) setting.In addition, root
According to the relative positional relationship in monitored object operating portion and sensor, when generation signal is transmitted to sensor from monitored object operating portion
Transmission function, and assigned using the transmission function as bj (f).For example, the case where there are measurement sensors in microphone
Down can be using such as flowering structure: according to can make amplitude at a distance from monitored object operating portion and sensor square proportionally
The time delay for decaying and remove apart from and seek with the velocity of sound transmission function, to find out bj (f), wherein the transmission function is signal
Function when being transmitted to sensor from monitored object operating portion.For by vibration and being deformed according to design information
(distortion) physical analogy is carried out to seek the structure of transmission function, there is that well known to a person skilled in the art a variety of sides
Method, however, it is possible to which these methods are widely applied to seek bj (f).In addition, the size of adjustment bj (f) is so that vector bj's (f) is big
Small (squared norm) is 1.
<periodic component/aperiodic ingredient study portion internal structure>
Figure 12 is the structural block diagram for indicating periodic component/aperiodic ingredient study portion 403 inside.
Periodic component/aperiodic ingredient study portion 403 includes multiple every frequency cycle ingredients/aperiodic ingredient study portion
802-1 to 802-k and sequence solution portion 803.
Every frequency cycle ingredient/aperiodic ingredient study portion 802-1 ... k, the sensing that will be measured in temporal frequency domain
Device signal x (f, τ) is separated into periodic component and aperiodic ingredient for each frequency f.Since sensor signal is according to each frequency
Rate changes at any time, and the period also difference (referring to Fig. 3) according to each frequency, therefore is learned with being divided into multiple frequency bands
It practises.Come learning cycle ingredient and aperiodic ingredient according to each frequency, is one of the features of the present invention.
In addition, design cycle ingredient or aperiodic ingredient can not be indexed in advance as which kind of in the separating treatment, for
Each frequency takes arbitrary value.For example, the periodic component of the 1st index in the 1st periodic component and frequency 2 indexed in frequency 1
It is from the signal of entirely different signal source.
Therefore, 803 pairs of sequence solution portion index, which implements replacement processing, makes if the index for the frequency content isolated is identical
As the signal from identical signal source.Due to exporting multiple frequency contents from same operating portion, it is therefore desirable to by each ingredient with
Operating portion is corresponding.Sequence solution processing is well known, if those skilled in the art can just implement, for example, executing association
Index is handled so that the correlation of the frame direction of the volume vi (f, τ) of every frame gets higher (each frequency exported from same operating portion
The correlation of ingredient is high, thus is corresponded to), implementable index replacement appropriate as a result,.Even if sequence solution portion 803 is defeated
Signal is different frequency out, can also be expected if index is identical for the signal from identical signal source.
<frequency content of every frequency/aperiodic ingredient study processing details>
Figure 13 is the processing details that periodic component/aperiodic ingredient study portion 802 for illustrating by every frequency executes
Flow chart.In order to from input signal separation cycle ingredient and aperiodic ingredient, need for isolated parameter (for example, strong
Degree information, in which information for becoming larger of which frequency content period).But it must separate to obtain the parameter first defeated
Enter signal.According to the signal isolated, learns certain frequency content and become larger on certain period/become smaller.Due to such study nothing
Method terminates in single treatment, therefore, executes such processing shown in Figure 13.
Central operation device 201 starts to learn each week according to the sensor signal x (f, τ) measured in temporal frequency domain
The processing (S1101) of phase ingredient and aperiodic ingredient.
In addition, central operation device 201 sets provisional parameter (S1102).In the provisional parameter setting processing, initially set
Fixed suitable parameter (random).Due to being random parameter, although precision is not high however it may be said that smart compared with input signal
Degree is improved.More specifically, firstly, using preset periodic component and the respective quantity of aperiodic ingredient as setting value
It reads.In addition, integral multiple of the period of each periodic component as the control period, also for each periodic component period from setting
Value is read.Each periodic component, aperiodic ingredient are indicated with same index i.In addition, by each periodic component, aperiodic ingredient when
Between the volume vi (f, τ) of frequency be set as 1, assign covariance matrix Ri (f, τ) as Hermite random matrix.Here, if
Ri (f, τ) is the matrix of M row M column.In addition, setting vi (f, τ) is scalar value.
Then, central operation device 201 is according to vi (f, τ) and Ri (f, τ), separates and extract each periodic component, aperiodic
Ingredient (S1103).Here, if τ is 1 to LLt.In addition, executing separation and Extraction using multichannel Wiener filter.Specifically,
Separation signal ci (f, τ) is sought by ci (f, τ)=wi (f, τ) x (f, τ).In addition, Ri (f, τ) is indicated using formula (4), and
Wiener filter wi (f, τ) is indicated using formula (5).In addition, central operation device 201 is calculated flat by formula (6) in S1103
Square error (Square error).In formula (6), if I is the unit matrix of M row M column.
[formula 4]
[formula 5]
wi(f, τ)=vi(f, τ) Ri(f, τ) Rx(f, τ)-1…(5)
[formula 6]
Rci(f, τ)=ci(f, τ) ci(f, τ)H+(I-Wi(f, τ)) vi(f, τ) Ri(f, τ) ... (6)
Then, central operation device 201 is updated separation parameter using formula (7) and updates (S1104).Formula (7) be it is maximum seemingly
The formula so estimated.That is, in the present invention, it is one of feature that service life information, which carries out maximal possibility estimation,.
[formula 7]
vi(f, τ)=trace (Ri(f, τ)-1Rci(f, τ)) ... (7)
In addition, i-th of signal is periodic component in S1104, and in the case where setting the period as Lt, central operation dress
Setting 201 deforms vi (f, τ).Specifically, firstly, for each time interval r in a period, seeking τ=Lt*n+r, (n is
Random natural number) τ set T (r).Then, for the τ for including in T (r), seek the average value of vi (f, τ), and set its as
vi(f,r).Then, for the τ for including in T (r), if vi (f, τ)=vi (f, r).In addition, seeking covariance square by formula (8)
Battle array Ri (f, τ) (being updated).
[formula 8]
Then, central operation device 201, which checks, executes whether the number that separation parameter updates has reached predetermined pre-
Number is determined, alternatively, whether being less than predetermined value by formula (6) calculated square error, in the case where having reached pre-determined number
Or in the case that square error is small, it is judged to learning terminating (S1105).In the case where being determined as end, processing is proceeded to
Learn the setting (S1106) terminated, and terminates to learn.In the case where not being determined as end, processing, which is transferred to, to be implemented at separation
It manages (S1103).
As above-mentioned, in study processing, divide discrete parameter by alternatively execute periodic component and aperiodic ingredient
It updates, improves the precision of separation.
<structure of abnormality detecting program>
Figure 14 is the structural block diagram for indicating abnormality detecting program 600.The abnormality detecting program abnormality detection abnormal by detection
The testing result that processing 501, the exception that will test out are sent to monitoring server 211 sends processing 601 and constitutes.
Figure 15 is the detailed block diagram for indicating abnormality detection processing 501.Abnormality detection processing 501 and shown in Fig. 7
Practising program 400 is identical structure, therefore the repetitive description thereof will be omitted.In addition, LLt need not be set in abnormality detection processing 501
For value identical with learning program 400, it is set as the natural several times in the control period of measuring device.
Firstly, 502 service life ingredient statistic DB404 of periodic component/aperiodic ingredient separation unit (is transported with monitored object
Accordingly register the periodic component of whole frequency bands in transfer part) in the statistic of periodic component registered, separation cycle ingredient and non-week
Phase ingredient.Average volume and x (t) after equalizing the volume of the periodic component ci (t) after decomposing on temporal frequency direction
Average volume after equalizing on temporal frequency direction is compared, and in certain proportion situation below, is considered as the periodic component
It is not present.By implementing the processing, even if the quantity of the periodic component of physical presence is less than the number of preset periodic component
Amount, can also extract the periodic component for being suitable for actual cycle composition quantity.
Abnormality determination unit 503 determines to extract using the information in the monitored object operating portion registered in normal components DB405
Periodic component whether include exception.That is, according to the probability distribution pj (x) in monitored object operating portion, inspection is extracted
Periodic component irrelevance, execute abnormal determination according to whether irrelevance is greater than predetermined value.
<periodic component/aperiodic ingredient separation unit detailed construction>
Figure 16 is periodic component/aperiodic ingredient separation unit 502 detailed block diagram for indicating embodiment of the present invention.
Every frequency cycle ingredient/aperiodic ingredient separation unit 902-1 to 902-k is by input signal x (f, τ) according to each frequency
Rate is separated into periodic component, aperiodic ingredient.The every frequency cycle ingredient/aperiodic ingredient separation unit 902-1 to 902-k is by filtering
Wave device is constituted.In order to obtain unwanted signal intensity (periodic component) in each period from periodic component statistic DB404
Information, setting can remove the filter of the unwanted signal, only obtain operating from monitored object by the filter
The signal in portion.Further, since aperiodic ingredient is changed according to time, place, therefore enter while undated parameter in real time
Row separation (referring to Fig.1 8).
Each periodic component ci (f, τ) is transformed to by frequency inverse transformation portion 903 by inverse Fourier transform or wavelet inverse transformation
Time-domain signal ci (t).Time-domain signal ci (t) after output transform.But the frequency inverse transformation portion 903 is not required structure.
In addition, periodic component/aperiodic ingredient separation unit 502 can also learn using with periodic component/aperiodic ingredient
The same structure in portion 403.Figure 17 is the figure for indicating such structure, with this configuration, can be real-timeed while carrying out study
Ground separates signal.In addition, the process of processing is explained, thus omit.
<every frequency cycle ingredient/aperiodic ingredient separation details>
Figure 18 is the flow chart for illustrating every frequency cycle ingredient/aperiodic ingredient separation unit 902 processing details.
Central operation device 201 executes separating treatment (S1201) when accumulating the input signal of scheduled frame length every time.
When being initially separated processing, central operation device 201 first from periodic component statistic DB404 readout interval at
The parameter (S1202) divided.Periodic component statistic DB404 will be registered in as the vi of the parameter of periodic component (f, τ) and Ri (f)
In.
On the other hand, for aperiodic ingredient, since signal source changes at any time, central operation device 201
It sets real-time update parameter (S1203).Here, for aperiodic ingredient, for example, vi (f, τ)=1 is set, if Ri (f) is Ai Er meter
Special random matrix.
Then, central operation device 201 executes separating treatment (S1204).Specifically, come according to vi (f, τ) and Ri (f, τ)
Each periodic component of separation and Extraction, aperiodic ingredient.Here, if τ is 1 to LLt, and multichannel Wiener filter is used, passes through ci
(f, τ)=wi (f, τ) x (f, τ) come seek separation signal ci (f, τ).In addition, indicating Ri (f, τ) using above-mentioned formula (4), and benefit
Wiener filter wi (f, τ) is indicated with above-mentioned formula (5).In addition, central operation device 201 is calculated by formula (6) in S1204
Square error.In formula (6), if I is the unit matrix of M row M column.
Then, central operation device 201 updates real-time update parameter (S1205) using above-mentioned formula (7).I-th of signal
It is periodic component, in the case where setting the period as Lt, central operation device 201 deforms vi (f, τ).Specifically, firstly, needle
To each time interval r in a period, the set T (r) of the τ of τ=Lt*n+r (n is random natural number) is sought.Then, for
The τ for including in T (r), seeks the average value of vi (f, τ), and sets it as vi (f, r).Then, for the τ for including in T (r), if vi
(f, τ)=vi (f, r).On the other hand, for aperiodic ingredient, using above-mentioned formula (8) seek covariance matrix Ri (f, τ) (into
Row updates).
Then, central operation device 201 is updated end judgement (S1206).For example, if executing real-time update parameter
The number of update has reached predetermined pre-determined number, then is judged to terminating.In the case where being determined as terminates (in S1206
For "Yes"), then processing is transferred to S1207 (setting that separation terminates), and terminates separating treatment.It is not determined as the case where terminating
Under the case where (in S1205 be "No"), processing returns to S1204 (implementing separating treatment).In addition it is also possible to according to calculated
Whether square error is less than predetermined value to determine to terminate.
<abnormal determination processing>
Figure 19 is the flow chart of the processing details for specification exception determination unit 503.
Central operation device 201 at the end of separating treatment 502 of periodic component and aperiodic ingredient, starts abnormal every time
Determine (S1301).
