CN110207974B - Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics - Google Patents
Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics Download PDFInfo
- Publication number
- CN110207974B CN110207974B CN201910618579.0A CN201910618579A CN110207974B CN 110207974 B CN110207974 B CN 110207974B CN 201910618579 A CN201910618579 A CN 201910618579A CN 110207974 B CN110207974 B CN 110207974B
- Authority
- CN
- China
- Prior art keywords
- time
- circuit breaker
- vibration signal
- fault
- energy distribution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- 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
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/327—Testing of circuit interrupters, switches or circuit-breakers
Abstract
The invention discloses a breaker fault identification method based on vibration signal time-frequency energy distribution characteristics, which comprises the steps of firstly obtaining vibration signals of each state of a breaker through experiments, and completing training of a deep self-coding network by using the time-frequency energy distribution characteristics of experimental data; and then, inputting the time-frequency energy distribution characteristics of the signals to be identified into the trained deep self-coding network, and accurately and reliably identifying the mechanical faults of the circuit breaker operating mechanism by using the network. The method and the device can judge the fault type of the circuit breaker, find potential faults in time, help maintainers to finish the maintenance and repair work of the circuit breaker more pertinently, and further reduce the accident loss caused by the circuit breaker.
Description
Technical Field
The invention relates to the technical field of breaker fault diagnosis, in particular to a breaker fault identification method based on vibration signal time-frequency energy distribution characteristics.
Background
Circuit breakers play a key role in control and protection in power systems. The statistical results at home and abroad show that the mechanical fault of the operating mechanism is the main fault type of the circuit breaker. The circuit breaker vibration signal contains abundant mechanical state information, which can be regarded as a result generated by the motion of various elements in the internal operating mechanism of the circuit breaker, and once the mechanical state of the circuit breaker changes, the motion state of the corresponding element changes, so that the vibration signal changes. Therefore, some slight changes on the mechanical structure can be reflected through the change of the vibration signal characteristic information, so that the mechanical fault type of the operating mechanism can be diagnosed.
The internal structure of the circuit breaker is complex, a plurality of elements are mutually associated, uncertain factors are numerous, the generated faults are often random, propagation, concurrency and the like, the composition of the whole circuit breaker has strong non-linearity and uncertainty, and monitoring and diagnosis cannot be carried out by means of an accurate physical model. Therefore, in the stage of feature extraction, the selection of the feature extraction method depends on professional knowledge and diagnosis experience, and often a certain method can only be adopted for specific mechanical faults, so that the uncertainty of feature extraction is high, the universality is poor, and the difficulty of fault diagnosis is improved. When the extracted characteristic information is used for carrying out fault identification on the circuit breaker, the traditional shallow networks such as a support vector machine and a BP neural network have insufficient capability of learning complex functions, and the problems of local optimization and the like exist; when complex signals such as circuit breaker vibration signals are processed, deep features of data cannot be sufficiently mined by the shallow layer networks, and the limitation is large when complex information is analyzed and processed. Due to the defects of feature extraction and fault identification, the reliability of the fault diagnosis of the circuit breaker is low at present, so that a circuit breaker fault diagnosis method with higher universality and more reliability needs to be introduced.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a method for identifying a fault of a circuit breaker based on vibration signal time-frequency energy distribution characteristics, which can determine the fault type of the circuit breaker, find a potential fault in time, help a maintainer to complete maintenance and repair work of the circuit breaker more specifically, and further reduce accident loss caused by the circuit breaker. The technical scheme is as follows:
a circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics is characterized by comprising the following steps:
step A: respectively collecting switching-on vibration signal experiment data of the operating mechanism in A times of experiment processes when the circuit breaker is in a normal state and the circuit breaker is in three fault states of jamming of the operating mechanism, jamming of an iron core and loosening of a base screw;
and B: extracting closing vibration signals to be processed in each state from the closing vibration signal experimental data, and calculating the time-frequency energy distribution characteristics of the experimental data in each state;
and C: training a deep self-coding network according to the time-frequency energy distribution characteristics of the experimental data in each state;
step D: collecting a closing vibration signal of a circuit breaker operating mechanism in actual engineering, extracting a vibration signal to be identified from the closing vibration signal, and calculating the time-frequency energy distribution characteristic of the vibration signal to be identified;
step E: and inputting the time-frequency energy distribution characteristics of the signal to be identified into the trained deep self-coding network, and outputting a fault identification result.
Further, the step B specifically includes:
step B1: denoising the acquired experimental data of the switching-on vibration signals of the circuit breaker operating mechanism in each state by utilizing wavelet transformation; comparing the amplitude of each sampling point of the denoised vibration signal in each state with a threshold alpha point by point, considering that closing vibration occurs when the average value of continuous a sampling points behind a certain point is greater than the threshold alpha, and selecting the point T before the point1T after second to this point2Taking the signal in second time as a switching-on vibration signal s to be processed in each stateb(t), b is the breaker state, b is 0 and represents the normal state, b is 1 and represents the operating mechanism jam trouble, b is 2 and represents the iron core jam, b is 3 and represents the base screw is loose;
step B2: switching-on vibration signal s to be processed in each stateb(t) m layers of wavelet packet conversion to obtain I-2mVibration signal component s of individual frequency bandbi(t), I ═ 1,2, …, I; dividing each frequency band signal into N sections with equal time length to obtain I multiplied by N time frequency sub-planes; calculating the energy of the vibration signal in each sub-plane:
in the formula, Eb,i,nRepresenting the energy of the time-frequency sub-plane in the nth time period of the ith frequency band in the state of a b-type breakern、tn+1The time of the start and the end of the nth time period respectively;
step B3: energy E of time-frequency sub-planeb,i,nAfter the Z-score standardization treatment is carried out, the vibration signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristics of the vibration signals
In the formula (I), the compound is shown in the specification,
representing the time-frequency energy distribution characteristics of the v-th experimental data in the b-type breaker state; v ═ 1,2, …, a.
Further, the step C specifically includes:
step C1: taking the time-frequency energy distribution characteristics of each state experimental data as a training sample set of the deep self-coding network:
setting the expected output layer of the deep self-coding network to be [1, 0, 0, 0] in a normal state; setting the expected output layer of the deep self-coding network to be [0, 1, 0, 0] under the jamming fault of the operating mechanism; setting the expected output layer of the deep self-coding network to be [0, 0, 1, 0] under the iron core jamming fault; setting the expected output layer of the deep self-coding network to be [0, 0, 0, 1] under the jamming fault of the operating mechanism;
step C2: ganglion point number K of input layer of depth self-coding network1Equal to the number of frequency sub-planes of the vibration signal, i.e. K1I × N; ganglionic point number K of setting depth self-coding network output layer2Equal to the desired length of the circuit breaker output layer, i.e. K24; the method comprises the steps that a depth self-coding network is arranged to comprise 2 hidden layers, the first hidden layer comprises I multiplied by N/2 neural nodes, and the second hidden layer comprises I multiplied by N/4 neural nodes; setting an activation function of the depth self-coding network as a sigmoid function;
step C3: importing a training sample set T into a depth self-coding network, pre-training the depth self-coding network by using an unsupervised layer-by-layer greedy algorithm, and then finely adjusting the depth self-coding network by using an error back propagation algorithm; and when the error is smaller than an error threshold value delta or the iteration times are larger than an iteration time threshold value epsilon in the fine tuning process, finishing the training of the depth self-coding network, and storing the trained depth self-coding network.
Further, the step D specifically includes:
step D1: acquiring a vibration signal during the closing operation of a circuit breaker in actual engineering according to a sampling frequency f, and denoising the vibration signal by utilizing wavelet transformation; after the completion, the amplitude of each sampling point of the denoised vibration signal is gradually compared with a threshold value alpha, when the average value of a continuous a sampling points behind a certain point is greater than the threshold value alpha, the closing vibration is considered to have occurred, and T before the point is selected1T after second to this point2The signal in the second time is used as a vibration signal g (t) to be identified;
step D2: after m layers of wavelet packet transformation are carried out on the vibration signal g (t) to be identified, I is 2mSignal component g of a frequency bandi(t), I ═ 1,2, …, I; dividing each frequency band signal into N sections with equal time length to obtain I multiplied by N time frequency sub-planes; calculating the energy G of the vibration signal in each sub-plane of the vibration signal to be identifiedi,n：
In the formula, Gi,nRepresents the energy of the time-frequency sub-plane in the nth time segment of the ith frequency band, tn、tn+1The time of the start and the end of the nth time period respectively;
the time-frequency sub-plane energy G of the signal to be identifiedi,nAfter the Z-score standardization treatment, the signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristic H of the signals to be identifiedw：
Further, the method for determining the fault state of the circuit breaker in the step E includes: when the output of the 1 st neural node of the output layer of the deep self-coding network is maximum, the circuit breaker is in a normal state at present; when the output of the 2 nd neural node of the output layer is maximum, the circuit breaker is in the jamming fault of the operating mechanism at present; when the output of the 3 rd neural node of the output layer is maximum, the circuit breaker is in an iron core jamming fault at present; when the output of the 4 th neural node of the output layer is maximum, the circuit breaker is at the base screw loosening fault at present.
Further, the number of each type of fault sample in step a is not less than 200.
Furthermore, the value of the threshold value alpha is 2-5.
Further, the error threshold δ is not greater than 0.005 and the iteration number threshold ε is not less than 20000.
Furthermore, the number of the divided sections N of the time period is 12, and the number m of the wavelet packet layers is 3.
Furthermore, in the comparison process, the number a of consecutive sampling points is 10.
The invention has the beneficial effects that:
1) the invention firstly carries out wavelet packet transformation and equal time segmentation on the vibration signal of the circuit breaker, and divides the vibration signal into a plurality of time-frequency sub-planes. After the circuit breaker has mechanical failure, as long as the time domain and the frequency domain of the vibration signal are changed, the energy of the corresponding time-frequency sub-plane is changed, so that the change of the time-frequency energy distribution characteristics is caused, therefore, the invention reserves the effective information in the circuit breaker vibration signal as much as possible, and has stronger universality.
2) The invention adopts the deep self-coding network to identify the fault, deeply excavates the potential association between the time-frequency energy distribution characteristic and the fault type, solves the local optimal problem, has higher fault diagnosis accuracy compared with the prior art, enables the maintainer to take targeted maintenance measures aiming at the breaker, and further reduces the accident loss caused by the breaker.
Drawings
Fig. 1 is a flow chart of the circuit breaker fault identification of the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and specific embodiments. The specific embodiment provides a circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics, the flow of which is shown in fig. 1, and the specific steps are as follows:
A. acquisition of circuit breaker experimental data
The breaker switching-on operation experiment is respectively carried out on the breaker in a normal state and three fault states of jamming of an operating mechanism, jamming of an iron core and loosening of a base screw, the breaker is acquired by an acceleration sensor at a sampling frequency f,
when the breaker is in a normal state, switching-on vibration signals of the operating mechanism in the A times of experiment processes are generated;
when the circuit breaker is in the jamming fault of the operating mechanism, switching-on vibration signals of the operating mechanism in the A-time experiment process are sent;
when the breaker is in an iron core jamming fault, switching-on vibration signals of the operating mechanism in the A-time experiment process are generated;
when the breaker is in a base screw loosening fault, the closing vibration signal of the operating mechanism is generated in the A-time experiment process.
The number N of each type of fault sample is not less than 200.
B. Extraction of breaker experimental data energy distribution characteristics
B1, denoising the breaker experiment data by utilizing wavelet transformation. After denoising is finished, comparing the amplitude of each sampling point of the vibration signal with a threshold value alpha point by point, considering that closing vibration occurs when the average value of 10 continuous sampling points behind a certain point is greater than the threshold value alpha, and selecting the point T before the point1T after second to this point2And taking the signal in the second time as a vibration signal s (t) to be processed.
B2, obtaining signal components s of 8 frequency bands after 3-layer wavelet packet transformation of vibration signals s (t) to be processedi(t), i ═ 1,2, …, 8. Dividing each frequency band signal into 12 segments with equal time length, then vibrating signal si(t) is divided into 8 × 12, 96 time-frequency sub-planes. Calculating the energy E of the vibration signal in each sub-plane by using the following formulai,n：
In the formula Ei,nRepresents the energy of the time-frequency sub-plane in the nth time segment of the ith frequency band, tn、tn+1Respectively, the start and end times of the nth time period.
B3: energy E of time-frequency sub-planei,nAfter the Z-score standardization treatment is carried out, the vibration signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristics of the vibration signals
In the formula
Representing the time-frequency energy distribution characteristics of the v-th experimental data in the b-type breaker state; b-0 represents normal state, b-1 represents jamming fault of the operating mechanism, b-2 represents jamming of the iron core, and b-3 represents loosening of the base screw; v ═ 1,2, …, a.
Repeating the steps B1 and B2 on the experimental data of each normal state to obtain the time-frequency energy distribution characteristics of A normal states
Repeating the steps B1 and B2 on the experimental data of the jamming faults of each operating mechanism to obtain the time-frequency energy distribution characteristics of the jamming faults of A operating mechanisms
Repeating the steps B1 and B2 on the experimental data of each iron core jamming fault to obtain the time-frequency energy distribution characteristics of A iron core jamming faults
Repeating the steps B1 and B2 on the experimental data of each base screw loosening fault to obtain the time-frequency energy distribution characteristics of A base screw loosening faults
C. Training of deep self-coding networks
C1, using the time-frequency energy distribution characteristics of the experimental data of each state in the step B3 as a training sample set of the deep self-coding network
Setting the expected output layer of the deep self-coding network to be [1, 0, 0, 0] in a normal state; setting the expected output layer of the deep self-coding network to be [0, 1, 0, 0] under the jamming fault of the operating mechanism; setting the expected output layer of the deep self-coding network to be [0, 0, 1, 0] under the iron core jamming fault; the expected output layer of the deep self-coding network under the jamming failure of the operating mechanism is set to be 0, 0, 0, 1.
C2 ganglion points K for setting depth self-coding network input layer1Equal to the number of frequency sub-planes of the vibration signal, i.e. K18 × 12 ═ 96; ganglionic point number K of setting depth self-coding network output layer2Equal to the desired length of the circuit breaker output layer, i.e. K24; the method comprises the steps that a depth self-coding network is arranged to comprise 2 hidden layers, the first hidden layer comprises 48 neural nodes, and the second hidden layer comprises 24 neural nodes; and setting an activation function of the deep self-coding network as a sigmoid function.
And C3, introducing the training sample T into the depth self-coding network, pre-training the depth self-coding network by using an unsupervised greedy algorithm layer by layer, and finely adjusting the depth self-coding network by using an error back propagation algorithm. And when the error is less than delta or the iteration times are more than epsilon in the fine tuning process, finishing the training of the depth self-coding network, and storing the trained depth self-coding network.
In this embodiment, the error threshold δ is not greater than 0.005, and the iteration threshold ∈ is not less than 20000.
D. Acquisition of breaker closing vibration signal time-frequency energy distribution characteristics
D1, collecting a vibration signal during the closing operation of the breaker in the actual engineering according to the sampling frequency f, wherein the model and the structural parameters of the breaker are the same as those of the breaker adopted in the experiment in the step A, and denoising the vibration signal by using wavelet transformation. After denoising is finished, comparing the amplitude of each sampling point of the signal with a threshold value alpha point by point, considering that closing vibration occurs when the average value of 10 continuous sampling points behind a certain point is greater than the threshold value alpha, and selecting the point T before the point1T after second to this point2And taking the signal in the second time as a vibration signal g (t) to be identified.
D2, obtaining signal components g of 8 frequency bands after 3-layer wavelet packet transformation of vibration signals g (t) to be identifiedi(t), i ═ 1,2, …, 8. Dividing each frequency band signal into 12 segments with equal time length, then vibrating signal gi(t) is divided into 8 × 12, 96 time-frequency sub-planes. Calculating the energy G of the vibration signal in each sub-plane of the vibration signal to be identified by using the following formulai,n：
In the formula, Ei,nRepresents the energy of the time-frequency sub-plane in the nth time segment of the ith frequency band, tn、tn+1Respectively, the start and end times of the nth time period.
D3, and identifying the time-frequency sub-plane energy E of the signal to be identifiedi,nAfter the Z-score standardization treatment, the signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristic H of the signals to be identifiedw：
Hw＝[G1,1,G1,2,…,G1,12,G2,1,G2,2,…,G2,12,…,G8,1,G8,2,…,G8,12]
E. The time-frequency energy distribution characteristic H of the signal to be identifiedwInputting the trained deep self-coding network in the step C3When the output of the 1 st neural node of the output layer is maximum, the circuit breaker is in a normal state at present; when the output of the 2 nd neural node of the output layer is maximum, the circuit breaker is in the jamming fault of the operating mechanism at present; when the output of the 3 rd neural node of the output layer is maximum, the circuit breaker is in an iron core jamming fault at present; when the output of the 4 th neural node of the output layer is maximum, the circuit breaker is at the base screw loosening fault at present.
The principle of the fault identification method of the invention is as follows:
the circuit breaker vibration signal is generated by multiple times of vibration superposition generated by collision and friction of internal elements of the circuit breaker, contains abundant mechanical state information, and can be correspondingly changed when a mechanical fault occurs. These variations can be categorized into three aspects, namely temporal characteristics, frequency characteristics, and amplitude characteristics. The change of the time characteristic refers to the change of the starting time and the duration time of a certain vibration in the vibration signal of the circuit breaker; the change of the frequency characteristic means that the frequency of a certain vibration changes; and the change of the amplitude characteristic means that the amplitude of the vibration of the circuit breaker changes at a certain time. In order to reflect the influence brought by the changes, the frequency domain of the vibration signal is divided through the wavelet packet, then the time domain of the signal is divided through an equal time segmentation method to obtain the time-frequency sub-planes of the vibration signal, the energy of the vibration amplitude in each time-frequency sub-plane is calculated, the time-frequency energy distribution characteristics are obtained and serve as characteristic information, and the changes of the time, the frequency and the vibration amplitude of the vibration signal can be comprehensively reflected.
The circuit breaker has a complex internal structure, a plurality of elements are mutually associated, uncertain factors are numerous, the generated faults often have the properties of randomness, transmissibility, concurrency and the like, the composition of the whole circuit breaker has strong non-linearity and uncertainty, and the traditional fault diagnosis effect is poor, so that the circuit breaker operation mechanism mechanical fault diagnosis method utilizes the approximation capability and the feature mining capability of a deep self-coding network to complex functions to diagnose the mechanical faults of the circuit breaker operation mechanism. The deep self-coding network is a network structure formed by stacking a plurality of self-coders, and the training process of the deep self-coding network is divided into two stages. The first stage is a pre-training stage, namely, training the self-encoder layer by layer from the bottom layer to the top layer, outputting a hidden layer of the self-encoder as the input of a next self-encoder after the training of the current self-encoder is finished, and then starting the training of the next self-encoder until the training process of all self-encoders is finished by utilizing the data propagation mode. The initial parameter setting of the network can be completed through a pre-training process. The second part is a fine adjustment stage, the network parameters obtained in the last stage are fine adjusted from the classification result layer to the input layer by adopting a supervised learning mode, and the deep self-coding network tends to be global optimal by adopting an error back propagation algorithm, so that the invention can effectively and reliably identify faults of the circuit breaker operating mechanism with a complex mechanical structure by inputting the time-frequency energy distribution characteristics into the deep self-coding network.
Claims (9)
1. A circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics is characterized by comprising the following steps:
step A: respectively collecting switching-on vibration signal experiment data of the operating mechanism in A times of experiment processes when the circuit breaker is in a normal state and the circuit breaker is in three fault states of jamming of the operating mechanism, jamming of an iron core and loosening of a base screw;
and B: extracting closing vibration signals to be processed in each state from the closing vibration signal experimental data, and calculating the time-frequency energy distribution characteristics of the experimental data in each state:
step B1: denoising the acquired experimental data of the switching-on vibration signals of the circuit breaker operating mechanism in each state by utilizing wavelet transformation; comparing the amplitude of each sampling point of the denoised vibration signal in each state with a threshold alpha point by point, considering that closing vibration occurs when the average value of continuous a sampling points behind a certain point is greater than the threshold alpha, and selecting the point T before the point1T after second to this point2Taking the signal in second time as a switching-on vibration signal s to be processed in each stateb(t), b is the breaker state, b is 0 and represents the normal state, b is 1 and represents the operating mechanism jamming trouble, b is 2 and represents the iron core jamming, b is 3 and representsLoosening the base screw;
step B2: switching-on vibration signal s to be processed in each stateb(t) m layers of wavelet packet conversion to obtain I-2mVibration signal component s of individual frequency bandbi(t), I ═ 1,2, …, I; dividing each frequency band signal into N sections with equal time length to obtain I multiplied by N time frequency sub-planes; calculating the energy of the vibration signal in each sub-plane:
in the formula, Eb,i,nRepresenting the energy of the time-frequency sub-plane in the nth time period of the ith frequency band in the state of a b-type breakern、tn+1The time of the start and the end of the nth time period respectively;
step B3: energy E of time-frequency sub-planeb,i,nAfter the Z-score standardization treatment is carried out, the vibration signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristics of the vibration signals
In the formula (I), the compound is shown in the specification,
representing the time-frequency energy distribution characteristics of the v-th experimental data in the b-type breaker state; v ═ 1,2, …, a;
and C: training a deep self-coding network according to the time-frequency energy distribution characteristics of the experimental data in each state;
step D: collecting a closing vibration signal of a circuit breaker operating mechanism in actual engineering, extracting a vibration signal to be identified from the closing vibration signal, and calculating the time-frequency energy distribution characteristic of the vibration signal to be identified;
step E: and inputting the time-frequency energy distribution characteristics of the signal to be identified into the trained deep self-coding network, and outputting a fault identification result.
2. The method for identifying the fault of the circuit breaker based on the vibration signal time-frequency energy distribution characteristics as claimed in claim 1, wherein the step C is specifically as follows:
step C1: taking the time-frequency energy distribution characteristics of each state experimental data as a training sample set of the deep self-coding network:
setting the expected output layer of the deep self-coding network to be [1, 0, 0, 0] in a normal state; setting the expected output layer of the deep self-coding network to be [0, 1, 0, 0] under the jamming fault of the operating mechanism; setting the expected output layer of the deep self-coding network to be [0, 0, 1, 0] under the iron core jamming fault; setting the expected output layer of the deep self-coding network to be [0, 0, 0, 1] under the condition of the loosening fault of the base screw;
step C2: ganglion point number K of input layer of depth self-coding network1Equal to the number of frequency sub-planes of the vibration signal, i.e. K1I × N; ganglionic point number K of setting depth self-coding network output layer2Equal to the desired length of the circuit breaker output layer, i.e. K24; the method comprises the steps that a depth self-coding network is arranged to comprise 2 hidden layers, the first hidden layer comprises I multiplied by N/2 neural nodes, and the second hidden layer comprises I multiplied by N/4 neural nodes; setting an activation function of the depth self-coding network as a sigmoid function;
step C3: importing a training sample set T into a depth self-coding network, pre-training the depth self-coding network by using an unsupervised layer-by-layer greedy algorithm, and then finely adjusting the depth self-coding network by using an error back propagation algorithm; and when the error is smaller than an error threshold value delta or the iteration times are larger than an iteration time threshold value epsilon in the fine tuning process, finishing the training of the depth self-coding network, and storing the trained depth self-coding network.
3. The method for identifying the fault of the circuit breaker based on the vibration signal time-frequency energy distribution characteristics as claimed in claim 2, wherein the step D specifically comprises the following steps:
step D1: acquiring a vibration signal during the closing operation of a circuit breaker in actual engineering according to a sampling frequency f, and denoising the vibration signal by utilizing wavelet transformation; after the completion, the amplitude of each sampling point of the denoised vibration signal is gradually compared with a threshold value alpha, when the average value of a continuous a sampling points behind a certain point is greater than the threshold value alpha, the closing vibration is considered to have occurred, and T before the point is selected1T after second to this point2The signal in the second time is used as a vibration signal g (t) to be identified;
step D2: after m layers of wavelet packet transformation are carried out on the vibration signal g (t) to be identified, I is 2mSignal component g of a frequency bandi(t), I ═ 1,2, …, I; dividing each frequency band signal into N sections with equal time length to obtain I multiplied by N time frequency sub-planes; calculating the energy G of the vibration signal in each sub-plane of the vibration signal to be identifiedi,n：
In the formula, Gi,nRepresents the energy of the time-frequency sub-plane in the nth time segment of the ith frequency band, tn、tn+1The time of the start and the end of the nth time period respectively;
the time-frequency sub-plane energy G of the signal to be identifiedi,nAfter the Z-score standardization treatment, the signals are arranged according to the frequency and time sequence to obtain the time-frequency energy distribution characteristic H of the signals to be identifiedw：
Hw＝[G1,1,G1,2,…,G1,N,G2,1,G2,2,…,G2,N,…,GI,1,GI,2,…,GI,N]。
4. The method for identifying the fault of the circuit breaker based on the vibration signal time-frequency energy distribution characteristics as claimed in claim 3, wherein the method for judging the fault state of the circuit breaker in the step E is as follows: when the output of the 1 st neural node of the output layer of the deep self-coding network is maximum, the circuit breaker is in a normal state at present; when the output of the 2 nd neural node of the output layer is maximum, the circuit breaker is in the jamming fault of the operating mechanism at present; when the output of the 3 rd neural node of the output layer is maximum, the circuit breaker is in an iron core jamming fault at present; when the output of the 4 th neural node of the output layer is maximum, the circuit breaker is at the base screw loosening fault at present.
5. The method for identifying the fault of the circuit breaker based on the time-frequency energy distribution characteristics of the vibration signals as claimed in claim 1, wherein the number of each type of fault samples in the step A is not less than 200.
6. The circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics as claimed in claim 1 or 3, wherein the value of the threshold value alpha is 2-5.
7. The method for identifying the fault of the circuit breaker based on the time-frequency energy distribution characteristics of the vibration signals as claimed in claim 2, wherein the error threshold δ is not more than 0.005, and the iteration threshold ε is not less than 20000.
8. The method for identifying the fault of the circuit breaker based on the vibration signal time-frequency energy distribution characteristics according to claim 1 or 3, wherein the number N of the divided time periods is 12, and the number m of wavelet packet layers is 3.
9. The method for identifying the fault of the circuit breaker based on the time-frequency energy distribution characteristics of the vibration signals as claimed in claim 1 or 3, wherein the number a of continuous sampling points is 10 in the comparison process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910618579.0A CN110207974B (en) | 2019-07-10 | 2019-07-10 | Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910618579.0A CN110207974B (en) | 2019-07-10 | 2019-07-10 | Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110207974A CN110207974A (en) | 2019-09-06 |
| CN110207974B true CN110207974B (en) | 2021-04-09 |
Family
ID=67796955
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910618579.0A Active CN110207974B (en) | 2019-07-10 | 2019-07-10 | Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110207974B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112327149A (en) * | 2020-10-19 | 2021-02-05 | 西安工程大学 | SSD-SRAE-based high-voltage circuit breaker fault diagnosis method |
| CN112665711A (en) * | 2020-12-21 | 2021-04-16 | 陕西宝光集团有限公司 | Method and device for detecting running state of equipment, electronic equipment and storage medium |
| CN112763904A (en) * | 2020-12-29 | 2021-05-07 | 广州航天海特系统工程有限公司 | Circuit breaker detection method, device, equipment and storage medium |
| CN114690038B (en) * | 2022-06-01 | 2022-09-20 | 华中科技大学 | Motor fault identification method and system based on neural network and storage medium |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040264094A1 (en) * | 2003-05-06 | 2004-12-30 | Rahman Md Azizur | Protective control method and apparatus for power devices |
| CN105868770B (en) * | 2016-03-23 | 2019-09-10 | 国网山东省电力公司电力科学研究院 | A kind of Fault Diagnosis for HV Circuit Breakers method based on unsupervised learning model |
| CN105956290B (en) * | 2016-05-06 | 2019-06-07 | 国网江苏省电力有限公司检修分公司 | A kind of Mechanical Failure of HV Circuit Breaker diagnostic method based on multi-data fusion technology |
| CN107607303A (en) * | 2017-09-13 | 2018-01-19 | 河海大学 | Mechanical Failure of HV Circuit Breaker recognition methods based on wavelet packet Yu SOM networks |
| CN108828440A (en) * | 2018-06-12 | 2018-11-16 | 江苏镇安电力设备有限公司 | High-voltage circuitbreaker defect diagnostic method based on wavelet-packet energy |
-
2019
- 2019-07-10 CN CN201910618579.0A patent/CN110207974B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN110207974A (en) | 2019-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110207974B (en) | Circuit breaker fault identification method based on vibration signal time-frequency energy distribution characteristics | |
| CN110674604B (en) | Transformer DGA data prediction method based on multi-dimensional time sequence frame convolution LSTM | |
| CN110428004B (en) | Mechanical part fault diagnosis method based on deep learning under data imbalance | |
| CN110610035B (en) | Rolling bearing residual life prediction method based on GRU neural network | |
| CN109000930B (en) | Turbine engine performance degradation evaluation method based on stacking denoising autoencoder | |
| CN109633431B (en) | On-load tap-changer fault identification method based on vibration signal feature extraction | |
| CN110261109B (en) | Rolling bearing fault diagnosis method based on bidirectional memory cyclic neural network | |
| CN111351665B (en) | Rolling bearing fault diagnosis method based on EMD and residual error neural network | |
| CN110764063A (en) | Radar signal sorting method based on combination of SDIF and PRI transformation method | |
| CN108304348B (en) | Bearing residual life prediction method based on binary wiener process | |
| CN110059845B (en) | Metering device clock error trend prediction method based on time sequence evolution gene model | |
| CN116226646B (en) | Method, system, equipment and medium for predicting health state and residual life of bearing | |
| CN110726898A (en) | Power distribution network fault type identification method | |
| CN103852525B (en) | Acoustic emission signal recognition methods based on AR-HMM | |
| CN110794254B (en) | Power distribution network fault prediction method and system based on reinforcement learning | |
| CN112508242A (en) | Method for constructing bearing fault location and classification model of wind power generator | |
| CN116619136A (en) | Multi-working-condition multi-source data cutter abrasion prediction method | |
| CN111079647A (en) | Circuit breaker defect identification method | |
| Zhang et al. | Retracted: Fault Diagnosis Method Based on EEMD and Multi-Class Logistic Regression | |
| CN116735198A (en) | Fan gear box fault diagnosis method and system based on speEMD and double index screening | |
| CN115630280A (en) | Rolling bearing fault diagnosis method based on CEEMD multi-scale diffusion entropy and PSO-ELM | |
| Li | Remaining useful life prediction of bearings using fuzzy multimodal extreme learning regression | |
| CN113222145A (en) | MODWT-EMD-based time sequence hybrid prediction method | |
| CN115032508B (en) | Distributed transmission line fault diagnosis method and system based on target identification | |
| AU2016330300B2 (en) | Method for determining diagnostic models for time series of a technical system and diagnostic method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |