CN102183710A - Method and system for identifying numbers of partial discharge sources of transformers - Google Patents
Method and system for identifying numbers of partial discharge sources of transformers Download PDFInfo
- Publication number
- CN102183710A CN102183710A CN2011100501285A CN201110050128A CN102183710A CN 102183710 A CN102183710 A CN 102183710A CN 2011100501285 A CN2011100501285 A CN 2011100501285A CN 201110050128 A CN201110050128 A CN 201110050128A CN 102183710 A CN102183710 A CN 102183710A
- Authority
- CN
- China
- Prior art keywords
- signal
- array
- partial discharge
- transformer
- source
- 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
Images
Abstract
The invention relates to a method and a system for identifying numbers of partial discharge sources of transformers and is used for monitoring and positioning the partial discharge of the transformers. The technical solution comprises the steps of receiving ultrasonic broadband signals emitted by the partial discharge sources by employing an ultrasonic sensor array and forming an array model, then converting the broadband signals received by the ultrasonic sensor array to narrow band signals through a TCT algorithm in broadband array signal processing algorithms, and finally identifying the numbers of the partial discharge sources of the transformers by employing an improved Gerschgorin circle method. The invention also provides the identification system. The method not only realizes accurate online identification of the numbers of the partial discharge sources in the transformers, but also realizes simple and easy execution with the number identification system, which is simple in structure.
Description
Technical field
The present invention relates to a kind of method and dedicated test system that is used for identification partial discharge of transformer source number, belong to the detection technique field.
Background technology
Along with the growth of power supply capacity, more and more higher to the requirement of Operation of Electric Systems reliability, and power transformer is the hinge equipment in the electric system, its operational reliability is directly connected to the safety and stablization of electric system.To the statistical results show of physical fault, insulation fault is the main cause that influences the normal operation of transformer, and shelf depreciation is to cause the major reason of insulation fault.
Current, the state maintenance of development power equipment has become a kind of inexorable trend, and not only there is important booster action shelf depreciation location to the assessment of the extent of injury of shelf depreciation accurately, and can provide the guidance of science for the repair based on condition of component of transformer, so that the maintainer carries out maintenance of equipment with a definite target in view, fix a breakdown rapidly, improve the maintenance level, also the design of improvement transformer device structure, raising manufacturing technology level are had directive significance simultaneously.
The traditional detection method of partial discharge of transformer mainly contains pulse current method, vapor-phase chromatography, supercritical ultrasonics technology, ultrahigh frequency method and flash spotting.Wherein, the influence that can avoid electromagnetic interference (EMI), location are convenient because of having, the advantage such as come to the same thing of online detection and offline inspection has obtained widespread use for supercritical ultrasonics technology.Utilizing supercritical ultrasonics technology to detect in each algorithm of partial discharge of transformer source position, it is a key issue that the source number is put in accurate identification office.In practical application, source number unknown number is often put in office, needs first identification office to put the number in source or suppose knownly, could estimation office puts the orientation in source.For example the most of algorithm in the Estimation of Spatial Spectrum technology all needs to know signal number, when the office of identification puts the source number and number is inconsistent really, peak value number and actual source number in the spatial spectrum curve are inequality, cause false dismissal or false-alarm, thus to the actual signal orientation really the fixed output quota life have a strong impact on.Therefore, the number identification in power transformer partial discharge source is a crucial research work.
At present, the common method of signal source number identification mainly comprises: AIC criterion, level and smooth rank technique, and Gai Shi circle criterion, canonical correlation technique, or the like.
Propose in order to solve the model selection problem by the AIC criterion of Akaike proposition and the MDL criterion that proposes by Schwartz and Rissanen.Based on AIC and MDL criterion, M.Wax and T. Kailath propose a kind of signal source number estimation method, and the estimation that obtained by the MDL criterion of proof is that the consistance of information source number actual value is estimated, are not consistent Estimation and AIC criterion obtains.This method utilization be the eigenwert of array covariance matrix, carry out the identification of signal source number according to signal characteristic value and noise characteristic value.Advantage is not need artificial to set a detection threshold in advance, thereby the subjective factor of having avoided the people is to information source number order estimation effect.AIC criterion is not that consistance is estimated, promptly in the occasion of big fast umber of beats, it still has the bigger probability of error; And the MDL criterion is better relatively; The HQ criterion occupy between the two, and main cause is to be caused by the penalty function item in the criterion.The MDL criterion is that consistance is estimated, just this criterion has preferable performance under the high s/n ratio situation, but this criterion is compared AIC the higher probability of error is arranged under the low signal-to-noise ratio situation.
Level and smooth rank technique is based on to be separated relevant based signal source and counts estimation problem, judges source number and structure according to the level and smooth order sequence that obtains, and maximum number is signal number in the level and smooth order.It has compared following advantage with the information theory method: when big, the fast umber of beats of signal to noise ratio (S/N ratio) was big, level and smooth order serial method was better than the performance of AIC and HQ algorithm; When incoming signal comprises several coherent source group times, the information theory method can only identification signal source number, can't carry out identification to structure.And level and smooth order algorithm not only can pick out the signal source sum, but also can pick out the signal source structure.But the calculating process of level and smooth rank technique is comparatively numerous and diverse, not easy programming.
Gai Shi circle criterion (GDE) is not an eigenwert of utilizing autocorrelation matrix, but utilizes its Gai Shi radius of circle to carry out signal number visual estimation meter.According to the Gai Shi circle theorem, the Gai Shi radius of circle of signal correspondence is bigger, and the Gai Shi radius of circle of noise correspondence is zero.In fact, because autocorrelation matrix estimates that the Gai Shi radius of circle of noise correspondence might not be zero.The degree that two class Gai Shi disks separate is decided by the sample number of estimate covariance matrix, and sample number is big more, and covariance matrix is accurate more, more can be separately with the Gai Shi disk, thereby just can from the Gai Shi disk, determine innings to put the source number.At present, the existing multiple improvement of Gai Shi circle criterion, wherein modal improvement algorithm (MGDE) ultimate principle is that related function between array element is replaced and weighted comprehensive improves identification precision.Gai Shi circle criterion performance under the high s/n ratio situation is fine, and performance is not ideal under the low signal-to-noise ratio.
Canonical correlation technique (CCT) is to suppose that narrow band signal independently incides the array of two apart, covariance matrix to whole array carries out certain conversion and it is carried out svd, then with the singular value descending sort, utilize sequence of singular value structure again, by this sequence and one relatively, just can judge information source number by the resulting decision threshold of known false-alarm probability.This algorithm performance under high s/n ratio is fine, and is also obviously good than level and smooth rank technique and AIC criterion under the low signal-to-noise ratio, can set different threshold values under coloured noise, has robustness.
Summary of the invention
The objective of the invention is to overcome the deficiencies in the prior art, the method for a kind of identification Partial Discharge in Power Transformer easily and accurately source number is provided, the present invention gives identification system simultaneously.
Problem of the present invention realizes with following technical proposals:
A kind of transformer partial discharge source number discrimination method, the ultrasonic broadband signal that it adopts the supersonic array sensor office of acceptance source of putting to send, and formation Array Model, the TCT algorithm of using then in the wideband Array Signal Processing algorithm converts the broadband signal that the supersonic array sensor is received to narrow band signal, and application enhancements Gai Shi circle method is carried out identification to partial discharge of transformer source number again.
Above-mentioned transformer partial discharge source number discrimination method, it may further comprise the steps:
A, the supersonic array sensor is close to the ultrasonic broadband signal that send in the transformer surface office of acceptance source of putting;
B, employing TCT broadband focusing algorithm are with the direction matrix under the different frequency
The definite method that transforms to optimum reference frequency f 0:f 0 is as follows:
Criterion according to the focusing error minimum:
Adopt the method for search to obtain optimum reference frequency
In the formula,
,
With
Expression respectively
With
Singular value,
With
The covariance matrix (focus method is seen below and chatted) of representing the frequency of central reference frequency after the denoising of array output data and j segment data respectively;
C, employing improve Gai Shi circle algorithm the narrow band signal after focusing on are carried out the number identification, determine the number of Partial Discharge Sources:
Suppose uniform array by
Individual equally spaced primitive is formed, and has
The discharge signal incident of individual office, can pass through following formulate after the ultrasonic broadband signal line focus of array received:
When noise was the space white Gaussian noise, the covariance matrix of this signal was:
Wherein,
Be the signal source covariance matrix,
Be the energy of white Gaussian noise,
Be
Unit matrix, in actual applications,
Be on average to approach by the signal covariance matrix that each time snap obtains,
If
Each column vector be
,
, then office puts the number identification and is undertaken by following step:
1. right
Carry out svd and suppose the singular value descending sort:
Wherein,
It is the resulting characteristic vector space of svd;
2. get
,
, and respectively to each
Utilize following method that information source number is carried out identification:
K gets 1,2 successively ..., M, the value of calculating following formula:
Wherein
Be on average to approach by the signal covariance matrix that each time snap obtains
Fast umber of beats,
Be
Decreasing function, set as the case may be, if work as
The time,
For first than zero little value, then the identifier of information source number is got
3. will
Inferior estimated result is weighted on average, and resulting result is the information source number that finally picks out.
A kind of transformer partial discharge source number identification system, constitute by supersonic array sensor, multi-channel high-speed data synchronous acquisition device and notebook computer, described supersonic array sensor is close to transformer case, its output signal connects the input end of multi-channel high-speed data synchronous acquisition device through signal amplifier, and the signal output part of described multi-channel high-speed data synchronous acquisition device connects the input interface of notebook computer.
Above-mentioned transformer partial discharge source number identification system, described supersonic array sensor are the planar structure of 16 array elements, 4 row * 4 row, and single array element is 7.8mm for the horizontal section diameter, the right cylinder of high 18mm, and array element center distance 8mm, material is a piezoelectric ceramics.
The supersonic array sensor that transformer case is close in utilization of the present invention receives the ultrasonic broadband signal of shelf depreciation, and adopt wideband array signal processing method TCT directly the type local-discharge ultrasonic broadband signal to be handled, in the number that guarantees to pick out under the prerequisite that tested transformer can be not impaired Partial Discharge Sources, this method not only can be implemented online accurate identification to transformer inside Partial Discharge Sources number, and used number identification system is simple in structure, and method is simple.
Description of drawings
The invention will be further described below in conjunction with accompanying drawing.
Fig. 1 is a TCT broadband focusing algorithm process flow diagram;
Fig. 2 is that source number identification process flow diagram is put in office;
Fig. 3 is 16 array element supersonic array sensor vertical cross-section sectional structure synoptic diagram;
Fig. 4 is that detection system constitutes synoptic diagram;
Fig. 5 is the detection system circuit theory diagrams;
Fig. 6 is that 16 passages are put source time-domain filtering oscillogram for single innings.
Each label is among the figure: A, supersonic array sensor; 1, cable; 2, gland nut; 3, stainless steel cover; 4, array element lead-in wire; 5, O-ring seal; 6, acoustic damping piece; 7, piezoelectricity array element; 8, Stainless Steel Shell; 9, silicon rubber filling; 10, stainless steel cuticula; 11, transformer; 12, signal amplifier; 13, multi-channel high-speed data synchronous acquisition device; 14, computer; 15, the source is put in office.
Each symbol is in the literary composition:
, the direction matrix;
, comprise the signal source direction angle
And the angle of pitch
The parameter of information;
, reference frequency is focusing center's frequency;
, the j segment data frequency;
, steering vector;
, white Gaussian noise energy;
, fast umber of beats;
, the sampling duration be the data matrix of t;
, the sampling duration be the signal space data matrix of t;
, the sampling duration be the spatial noise data matrix of t;
, covariance matrix;
, matrix eigenvector;
, signal covariance matrix;
,
Column vector;
, equidistant primitive number;
,
Unit matrix;
, far field narrow band signal number;
, the j section receives the DFT conversion of data matrix;
, the j section receives the DFT conversion of data direction matrix;
, the j section receives the DFT conversion of data-signal spatial data matrix;
, the j section receives the DFT conversion of data noise spatial data matrix;
, focussing matrix;
, the Frobenius mould.
Embodiment
The capacitance of each electric capacity is among Fig. 5
,
,
And the resistance value of each resistance is
,
,
Put source number identification at office and put the importance of location for ultrasound wave office, the present invention aims to provide the number identification system and the method thereof of the wideband Array Signal Processing in a kind of partial discharge of transformer source.
This invention adopts the supersonic array sensor to receive the ultrasonic broadband signal of shelf depreciation, form Array Model, and use broadband in the wideband Array Signal Processing algorithm and focus on, improve Gai Shi circle scheduling algorithm and realize number identification to Partial Discharge Sources, for accurately assessment transformer partial discharge source number and maintenance strategy provide effective reference.
For achieving the above object, the invention provides the number identification system and the method thereof of the wideband Array Signal Processing in a kind of partial discharge of transformer source, concrete technical scheme is as follows:
1. the supersonic array number identification system in partial discharge of transformer source.
This number identification system is constructed as follows: supersonic array sensor, signal amplifier (comprising charge amplifier, filter amplifier), multi-channel high-speed data synchronous acquisition device and notebook computer constitute.
The supersonic array sensor is the planar structure of 16 array elements, 4 row * 4 row, and single array element is diameter 7.8mm, long 18mm right cylinder, and array element distance 8mm, material is a piezoelectric ceramics.
When the source number was put in identification office, the supersonic array sensor was close to the shell of transformer, and each array element detects and obtains the type local-discharge ultrasonic signal, received and form the planar array signal of 4 row * 4 row.
The array signal of 16 array elements connects 16 electric charges (voltage) amplifier respectively by cable, connects filter amplifier again, receives 16 channel data synchronous acquisition devices then, obtains the supersonic array time-domain signal of partial discharge of transformer.
2. partial discharge of transformer source supersonic array number discrimination method.
(1) partial discharge of transformer wideband supersonic array signal obtains.Adopt the supersonic array sensor of 4 row * 4 row directly to receive the shelf depreciation ultrasonic signal, and, on 16 passage high-speed data synchronous acquisition devices, obtain wideband supersonic array signal by after the filtering amplification from the transformer outer wall.
The development of supersonic array sensor
The supersonic array sensor adopts planar array, and its basic structure mainly comprises: piezoelectric transducer array; The sound absorption backing; Acoustic matching layer; Other is as contact conductor, shell etc.Above-mentioned 4 parts play a part in sensor separately, form an organic whole.
A, transducer array design
Transducer array is the array of being lined up according to certain rules by several piezoelectric transducers.The single transducer of forming transducer array is called array element, so array element is the simple structure form of transducer array.Transducer array is called sonar transducer array in underwater acoustics, be called ultrasonic probe in ultrasonics.The signal space processor of battle array and Beam-former (device of being made up of time-delay, weighting, addition and circuit such as multiply each other) formation makes sound detection equipment obtain required orientation accuracy and spatial gain.Its function, principle of work and version are similar to the antenna array in the radar.According to the not same-action that transmits and receives sound wave in equipment, battle array is divided into the emission battle array and receives battle array.If same battle array double as receives, emission, then be called the transceiver battle array.
The transducer array unit design is cylindrical, and array design is a square.The cylinder design of array element mainly is the variation error that reduces the pressure of piezoelectric sensor sound wave in the vibration of sensor edge and inside, sound wave relative cylindrical center in the vibration at medium inside and edge is equidistant in cylindrical, can effectively reduce the piezoelectricity transformed error.The squared design of array is for guaranteeing the signal conformance of sensor at length and width both direction.
The frequency of popping one's head in the ultrasonic diagnosis is by the decision of the thickness of piezoelectric crystal.When a certain alternating voltage added on the piezoelectric, a vibration was sent on the piezoelectric crystal surface, and to it the contact medium in emittance.The size of emittance is relevant with the characteristic of boundary.Unless characteristic impedance equates, otherwise in the ordinary course of things, always some energy is reflected by piezoelectric crystal, returns towards the piezoelectric crystal surface reflection, propagates in the other direction to piezoelectric crystal.If from the vibrations that the front surface of piezoelectric crystal sends, be reflected after piezoelectric crystal when reaching back of the body surface, the vibration of its meeting and the generation of back of the body surface superposes mutually.According to the superposition principle of sound wave, have only two branches of working as to vibrate same phase time, itself and amplitude are maximum, could launch strong ultrasound wave.
If wish that two surfaces of transducer can both free vibration, when crystal thickness just in time equals half wavelength, produce resonance, i.e. these two vibrations form homophase on the surface, and stress is strengthened mutually, and vibration has peak swing.The fundamental resonance frequency that is called transducer corresponding to the frequency of half-wavelength thickness.If longitudinal wave propagation speed in certain piezoelectric
,
Be the thickness of material,
, promptly the product of the fundamental frequency of certain piezoelectric and thickness is a constant (being called frequency constant on the acoustics).If the sticking lastblock hard material in the transducer rear surface, its specific acoustic impedance is more much higher than transducer, can think that so carrying on the back the surface is clamped motionless.And the much lower mineral oil of front and acoustic impedance contacts, and promptly the front almost can free vibration, is single side radiation in this case, and at this moment resonance occurs in thickness
Satisfy
Under the situation.
The directive property of battle array is meant its transmitting response or receives the specific character of the amplitude of response with azimuthal variation.Usually it has a maximum value on certain reference direction.According to the sound field theory, the formation of the directive property of emission battle array response is because its each several part emitting sound wave is interfered the result of stack in the free field far-field region; The formation that receives the directive property of battle array response is because the reception battle array is in the far-field region of sound source to be measured, and arriving the total force that receives the lip-deep sound wave generation of battle array is that the total force that each wavelet produces is a result of sub-wave interference stack.
The directive property of battle array is its a kind of attribute in the far field.For the emission battle array, the each several part emitting sound wave can be similar to apart from the sound ray that arrives the far-field region through " equivalent infinity " regards a branch of parallel lines as.Therefore, each sound wave is interfered stack with same frequency, various amplitude, out of phase on observation point, and the amplitude of total acoustic pressure is the function of parallel sound ray at dimensional orientation.It is far-field region in emission space that the emission formation becomes the condition of directive property.For receiver, only when it is in apart from the measured signal source " equivalent infinity " when locating, shining the lip-deep sound wave of receiver just can be similar to and regard a branch of parallel sound ray as, at this moment, the total force that each bar sound ray interferes stack to produce on the receiver surface is the function of parallel sound ray at dimensional orientation, thereby its open circuit output voltage also changes with the incident angle of incident sound ray.This shows that gust condition that presents directive property that receives is the far-field region in sound source to be measured.
For the emission battle array, the non-normalized and normalized directivity function of acoustic pressure is used respectively
With
Expression.According to the definition of above-mentioned directional response, the expression formula of directivity function is respectively:
In the formula
,
Representing in the acoustic centre of source with battle array respectively is the spherical coordinates initial point, apart from the spherical coordinates initial point is
The far-field region sphere on, any direction
Go up the amplitude and the main beam maximal value direction of acoustic pressure again
Go up the amplitude of acoustic pressure again.Standard (IEC-565) regulation according to International Electrotechnical Commission: the directive property of drawing and measure battle array is used coordinate system, then
For sound ray with
The angle of between centers.
The normalized directivity function of the sound intensity is used usually
Expression, that is:
In the formula
,
Represent respectively
With
The effective value of the sound intensity on the direction.
According to the relation of sound wave at the free field far-field region acoustic pressure and the sound intensity:
In the formula
,
Be respectively the density and the velocity of sound of medium, can prove:
For receiving battle array, normalization output voltage directivity function and normalized power directivity function are used respectively
,
Expression, their expression formula is respectively:
In the formula
Representative is positioned at sound source to be measured far-field region, sound wave edge when receiving battle array
During direction incident, receive the battle array output voltage values.
Represent sound wave along receiving battle array peak response direction
During incident, receive the battle array output voltage amplitude.In like manner,
,
Represent respectively
With
Receive the amplitude of battle array output voltage square on the direction.
The design of b, backing
The purpose of backing sound absorption is in order to prevent to enter the acoustic reflection back pressure electricity array in the backing and to reduce acoustic coupling between array element.But can cause emission efficiency and receiving sensitivity to reduce greatly like this.As seen broadband and high sensitivity are a pair of contradiction, thereby need take all factors into consideration the contradiction between bandwidth and receiving sensitivity, seek to be fit to backing structure and the formation that probe requires.
At present, the compound substance that common backing all adopts tungsten powder and epoxy resin to make because this compound substance is moderate than acoustic impedance, when guaranteeing the probe bandwidth, can also satisfy certain sensitivity.And acoustic attenuation is big, can reduce the acoustic coupling between array element.Thereby this paper also adopts this compound substance, but needs its proportioning and formation of research in addition.
The density of compound substance is:
In the formula:
For the density of epoxy resin, get
For the density of tungsten powder, get
Be the percent by volume of tungsten powder in compound substance; The ratio of tungsten powder in compound substance
The velocity of sound of compound substance along with
Increase,
Below, velocity of longitudinal wave reduces gradually,
More than, just begin rapid rise.
By ultrasonic propagation The Characteristic Study and the experiment to compound substance, the reason of the acoustic attenuation of compound substance is that the scattering by wherein metallic particles is caused.
According to theoretical and empirical curve as can be known, acoustic attenuation constant
Along with frequency increases and increases, and be tending towards maximal value, have a declining tendency then; Acoustic attenuation constant
Along with percent by volume
Increase and increase earlier, be tending towards smooth then, reach maximal value, have a declining tendency at last.
Get
Be the wave beam of sound wave in epoxy resin,
Radius for metallic particles.In addition, exist
With
The time lead powder epoxy resin composite material acoustic attenuation, maximal value about
The zone, and with
Be increased at 15% o'clock from 5%, acoustic attenuation obviously increases and decreases.
According to top conclusion, get
So, select the tungsten powder particles radius to be:
The epoxy resin velocity of sound after if choosing is solidified
, if get
, then
The acoustic impedance of backing, the velocity of sound and acoustic attenuation, the proportioning by regulating ring epoxy resins and tungsten powder changes.If but tungsten/powder epoxy resin is than too high, epoxy resin is more than needed in backing manufacturing process, has promptly formed the epoxy resin layer that does not have tungsten powder.Thereby, with piezoelectric ceramics
Powder joins.In the manufacturing process to tungsten/
Composite powder vacuum pouring epoxy resin.In the backing that completes, the ratio of tungsten/piezoelectric ceramics/epoxy resin is about 9:7:1, and acoustic impedance is
, acoustic attenuation coefficient is about
The design of c, matching layer
Increase the frequency band of the acoustic impedance energy broadening transducer of backing, but also can reduce the receiving sensitivity of transducer simultaneously,, adopt matching layer for relaxing these contradiction.
The acoustic propagation characteristic of analyzing acoustic matching layer has two kinds of methods: a kind of is to set about from its dynamo-electric equivalent electrical circuit; A kind of is to set about from the propagation characteristic of sound wave acoustic matching layer.This paper adopts a kind of method decision in back acoustic matching layer structure.
If incident wave is a plane wave.For single matching layer, can get its sound intensity transmission coefficient according to ACOUSTIC WAVE EQUATION and boundary condition
In order to make the transmission coefficient maximum, the impedance of acoustic matching layer
Need to satisfy:
At this moment, in the time of can getting the acoustic impedance coupling
Relation with frequency of sound wave:
In the formula:
,
For
Fundamental frequency when maximum.
For double-deck matching layer, in the time of can getting the transmission coefficient maximum equally, the resistance value of acoustic matching layer need satisfy:
Transmission coefficient when at this moment, double-deck acoustic impedance is mated
With the pass of frequency be:
In the formula:
, promptly
For three layers of matching layer, in the time of can getting the transmission coefficient maximum equally, the resistance value of acoustic matching layer need satisfy:
Transmission coefficient when at this moment, three layers of acoustic impedance are mated
With the pass of frequency be:
In the formula:
,
,
The acoustic impedance of piezoelectric crystal
, centre frequency
The acoustic impedance of transformer oil
In frequency
The wavelength of following acoustic matching layer is
But owing to actual conditions are, the acoustic matching layer material has decay, and the acoustic impedance material of optimum matching is difficult to obtain, and number of plies increase, and technology difficulty is bigger.The centre frequency 120kHz of system requirements adopts two-layer acoustic matching layer can reach the range of signal that bandwidth covers 80～180kHz.The thickness of each layer matching layer is
The supersonic array sensor that research is made as shown in Figure 3.
(2) based on the partial discharge of transformer source number identification of wideband supersonic array signal processing algorithm.
Adopt TCT (Two-sided Correlation Transformation) algorithm, improvement Gai Shi circle algorithm in the wideband array signal space spectrum estimation theory to realize focusing processing and the number identification that office relaxes the band ultrasonic signal.Specific algorithm is as follows:
The broadband focusing algorithm
The ultrasonic frequency that shelf depreciation produces mainly concentrates on 50kHz to 300kHz frequency range, and its crest frequency between 70kHz to 200kHz, is a broadband signal mainly.But Gai Shi circle algorithm itself can only carry out the number identification in narrow band signal source, the rigid ultrasonic signal that shelf depreciation is produced is lost as the spatial information (si) feature of enriching that the narrow band signal processing can make original signal have, and causes the number identification precision not enough even can not realize.Therefore need at first to convert the signal data that receives to the narrow band signal data, use the narrow band signal Processing Algorithm again and carry out the number identification.
The broadband focusing algorithm mainly is divided into two big classes, and the first kind is based on irrelevant Signal Processing method (ISM), and second class is based on the disposal route (CSM) of coherent signal.This class basic idea is that the data on nonoverlapping Frequency point in the frequency band are focused on the reference frequency point, obtains single-frequency point data covariance after the focusing, uses the narrow band signal Processing Algorithm again and carries out the DOA estimation.CSM class algorithm is compared with ISM class algorithm has the high characteristics of estimated accuracy.Typical algorithm mainly contains: bilateral associated change method (TCT), gyrator space law (RSS), signal subspace converter technique (SST) etc.
The essence of above-mentioned algorithm all is to seek a focussing matrix
, the TCT algorithm is compared the focusing error minimum with other algorithm on the performance, best performance, and operand is moderate, and the data acquisition storage speed is suitable with the late time data processing speed, can satisfy the needs of handling real time data, therefore chooses TCT broadband focusing algorithm.
The focusing of wideband array signal is handled and is meant on the basis that guarantees the ultrasound wave broadband signal information that shelf depreciation produces, the searching focussing matrix
, with the direction matrix under the different frequency
Transform on the optimum reference frequency f 0, and then handle in conjunction with the arrowband algorithm.
In the ideal case, f 0 is arranged:
So the focusing error of all data is:
Obviously,
Satisfy formula:
Can make the focusing error minimum shown in the following formula.
In fact, because the focusing transform matrix is the normalization constraint, desirable focusing performance is can not be getable, so consider following replacement scheme, promptly chooses reference frequency
Make formula:
The focusing error minimum.
When fixing
The time, can obtain focussing matrix
, at this moment focusing error can be written as:
In the formula
With
Expression respectively
With
Singular value since following formula in
With the focusing frequency-independent, focusing error can be rewritten as:
Make the condition of following formula minimum be under the situation of constant ignoring:
Consider non-ideality, the criterion of focusing error minimum can be revised as:
In the formula,
Following formula is the one dimension optimization problem, is easy to obtain optimum reference frequency with the method for searching for
Improve Gai Shi circle algorithm
Tradition Gai Shi circle algorithm:
Suppose uniform array by
Individual equally spaced primitive is formed, and has
Individual far field narrow band signal from
Incident, then the signal phasor of array received can pass through following formulate:
When noise is under the situation of space white Gaussian noise, and covariance matrix can be write as again
Wherein,
Be the signal source covariance matrix,
Be the energy of white Gaussian noise,
Be
Unit matrix.In actual applications,
Be on average to approach by the signal covariance matrix that each time snap obtains.
The covariance matrix of received signal
Remove last column and last row formation matrix
,
Proper vector constitute unitary matrix
, generate unitary matrix
, do unitary transformation with the unitary matrix covariance matrix to received signal that generates:
Wherein
,
Be array guiding matrix
Remove the later matrix of last column, so
,
If
In
Corresponding is noise subspace, so it with
Be quadrature,
Be zero.Otherwise, if
In
Corresponding is signal subspace, so it with
Be actually in a space,
It is a relatively large number.
Use formula in actual applications
Differentiate information source number, wherein
Be on average to approach by the signal covariance matrix that each time snap obtains
Fast umber of beats.
Be
Decreasing function, need to set as the case may be.
If work as
The time,
For first than zero little value, then the estimated value of information source number is got
The essence of improving Gai Shi circle algorithm is that the covariance matrix to array data has carried out feature decomposition completely, therefore compares with traditional Gai Shi circle algorithm to have higher identification precision and identification capacity.Export (being the signal after focusing in this method) for array:
When noise was the space white Gaussian noise, the covariance matrix of received signal was:
If
Each column vector be
, then
Wherein,
Each row vector for A (Θ).
Covariance matrix to the array received signal
Carry out svd and suppose the singular value descending sort:
Wherein,
It is the resulting characteristic vector space of svd.
Get
Wherein,
,
Can be 1 to arrive
In any one the number.Obviously,
With the information source criterion in the Gai Shi circle criterion
Structure is identical.Identical with the ultimate principle of Gai Shi circle criterion because characteristics of noise subspace and steering vector quadrature all, and the proper subspace of signal in fact with steering vector in same space, so
In corresponding to noise
Be zero, and corresponding to signal
It is a relatively large number.
From
Expression formula as can be seen, this algorithm identification precision is by eigenvector
And array element
With array element
Between related function decide.In order to improve identification precision, vector
Can be the related function of any two array elements, promptly
Any one row.So just can obtain
Individual vector
,
Respectively to each
With being similar to Gai Shi circle criterion information source number is done identification:
Wherein
It is fast umber of beats.
Be
Decreasing function, need to set as the case may be.K gets 1,2 successively ..., M is if work as
The time,
For first than zero little value, then the identifier of information source number is got
Will
Inferior estimated result is weighted on average, and resulting result is the information source number that finally picks out.
The array element of 16 array element supersonic array sensors is 4 row, 4 row square arrangement, array element distance 8mm, array element received signal centre frequency 120kHz.
The supersonic array number identification system in partial discharge of transformer source, by supersonic array sensor, equidistant low cable, signal amplifier (voltage amplifier, filter amplifier), the multi-channel high-speed data synchronous acquisition device formation of decreasing, each unit installation connects as follows:
The installation of supersonic array sensor
16 array element supersonic array sensor design centre frequencies are 120kHz, array element distance 8mm.
In transformer inside, because insulating material character, design and the problem of making and the factor of impurity bubble, easily cause non-uniform electric, some place exists electric field too concentrated, this just very easily produces shelf depreciation, will select according to transformer built-in electrical insulation thin spot in the time of preliminary placement sensor.Near the place of transformer built-in electrical insulation thin spot the supersonic array sensor is installed at the transformer outer wall, is coupled with good quality butter between supersonic array sensor and the transformer case.
Connect detection system
Be output as quantity of electric charge weak output signal because the piezoelectric crystal that uses surpasses, must improve signal to noise ratio (S/N ratio) through amplifying the influence that could reduce undesired signal.And the carrying load ability of piezoelectric crystal is poor, requires pre-amplification circuit that bigger defeated people's impedance is arranged, and noise is low.Therefore 16 array elements connect 16 electric charges (voltage) amplifier respectively by cable, connect filter amplifier again, receive 16 passage high-speed data synchronous acquisition devices then, connect computing machine then.
Data analysis and processing
What we can obtain in the experiment is 16 data files that 16 channel data synchronous acquisition devices are preserved.Construct thus three-dimensional data matrix x (m, n, t), m wherein, n, t are respectively the capable ordinal number of supersonic array sensor, the row ordinal number, the sampling time counts.Whole broadband signal is divided into the K section by the time that receives, and the data that respectively each array element received are carried out quick FFT conversion then, and each array element is carried out K time.(choosing of segments K value: for guaranteeing K segment data length unanimity, K at first will guarantee to be sampled the data sum and divide exactly; The K value is big more, and promptly the burnt performance of poly is good more more for segments, and operand also is doubled and redoubled simultaneously.Finally choose K=4 based on these 2.Both guarantee focusing performance, also considered data processing speed.）
Focus on and handle carry out TCT through the data output matrix after the FFT conversion, be divided into and be K time, each time all to focusing center's frequency.(f 0 chooses optimum reference frequency f 0 according to the minimum accurate side calculating of energy loss error; According to formula
), obtain focussing matrix
The line focus matrix is handled, and broadband signal converts the treatable narrow band signal of down-stream to.According to improving Gai Shi circle algorithm principle, data matrix covariance is decomposed, and proper vector and each row of signal covariance matrix are multiplied each other obtain 16 sequences
Because characteristics of noise subspace and steering vector be quadrature all, and the proper subspace of signal in fact with steering vector in same space, so each
In corresponding to noise
Be zero, and corresponding to signal
It is a relatively large number.Choose a judgement benchmark
, then greater than
Number is each
The source number is put in the office that picks out, with 16
The number of identification is weighted the office that on average is final identification and puts the source number.Experiment obtains covariance matrix by the array data that receives
And proper vector
, make number after proper vector multiplied each other with each row of covariance matrix respectively and estimate to obtain the number identification result 16 times, at last the number identification result is exported with form.
Front end multiplying arrangement (charge/voltage amplifier) enlargement factor is 10-100 times, and back level multiplying arrangement (filter amplifier) enlargement factor is 1-200 times, regulates as required during test.During the inner generation of transformer shelf depreciation, its acoustic emission spectrum is than broad, and energy mostly concentrates on the 40KHz-200KHz frequency range, so the high-speed data acquiring device frequency band is selected 40KHz-250KHz, the setting sample frequency is 10M, and sampling length is 512K.Sending the data of coming to from capture card is binary number, through being converted into floating number, analyzes respectively on time domain and frequency domain.
Analysis on the time domain mainly is to obtain the time of arrival of shelf depreciation ultrasonic signal and number of times, by judging and the peak value of signal calculated obtains.Under laboratory condition, the noise level that collects signal is the millivolt level, the defeated artificial volt level of shelf depreciation ultrasonic signal, amplitude during discharge is apparently higher than noise signal, a discharge period is no more than 10 μ s, because sampling rate is 10MHz, institute thinks 100 points, therefore can obtain shelf depreciation time and number of times by calculating by the amplitude and the sequence number of these 100 points.
If on frequency domain, analyze the power spectrum of data owner analytic signal.The final TCT algorithm of using converts the shelf depreciation broadband signal to narrow band signal, and application enhancements Gai Shi circle algorithm picks out the number of signal source again.Source number identification algorithm flow process such as Fig. 2 are put in office.
For verifying correctness of the present invention, experimental study has been carried out in the chamber.The piezoelectricity transmitting probe that the source is put in the local office of simulation is a high-voltage pulse generator, the high-voltage pulse of output amplitude 4.5kV, pulsewidth 3 μ s, give off ultrasound wave in order to drive the piezoelectricity transmitting probe, can produce that amplitude is big, the ultrasonic signal of stable performance, and put it into and be full of mineral oil, be of a size of in the fuel tank of 100*100*120 cm3, with shelf depreciation situation in the simulated oil.
Sensor array places the fuel tank outer wall, measures with identification office at the A place and puts the source number.Filtering amplification and data collector adopt program control amplifilter of USB-13120pt01 type 16 passages and USB-11140pt10 type 16 port number device for picking respectively, can realize parallel data acquisition, sample frequency is 10MHz, the filtering cutoff frequency is 40KHz to 250KHz, and signal amplification factor is 16 times.
Office is set, and to put the source number be 1, and record data trigger with electric signal, and fast umber of beats is 6400.The received filtered waveform of supersonic array sensor as shown in Figure 6.At first each passage time-domain signal is transformed to frequency domain, write then and focus on to have handled and signal source number identification algorithm program, it is as shown in table 1 to obtain putting source number identification result single innings, and its number identification accuracy is 87.5%.
The two offices of same procedure acquisition put source number identification accuracy and are 81.3%, and are as shown in table 2.
Claims (4)
1. transformer partial discharge source number discrimination method, it is characterized in that, the ultrasonic broadband signal that it utilizes the supersonic array sensor office of acceptance source of putting to send, and formation Array Model, the TCT algorithm of using then in the wideband Array Signal Processing algorithm converts the broadband signal that the supersonic array sensor is received to narrow band signal, and application enhancements Gai Shi circle method is carried out identification to partial discharge of transformer source number again.
2. according to the described transformer partial discharge of claim 1 source number discrimination method, it is characterized in that it carries out according to the following steps:
A, the supersonic array sensor is close to the ultrasonic broadband signal that send in the transformer surface office of acceptance source of putting;
B, employing TCT broadband focusing algorithm are with the direction matrix under the different frequency
Transform to optimum reference frequency f
0: f
0Definite method as follows:
Criterion according to the focusing error minimum:
Adopt the method for search to obtain optimum reference frequency
In the formula,
,
With
Expression respectively
With
Singular value,
With
The covariance matrix of representing the frequency of central reference frequency after the denoising of array output data and j segment data respectively;
C, employing improve Gai Shi circle algorithm the narrow band signal after focusing on are carried out the number identification, determine the number of Partial Discharge Sources;
Suppose uniform array by
Individual equally spaced primitive is formed, and has
The discharge signal incident of individual office, can pass through following formulate after the ultrasonic broadband signal line focus of array received:
When noise was the space white Gaussian noise, the covariance matrix of this signal was:
Wherein,
Be the signal source covariance matrix,
Be the energy of white Gaussian noise,
Be
Unit matrix, in actual applications,
Be on average to approach by the signal covariance matrix that each time snap obtains,
If
Each column vector be
,
, then office puts the number identification and is undertaken by following step:
1. right
Carry out svd and suppose the singular value descending sort:
Wherein,
It is the resulting characteristic vector space of svd;
2. get
,
, and respectively to each
Utilize following method that information source number is carried out identification:
K gets 1,2 successively ..., M, the value of calculating following formula:
Wherein
Be on average to approach by the signal covariance matrix that each time snap obtains
Fast umber of beats,
Be
Decreasing function, set as the case may be, if work as
The time,
For first than zero little value, then the identifier of information source number is got
3. will
Inferior estimated result is weighted on average, and resulting result is the information source number that finally picks out.
3. transformer partial discharge source number identification system, it is characterized in that, described system is made of supersonic array sensor, multi-channel high-speed data synchronous acquisition device (13) and notebook computer (14), described supersonic array sensor is close to transformer (11) shell, its output signal connects the input end of multi-channel high-speed data synchronous acquisition device (13) through signal amplifier (12), and the signal output part of described multi-channel high-speed data synchronous acquisition device (13) connects the input interface of notebook computer (14).
4. according to the described transformer partial discharge of claim 3 source number identification system, it is characterized in that described supersonic array sensor is the planar structure of 16 array elements, 4 row * 4 row, single array element is 7.8mm for the horizontal section diameter, the right cylinder of high 18mm, array element center distance 8mm, material is a piezoelectric ceramics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100501285A CN102183710A (en) | 2011-03-02 | 2011-03-02 | Method and system for identifying numbers of partial discharge sources of transformers |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011100501285A CN102183710A (en) | 2011-03-02 | 2011-03-02 | Method and system for identifying numbers of partial discharge sources of transformers |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102183710A true CN102183710A (en) | 2011-09-14 |
Family
ID=44569913
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011100501285A Pending CN102183710A (en) | 2011-03-02 | 2011-03-02 | Method and system for identifying numbers of partial discharge sources of transformers |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102183710A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102662132A (en) * | 2012-05-30 | 2012-09-12 | 华北电力大学（保定） | Method and system for locating partial discharge of electrical equipment |
| CN103529364A (en) * | 2013-09-20 | 2014-01-22 | 华北电力大学（保定） | In-oil multi-local discharge source positioning method based on ultrasonic thinned array sensor |
| CN104251961A (en) * | 2013-06-27 | 2014-12-31 | 国家电网公司 | Partial discharging signal source positioning device and system of gas-insulation completely-enclosed combined electrical appliance |
| CN104407278A (en) * | 2014-11-07 | 2015-03-11 | 国家电网公司 | Extraction method and device of ultrasonic wave characteristic parameters |
| CN104614653A (en) * | 2015-02-09 | 2015-05-13 | 国家电网公司 | Array antenna based multiple local discharge point positioning and distinguishing method for local discharge detection device |
| CN105334480A (en) * | 2014-08-11 | 2016-02-17 | 国家电网公司 | Spatial position determining method and device for sensor array with four array elements |
| CN106556783A (en) * | 2016-12-05 | 2017-04-05 | 西安交通大学 | A kind of shelf depreciation direction-finding method in transformer station based on superfrequency phased array principle |
| CN107315886A (en) * | 2017-07-06 | 2017-11-03 | 国网重庆市电力公司电力科学研究院 | A kind of method and apparatus of transformer room's exterior three dimensional spatial noise prediction |
| CN108333480A (en) * | 2018-01-04 | 2018-07-27 | 国家电网公司华中分部 | A kind of localization method of substation's shelf depreciation positioning system |
| CN111203374A (en) * | 2018-11-21 | 2020-05-29 | 美国西门子医疗系统股份有限公司 | Composite acoustic absorber for ultrasound transducer array |
| CN112466627A (en) * | 2020-11-11 | 2021-03-09 | 西安西电变压器有限责任公司 | Design method and system of transformer oil tank and transformer oil tank |
| CN113030983A (en) * | 2021-03-17 | 2021-06-25 | 中国科学院声学研究所 | Near-field point-by-point focusing DOA method based on depth sounding side-scan sonar |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3728619A (en) * | 1971-06-21 | 1973-04-17 | Westinghouse Electric Corp | Acoustical corona locator having rotatable and pivotable transducers |
| CN101702000A (en) * | 2009-11-26 | 2010-05-05 | 华北电力大学（保定） | Phased ultrasound locating method and system of partial discharge of transformer |
-
2011
- 2011-03-02 CN CN2011100501285A patent/CN102183710A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3728619A (en) * | 1971-06-21 | 1973-04-17 | Westinghouse Electric Corp | Acoustical corona locator having rotatable and pivotable transducers |
| CN101702000A (en) * | 2009-11-26 | 2010-05-05 | 华北电力大学（保定） | Phased ultrasound locating method and system of partial discharge of transformer |
Non-Patent Citations (1)
| Title |
|---|
| 谢庆: "基于空间谱估计的变压器局放超声阵列定位方法研究", 《华北电力大学博士学位论文》, 15 September 2010 (2010-09-15) * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102662132A (en) * | 2012-05-30 | 2012-09-12 | 华北电力大学（保定） | Method and system for locating partial discharge of electrical equipment |
| CN104251961A (en) * | 2013-06-27 | 2014-12-31 | 国家电网公司 | Partial discharging signal source positioning device and system of gas-insulation completely-enclosed combined electrical appliance |
| CN103529364B (en) * | 2013-09-20 | 2016-05-25 | 华北电力大学（保定） | Many Partial Discharge Sources localization method in oil based on ultrasonic thinned array sensor |
| CN103529364A (en) * | 2013-09-20 | 2014-01-22 | 华北电力大学（保定） | In-oil multi-local discharge source positioning method based on ultrasonic thinned array sensor |
| CN105334480B (en) * | 2014-08-11 | 2018-09-18 | 国家电网公司 | The determination method and apparatus of the sensor array spatial position of four array elements |
| CN105334480A (en) * | 2014-08-11 | 2016-02-17 | 国家电网公司 | Spatial position determining method and device for sensor array with four array elements |
| CN104407278A (en) * | 2014-11-07 | 2015-03-11 | 国家电网公司 | Extraction method and device of ultrasonic wave characteristic parameters |
| CN104614653A (en) * | 2015-02-09 | 2015-05-13 | 国家电网公司 | Array antenna based multiple local discharge point positioning and distinguishing method for local discharge detection device |
| CN106556783A (en) * | 2016-12-05 | 2017-04-05 | 西安交通大学 | A kind of shelf depreciation direction-finding method in transformer station based on superfrequency phased array principle |
| CN106556783B (en) * | 2016-12-05 | 2019-07-19 | 西安交通大学 | Shelf depreciation direction-finding method based on superfrequency phased array principle in a kind of substation |
| CN107315886A (en) * | 2017-07-06 | 2017-11-03 | 国网重庆市电力公司电力科学研究院 | A kind of method and apparatus of transformer room's exterior three dimensional spatial noise prediction |
| CN108333480A (en) * | 2018-01-04 | 2018-07-27 | 国家电网公司华中分部 | A kind of localization method of substation's shelf depreciation positioning system |
| CN111203374A (en) * | 2018-11-21 | 2020-05-29 | 美国西门子医疗系统股份有限公司 | Composite acoustic absorber for ultrasound transducer array |
| CN112466627A (en) * | 2020-11-11 | 2021-03-09 | 西安西电变压器有限责任公司 | Design method and system of transformer oil tank and transformer oil tank |
| CN113030983A (en) * | 2021-03-17 | 2021-06-25 | 中国科学院声学研究所 | Near-field point-by-point focusing DOA method based on depth sounding side-scan sonar |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102183710A (en) | Method and system for identifying numbers of partial discharge sources of transformers | |
| CN101702000A (en) | Phased ultrasound locating method and system of partial discharge of transformer | |
| CN102662132B (en) | A kind of local discharge of electrical equipment localization method and system | |
| CN103529364B (en) | Many Partial Discharge Sources localization method in oil based on ultrasonic thinned array sensor | |
| CN103278810B (en) | Method for extracting dimension characteristics of underwater target based on space similarity | |
| CN110926386B (en) | Transformer winding ultrasonic detection imaging method based on improved wavelet threshold denoising | |
| CN105066918A (en) | Ultrasonic underwater target thickness measuring system and thickness measuring method | |
| CN111024209B (en) | Line spectrum detection method suitable for vector hydrophone | |
| CN106404911B (en) | True time delay single mode Lamb wave phased array system for plate structure detection | |
| CN104502904B (en) | Torpedo homing beam sharpening method | |
| CN110058245A (en) | Low frequency active dragging line battle array sonar shallow sea based on cloud model detects efficiency estimation method | |
| CN106526563A (en) | Quintuple volume array multi-target orientation estimation method based on cross-correlation virtual array | |
| CN104280718A (en) | Transformer station abnormal sound monitoring and positioning method | |
| CN101855547A (en) | Detection of ingress of water in an intermediate layer using acoustic resonance technology | |
| CN109407098A (en) | A kind of implementation method for the parallel processing algorithm handling orthogonal pulses string signal | |
| CN108896974A (en) | A kind of improved MIMO array high-resolution Estimation of Spatial Spectrum method | |
| CN109061654A (en) | Single ring array active 3-D positioning method under a kind of deep-marine-environment | |
| CN103582912A (en) | Transducer for phased array acoustic systems | |
| CN105676225B (en) | Ranging system and method in opaque troubled liquor | |
| CN109238440A (en) | The acoustic stimulation insertion loss measurement method of focusing technology when based on optimal sky | |
| CN114201991A (en) | Partial discharge signal detection method and system based on ultrasonic sensor array | |
| CN112883628A (en) | Method and system for positioning abnormal sound source of transformer substation equipment | |
| CN117148273B (en) | Underwater sound source positioning method for matching probability measure | |
| Fang et al. | The echolocation transmission beam of free-ranging Indo-Pacific humpback dolphins (Sousa chinensis) | |
| Zhou et al. | Partial discharge sources detection and location with an electromagnetic vector sensor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20110914 |