Then, the periodic signal source for carrying out determination processing index is set as 1 (S1302) by central operation device 201.Cause
This, executes abnormal determination processing since the smallest periodic component of index in the periodic component isolated.
First eigenvalue of the transmission function of the periodic component of 201 extraction process object of central operation device as Ri (f)
di(f)(S1303)。
Then, central operation device 201 calculates the monitored object operating portion registered in normal components DB405 using formula (9)
Transmission function aj (f) at a distance from the transmission function (the first eigenvalue di (f)) of the periodic component of process object, and select to count
The smallest j of the distance of calculating (S1304).
[formula 9]
E (i, j, f)=1- | aj(f)Hdi(f)|2…(9)
Central operation device 201 calculates the degree of approach of signal source by formula (10), and determines whether it is predetermined
Threshold value TH1 or less (S1305).It is in threshold value TH1 situation below in the degree of approach of signal source (is the feelings of "Yes" in S1305
Condition), processing is transferred to S1306, and in the case where being greater than TH1 (the case where being "No" in S1306), processing is transferred to S1309.
[formula 10]
In S1306, central operation device 201 extracts sonority features amount x from the time-domain signal ci (t) of periodic component i
(S1306)。
Then, the sonority features amount extracted is updated in normal components DB405 and stores just by central operation device 201
In the x of the probability distribution pj (x) of Chang Chengfen, likelihood (likelihood) pj (x) (S1307) is calculated.
Then, central operation device 201 determines whether calculated likelihood is predetermined threshold value TH2 or more
(S1308).In the case where likelihood is threshold value TH2 or more (the case where being "Yes" in S1308), processing is transferred to S1309, small
In the case where TH2 (the case where being "No" in S1308), processing is transferred to S1311.
In S1309, the periodic component for the process object that central operation device 201 is judged to extracting is normal, and is directed to
1 (S1309) of periodic signal source index addition.
Then, central operation device 201 determine periodic signal source index whether be the periodic signal source quantity that extracts with
Under (S1310).In for periodic signal source quantity situation below (the case where being "Yes" in S1310), processing is transferred to next
The judgement (S1303) of a periodic component, in addition to this in the case where, notice is normal and ends processing (S1312).
On the other hand, in the case where being determined as that calculated likelihood is less than predetermined threshold value TH2 in S1308, in
Centre arithmetic unit 201 is determined as that the j-th monitored object operating portion nearest from i-th is abnormal, and ends processing (S1311).
<summary>
(i) embodiment according to the present invention is measured by the multiple measurement sensors being arranged in measuring device
Sensor signal is input into abnormal detector.Volume is contaminated in the sensor signal with the control period of measuring device
The changed periodic component of constant times and with the aperiodic ingredient that independently changes of control period.Firstly, abnormal examination fills
The information (setting the period of frequency content as the constant times in control period) in the control period by using measuring device is set for biography
Sensor signal execute maximum likelihood presumption (initially, as parameter assign suitable volume vi (f, τ) and covariance matrix Ri (f,
τ), parameter is executed to update until the square error of periodic component ci (f, τ) becomes smaller than predetermined value or executes pre-determined number ginseng
Number updates), separation cycle ingredient and the aperiodic ingredient.Then, abnormal detector, which is calculated, indicates separation with probabilistic manner
The probability distribution of the variation of the characteristic quantity of periodic component out, and be registered in together with the transmission function in monitored object operating portion
In database (being study processing above).In addition, there are multiple operating portions, according to preset monitored object
The time constant information of the periodic component of the transmission function in operating portion and the acoustic signal from monitored object operating portion is extracted and
From the periodic component of the acoustic signal in monitored object operating portion.
Then, abnormal detector is with the sensor signal other than sensor signal used in handling in study to check
Object checks for greatly deviating from from the periodic component learnt using the result (information of probability distribution) of study processing
Thus signal detects the exception of the sensor signal as check object (the above are abnormality detection processing).In this way, due to making
With maximal possibility estimation from sensor signal separation cycle ingredient and aperiodic ingredient come the exception of detection cycle ingredient, therefore energy
Enough operation noises for following the changed measuring device of sound field environment in a short time, and noise can be effectively inhibited to mention
Take the signal of monitored object.
(ii) in study processing, the sensor signal received is divided into multiple frequency bands by abnormal detector, will be each
The Signal separator of frequency band is periodic component and aperiodic ingredient (separating treatment of each frequency band), and according to the signal of each frequency band
Correlation extracts the periodic component (sequence solution processing) in monitored object operating portion.Processing in this way, even for every
The signal that a frequency time of origin changes, can also be separated into periodic component and aperiodic ingredient.
(iii) in embodiments of the present invention, detect signal exception in the case where also execute with above-mentioned study at
The same processing of reason.That is, abnormal detector is directed to subject sensor signal (input signal), using by study
The parameter of periodic component obtained by reason, execute maximal possibility estimation, thus by subject sensor Signal separator be periodic component and
Aperiodic ingredient.Then, abnormal detector is handled according to the sonority features amount for the periodic component being isolated and in study
In calculated monitored object operating portion the probability distribution (week that the periodic component that inspection is isolated is registered from normal components DB
The case where phase ingredient deviates), detect the exception of the sensor signal as object.
In addition, subject sensor signal is divided into multiple frequency bands in abnormality detection processing by abnormal detector.Then,
The filter of the signal in operating portion of the abnormal detector using removal other than monitored object operating portion, by the letter of each frequency band
Number it is separated into periodic component and aperiodic ingredient, and takes out the signal of the periodic component in monitored object operating portion.In this way, make
For check object, the periodic component of desired signal (signal from monitored object operating portion) efficiently can be only extracted.
(iv) present invention can also be realized by the program code of the software of the function of realization embodiment.Such case
Under, the storage medium for having recorded program code, computer (or the CPU, MPU) reading of the system or device are provided to system or device
The program code stored in storage medium out.In this case, program code read from storage medium itself is realized aforementioned
Embodiment function, the program code itself and store the storage medium of the program code as a result, and constitute the present invention.
As the storage medium for supplying such program code, floppy disk, CD-ROM, DVD-ROM, hard disk, light can be used for example
Disk, magneto-optic disk, CD-R, tape, non-volatile memory card, ROM etc..
In addition it is also possible to which the instruction according to program code such as OS (operating system) run on computers, carries out one
Point or whole actual treatments, realize the function of embodiment above-mentioned through this process.In addition it is also possible to will be from depositing
After the program code that storage media is read is written to the memory on computer, CPU of computer etc. is according to the finger of the program code
Show and carry out some or all actual treatment, realizes the function of embodiment above-mentioned through this process.
Further, it is also possible to realize the program code of the software of function of embodiment by sending via network, deposited
Store up in the storage mediums such as storage units or CD-RW, CD-R such as hard disk, memory of system or device, in use, the system or
The computer (or CPU, MPU) of device reads the storage unit, the program code stored in the storage medium and execution.
Finally, it is to be understood that the processing and technology described here is not to be substantially associated with any specific device,
But component is able to carry out installation by any appropriate combination.General purpose is able to use according to the introduction described here
A plurality of types of equipment.For the step of executing the method described here, constructing dedicated unit may be beneficial.In addition, passing through reality
The appropriately combined of multiple structural elements disclosed in mode is applied, various inventions are capable of forming.For example, can be disclosed from embodiment
Entire infrastructure element in delete several structural elements.In addition it is also possible to which the structural element of different embodiments is carried out group
It closes.The present invention is associated with concrete example and is described, however these are only intended to illustrate rather than progress in whole viewpoints
It limits.It will be appreciated by those skilled in the art that for implementing the present invention, there are the multiple of suitable hardware, software and firmware
Combination.For example, using the large-scale program such as compilation, C/C++, perl, Shell, PHP, Jave (registered trademark) or script
Language, to install described software.
In addition, in the above-described embodiment, considering the needs in explanation and having used control line, information wire, but making
Whole control lines, information wire need not be shown on product.Whole structures can be connected with each other.
In addition, there is the personnel of general knowledge for the art, other installations of the invention are according to disclosed herein
The investigation of the specification and embodiments of invention and become clear.In addition, specification and concrete example are only typical cases, it is of the invention
Scope and spirit indicate in subsequent request range.
Symbol description
100 measuring devices;
101 turntables;
102 operating portions;
103 monitored object operating portions;
Sensor is used in 104 measurements;
200 abnormal detectors;
201 central operation devices;
202 nonvolatile memories;
203 volatile memory;
211 monitoring servers.
Claims (14)
1. a kind of abnormal detector, detection passes through the sensor that multiple measurement sensors are measured possessed by measuring device
The exception of signal, which is characterized in that
The abnormal detector includes
Memory, the sensor signal measured described in storage, i.e., volume is with the constant in the control period of the measuring device
The sensor signal that the periodic component of variation and the aperiodic ingredient independently changed with the control period mix again;And
Processor detects the exception of the sensor signal,
The processor executes following processing:
The sensor signal is read from the memory, the sensor signal is executed using the information in the control period
Thus maximal possibility estimation separates the periodic component and the aperiodic ingredient, and calculating indicates the separation with probabilistic manner
The study of the probability distribution of the variation of the characteristic quantity of periodic component out is handled;And
The sensor signal other than the sensor signal used in study processing is read as inspection from the memory
Object uses the result of the study processing to detect the abnormal abnormal examination as the sensor signal of the check object
Processing.
2. abnormal detector according to claim 1, which is characterized in that
The sensor signal includes from multiple operating portions comprising monitored object operating portion that the measuring device has
Acoustic signal,
In study processing, the processor is according to the transmission function in preset monitored object operating portion and comes
From the time constant information of the periodic component of the acoustic signal in monitored object operating portion, extracts and transported from the monitored object
The periodic component of the acoustic signal of transfer part.
3. abnormal detector according to claim 2, which is characterized in that
In study processing, the processor executes following processing: the sensor signal is divided into multiple frequency bands, and
It is the processing of the periodic component and the aperiodic ingredient by the Signal separator of each frequency band;And the letter according to each frequency band
Number correlation extraction described in monitored object operating portion periodic component sequence solution processing.
4. abnormal detector according to claim 2, which is characterized in that
In abnormality detection processing, the processor uses the ginseng by the periodic component obtained by study processing
Number executes maximal possibility estimation for the sensor signal as the check object, as a result, will be as the check object
Sensor signal is separated into the periodic component and the aperiodic ingredient, and according to the sound equipment for the periodic component being isolated
Characteristic quantity and the study processing in calculated monitored object operating portion the probability distribution, detection be used as described in
The exception of the sensor signal of check object.
5. abnormal detector according to claim 4, which is characterized in that
In abnormality detection processing, the processor executes following processing: the sensor as the check object is believed
Number it is divided into multiple frequency bands, and using the filter for the signal for removing the operating portion other than monitored object operating portion,
It is the periodic component and the aperiodic ingredient by the Signal separator of each frequency band, and takes out the week in monitored object operating portion
The signal of phase ingredient.
6. abnormal detector according to claim 2, which is characterized in that
The processor also executes, and is at normal or abnormal abnormality detection by the movement for indicating monitored object operating portion
The processing on the display picture of display device as the result is shown of reason.
7. abnormal detector according to claim 2, which is characterized in that
The processor also executes following processing: setting monitored object operating portion from the multiple operating portion, and is showing
The first of the respective intensity for accordingly setting the multiple measurement sensor with the monitored object operating portion is shown in device
Beginning setting screen.
8. a kind of method for detecting abnormality, detection passes through the sensor that multiple measurement sensors are measured possessed by measuring device
The exception of signal, which is characterized in that
The method for detecting abnormality comprises the steps of:
The step of processor reads the sensor signal from the memory for the sensor signal measured described in storage, the biography
Sensor signal be volume with the periodic component of the constant times variation in the control period of the measuring device and with the control period
The sensor signal that the aperiodic ingredient independently changed mixes;
The processor executes maximal possibility estimation, separation to the sensor signal by using the information in the control period
The periodic component and the aperiodic ingredient, and calculate the characteristic quantity that the periodic component isolated is indicated with probabilistic manner
The learning procedure of the probability distribution of variation;And
The processor reads the sensor letter other than the sensor signal used in the learning procedure from the memory
Number be used as check object, use the processing result of the learning procedure to detect as the sensor signal of the check object
Abnormal abnormal examination step.
9. method for detecting abnormality according to claim 8, which is characterized in that
The sensor signal includes from multiple operating portions comprising monitored object operating portion that the measuring device has
Acoustic signal,
In the learning procedure, the processor is according to the transmission function in preset monitored object operating portion and comes
From the time constant information of the periodic component of the acoustic signal in monitored object operating portion, extracts and transported from the monitored object
The periodic component of the acoustic signal of transfer part.
10. method for detecting abnormality according to claim 9, which is characterized in that
In the learning procedure, the processor executes following processing: the sensor signal is divided into multiple frequency bands, and
It is the processing of the periodic component and the aperiodic ingredient by the Signal separator of each frequency band;And the letter according to each frequency band
Number correlation extraction described in monitored object operating portion periodic component sequence solution processing.
11. method for detecting abnormality according to claim 9, which is characterized in that
In the anomalies detecting step, the processor using the period obtained by processing by the learning procedure at
The parameter divided executes maximal possibility estimation for the sensor signal as the check object, will be used as the inspection as a result,
The sensor signal of object is separated into the periodic component and the aperiodic ingredient, and according to the periodic component being isolated
Sonority features amount and the study processing in calculated monitored object operating portion the probability distribution, detection make
For the exception of the sensor signal of the check object.
12. method for detecting abnormality according to claim 11, which is characterized in that
In the anomalies detecting step, the sensor signal as the check object is divided into multiple frequencies by the processor
Band, and using the filter for the signal for removing the operating portion other than monitored object operating portion, by the signal of each frequency band
It is separated into the periodic component and the aperiodic ingredient, and takes out the signal of the periodic component in monitored object operating portion.
13. method for detecting abnormality according to claim 9, which is characterized in that
The method for detecting abnormality also comprises the following steps:
The movement for indicating monitored object operating portion is the knot of normal or abnormal abnormality detection processing by the processor
Fruit is shown on the display picture of display device.
14. method for detecting abnormality according to claim 9, which is characterized in that
The method for detecting abnormality also comprises the following steps:
The processor from the multiple operating portion set monitored object operating portion, and show in a display device for
The monitored object operating portion accordingly sets the initial setting screen of the respective intensity of the multiple measurement sensor.
Applications Claiming Priority (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2013218008A JP6236282B2 (en) | 2013-10-21 | 2013-10-21 | Abnormality detection apparatus, abnormality detection method, and computer-readable storage medium |
JP2013-218008 | 2013-10-21 | ||
PCT/JP2014/077884 WO2015060273A1 (en) | 2013-10-21 | 2014-10-21 | Abnormality detection device, abnormality detection method, and computer-readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105637331A CN105637331A (en) | 2016-06-01 |
CN105637331B true CN105637331B (en) | 2019-01-01 |
Family
ID=52992867
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201480056176.7A Active CN105637331B (en) | 2013-10-21 | 2014-10-21 | Abnormal detector, method for detecting abnormality |
Country Status (3)
Country | Link |
---|---|
JP (1) | JP6236282B2 (en) |
CN (1) | CN105637331B (en) |
WO (1) | WO2015060273A1 (en) |
Families Citing this family (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108291837B (en) * | 2015-12-09 | 2020-02-14 | 三菱电机株式会社 | Degraded portion estimation device, degraded portion estimation method, and mobile body diagnosis system |
JP5954604B1 (en) * | 2015-12-14 | 2016-07-20 | 富士ゼロックス株式会社 | Diagnostic device, diagnostic system and program |
JP6620290B2 (en) * | 2017-07-26 | 2019-12-18 | 株式会社Ihi原動機 | Vibro-acoustic analysis method and apparatus, and apparatus abnormal part estimation method and apparatus |
CN108064456B (en) * | 2017-09-30 | 2021-09-03 | 深圳市欢太科技有限公司 | Abnormal frequency point detection method and device and computer storage medium |
JP7140194B2 (en) * | 2018-08-10 | 2022-09-21 | 日本電信電話株式会社 | Anomaly detector, probability distribution learning device, autoencoder learning device, program |
JP7272631B2 (en) * | 2019-03-08 | 2023-05-12 | 株式会社アニモ | Sound or vibration determination method and information processing system for determination |
JP7442390B2 (en) * | 2020-06-01 | 2024-03-04 | 株式会社日立ビルシステム | Bearing inspection equipment and bearing inspection method |
CN114999537A (en) * | 2022-06-22 | 2022-09-02 | 歌尔科技有限公司 | Abnormal sound signal detection method, device and equipment and readable storage medium |
CN117056663B (en) * | 2023-10-13 | 2024-02-13 | 腾讯科技(深圳)有限公司 | Data processing method and device, electronic equipment and storage medium |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003156415A (en) * | 2001-11-21 | 2003-05-30 | Shogo Tanaka | Method of inspecting soundness of large-sized structure by adaptive parameter estimation method using physical model, and device therefor |
CN101804932A (en) * | 2009-02-17 | 2010-08-18 | 株式会社日立制作所 | Apparatus for diagnosis of abnormality and abnormality diagnostic method |
CN102809493A (en) * | 2011-06-02 | 2012-12-05 | 三菱电机株式会社 | Abnormal sound diagnosis device |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2012018066A (en) * | 2010-07-07 | 2012-01-26 | Panasonic Electric Works Sunx Co Ltd | Device for inspecting abnormality |
-
2013
- 2013-10-21 JP JP2013218008A patent/JP6236282B2/en active Active
-
2014
- 2014-10-21 CN CN201480056176.7A patent/CN105637331B/en active Active
- 2014-10-21 WO PCT/JP2014/077884 patent/WO2015060273A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2003156415A (en) * | 2001-11-21 | 2003-05-30 | Shogo Tanaka | Method of inspecting soundness of large-sized structure by adaptive parameter estimation method using physical model, and device therefor |
CN101804932A (en) * | 2009-02-17 | 2010-08-18 | 株式会社日立制作所 | Apparatus for diagnosis of abnormality and abnormality diagnostic method |
CN102809493A (en) * | 2011-06-02 | 2012-12-05 | 三菱电机株式会社 | Abnormal sound diagnosis device |
Also Published As
Publication number | Publication date |
---|---|
CN105637331A (en) | 2016-06-01 |
WO2015060273A1 (en) | 2015-04-30 |
JP2015081768A (en) | 2015-04-27 |
JP6236282B2 (en) | 2017-11-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105637331B (en) | Abnormal detector, method for detecting abnormality | |
US11581005B2 (en) | Methods and systems for improved signal decomposition | |
EP3440598B1 (en) | Frequency based audio analysis using neural networks | |
CN110503971A (en) | Time-frequency mask neural network based estimation and Wave beam forming for speech processes | |
CN110491403A (en) | Processing method, device, medium and the speech enabled equipment of audio signal | |
CN103718242B (en) | Adopt the system and method for the treatment of voice signal of spectrum motion transform | |
CN107607065A (en) | A kind of impact echo signal analysis method based on variation mode decomposition | |
US9690749B2 (en) | Smart data sampling and data reconstruction | |
Kram et al. | UWB channel impulse responses for positioning in complex environments: A detailed feature analysis | |
CN103559888A (en) | Speech enhancement method based on non-negative low-rank and sparse matrix decomposition principle | |
CN108414240A (en) | A kind of method and apparatus of detection machine abnormal vibrations | |
US20210319802A1 (en) | Method for processing speech signal, electronic device and storage medium | |
US11393443B2 (en) | Apparatuses and methods for creating noise environment noisy data and eliminating noise | |
Ristea et al. | Estimating the magnitude and phase of automotive radar signals under multiple interference sources with fully convolutional networks | |
CN114077854B (en) | phi-OTDR underwater acoustic signal processing method and device based on self-adaptive VMD | |
Brankovic et al. | A machine learning-based seismic data compression and interpretation using a novel shifted-matrix decomposition algorithm | |
Song et al. | A novel noise reduction technique for underwater acoustic signals based on dual‐path recurrent neural network | |
JP6747447B2 (en) | Signal detection device, signal detection method, and signal detection program | |
WO2020250797A1 (en) | Information processing device, information processing method, and program | |
CN112735466A (en) | Audio detection method and device | |
Wang et al. | Fast ship radiated noise recognition using three-dimensional mel-spectrograms with an additive attention based transformer | |
CN110161283A (en) | Reduce the method and apparatus of acceleration signal measurement noise | |
US11297418B2 (en) | Acoustic signal separation apparatus, learning apparatus, method, and program thereof | |
CN115758237A (en) | Bearing fault classification method and system based on intelligent inspection robot | |
KR102275571B1 (en) | Method for providing smart device based fault diagnosis service using sound and vibration |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |