CN115468645A - Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization - Google Patents
Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization Download PDFInfo
- Publication number
- CN115468645A CN115468645A CN202211035522.6A CN202211035522A CN115468645A CN 115468645 A CN115468645 A CN 115468645A CN 202211035522 A CN202211035522 A CN 202211035522A CN 115468645 A CN115468645 A CN 115468645A
- Authority
- CN
- China
- Prior art keywords
- color
- changer
- load tap
- characteristic
- vibration signal
- 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
- 238000000034 method Methods 0.000 title claims abstract description 51
- 238000001228 spectrum Methods 0.000 title claims abstract description 34
- 238000005457 optimization Methods 0.000 title claims abstract description 28
- 238000003745 diagnosis Methods 0.000 title claims abstract description 22
- 230000011218 segmentation Effects 0.000 title claims abstract description 18
- 239000011159 matrix material Substances 0.000 claims abstract description 55
- 239000013598 vector Substances 0.000 claims abstract description 48
- 230000009467 reduction Effects 0.000 claims abstract description 18
- 238000011426 transformation method Methods 0.000 claims abstract description 8
- 238000013528 artificial neural network Methods 0.000 claims abstract description 3
- 230000009471 action Effects 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 7
- 230000001133 acceleration Effects 0.000 claims description 5
- 238000000354 decomposition reaction Methods 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 230000000295 complement effect Effects 0.000 claims description 2
- 230000000694 effects Effects 0.000 claims description 2
- 238000000605 extraction Methods 0.000 abstract description 5
- 230000006870 function Effects 0.000 description 11
- 230000008569 process Effects 0.000 description 9
- 238000011176 pooling Methods 0.000 description 8
- 238000012549 training Methods 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 3
- 230000009466 transformation Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 108010076504 Protein Sorting Signals Proteins 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000033001 locomotion Effects 0.000 description 1
- 230000005389 magnetism Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000010248 power generation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
- G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)
Abstract
The application relates to an intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization, which comprises the following steps: A. collecting vibration signals of the on-load tap-changer and carrying out denoising treatment; B. extracting energy and kurtosis characteristics of each segmented signal; C. extracting color moments to describe color features of the time-frequency image; D. providing a total correlation arrangement optimization method for optimizing the characteristic parameters, and E, providing a matrix principal component transformation method for further reducing the dimension of the optimized color characteristics, and keeping the similarity of the characteristic components of the original data and simultaneously minimizing the deformation caused by dimension reduction; F. and (4) forming an on-load tap-changer state characteristic vector by using the energy characteristic and kurtosis characteristic of the segmented vibration signal and the time-frequency map color characteristic after dimension reduction, and completing fault identification through a one-dimensional convolution neural network. The vibration signal characteristic extraction method can effectively capture and depict the running state characteristics of the on-load tap-changer equipment.
Description
Technical Field
The application relates to the field of fault diagnosis of electrical equipment, in particular to an intelligent fault diagnosis method for an on-load tap-changer based on vibration signal segmentation time-frequency spectrum optimization.
Background
Since the 20 th century and the 50 th century, china began to independently develop a transformer on-load voltage regulation technology, over 70 years ago, mechanical on-load tap changers (OLTCs) became mature gradually and are assembled to important transformation nodes to realize power grid on-load voltage regulation. However, in the construction of a novel power system for rapidly developing new energy power generation at present, the on-load tap changer of the on-load tap changer has the defects of wide voltage regulation range, high running load and frequent switching operation, so that the running risk of equipment is increased. In addition, the distributed power supply grid connection has the characteristics of impact and intermittence, the voltage needs to be frequently regulated after the grid connection so as to ensure the active and reactive power distribution and the power supply quality of the system, and a severe challenge is provided for the operation reliability of the on-load tap-changer of the transformer.
The on-load tap-changer has the advantages of enriching equipment state information and non-invasive monitoring along with signal changes of current, temperature rise, sound waves, vibration and the like in operation and operation, and a method for identifying equipment state abnormity and faults by taking the characteristics of the monitored vibration signals as judgment basis has obtained certain results in the current research. All mechanical motions of the on-load tap-changer parts are accompanied with vibration, vibration signal data are processed by a scientific method, and the running state characteristics of the equipment are extracted, so that the running state change of the equipment is sensitively sensed. Most of the existing on-load tap-changer fault diagnosis methods based on vibration signals focus on the overall time-frequency characteristics of the signals, only wavelet, EMD, EEMD, hilbert transformation and other methods are used for calculating certain frequency domain characteristics, equipment action time sequence information contained in local characteristics is ignored, insufficient consideration is given to accumulated energy of the vibration signals, and the correlation with the action mechanism of the internal structure of the on-load tap-changer is not tight. Meanwhile, the characteristic parameters obtained by signal processing often contain redundant information, and a large number of non-screened useless and repeated characteristics cause the problems of low fault diagnosis accuracy, low identification efficiency and the like.
Disclosure of Invention
The method is characterized in that the sectional time-frequency spectrum characterization equipment running state characteristics are extracted based on the corresponding relation between the vibration signals and the action of the change-over switch, and fault diagnosis is carried out by means of the self-learning capability of deep learning after the optimal characteristics are arranged, wherein the method is mainly used for fault diagnosis of the on-load tap-changer based on vibration signal multi-section characteristics and frequency spectrum texture extraction.
In order to achieve the above purpose, the present application provides the following technical solutions:
the embodiment of the application provides an on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization, which is characterized by comprising the following steps,
A. an acceleration sensor is adopted to collect vibration signals of the on-load tap-changer, and denoising processing is carried out by utilizing local mean decomposition;
B. taking an absolute value of the denoised vibration signal, performing accumulated integration on the absolute value as a time axis to obtain a plurality of sections of energy changes represented by an accumulated integration curve, segmenting the vibration signal according to an inflection point of the curve amplitude to obtain different action stages corresponding to a switch, and extracting energy and kurtosis characteristics of each section of signal;
C. carrying out short-time Fourier transform on the vibration signals to obtain each segmented time-frequency map, and extracting color moments to describe color features of the time-frequency image;
D. a total correlation arrangement optimization method is provided, and feature parameters are optimized to obtain a low-redundancy optimized color feature vector reflecting the operation state of the tap changer;
E. providing a matrix principal component transformation method, performing further dimension reduction on the preferred color characteristics, and keeping the similarity of the characteristic components of the original data and simultaneously minimizing the deformation caused by the dimension reduction;
F. and (4) forming an on-load tap-changer state characteristic vector by using the energy characteristic and kurtosis characteristic of the segmented vibration signal and the time-frequency map color characteristic after dimension reduction, and completing fault identification through a one-dimensional convolution neural network.
In the step A, the acceleration sensor is adsorbed on a shell of the on-load tap-changer, and a stator current signal of a driving motor is used as a capture card trigger source; after the signal is decomposed by the local mean value, each decomposed component is screened according to the correlation coefficient, and reconstruction is carried out to achieve the denoising effect.
In the step B, a method for segmenting the vibration signal by using an accumulated integral curve is provided, and accumulated integral is performed on the denoised vibration signal along a time axis after an absolute value is taken, so as to obtain an accumulated integral curve s (t), and the calculation method is as follows:
wherein x (t) is a vibration signal after denoising, and t is time;
the vibration signal is divided into three sections by capturing two important inflection points of the curve, the three sections respectively correspond to a single resistance section, a bridge section and a double resistance section of the action of a change-over switch in the on-load tap-changer, and each section is respectively embodied as an impact cluster in the vibration signal.
And in the step C, constructing a vibration segmentation signal time-frequency map by using short-time Fourier transform, extracting a first moment, a second moment and a third moment of colors to describe the color distribution of the image, and obtaining a P-dimensional color characteristic vector.
In the step D, a total correlation permutation optimization method is provided for feature parameter optimization, Q signal samples under the same working condition are taken, color feature vectors are calculated, and a color feature sample matrix Y is constructed Q×P (ii) a Will Y Q×P The P column vectors are recorded as color feature sample vectors, and the total correlation coefficient μ of each color feature sample vector is calculated by the following calculation method:
v and w are column numbers where the two color sample vectors are located; m is a color characteristic sample matrix row number; y is mv ,y mw Is the vector value of the mth row, the vth column and the w column,
is the mean of the corresponding vectors;
defining theta as a total correlation index, screening color features with the mu values smaller than theta, and listing the color features into a preferred feature set of the working condition; the preferred characteristic sets of all the working conditions are merged to obtain the color of the on-load tap-changerThe color characteristic is optimized to be a total set, the number of characteristic parameters in the total set is U, and an optimized matrix E is obtained by reconstructing a color characteristic sample matrix by utilizing the U characteristic parameters Q×U 。
In the step E, a matrix principal component transformation method is provided, dimension reduction is further carried out on the preferred color characteristics, the similarity of the characteristic components of the original data is kept, meanwhile, the deformation caused by dimension reduction is minimized, and a preferred matrix E is calculated Q×U Distance matrix D of middle row vector, wherein elements in D are D rs ：
d rs ＝|x r -x s |
Wherein x r ,x s Is E Q×U R, s =1, \ 8230;, Q;
constructing a matrix G with elements G rs ，g rs The calculation method comprises the following steps:
g rs ＝d rs -d r1 -d s1 +d 11
selecting the first Z maximum eigenvalues and eigenvectors of G, and calculating an optimal matrix E Q×U Principal component matrix H of Q×Z ；
Wherein I is a matrix formed by the first Z orthonormal eigenvectors reserved by the matrix G; alpha is a diagonal matrix of Z largest eigenvalues.
And in the step F, the energy characteristic and the kurtosis characteristic of the segmented vibration signal and the time-frequency map color characteristic after dimension reduction form an on-load tap-changer state characteristic vector, the on-load tap-changer state characteristic vector and the kurtosis characteristic are complementary to each other to accurately describe the equipment operation state characteristic, and the fault identification is completed through a one-dimensional convolutional neural network.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides a discrete signal segmentation method, which reflects the change trend of signal amplitude by using an integral curve of a discrete signal absolute value to realize the segmentation of a vibration impact cluster. And solving a power spectrum of the vibration signal of the on-load tap-changer by utilizing short-time Fourier transform, and calculating the segmented energy and kurtosis characteristics so as to extract local characteristics of each stage of the switch action.
The invention provides a total correlation arrangement optimization method, which is used for screening out redundant characteristic parameters by representing the repeatability of equipment state information carried by the characteristic parameters according to the sum of correlation coefficients between characteristic sample vectors and characteristics of the same type. And taking the union of the preferable feature sets under all working conditions of the equipment as the color feature of the on-load tap-changer.
The invention provides a matrix principal component transformation method, which is used for extracting principal components of a color feature sample matrix based on a distance matrix, and minimizing deformation caused by dimension reduction while keeping data feature component similarity.
Drawings
To more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
Fig. 1 is a flow diagram of on-load tap-changer fault diagnosis;
FIG. 2 is a vibration signal after LMD denoising;
FIG. 3 is a waveform of an amplitude integral of a vibration signal;
FIG. 4 is a power spectrum of a segmented vibration signal, FIG. 4 (a) is a vibration signal segment 1 power spectrum, FIG. 4 (b) is a vibration signal segment 2 power spectrum, and FIG. 4 (c) is a vibration signal segment 3 power spectrum;
FIG. 5 is a segmented vibration signal STFT time-frequency spectrum, FIG. 5 (a) is a vibration signal segment 1 time-frequency spectrum, FIG. 5 (b) is a vibration signal segment 2 time-frequency spectrum, and FIG. 5 (c) is a vibration signal segment 3 time-frequency spectrum;
FIG. 6 is a flow chart of a total correlation permutation optimization method;
FIG. 7 is the 1D-CNN diagnostic result.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
The terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising a," "8230," "8230," or "comprising" does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
Referring to fig. 1, an embodiment of the present invention includes the steps of:
A. the method comprises the steps of collecting vibration signals of the on-load tap-changer in different running states by using an acceleration sensor, and carrying out local mean value decomposition (LMD) denoising on the vibration signals by considering on-site environment noise interference and complexity of an internal mechanism of the on-load tap-changer. Firstly, LMD decomposition is carried out on signals to obtain a group of PF components of different frequency bands, the degree of association between decomposed signals and original signals is measured according to the correlation coefficient of each component, the first three-order component and the 4-6-order component with higher correlation coefficient in each state are selected to be reconstructed to obtain denoised signals, and the signal-to-noise ratio of the vibration signals is improved.
B. The method comprises the steps of performing accumulated integration on an absolute value of a vibration signal as a time axis to obtain an accumulated integration curve, dividing the vibration signal into three sections by capturing two important corners of the curve, respectively corresponding to a single-resistor section, a bridge section and a double-resistor section of a change-over switch in the on-load tap-changer, solving the energy characteristics of each section of the vibration signal after the section, and then calculating a power spectrum of the vibration signal to extract kurtosis characteristics.
C. And further carrying out short-time Fourier transform (STFT) on the segmented vibration signals to obtain a time-frequency map, wherein the time-domain vibration waveform corresponds to the time-frequency map on a time scale, the corresponding relation between different time and frequency energy intensities of the original signals is reflected by colors, and abundant color features in the images are extracted by utilizing color moments.
D. In order to balance the specific gravity of the characteristic quantity and improve the fault diagnosis efficiency and accuracy, the invention firstly provides a total correlation arrangement optimization method to optimize the color characteristic parameters of the STFT time-frequency spectrum, obtains a color characteristic optimization set based on vibration signal samples under different working conditions, and screens out redundant characteristic parameters while keeping the color characteristic fault sensitivity.
E. For the preferred characteristics obtained by the total correlation permutation preferred method, the invention provides a matrix principal component transformation method for further dimension reduction, which aims to fit a high-dimensional matrix to a low-dimensional matrix, and minimize the deformation caused by dimension reduction while keeping the characteristic component similarity of the original matrix.
F. And forming an on-load tap-changer state characteristic vector by using the energy characteristic and the kurtosis characteristic of the segmented vibration signals and the time-frequency map color characteristic after dimension reduction, complementarily and accurately describing the vibration signal characteristics of different equipment in the running state, and completing fault identification through a one-dimensional convolutional neural network (1D-CNN).
In the step A, a SYXZ-10/100-9 type on-load tap-changer is taken as an example to collect vibration signals, and a piezoelectric vibration sensor (YD-37) is adsorbed on a top cover of the switch through strong magnetism. The fault simulation platform is used for simulating normal switching of the on-load tap-changer and common fault states such as looseness of a base, jamming of a rotating shaft, refusing of the switch, fatigue of a spring and the like. Under each working condition, signals are collected by 40 groups, 200 groups of sample data are totally collected, each group of data comprises 25500 sampling points, and training samples and test samples are randomly extracted according to the proportion of 1.
LMD is as a kind of self-adaptation time frequency analysis algorithm, it carries on the product operation to envelope signal and pure frequency modulation signal according to the signal extreme point and gets a PF component with the instantaneous frequency of the actual physical meaning, decomposes all PF components through the loop iteration, thus gets the time frequency distribution of the original signal, the concrete process is as follows:
A1. calculating all local extreme points n of the original vibration signal x (t) i (i =1,2 \8230;) and the average m of adjacent local extrema points is determined i Adjacent mean value points m i Connecting with straight line and smoothing by sliding average method to obtain local mean curve m 11 (t) and calculating an envelope estimation value a i ：
a i ＝|n i -n i+1 |/2
A2. All a to be obtained i The connection is smoothed by a moving average method to obtain an envelope estimation function a 11 (t) applying a local mean function m 11 (t) removing from the original signal x (t) to obtain a signal h 11 (t), envelope estimation function a 11 (t) to h 11 (t) demodulating to obtain a frequency modulated signal s 11 (t)。
A3. If s 11 Envelope estimation function a of (t) 12 (t) ≠ 1, then it represents s 11 (t) if the signal is not a pure FM signal, it is used as a new original signal and the above steps are repeated until s 1n (t) until the pure frequency modulation signal is obtained, the iteration process is as follows:
multiplying the envelope estimation function generated in the iterative algorithm to obtain an envelope signal a 1 (t)：
A4. Envelope signal a 1 (t) and a pure FM signal s 1n (t) performing a product operation to obtain a first PF component PF 1 (t) removing the resulting PF component PF from the original signal x (t) 1 (t) repeating the above steps with the residual value as the original signal until the obtained residual value is a monotonic function.
A5. And decomposing the vibration time domain signal by LMD to obtain a 6-order PF component. The components of each order contain different main signal components, and the most suitable component of each state is screened by calculating a correlation coefficient. According to the correlation coefficient of each PF component, PF 1-PF 4, PF 1-3 and PF6 are reserved for normal switching and refusal, base looseness, rotating shaft jamming and spring fatigue are used as research objects, and PF components obtained after denoising are reconstructed to obtain signals as shown in FIG. 2.
In the step B, the segmentation feature extraction method based on the vibration signal comprises the following specific steps:
B1. the vibration waveform of the on-load tap-changer has obvious cluster distribution, and according to the gear shifting principle of the on-load tap-changer, the vibration waveform respectively corresponds to three stages of a single resistance section, a bridging section, a double resistance section and the like of the action of a change-over switch, and integrates the vibration signal along a time axis to obtain an accumulated integral curve s (t), wherein the curve waveform is shown in figure 3, and two important curve inflection points t are arranged at two important curve inflection points 1 ＝0.031s、t 2 The calculation process of s (t) =0.052s is as follows:
B2. dividing the vibration waveform line into three sections according to the inflection points of two important waveforms, and dividing the vibration signal x into three sections 1 (t)、x 2 (t)、x 3 (t), each segment is an impact cluster, and the energy E of each impact cluster is calculated 1 、E 2 、 E 3 ：
B3. For the segmented vibration signal x 1 (t)、x 2 (t)、x 3 (t) respectively carrying out discrete Fourier transform and respectively calculating to obtain power spectrum S of each section of vibration signal i (w) segmentation of the vibration signalThe power spectrum is shown in fig. 4.
Wherein x is d (t) is a vibration signal; n represents the number of Fourier transform points; d =1,2,3.
The method is characterized in that a power spectrum kurtosis index K of a segmented vibration signal is taken as a dimensionless frequency domain index for describing waveform kurtosis, is extremely sensitive to an impact signal and is widely applied to equipment fault diagnosis, and K is expressed as:
wherein x is d Is a segmented vibration signal sequence; delta. For the preparation of a coating x Representing a vibration signal standard deviation; n represents the number of Fourier transform points;
is the average value of the sectional vibration signals; d =1,2,3.
And step C, constructing a vibration signal time-frequency spectrum by using the STFT to extract color characteristics. The vibration characteristics of the on-load tap-changer in different operating states are mainly embodied in that the color of a time-frequency graph changes, the higher the vibration amplitude is, the warmer the stripe tone of the time-frequency graph is reflected, the rich color characteristics in a color moment description time-frequency image are calculated to serve as the basis characteristics for judging the operating state of the on-load tap-changer, and the process is as follows:
C1. the time domain vibration signal is divided into a plurality of small segments with the same length according to the length, the time domain signal in the time window is assumed to be a stable signal, fourier transform is carried out on the vibration signal in the time window, and after the Fourier transform is completed on the signal of each segment of the time window, the time-varying frequency spectrum of the whole signal can be obtained. The time-varying frequency spectrum of the segmented vibration signal is shown in fig. 5, and each segmented signal has good specificity and represents the local characteristics of the vibration signal.
The STFT transform formula is as follows:
wherein, the signal amplitude at the g moment is represented; representing a sliding window function of the time domain signal; l represents a window function length.
C2. Since the color distribution information is mainly concentrated in the low order moments, it is sufficient to express the color distribution of the image using only the first, second and third moments of color, the first three moments being expressed as:
first moment, reflecting image brightness:
second moment, reflecting the image color distribution range:
third moment, reflecting the image color distribution symmetry:
wherein N is the number of image pixels; p is j Is the jth pixel color value in the image; a =1,2,3.
The first three moments of color total 9 parameters, and the color features of the time-frequency diagram are characterized by the average intensity, variance and skewness of the color components.
In step D, for the 27-dimensional color feature vector obtained by the STFT time-frequency spectrum, the present invention proposes to optimize the feature parameters by using a total correlation permutation optimization method for the first time to obtain a low redundancy feature vector reflecting the operating state of the on-load tap-changer, and the flow is shown in fig. 6, and the specific calculation steps are as follows:
D1. taking 20 sample signals as an example, 20 signal sample meters under the same working condition of on-load tap-changer vibration are randomly selectedCalculating 27-dimensional color feature vectors to form a color feature sample matrix Y 20×27 In Y, the element is Y op . Taking 27 column vectors of the color feature sample matrix as a color feature sample vector Y 1 ，Y 2 ，… Y 27 Calculating the correlation coefficient R of each color feature sample vector and the rest 26 vectors respectively vw ：
Wherein v and w are column numbers of two color sample vectors, and v, w =1,2 \823027; m is a color characteristic sample matrix line number, and m =1,2 \823020; y is mv ,y mw Is the vector value of the mth row, the vth column and the w column,
is the mean of the corresponding vectors.
D2. And respectively calculating the sum of each color feature sample vector and the correlation coefficients of the rest 26 vectors, recording the sum as a total correlation coefficient mu, representing the repeatability between the color feature corresponding to the vector and the reflected equipment information and other features, wherein the lower mu is, the more the color feature can represent the vibration characteristic of the on-load tap-changer under the working condition.
D3. Defining theta as a total correlation index, screening color features with the mu value smaller than theta, listing the color features in the color feature optimization set of the working condition, and taking the theta value to be 2.3746 according to the mechanical dispersity and the measured data of the tested equipment.
D4. And D1-D3 operation is carried out on the vibration signal sample data of the on-load tap-changer under each working condition, and the color characteristic optimal set obtained under each working condition is subjected to union working to obtain the color characteristic optimal total set of the on-load tap-changer. Through the optimization, 15 characteristic elements in the color characteristic optimization total set are selected finally to serve as new color characteristic vectors, and a color characteristic sample matrix is reconstructed to obtain an optimization matrixE 20×15 。
In step E, in order to further reduce the dimension of the vibration signal color characteristic vector, the invention provides a matrix principal component transformation method, the basic aim of which is to optimize a preferred matrix E obtained by optimizing a total correlation coefficient 20×15 Fitting to a low-dimensional matrix H 20×Z In (3), the similarity of the preferred matrix data is matched while minimizing the amount of deformation caused by dimension reduction. The method comprises the following specific steps:
E1. calculating a preferred matrix E 20×15 The row vector distance matrix D, the element of D is D rs 。
d rs ＝|x r -x s |
Wherein x is r ,x s Is a preference matrix E 20×15 R, s =1, \8230, 20.
E2. Constructing a matrix G, wherein the element in G is G rs ：
g rs ＝d rs -d r1 -d s1 +d 11
E3. Calculating the first 10 maximum eigenvalues and eigenvectors of G, and calculating a preference matrix E 20×15 Principal component matrix H of 20×10 The 15-dimensional preferred color vector is reduced to 10-dimensional.
Wherein, I is a matrix formed by the first 10 orthonormal eigenvectors reserved by the matrix G; alpha is the diagonal matrix of 10 largest eigenvalues.
And F, enabling the 1D-CNN model to consist of an input layer, a convolution layer, a pooling layer, a full-connection layer and an output layer. The weight matrix of each layer is convoluted with the characteristic matrix, the convolution result of the previous layer is output to be the next neuron after the operation of the activation function so as to construct the corresponding characteristic of the next layer, and the method comprises the following specific steps:
F1. the convolution layer convolves input data with a convolution kernel, sets the convolution kernel size to 1 × 15, and selects ReLU as an activation function to construct the nonlinearity of a feature vector addition model. The same convolution kernel shares parameters in the convolution process to obtain a class of characteristics, and the formula of the calculation process is as follows:
wherein
Represents the l layer input; n is a radical of j Representing an input feature vector; l represents the l-th network; k represents a convolution kernel; b represents the bias of the convolution kernel.
F2. The pooling layer performs scaling mapping on input data through pooling cores, reduces the dimension of the data and simultaneously extracts features, and adopts the maximum pooling with the size of 1 multiplied by 2 and the step length of 2. Pooling comprises average pooling and maximum pooling, with a transformation function of:
wherein W is the convolution kernel width;
the value of the Tth neuron in the ith feature of the l th layer is obtained; p is i l+1 (j) The value for the l +1 th neuron.
And F3. The output layer of the CNN is used for fully connecting the output of the last pooling layer, and then a Soft-Max classifier is used for solving the multi-classification problem.
F4. Two feature extraction layers are selected, the number of convolution kernels of the feature extraction layers is set to be 32 and 64 respectively, the number of nodes of two full-connection layers is set to be 256 and 64 respectively, and the training times are set to be 500 times.
F5. The method comprises the steps of randomly extracting 50% of vibration signal data, putting 20 groups of vibration signal data in each working condition as samples into 1D-CNN for training, classifying the rest 50% of test group data samples after training is finished, and obtaining a diagnosis result as shown in FIG. 7, wherein a normal switching sample label is 1, a base loose sample label is 2, a rotating shaft jamming label is 3, a switch refusing action is 4, a spring fatigue label is 5, only one rotating shaft jamming is mistakenly classified as a refusing action, the classification accuracy is obviously improved, and the condition of missing report does not exist.
The experimental result shows that the vibration signal characteristics of the on-load tap-changer extracted according to the steps have good robustness and specificity, can effectively capture and depict the running state characteristics of the equipment, and plays an important role in the fields of on-load tap-changer running state evaluation and fault diagnosis.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (7)
1. An intelligent diagnosis method for faults of an on-load tap-changer based on vibration signal segmentation time-frequency spectrum optimization is characterized by comprising the following steps,
A. an acceleration sensor is adopted to collect vibration signals of the on-load tap changer, and denoising processing is carried out by utilizing local mean decomposition;
B. taking an absolute value of the denoised vibration signal, performing accumulated integration on the absolute value as a time axis to obtain a plurality of sections of energy changes represented by an accumulated integration curve, segmenting the vibration signal according to an inflection point of the curve amplitude to obtain different action stages corresponding to a switch, and extracting energy and kurtosis characteristics of each section of signal;
C. carrying out short-time Fourier transform on the vibration signals to obtain each segmented time-frequency map, and extracting color moments to describe color features of the time-frequency image;
D. a total correlation arrangement optimization method is provided, and feature parameters are optimized to obtain low-redundancy optimized color feature vectors reflecting the operation state of the tap changer;
E. a matrix principal component transformation method is provided, further dimension reduction is carried out on the preferred color characteristics, the similarity of the characteristic components of the original data is kept, and the deformation caused by dimension reduction is minimized;
F. and forming an on-load tap-changer state characteristic vector by using the energy characteristic and the kurtosis characteristic of the segmented vibration signals and the time-frequency map color characteristic after dimension reduction, and completing fault identification through a one-dimensional convolution neural network.
2. The on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization according to claim 1, wherein in the step A, the acceleration sensor is adsorbed on the on-load tap-changer shell, and the current signal of the stator of the driving motor is used as a capture card trigger source; after the signal is subjected to local mean decomposition, each decomposed component is screened according to the correlation coefficient, and reconstruction is carried out to achieve the denoising effect.
3. The intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization as claimed in claim 1, wherein in step B, a method for segmenting vibration signals by using an accumulated integral curve is provided, an absolute value of the denoised vibration signals is taken and then accumulated integration is performed along a time axis to obtain an accumulated integral curve s (t), and the calculation method is as follows:
wherein x (t) is a vibration signal after denoising, and t is time;
the vibration signal is divided into three sections by capturing two important inflection points of a curve, the three sections respectively correspond to a single resistance section, a bridge section and a double resistance section of the action of a change-over switch in the on-load tap-changer, and each section is respectively embodied as an impact cluster in the vibration signal.
4. The on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization according to claim 1, wherein in the step C, a vibration segmentation signal time-frequency spectrum is constructed by using short-time fourier transform, and a first moment, a second moment and a third moment of color are extracted to describe image color distribution, so as to obtain a P-dimensional color feature vector.
5. The on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization as claimed in claim 1, wherein in the step D, a total correlation arrangement optimization method is proposed for feature parameter optimization, Q signal samples under the same working condition are taken, color feature vectors are calculated to construct a color feature sample matrix Y Q×P (ii) a Will Y Q×P The P column vectors are recorded as color feature sample vectors, and the total correlation coefficient μ of each color feature sample vector is calculated by the following calculation method:
wherein v and w are column numbers of the two color sample vectors; m is a color characteristic sample matrix row number; y is mv ,y mw Is the vector value of the mth row, the vth column and the w column,
is the mean of the corresponding vectors;
defining theta as a total correlation index, screening color features with the mu value smaller than theta, and listing the color features into an optimal feature set of the working condition; obtaining a color characteristic preferred total set of the on-load tap-changer by solving and collecting preferred characteristic sets of all working conditions, wherein the number of characteristic parameters in the total set is U, and reconstructing a color characteristic sample matrix by utilizing the U characteristic parameters to obtain a preferred matrix E Q×U 。
6. The on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization according to claim 5, wherein in the step E, a matrix principal component transformation method is provided, dimension reduction is further performed on the optimized color features, the similarity of original data feature components is kept, meanwhile, the deformation caused by dimension reduction is minimized, and an optimized matrix E is calculated Q×U Distance matrix D of middle row vector, wherein the element in D is D rs ：
d rs ＝|x r -x s |
Wherein x r ,x s Is E Q×U R, s =1, \ 8230;, Q;
constructing a matrix G, wherein the element in G is G rs ，g rs The calculation method comprises the following steps:
g rs ＝d rs -d r1 -d s1 +d 11
selecting the first Z maximum eigenvalues and eigenvectors of G, and calculating an optimal matrix E Q×U Principal component matrix H of Q×Z ；
Wherein I is a matrix formed by the first Z orthonormal eigenvectors reserved by the matrix G; alpha is a diagonal matrix of Z maximum eigenvalues.
7. The on-load tap-changer fault intelligent diagnosis method based on vibration signal segmentation time-frequency spectrum optimization according to claim 1, wherein in the step F, energy characteristics and kurtosis characteristics of the segmented vibration signals and dimension-reduced time-frequency spectrum color characteristics form an on-load tap-changer state characteristic vector, the on-load tap-changer state characteristic vector and the kurtosis characteristics are complementary to each other to accurately describe equipment operation state characteristics, and fault identification is completed through a one-dimensional convolutional neural network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211035522.6A CN115468645A (en) | 2022-08-26 | 2022-08-26 | Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211035522.6A CN115468645A (en) | 2022-08-26 | 2022-08-26 | Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115468645A true CN115468645A (en) | 2022-12-13 |
Family
ID=84368775
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211035522.6A Pending CN115468645A (en) | 2022-08-26 | 2022-08-26 | Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115468645A (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102012103263A1 (en) * | 2012-04-16 | 2013-10-17 | Maschinenfabrik Reinhausen Gmbh | Method and system for investigating a tapping change device of a transformer |
| CN105510023A (en) * | 2015-11-24 | 2016-04-20 | 国网内蒙古东部电力有限公司电力科学研究院 | Divergence-index-based fault diagnosis method of variable-working-condition wind-power planetary gearbox |
| CN106023185A (en) * | 2016-05-16 | 2016-10-12 | 国网河南省电力公司电力科学研究院 | Power transmission equipment fault diagnosis method |
| CN106950047A (en) * | 2017-03-09 | 2017-07-14 | 西安交通大学 | The visual analysis method of vibration acceleration signal frequency spectrum |
| CN108845250A (en) * | 2018-06-01 | 2018-11-20 | 国网江苏省电力有限公司南京供电分公司 | The load ratio bridging switch fault recognition method extracted based on vibration signal characteristics |
| CN112268615A (en) * | 2020-11-27 | 2021-01-26 | 华侨大学 | Electromechanical equipment vibration signal feature extraction method |
| KR20210080660A (en) * | 2019-12-20 | 2021-07-01 | 한국전력공사 | Method for detecting mechanical anomaly of tap-changer |
-
2022
- 2022-08-26 CN CN202211035522.6A patent/CN115468645A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102012103263A1 (en) * | 2012-04-16 | 2013-10-17 | Maschinenfabrik Reinhausen Gmbh | Method and system for investigating a tapping change device of a transformer |
| CN105510023A (en) * | 2015-11-24 | 2016-04-20 | 国网内蒙古东部电力有限公司电力科学研究院 | Divergence-index-based fault diagnosis method of variable-working-condition wind-power planetary gearbox |
| CN106023185A (en) * | 2016-05-16 | 2016-10-12 | 国网河南省电力公司电力科学研究院 | Power transmission equipment fault diagnosis method |
| CN106950047A (en) * | 2017-03-09 | 2017-07-14 | 西安交通大学 | The visual analysis method of vibration acceleration signal frequency spectrum |
| CN108845250A (en) * | 2018-06-01 | 2018-11-20 | 国网江苏省电力有限公司南京供电分公司 | The load ratio bridging switch fault recognition method extracted based on vibration signal characteristics |
| KR20210080660A (en) * | 2019-12-20 | 2021-07-01 | 한국전력공사 | Method for detecting mechanical anomaly of tap-changer |
| CN112268615A (en) * | 2020-11-27 | 2021-01-26 | 华侨大学 | Electromechanical equipment vibration signal feature extraction method |
Non-Patent Citations (4)
| Title |
|---|
| 李云鹏: "基于振动-电流联合特征的弹簧高压断路器储能状态辨识方法", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》, no. 3, 15 March 2022 (2022-03-15), pages 18 - 47 * |
| 石永芳 等: "非参数互信息和多维尺度分析在滚动轴承故障诊断中的应用", 《机床与液压》, no. 20, 31 December 2019 (2019-12-31), pages 188 - 189 * |
| 苗刚 等: "基于J散度的模式分类方法在故障诊断中的应用", 《中国机械工程》, no. 4, 30 April 2007 (2007-04-30), pages 432 - 433 * |
| 赵书涛 等: "基于振动信号分析的有载分接开关故障诊断研究综述", 《华北电力大学学报(自然科学版)》, no. 5, 31 October 2021 (2021-10-31), pages 61 - 71 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105841961A (en) | Bearing fault diagnosis method based on Morlet wavelet transformation and convolutional neural network | |
| CN110543860B (en) | Mechanical fault diagnosis method and system based on TJM (machine learning model) transfer learning | |
| CN111259921A (en) | Transformer sound anomaly detection method based on improved wavelet packet and deep learning | |
| CN109389171B (en) | Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology | |
| CN110705525A (en) | Method and device for diagnosing rolling bearing fault | |
| CN112881942B (en) | Abnormal current diagnosis method and system based on wavelet decomposition and empirical mode decomposition | |
| CN112633195A (en) | Myocardial infarction identification and classification method based on frequency domain features and deep learning | |
| CN109918417B (en) | Time sequence data self-adaptive segmentation, dimension reduction and characterization method based on wavelet transformation and application | |
| CN113673346A (en) | Motor vibration data processing and state recognition method based on multi-scale SE-Resnet | |
| CN113627375A (en) | Planetary gear fault diagnosis method and system, storage medium and computing device | |
| CN115146842B (en) | Multi-element time sequence trend prediction method and system based on deep learning | |
| CN114112992B (en) | Detection method and device for blue pigment of cream and storage medium | |
| CN111753891A (en) | Rolling bearing fault diagnosis method based on unsupervised feature learning | |
| CN113920375B (en) | Fusion characteristic typical load identification method and system based on combination of Faster R-CNN and SVM | |
| CN113865859A (en) | Multi-scale multi-source heterogeneous information fusion gearbox state fault diagnosis method | |
| CN112270255A (en) | Electroencephalogram signal identification method and device, electronic equipment and storage medium | |
| CN111160392A (en) | Hyperspectral classification method based on wavelet width learning system | |
| CN115166514A (en) | Motor fault identification method and system based on self-adaptive spectrum segmentation and denoising | |
| CN111639697A (en) | Hyperspectral image classification method based on non-repeated sampling and prototype network | |
| CN113052099B (en) | SSVEP classification method based on convolutional neural network | |
| CN115468645A (en) | Intelligent on-load tap-changer fault diagnosis method based on vibration signal segmentation time-frequency spectrum optimization | |
| CN116776245A (en) | Three-phase inverter equipment fault diagnosis method based on machine learning | |
| CN109460788B (en) | Hyperspectral image classification method based on low-rank-sparse information combination network | |
| CN116644273A (en) | Fault diagnosis method and system based on interpretability multiplication convolution network | |
| CN116312860A (en) | Agricultural product soluble solid matter prediction method based on supervised transfer learning |
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 |