CN110244204A - A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values - Google Patents
A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values Download PDFInfo
- Publication number
- CN110244204A CN110244204A CN201910567311.9A CN201910567311A CN110244204A CN 110244204 A CN110244204 A CN 110244204A CN 201910567311 A CN201910567311 A CN 201910567311A CN 110244204 A CN110244204 A CN 110244204A
- Authority
- CN
- China
- Prior art keywords
- switchgear
- signal
- characteristic values
- diagnosing faults
- electric wave
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1209—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing using acoustic measurements
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses switchgear method for diagnosing faults, system and the media of a kind of multiple characteristic values, switchgear method for diagnosing faults of the present invention include acquire human ear heard sound, ultrasonic wave that tested switchgear issues, three kinds of waveform signals of electric wave, extract characteristic signals for three kinds of waveform signals respectively, characteristic signal be normalized, carries out dimensionality reduction;Characteristic signal after dimensionality reduction is inputted into trained machine learning model and obtains the fault diagnosis result of tested switchgear;The machine learning model establishes the mapping relations between dimensionality reduction and fused characteristic signal, fault diagnosis result by preparatory training.The present invention by extract human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave, extract corresponding feature, and feature is subjected to dimensionality reduction, recycling machine learning techniques are trained to data and comprehensive diagnos, the accuracy and reliability that switchgear fault detection can be significantly improved have huge application value in switchgear fault diagnosis field.
Description
Technical field
The present invention relates to Fault Diagnosis for Electrical Equipment technical fields, and in particular to a kind of switchgear failure of multiple characteristic values is examined
Disconnected method, system and medium.
Background technique
Equipment of the switchgear as terminaloriented user will generate important shadow if a failure occurs to people's lives
It rings and brings about great losses, therefore switchgear fault diagnosis technology is as the actual practicability technology of an incorporation engineering, in electricity
Key player is play in power production and application.Currently, the switchgear fault detection method being most widely used is mainly ultrasound
Wave detection method, the methods of electric wave detection method, they play certain effect in partial discharge of switchgear detection field.
In actually detected at the scene, using above two method, have the following disadvantages: one, due to all kinds of detectors
Testing result single index and threshold alarm can only be provided so that comprehensive descision needs to cause to detect by artificial experience
As a result accuracy, reliability is unable to get guarantee;Two, it only uses ultrasonic wave or ground electric wave carries out fault diagnosis, detection side
Formula is more single, it is difficult to carry out comprehensive assessment, testing result is be easy to cause to judge by accident.
Since the local discharge signal (ultrasonic wave, electric wave etc.) of switchgear also has certain specific modes, such as waveform
Amplitude, phase, frequency etc., these features tend to the operating status for largely reflecting equipment.However, these
Feature sometimes can be extremely complex, and the mankind, which are difficult to make, accurately to be judged, traditional mode identification method is also difficult to handle.And it is deep
The strong point of degree learning method is just being the ability of feature extraction and pattern-recognition, especially when facing challenge, one
As can reach the recognition effects of remote super other methods.
When discharge fault occurs for high-tension switch cabinet other than generating ultrasonic wave and ground electric wave, also along with human ear heard sound
Signal.The acquisition and processing of audible acoustical signal are currently widely used for mechanical oscillation fault detection, complex environment monitoring etc.,
But the research for the detection of high-tension switch cabinet discharge fault less at present, has scholar to propose using human ear audible acoustical signal in recent years
The research for carrying out the detection of switchgear discharge fault, achieves preferable effect.
If can by ultrasonic wave, three kinds of switchgears of electric wave and human ear heard sound issue signal use deep learning side
Method carries out comprehensive analysis to switchgear discharge fault situation, this is by the great accuracy for improving switchgear fault detection and reliably
Property, it has broad application prospects.
Summary of the invention
The technical problem to be solved in the present invention: in view of the above problems in the prior art, a kind of switch of multiple characteristic values is provided
Cabinet method for diagnosing faults, system and medium, the present invention by extract human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave,
Corresponding feature is extracted, and feature is subjected to dimensionality reduction, recycling machine learning techniques are trained to data and comprehensive diagnos, energy
The accuracy and reliability for enough significantly improving switchgear fault detection have huge using valence in switchgear fault diagnosis field
Value.
In order to solve the above-mentioned technical problem, the technical solution adopted by the present invention are as follows:
A kind of switchgear method for diagnosing faults of multiple characteristic values, implementation steps include:
1) acquire human ear heard sound, ultrasonic wave that tested switchgear issues, three kinds of waveform signals of electric wave;
2) respectively for human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave extract characteristic signals;
3) characteristic signal is normalized, then carries out dimensionality reduction;
4) characteristic signal after dimensionality reduction is inputted into trained machine learning model and obtains the fault diagnosis of tested switchgear
As a result;The machine learning model is established between dimensionality reduction and fused characteristic signal, fault diagnosis result by preparatory training
Mapping relations.
Preferably, the step of extracting characteristic signal for human ear heard sound in step 2) includes: for human ear heard sound
Waveform signal carries out adding window and sub-frame processing multiplied by time window respectively;It is special that time domain is extracted to the waveform model of adding window and sub-frame processing
It seeks peace characteristic signal of the frequency domain character as human ear heard sound, the temporal signatures include short-time average energy, short-time average width
Spend at least one of function, short-time average zero-crossing rate, short-time autocorrelation function, short time average magnitude difference function, the frequency domain
Feature includes mel cepstrum function.
Preferably, the step of ultrasonic wave extraction characteristic signal is directed in step 2) includes: the waveform signal for ultrasonic wave
Ultrasonic wave time-frequency figure is obtained after carrying out time-frequency convert, using ultrasonic wave time-frequency figure as the characteristic signal of ultrasonic wave.
Preferably, the step of extracting characteristic signal for ground electric wave in step 2) includes: the waveform signal for ground electric wave
Ground electric wave time-frequency figure is obtained after carrying out time-frequency convert, using ground electric wave time-frequency figure as the characteristic signal of ground electric wave.
Preferably, further include the steps that the waveform signal for ultrasonic wave is filtered amplification before step 2), and filter
Use frequency filtering for the narrow band filter of 40 ± 2kHz, amplification uses power to put for the signal of one of 60dB, 80dB, 100dB
Big device.
Preferably, further include the steps that the waveform signal for ground electric wave is filtered amplification before step 2), and filter
Use frequency filtering for the broadband filter of 3-60MHz, amplification is using the two-stage signal for using power being respectively 20dB and 40dB
Amplifier.
The present invention also provides a kind of switchgear fault diagnosis systems of multiple characteristic values, including sensor unit, preposition processing
Unit and signal processing unit, the sensor unit are tested human ear heard sound, the ultrasound that switchgear issues for acquisition testing
Wave, three kinds of waveform signals of electric wave and exported by preprocessing unit to signal processing unit, the signal processing unit quilt
Programming or the step of configure switchgear method for diagnosing faults to execute the multiple characteristic values or the signal processing unit
It is stored on storage medium and is programmed or configures to execute the computer journey of the switchgear method for diagnosing faults of the multiple characteristic values
Sequence.
Preferably, the preprocessing unit includes the first filtering for ultrasonic signal filter and amplification and denoising
Amplifying circuit and for ground electric wave signal filter and amplification and the second filter amplification circuit of denoising, first filter and amplification
Circuit is by the narrow band filter that frequency filtering is 40 ± 2kHz and power is used to put for the signal of one of 60dB, 80dB, 100dB
Big device is formed;Second filter amplification circuit is distinguished by the broadband filter that frequency filtering is 3-60MHz and using power
It is formed for the two-stage signal amplifier of 20dB and 40dB.
The present invention also provides a kind of switchgear fault diagnosis system of multiple characteristic values, including computer equipment, the computers
Equipment is programmed or configures the step of to execute the switchgear method for diagnosing faults of the multiple characteristic values or the computer equipment
Storage medium on be stored with and be programmed or configure to execute the computer of the switchgear method for diagnosing faults of the multiple characteristic values
Program.
The present invention also provides a kind of computer readable storage medium, it is stored with and is programmed on the computer readable storage medium
Or it configures to execute the computer program of the switchgear method for diagnosing faults of the multiple characteristic values.
Compared to the prior art, the present invention has an advantage that
1, the present invention by extract human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave, this three is extracted corresponding
Feature, and feature is subjected to dimensionality reduction, recycles machine learning techniques to be trained to data and comprehensive diagnos, can obviously mention
The accuracy and reliability of high switchgear fault detection have huge application value in switchgear fault diagnosis field, reduce
Testing result False Rate, greatly improves detection efficiency.
2, present invention utilizes machine learning models to be trained, and can be competent at complicated multiple features fault diagnosis demand,
Improve the intelligent level and Detection accuracy of detection.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without any creative labor, according to
These attached drawings obtain other attached drawings and still fall within scope of the invention.
Fig. 1 is the basic procedure schematic diagram of present invention method.
Fig. 2 is the schematic illustration of present invention method.
Fig. 3 is the structural schematic diagram of the convolutional neural networks CNN of the embodiment of the present invention.
Fig. 4 is the system structure diagram of the embodiment of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, the present invention is made into one below in conjunction with attached drawing
Step ground detailed description.
As shown in Figure 1, the implementation steps of the switchgear method for diagnosing faults of the present embodiment multiple characteristic values include:
1) acquire human ear heard sound, ultrasonic wave that tested switchgear issues, three kinds of waveform signals of electric wave;
2) respectively for human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave extract characteristic signals;
3) characteristic signal is normalized, then carries out dimensionality reduction;
4) characteristic signal after dimensionality reduction is inputted into trained machine learning model and obtains the fault diagnosis of tested switchgear
As a result;The machine learning model is established between dimensionality reduction and fused characteristic signal, fault diagnosis result by preparatory training
Mapping relations.
In the present embodiment selection based on human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave realize that switchgear is former
Hinder diagnosis concrete reason it is as follows: currently, the switchgear fault detection method being most widely used be mainly ultrasonic Detection Method,
The methods of ground electric wave detection method, they play certain effect in partial discharge of switchgear detection field.Partial discharge of switchgear
When, charging point is accelerated due to electric field action, and mechanical oscillation occur and generate ultrasonic wave, therefore can be according to ultrasonic wave
The feature of signal judges the degree and switchgear fault type of shelf depreciation;Partial discharge of switchgear also will form over the ground simultaneously
Electric current is propagated, the magnetic of the electric field and its induction that are generated according to Maxwell's theorem, partial discharge phenomenon in the metal surface of equipment
Field propagates form electromagnetic wave outward, and electromagnetic wave induces high-frequency current in switchgear outer surface, can be measured using capacitive coupling
Discharge pulse signal, thus detection switch cabinet discharge fault.Switchgear in addition to generate ultrasonic wave and ground electric wave other than, also along with
Human ear audible acoustical signal.There are mechanical oscillation when operating normally for switchgear, voice signal at this time be " drone " sound, volume one
As it is smaller.And when switchgear starts to occur partial discharge phenomenon, it will appear the discharging sound of " taste ", sound in shelf depreciation region
Signal to noise ratio " drone " normal operation sound it is slightly larger, and with the intensification of degree of discharge, the insulating layer for eventually resulting in switchgear is hit
It wears.Can be heard when insulation breakdown it is of short duration it is sharp " patter " sound.Therefore, become according to the audio frequency characteristics of switchgear voice signal
Change, can determine whether out whether switchgear breaks down.The acquisition and processing of audible acoustical signal are currently widely used for mechanical oscillation event
Barrier detection, complex environment monitoring etc., but the research for the detection of high-tension switch cabinet discharge fault less at present, have in recent years
Scholar proposes the research that the detection of switchgear discharge fault is carried out using human ear audible acoustical signal, achieves preferable effect.If
Can by ultrasonic wave, the signal that issues of three kinds of switchgears of electric wave and human ear heard sound switchgear is put using the method for deep learning
Electrical fault conditions carry out comprehensive analysis, this has the great accuracy and reliability for improving switchgear fault detection wide
Application prospect.
For electromagnetism interference, clutter reduction, thus the analog electrical signal of outputting high quality, in the present embodiment, step 2)
Further include the steps that the waveform signal for ultrasonic wave is filtered amplification before, and filtering uses frequency filtering for 40 ± 2kHz
Narrow band filter, amplification use power for the signal amplifier of one of 60dB, 80dB, 100dB.Unquestionably, filter and amplification
It can according to need and realized by the way of software or hardware, hardware realization is used in the present embodiment, therefore it is actually occurred
Before timing node is step 1);If using software realization, before practical time of origin node is step 2).
For electromagnetism interference, clutter reduction, thus the analog electrical signal of outputting high quality, in the present embodiment, step 2)
Further include the steps that the waveform signal for ground electric wave is filtered amplification before, and filtering uses frequency filtering for 3-60MHz
Broadband filter, amplification using using power be respectively 20dB and 40dB two-stage signal amplifier.Unquestionably, filtering is put
It can according to need greatly and realized by the way of software or hardware, hardware realization, therefore its practical hair are used in the present embodiment
Before raw timing node is step 1);If using software realization, before practical time of origin node is step 2).
As shown in Fig. 2, the step of extracting characteristic signal for human ear heard sound in step 2) includes: for human ear heard sound
Waveform signal carry out adding window and sub-frame processing multiplied by time window respectively;Time domain is extracted to the waveform model of adding window and sub-frame processing
The characteristic signal of feature and frequency domain character as human ear heard sound, the temporal signatures include short-time average energy, short-time average
At least one of amplitude function, short-time average zero-crossing rate, short-time autocorrelation function, short time average magnitude difference function, the frequency
Characteristic of field includes mel cepstrum function.
In the present embodiment, adding window and sub-frame processing are carried out multiplied by time window respectively for the waveform signal of human ear heard sound
When, specifically used Hamming window, function expression is shown below;
In above formula, w (n) is Hamming window window function, and N is that window is long (N=200 in the present embodiment), and n is frame number, is added
Window and sub-frame processing obtain the data frame s (n, h) after framing, also referred to as s (n), and wherein n is frame number, and h is frame length.
The calculating function expression of short-time average energy is shown below:
In above formula, EnFor short-time average energy, n is frame number, and N is that window is long, and s (m) is adding window and the number that sub-frame processing obtains
According to frame, m is to calculate intermediate quantity, and w (n-m) is Hamming window function.
Short-time average magnitude function can be used to measure the variation of sound amplitude, the calculating function table of short-time average magnitude function
It is shown below up to formula:
In above formula, MnFor the value of short-time average magnitude function, n is frame number, and N is that window is long, and s (n) is adding window and sub-frame processing
The data frame obtained, m are to calculate intermediate quantity, and w (n-m) is Hamming window function.
The calculating function expression of short-time average zero-crossing rate is shown below:
In above formula, ZnFor the value of short-time average zero-crossing rate, n is frame number, and N is that window is long, and s (m) is that adding window and sub-frame processing obtain
The data frame arrived, m are to calculate intermediate quantity, and s (m-1) is the data frame that one unit of time delay obtains.
Sgn (x) is sign function, and x is independent variable, is defined as follows shown in formula:
Auto-correlation function is used for the similitude of gauge signal temporal waveform, short-time autocorrelation function in the present embodiment
Function expression is calculated to be shown below:
In above formula, RnIt (k) is the value of short-time autocorrelation function, s (m) is adding window and the data frame that sub-frame processing obtains, and m is
Intermediate quantity is calculated, s (m+k) is the data frame that time shift-k units obtain, and w (n-m) and w (n-k-m) are respectively to s (m) and s (m
+ k) overturn and shifted after value, n is frame number, and m is to calculate intermediate quantity, and k is the independent variable of short-time autocorrelation function.
The calculating function expression of short time average magnitude difference function is shown below:
In above formula, rnIt (k) is the value of short time average magnitude difference function, n is frame number, and N is that window is long, and x (n) is voice signal, x
It (n+k) is the voice signal of time shift-k units.
Mel cepstrum function (MFCC) is the frequency spectrum for analyzing sound according to the hearing test result of people, and being usually used at present can
The research identified on hearing the news calculates step are as follows: 1) carries out Short Time Fourier Transform to each frame data s (n), obtain frequency domain data
S (n, j) (j is the j-th strip spectral line in frequency domain);2) line energy and the energy by Meier filter, gained spectral line energy are calculated
Amount seeks Meier spectrum energy and obtains energy parameter E (n, j) by Meier filter;3) the energy parameter E of Meier filter
(n, j) takes progress discrete cosine transform (DCT) after logarithm to obtain mel cepstrum function (MFCC), is shown below:
In above formula, mfcc (n, l) is to carry out discrete cosine transform (DCT) to obtain mel cepstrum function (MFCC), l DCT
Spectral line later, E (n, j) are that Meier spectrum energy obtains energy parameter, and M is the number of Meier filter, and m is to calculate intermediate quantity, n
For frame number.
As shown in Fig. 2, the step of being directed to ultrasonic wave extraction characteristic signal in step 2) includes: the waveform letter for ultrasonic wave
Number carry out time-frequency convert after obtain ultrasonic wave time-frequency figure, using ultrasonic wave time-frequency figure as the characteristic signal of ultrasonic wave.
As shown in Fig. 2, the step of extracting characteristic signal for ground electric wave in step 2) includes: the waveform letter for ground electric wave
Number carry out time-frequency convert after obtain ground electric wave time-frequency figure, using ground electric wave time-frequency figure as electric wave characteristic signal.
Be normalized in step 3) calculating data can be limited to after treatment it is certain within the scope of, while can
To accelerate convergence speed of the algorithm.In the present embodiment, the function expression for normalizing calculating is shown below:
Y=(X-Xmin)/(Xmax-Xmin)
In above formula, Y is normalization as a result, X is the eigenmatrix extracted from audible acoustical signal, and Xmin is characterized parameter
Minimum value, Xmax are characterized the maximum value of parameter.
Dimensionality reduction is carried out in the present embodiment, in step 3) to specifically refer to characteristic signal using principal component analytical method
(principal component analysis, PCA) carries out dimensionality reduction.Principal component analysis (principal component
Analysis, PCA) it is a kind of most common Feature Dimension Reduction algorithm, detailed process are as follows: average value processing is carried out to matrix Y and is obtained
Z;Covariance matrix D:D=ZZT is calculated, the characteristic value of matrix D and corresponding feature vector can be obtained by singular value decomposition;
The corresponding characteristic value of feature vector is resequenced in descending order, k row composition matrix E before taking, the as data after dimensionality reduction to k dimension.
In the present embodiment, the machine learning model in step 4) is convolutional neural networks CNN.As shown in figure 3, the present embodiment
The hierarchical structure of middle convolutional neural networks CNN includes data input layer, convolutional calculation layer, ReLU excitation layer, pond layer, full connection
Layer etc..A total of 8 layers, first 5 layers have been divided into two parts, and one of part is used to carry out mentioning for ultrasonic signal low-level image feature
It takes, another part is then used to extract the low-level image feature of ground electric wave signal.Then it is combined, comprehensively considers two sons
The feature combination that part is abstracted, continues to extract more advanced another characteristic.Full articulamentum has 4096 neurons, in order to keep away
Exempt from gradient disappearance problem, which has used ReLu activation primitive.This is because compared with general sigmoid activation primitive,
ReLu shows better performance in gradient descent method.
In conclusion the switchgear fault diagnosis of the present embodiment multiple characteristic values is integrated with microphone array due to compound sensor
Column, ultrasonic sensor, radio wave sensor, and therefrom extract different feature sets, can be to detection using multiple characteristic values
Data carry out comprehensive assessment, reduce the difficulty of testing result False Rate and live detection work, greatly improve detection effect
Rate;The convolutional neural networks (CNN) that the present embodiment is utilized in deep learning are trained, and convolutional neural networks can be competent at multiple
Miscellaneous multiple features fault diagnosis demand, improves the intelligent level and Detection accuracy of detection.
As shown in figure 4, the present embodiment also provides a kind of switchgear fault diagnosis system of multiple characteristic values, including sensor list
Member 1, preprocessing unit 2 and signal processing unit 3, sensor unit 1 are tested the human ear that switchgear issues for acquisition testing
Heard sound, ultrasonic wave, three kinds of waveform signals of electric wave and by the output of preprocessing unit 2 to signal processing unit 3, at signal
The step of reason unit 3 is programmed or is configured to execute the switchgear method for diagnosing faults of the aforementioned multiple characteristic values of the present embodiment, or
It is stored on the storage medium of signal processing unit 3 and is programmed or configures to execute the switchgear of the aforementioned multiple characteristic values of the present embodiment
The computer program of method for diagnosing faults.
Sensor unit 1 includes microphone array, ultrasonic sensor and ground radio wave sensor, is produced for acquiring switchgear
Raw human ear heard sound, ultrasonic wave, electric wave, and by collected human ear heard sound, ultrasonic wave, electric wave be converted into phase respectively
The analog electrical signal answered.In the present embodiment, the frequency response range of microphone array is 20~20kHz, sample rate 8kHz;It is super
The sample magnitude range of sonic sensor is -20~65dBmV, centre frequency 40kHz;The sample magnitude of ground radio wave sensor
Range is 0~60dBmV, and signal bandwidth is 3~60MHz.
In the present embodiment, sensor unit 1 is to be adopted by microphone array, ultrasonic sensor and ground radio wave sensor three
The compound sensor to be become one with micro-electromechanical system (MEMS) (Microelectromechanical Systems), so that multiple
Close sensor can integrate in narrow space there are three types of sensor avoided so that carrying convenience and multipurpose use can be carried out
Carry the disadvantage of excessive detector.In the present embodiment, microphone array, ultrasonic sensor and the ground electric wave of sensor unit 1
It is handled between sensor three using screening arrangement, can inhibit electromagnetic interference, prevent the crosstalk between multiple signals.
In the present embodiment, preprocessing unit 2 includes the first filter for ultrasonic signal filter and amplification and denoising
Wave amplifying circuit 21 and for ground electric wave signal filter and amplification and the second filter amplification circuit 22 of denoising, first filter
Wave amplifying circuit 21 is by the narrow band filter that frequency filtering is 40 ± 2kHz and uses power for one of 60dB, 80dB, 100dB
Signal amplifier formed;Second filter amplification circuit 22 is by the broadband filter that frequency filtering is 3-60MHz and uses function
Rate is respectively that the two-stage signal amplifier of 20dB and 40dB is formed.In the present embodiment, preprocessing unit 2 further includes A/D conversion
Circuit, A/D conversion circuit are used to analog electrical signal being converted to digital electric signal.
As shown in figure 4, signal processing unit 3 is made of detector 31, background host computer 32 in the present embodiment, detector 31
Input terminal is connected with the output end of preprocessing unit 2, and detector 31 executes feature extraction list by built-in microprocessor 311
The code of member, Feature Dimension Reduction unit completes feature extraction and Feature Dimension Reduction, then will be obtained after dimensionality reduction by communication unit 312
Characteristic signal export to background host computer 32, the characteristic signal after dimensionality reduction is inputted trained machine learning mould by background host computer 32
Type obtains the fault diagnosis result of tested switchgear, and fault diagnosis result is returned to detector 31, and pass through detector 31
The display unit 313 of upper connection shows output.By the above-mentioned means, on the one hand, can be minimized convenient for detector 31, so that producing
Product are more portable;On the other hand, it is also convenient for realizing by background host computer 32 to the convolutional neural networks (CNN) in deep learning
It is hardware-accelerated.Furthermore it also can according to need using individual computer node or more computer nodes and realize.
In the present embodiment, the communication unit 312 of detector 31 includes wireless communication module and wire communication module;Its
In, wireless communication module can use as needed the sub- communication module of GPRS, Zigbee communication submodule, WIFI communication submodule and
One or more among Bluetooth communication submodule;Wire communication module can use as needed 10/100BASE-T adaptively with
Too network submodular, RS-485 interface sub-module, RS-232 interface submodule, it is a kind of or more among optical fiber interface submodule
Kind.The display unit 313 of detector 31 uses LCD MODULE, for showing fault diagnosis result.
The course of work of the switchgear fault diagnosis system of multiple characteristic values is as follows in the present embodiment: sensor unit 1 acquires
Switchgear generate ultrasonic signal, electric wave signal and audible acoustical signal, and by the collected ultrasonic signal, it is electric
Wave signal and audible acoustical signal are converted into corresponding analog electrical signal respectively;Multiple simulations after 2 pairs of preprocessing unit conversions
Electric signal is filtered amplification and denoising respectively;Signal processing unit 3 respectively carries out three kinds of signals by detector 31
Multi-feature extraction and Feature Dimension Reduction processing, using convolutional neural networks (CNN) model in deep learning, complex characteristic is uploaded
Classification is trained and diagnosed fault to background host computer 32;Relevant diagnostic information is sent to detector 31, detector 31 passes through
The screen of display unit 313 shows fault diagnosis result.
In addition, the present embodiment also provides a kind of switchgear fault diagnosis system of multiple characteristic values, including computer equipment, it should
Computer equipment is programmed or configures the step of to execute the switchgear method for diagnosing faults of the aforementioned multiple characteristic values of the present embodiment, or
It is stored on the storage medium of person's computer equipment and is programmed or configures to execute the switch of the aforementioned multiple characteristic values of the present embodiment
The computer program of cabinet method for diagnosing faults.
In addition, the present embodiment also provides a kind of computer readable storage medium, stored on the computer readable storage medium
Have and is programmed or configures to execute the computer program of the switchgear method for diagnosing faults of the aforementioned multiple characteristic values of the present embodiment.
The above is only a preferred embodiment of the present invention, protection scope of the present invention is not limited merely to above-mentioned implementation
Example, all technical solutions belonged under thinking of the present invention all belong to the scope of protection of the present invention.It should be pointed out that for the art
Those of ordinary skill for, several improvements and modifications without departing from the principles of the present invention, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (10)
1. a kind of switchgear method for diagnosing faults of multiple characteristic values, it is characterised in that implementation steps include:
1) acquire human ear heard sound, ultrasonic wave that tested switchgear issues, three kinds of waveform signals of electric wave;
2) respectively for human ear heard sound, ultrasonic wave, three kinds of waveform signals of electric wave extract characteristic signals;
3) characteristic signal is normalized, then carries out dimensionality reduction;
4) characteristic signal after dimensionality reduction is inputted into trained machine learning model and obtains the fault diagnosis result of tested switchgear;
The machine learning model establishes the mapping between dimensionality reduction and fused characteristic signal, fault diagnosis result by preparatory training
Relationship.
2. the switchgear method for diagnosing faults of multiple characteristic values according to claim 1, which is characterized in that be directed in step 2)
It includes: to be added respectively multiplied by time window for the waveform signal of human ear heard sound that human ear heard sound, which extracts the step of characteristic signal,
Window and sub-frame processing;Temporal signatures and frequency domain character are extracted as human ear heard sound to the waveform model of adding window and sub-frame processing
Characteristic signal, the temporal signatures include short-time average energy, short-time average magnitude function, short-time average zero-crossing rate, in short-term from
At least one of correlation function, short time average magnitude difference function, the frequency domain character include mel cepstrum function.
3. the switchgear method for diagnosing faults of multiple characteristic values according to claim 2, which is characterized in that be directed in step 2)
The step of ultrasonic wave extraction characteristic signal include: for ultrasonic wave waveform signal carry out time-frequency convert after obtain ultrasonic wave time-frequency
Figure, using ultrasonic wave time-frequency figure as the characteristic signal of ultrasonic wave.
4. the switchgear method for diagnosing faults of multiple characteristic values according to claim 3, which is characterized in that be directed in step 2)
Ground electric wave extract characteristic signal the step of include: for ground electric wave waveform signal carry out time-frequency convert after obtain ground electric wave time-frequency
Figure, using ground electric wave time-frequency figure as the characteristic signal of ground electric wave.
5. the switchgear method for diagnosing faults of multiple characteristic values according to claim 1, which is characterized in that before step 2) also
Include the steps that the waveform signal for ultrasonic wave is filtered amplification, and filtering uses frequency filtering for the narrowband of 40 ± 2kHz
Filter, amplification use power for the signal amplifier of one of 60dB, 80dB, 100dB.
6. the switchgear method for diagnosing faults of multiple characteristic values according to claim 1, which is characterized in that before step 2) also
Include the steps that the waveform signal for ground electric wave is filtered amplification, and filtering uses frequency filtering for the broadband of 3-60MHz
Filter, amplification is using the two-stage signal amplifier for using power being respectively 20dB and 40dB.
7. a kind of switchgear fault diagnosis system of multiple characteristic values, which is characterized in that including sensor unit (1), preposition processing
Unit (2) and signal processing unit (3), human ear of the sensor unit (1) for acquisition testing to be tested switchgear sending can
On hearing the news, ultrasonic wave, three kinds of waveform signals of electric wave and by preprocessing unit (2) output give signal processing unit (3), it is described
The switchgear event that signal processing unit (3) is programmed or is configured with multiple characteristic values described in any one of perform claim requirement 1~6
It is stored on the storage medium of the step of hindering diagnostic method or the signal processing unit (3) and is programmed or configures to execute
The computer program of the switchgear method for diagnosing faults of multiple characteristic values described in any one of claim 1~6.
8. the switchgear method for diagnosing faults of multiple characteristic values according to claim 7, which is characterized in that the preposition processing
Unit (2) includes the first filter amplification circuit (21) for ultrasonic signal filter and amplification and denoising and for ground electric wave
Signal filter and amplification and the second filter amplification circuit (22) of denoising, first filter amplification circuit (21) is by filtering frequency
Rate is the narrow band filter of 40 ± 2kHz and power is used to be formed for the signal amplifier of one of 60dB, 80dB, 100dB;Institute
State the second filter amplification circuit (22) by frequency filtering be 3-60MHz broadband filter and using power be respectively 20dB and
The two-stage signal amplifier of 40dB is formed.
9. a kind of switchgear fault diagnosis system of multiple characteristic values, including computer equipment, which is characterized in that the computer equipment
It is programmed or configures with the step of the switchgear method for diagnosing faults of multiple characteristic values described in any one of perform claim requirement 1~6
Suddenly it is stored with to be programmed or configure or on the storage medium of the computer equipment and requires any one of 1~6 with perform claim
The computer program of the switchgear method for diagnosing faults of the multiple characteristic values.
10. a kind of computer readable storage medium, which is characterized in that be stored on the computer readable storage medium be programmed or
Configuration is with the computer program of the switchgear method for diagnosing faults of multiple characteristic values described in any one of perform claim requirement 1~6.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910567311.9A CN110244204A (en) | 2019-06-27 | 2019-06-27 | A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910567311.9A CN110244204A (en) | 2019-06-27 | 2019-06-27 | A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110244204A true CN110244204A (en) | 2019-09-17 |
Family
ID=67889891
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910567311.9A Pending CN110244204A (en) | 2019-06-27 | 2019-06-27 | A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110244204A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110519129A (en) * | 2019-09-26 | 2019-11-29 | 广东电网有限责任公司 | A kind of online rate real-time monitoring system of power distribution automation equipment |
CN110595609A (en) * | 2019-09-29 | 2019-12-20 | 国网河南省电力公司南召县供电公司 | Diagnosis device for rapidly judging fault of power transformation equipment through sound abnormity |
CN110808068A (en) * | 2019-11-11 | 2020-02-18 | 北京软通智城科技有限公司 | Sound detection method, device, equipment and storage medium |
CN111537853A (en) * | 2020-06-29 | 2020-08-14 | 国网山东省电力公司菏泽供电公司 | Intelligent detection method for partial discharge of switch cabinet based on multi-source heterogeneous data analysis |
CN111856224A (en) * | 2020-06-29 | 2020-10-30 | 国网河南省电力公司周口供电公司 | Partial discharge multi-source fusion detection system and method for high-voltage switch cabinet |
CN112329637A (en) * | 2020-11-06 | 2021-02-05 | 华北电力大学 | Load switch event detection method and system by using mode characteristics |
CN113255475A (en) * | 2021-05-07 | 2021-08-13 | 华中科技大学 | Loader offline fault identification method and system and electronic equipment |
CN113297922A (en) * | 2021-04-30 | 2021-08-24 | 广西电网有限责任公司电力科学研究院 | High-voltage switch cabinet fault diagnosis method and device and storage medium |
CN113390957A (en) * | 2021-04-27 | 2021-09-14 | 杭州电子科技大学 | Anti-crosstalk eddy current nondestructive flaw detection system based on magnetic sensing probe |
CN113866340A (en) * | 2021-08-31 | 2021-12-31 | 国网上海市电力公司 | Switch cabinet impulse voltage test and diagnosis method |
CN114019310A (en) * | 2021-11-08 | 2022-02-08 | 国网湖南省电力有限公司 | Method, equipment and medium for prejudging open circuit fault of energy storage converter |
CN115372764A (en) * | 2022-07-12 | 2022-11-22 | 国网湖北省电力有限公司黄石供电公司 | Fault diagnosis method for switch cabinet insulation component based on full-audio monitoring |
CN116930666A (en) * | 2023-09-15 | 2023-10-24 | 深圳凯升联合科技有限公司 | Intelligent diagnosis system and diagnosis method for low-voltage complete switch cabinet |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105021958A (en) * | 2014-07-27 | 2015-11-04 | 国家电网公司 | Switch cabinet partial discharge data recording and analyzing method based on multi-sensor detection |
CN105589020A (en) * | 2016-01-27 | 2016-05-18 | 深圳供电局有限公司 | Detector and method used for power distribution equipment inspection and live-line detection |
KR20160074022A (en) * | 2014-12-17 | 2016-06-28 | 전주비전대학교산학협력단 | Partial discharge diagnostic system using emf sensor and acoustic sensor |
CN206848410U (en) * | 2017-04-28 | 2018-01-05 | 山东迅风电子有限公司 | A kind of Partial discharge detector |
CN108052946A (en) * | 2017-12-11 | 2018-05-18 | 国网上海市电力公司 | A kind of high pressure cabinet switch automatic identifying method based on convolutional neural networks |
CN108169639A (en) * | 2017-12-29 | 2018-06-15 | 南京康尼环网开关设备有限公司 | Method based on the parallel long identification switch cabinet failure of Memory Neural Networks in short-term |
CN108303624A (en) * | 2018-01-31 | 2018-07-20 | 舒天才 | A kind of method for detection of partial discharge of switch cabinet based on voice signal analysis |
-
2019
- 2019-06-27 CN CN201910567311.9A patent/CN110244204A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105021958A (en) * | 2014-07-27 | 2015-11-04 | 国家电网公司 | Switch cabinet partial discharge data recording and analyzing method based on multi-sensor detection |
KR20160074022A (en) * | 2014-12-17 | 2016-06-28 | 전주비전대학교산학협력단 | Partial discharge diagnostic system using emf sensor and acoustic sensor |
CN105589020A (en) * | 2016-01-27 | 2016-05-18 | 深圳供电局有限公司 | Detector and method used for power distribution equipment inspection and live-line detection |
CN206848410U (en) * | 2017-04-28 | 2018-01-05 | 山东迅风电子有限公司 | A kind of Partial discharge detector |
CN108052946A (en) * | 2017-12-11 | 2018-05-18 | 国网上海市电力公司 | A kind of high pressure cabinet switch automatic identifying method based on convolutional neural networks |
CN108169639A (en) * | 2017-12-29 | 2018-06-15 | 南京康尼环网开关设备有限公司 | Method based on the parallel long identification switch cabinet failure of Memory Neural Networks in short-term |
CN108303624A (en) * | 2018-01-31 | 2018-07-20 | 舒天才 | A kind of method for detection of partial discharge of switch cabinet based on voice signal analysis |
Non-Patent Citations (2)
Title |
---|
史塨毓 等: "基于可闻声源多模态特征的开关柜绝缘故障识别", 《南京工程学院学报》 * |
朱正国 等: "基于卷积神经网络的开关柜局部放电监测", 《电网与清洁能源》 * |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110519129B (en) * | 2019-09-26 | 2023-10-10 | 广东电网有限责任公司 | Online rate real-time monitoring system of power distribution automation equipment |
CN110519129A (en) * | 2019-09-26 | 2019-11-29 | 广东电网有限责任公司 | A kind of online rate real-time monitoring system of power distribution automation equipment |
CN110595609A (en) * | 2019-09-29 | 2019-12-20 | 国网河南省电力公司南召县供电公司 | Diagnosis device for rapidly judging fault of power transformation equipment through sound abnormity |
CN110808068A (en) * | 2019-11-11 | 2020-02-18 | 北京软通智城科技有限公司 | Sound detection method, device, equipment and storage medium |
CN111537853A (en) * | 2020-06-29 | 2020-08-14 | 国网山东省电力公司菏泽供电公司 | Intelligent detection method for partial discharge of switch cabinet based on multi-source heterogeneous data analysis |
CN111856224A (en) * | 2020-06-29 | 2020-10-30 | 国网河南省电力公司周口供电公司 | Partial discharge multi-source fusion detection system and method for high-voltage switch cabinet |
CN112329637A (en) * | 2020-11-06 | 2021-02-05 | 华北电力大学 | Load switch event detection method and system by using mode characteristics |
CN113390957A (en) * | 2021-04-27 | 2021-09-14 | 杭州电子科技大学 | Anti-crosstalk eddy current nondestructive flaw detection system based on magnetic sensing probe |
CN113297922A (en) * | 2021-04-30 | 2021-08-24 | 广西电网有限责任公司电力科学研究院 | High-voltage switch cabinet fault diagnosis method and device and storage medium |
CN113255475A (en) * | 2021-05-07 | 2021-08-13 | 华中科技大学 | Loader offline fault identification method and system and electronic equipment |
CN113866340A (en) * | 2021-08-31 | 2021-12-31 | 国网上海市电力公司 | Switch cabinet impulse voltage test and diagnosis method |
CN114019310A (en) * | 2021-11-08 | 2022-02-08 | 国网湖南省电力有限公司 | Method, equipment and medium for prejudging open circuit fault of energy storage converter |
CN114019310B (en) * | 2021-11-08 | 2024-02-20 | 国网湖南省电力有限公司 | Open-circuit fault pre-judging method, equipment and medium for energy storage converter |
CN115372764A (en) * | 2022-07-12 | 2022-11-22 | 国网湖北省电力有限公司黄石供电公司 | Fault diagnosis method for switch cabinet insulation component based on full-audio monitoring |
CN116930666A (en) * | 2023-09-15 | 2023-10-24 | 深圳凯升联合科技有限公司 | Intelligent diagnosis system and diagnosis method for low-voltage complete switch cabinet |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110244204A (en) | A kind of switchgear method for diagnosing faults, system and the medium of multiple characteristic values | |
CN106405339B (en) | Based on the associated transmission line malfunction reason discrimination method of low-and high-frequency wavelet character | |
Deng et al. | Towards heart sound classification without segmentation via autocorrelation feature and diffusion maps | |
CN111325095B (en) | Intelligent detection method and system for equipment health state based on acoustic wave signals | |
CN110534118A (en) | Transformer/reactor method for diagnosing faults based on Application on Voiceprint Recognition and neural network | |
Istrate et al. | Information extraction from sound for medical telemonitoring | |
CN109616140B (en) | Abnormal sound analysis system | |
CN107748314A (en) | Transformer Faults Analysis system based on sound wave shock detection | |
CN102163427B (en) | Method for detecting audio exceptional event based on environmental model | |
CN110275094A (en) | A kind of high-tension cable shelf depreciation acousto-optic associated detecting method and system | |
CN109901031A (en) | Signal De-noising Method, information data processing terminal for local discharge signal | |
CN109856517A (en) | A kind of method of discrimination of extra-high voltage equipment Partial Discharge Detection data | |
CN108169639B (en) | Method for identifying switch cabinet fault based on parallel long-time and short-time memory neural network | |
CN106203380B (en) | Ultrasonic wave gesture identification method and system | |
CN103413113A (en) | Intelligent emotional interaction method for service robot | |
CN112735473B (en) | Method and system for identifying unmanned aerial vehicle based on voice | |
CN110133610A (en) | ULTRA-WIDEBAND RADAR action identification method based on time-varying distance-Doppler figure | |
CN101494049A (en) | Method for extracting audio characteristic parameter of audio monitoring system | |
CN108414079A (en) | A kind of transformer noise monitoring system | |
CN106782505A (en) | A kind of method based on electric discharge voice recognition high-tension switch cabinet state | |
You et al. | Novel feature extraction method for cough detection using NMF | |
CN114325256A (en) | Power equipment partial discharge identification method, system, equipment and storage medium | |
CN116778956A (en) | Transformer acoustic feature extraction and fault identification method | |
CN113221673B (en) | Speaker authentication method and system based on multi-scale feature aggregation | |
Rougui et al. | Audio sound event identification for distress situations and context awareness |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190917 |