Circuit breaker fault diagnosis method

Abstract

The invention discloses a circuit breaker fault diagnosis method based on a segmented CEEMDAN optimization algorithm, and the method comprises the steps: obtaining vibration signals of a circuit breaker under the conditions of normal switching-on, switching-on operation refusing, base loosening and connecting rod falling through a vibration sensor; segmentation processing is carried out on the vibration signals in the time domain, CEEMDAN decomposition is carried out on each segment, and then the kurtosis of each modal component is obtained. And then, performing pattern recognition on each segment by using an SVM algorithm. And finally, determining which condition the vibration signal belongs to through voting. The circuit breaker fault diagnosis method comprises the following steps: segmenting a signal, then processing the signal, and then carrying out feature extraction on the decomposed signal. According to the method, the state of each segmented signal is judged in a segmented manner, and the state of the whole signal is judged according to the state of each segmented signal, so that the probability of wrong judgment of the whole signal caused by errors in a certain process in the signal extraction process is reduced. The circuit breaker fault diagnosis error-tolerant rate can be increased, the precision and speed requirements for circuit breaker fault diagnosis in operation on site can be met, and rapid and accurate identification is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of electrical equipment fault detection, and in particular to a method for diagnosing circuit breaker faults. Background Art

[0002] High-voltage circuit breakers are the most commonly used and common power equipment in power systems. When a power system fails, they can quickly cut off the fault, protect the power system from damage, and ensure the stable operation of the power system. If a circuit breaker fails, the loss caused is far greater than the value of the circuit breaker itself. Therefore, circuit breaker fault diagnosis is a hot topic in research.

[0003] In order to diagnose the fault of the circuit breaker, choosing a suitable signal is the first step. At present, the more popular circuit breaker fault diagnosis methods are mainly based on the research of the current signal of the opening and closing coil, the mechanical vibration signal, the sound signal and so on. Among them, the vibration signal contains the state information of most of the components in the operating mechanism. The state-related information of some key components in the operating mechanism of the high-voltage circuit breaker can be mined from the vibration signal by using a suitable signal processing method, and then the state-related information can be parsed by using a suitable state discrimination method to identify the actual state of these components. Therefore, the current circuit breaker fault diagnosis method mainly uses vibration signals. In order to extract relevant state information from the signal, it is necessary to select a suitable signal processing method. The signal processing method is mainly based on empirical wavelet transform, variational mode decomposition, empirical mode decomposition, Wigner distribution, etc. For the processed signal, the signal features are extracted to realize fault diagnosis. In order to use these features to better realize the circuit breaker fault diagnosis, methods such as support vector machine and neural network based on machine learning are also used to greatly improve the accuracy of fault diagnosis.

[0004] The overall characteristics of the collected vibration signal are analyzed and processed, and the fault type of the circuit breaker is determined by building an overall fault database and comparing it with the characteristic signal. This method of overall signal processing is relatively old and single, and is limited by the form of feature extraction. While analyzing the entire collected signal, it is difficult to find the detailed signal fault characteristics, and the accuracy of the comparison is poor, making it difficult to accurately determine the fault type. Therefore, in order to tap the rich information contained in the vibration signal, it is particularly important to use segmented signals for fault diagnosis. Summary of the invention

[0005] The purpose of the present invention is to provide a circuit breaker fault diagnosis method based on CEEMDAN to decompose the segmented circuit breaker vibration signal. According to the physical process and prior knowledge experienced during the circuit breaker operation, the linkage process of mechanical parts during the circuit breaker operation is divided into four parts. The time domain waveforms corresponding to the vibration signals collected under normal conditions and three fault conditions are segmented respectively. CEEMDAN decomposition is performed on the segmented vibration signals respectively, and the kurtosis of each segment of the decomposed vibration signal is obtained as the eigenvalue, and the SVM algorithm is used for pattern recognition. Finally, a voting mechanism is introduced to vote on the judgment results of each segment, and finally determine which state the vibration signal belongs to.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for circuit breaker fault diagnosis based on segmented CEEMDAN optimization algorithm, through a vibration sensor, obtains the vibration signal of the circuit breaker in the cases of normal closing, closing refusal, loose base and connecting rod falling off. The vibration signal in the time domain is processed in segments, and each segment is decomposed by CEEMDAN, and then the kurtosis of each modal component is obtained. Then, the SVM algorithm is used for pattern recognition for each segment. Finally, it is decided by voting to which signal the vibration signal belongs. This is a circuit breaker fault diagnosis method, which first segments the signal and then processes the signal, and then extracts features from the decomposed signal. First, the state of each segment signal is determined in segments, and finally the state of the overall signal is determined according to the state of each segment, which reduces the probability of overall signal judgment error due to errors in a certain process during the signal extraction process. It can increase the fault tolerance rate of circuit breaker fault diagnosis, and can meet the accuracy and speed requirements of on-site fault diagnosis of circuit breakers in operation, and quickly and accurately identify. The steps are as follows:

[0008] (1) Install the circuit breaker vibration sensor device, adjust the sensor parameters, collect the circuit breaker vibration signal, and start the test and calculation analysis process;

[0009] (2) According to the physical process of circuit breaker operation, the waveform in the time domain is observed, and the horizontal axis of the waveform in the time domain is converted into the horizontal axis of the sampling point number. The sampling point number moment of 5% of the interval from the trough to the peak after each physical moment is taken as the segmentation point, and divided into four segments.

[0010] (3) Perform CEEMDAN decomposition on each physical time segment and extract the first four modal components as research objects.

[0011] (4) The kurtosis of the decomposed signal is calculated and the statistical significance of the kurtosis is used as the characteristic value for pattern recognition, which can further improve the fault recognition rate during the operation of the circuit breaker and can be specific to the action type.

[0012] (5) Use the SVM algorithm for pattern recognition and compare the kurtosis of each segment with the standard library of normal, closing refusal, base looseness, and connecting rod detachment to identify whether the measured signal segment is normal or has some kind of fault.

[0013] (6) After determining the fault condition of each signal segment, a voting mechanism is introduced to determine the fault condition of the overall signal.

[0014] (7) The method is verified using the collected data samples to obtain the accuracy of fault diagnosis of the circuit breaker, which serves as a sign of the practicality of the method.

[0015] Step (1) selects an AD50S piezoelectric acceleration sensor to collect vibration signals. The sensor is made of piezoelectric ceramics as piezoelectric materials, stainless steel as shell materials, sensitivity: 50mV / g, frequency response: 0.5-15000Hz, installation resonant frequency: 27kHz, range 0-100g, bias voltage: 8-12V, working environment temperature range: -40-120°C. The vibration sensor is installed on the outer casing of the circuit breaker box.

[0016] Step (2) takes the number of sampling points at the trough of each physical moment plus 5% of the sampling points in the interval from the trough to the peak as the segmentation points at the start stage, and divides the signal into four segments. In the time domain diagram, there is a common time from the complete end of the previous physical process of the circuit breaker to the beginning of the next physical process. Therefore, a period of time after the trough moment (5% of the sampling points in the interval from the trough to the peak) is taken to reduce the waveform aliasing between the physical processes and improve the accuracy of the segmentation. The standard signals of normal, closing refusal, base looseness and connecting rod detachment are segmented, and the number of sampling points corresponding to the impact peak value corresponding to each physical process is used as the x quantity to obtain the segmented signal x(t). The time domain signal of the normal operation of the circuit breaker ( Figure 1 ) Take the signal trough of each actual action interval as the discontinuity point, x1∈(1, 275), x2∈(275, 615), x3∈(615, 1005), x4∈(1005, 8005). Determine the segmentation point, and the total process can be divided into the following processes: 1. t1-t2 Closing contactor action. 2. t2-t3 Closing coil core hits the tripping device. 3. t3-t4 The process from moving and static contacts to full contact. 4. t4-t5 Movement of the connecting rod mechanism.

[0017] Step (3) performs CEEMDAN decomposition on the vibration signal x(t) of the time segments divided in step (2) to generate modal components. CEEMDAN is an adaptive decomposition algorithm that effectively solves the problem of modal aliasing. White noise with a standard normal distribution is added to the target signal, and the components obtained by multiple empirical mode decompositions are averaged as the final actual components. The problem of noise residual in the reconstructed signal can be effectively eliminated, and the modal aliasing problem can be well handled.

[0018] Definition E j (·) is the jth modal component obtained by EMD decomposition, and the kth modal component obtained by CEEMDAN decomposition is recorded as V i (t) represents the addition of a white noise sequence with a standard normal distribution in the i-th trial, ε j-1 Represents the signal-to-noise ratio coefficient of the jth modal component. The original signal s(t) plus the noise component constitutes s(t)+ε0V i (t) Carry out I tests. The first-order modal component is obtained by EMD decomposition And the residual signal r1(t) of the first stage is:

[0019]

[0020]

[0021] Then we can get the i-th residual signal r k (t) and the i+1th order modal component for:

[0022]

[0023]

[0024] Step (4) takes the first four modal components of each segment as the object of investigation, and calculates the kurtosis of the first four modal components of each segment. Kurtosis is a description of whether the peak of the distribution composed of the sample is abrupt or flat. Calculate the kurtosis of the time series x. Kurtosis is used to measure the deviation of x from a certain distribution. The kurtosis of the normal distribution is 3. When the peak of the time series curve is higher than that of the normal distribution, the kurtosis is greater than 3; when it is lower than that of the normal distribution, the kurtosis is less than 3.

[0025] Step (5) In many cases, when the data distribution is complex or the noise in the data is serious, there is a risk of error when the decision is based on the closest sample. The working environment of the circuit breaker is very complex and harsh, and there are noise and other influencing factors around it. SVM transforms the problem into a convex quadratic programming problem and has good processing capabilities for small sample situations. In order to solve the problem of linear inseparability of the sample set, slack variables and penalty factors are added and the samples are mapped to a high-dimensional space through a kernel function to make them linearly separable. The optimal classification hyperplane is constructed, and the Gaussian radial basis function is used as the kernel function. The problem is transformed into a constraint problem:

[0026]

[0027] Combined with the Lagrangian function, the nonlinear support vector machine classifier can be expressed as:

[0028]

[0029] Where: ω is the weight coefficient of the hyperplane; C is the penalty factor; δ i is the slack variable, δ i ≥0; b is the classification threshold; a i is the pull-type multiplier obtained in sample training; k(x i ,x) is the kernel function.

[0030] Step (6) obtains a total of 64 groups of data corresponding to each group through steps (3) to (4), and constructs a data resource library for pattern recognition based on the calculation results of the source signal. Then input 200 groups of source signals measured by sensors under fault or normal conditions, and also go through steps (2) to (5) to obtain 16 groups of data for each corresponding signal. Use the SVM algorithm to calculate the distance measurement of the corresponding segment, so as to determine the fault type to which each segment belongs.

[0031] Step (7) obtains the result of the fault type to which each segment belongs, introduces a voting mechanism, and gives an initial weight to each segment. When using data samples for training, the accuracy is calculated every fifty trials, and the weight is changed for each segment. The accuracy of the fifty groups of data is approximated as the correct diagnosis probability. The changed weight is the previous weight plus the correct diagnosis probability multiplied by the previous weight. After training, the high-weighted segment indicates that the segment signal has a higher accuracy rate for fault classification and has a higher decision power for the fault type to which the overall signal belongs. Determine what type of fault the overall signal belongs to.

[0032] The weight update formula is:

[0033] ω (i) =ω (i-1) +Pω (i-1)

[0034] Step (8) verifies the method using the collected data samples to obtain the accuracy of fault diagnosis for the circuit breaker. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Time domain segment diagram of normal, closing refusal, connecting rod falling off, base loosening

[0036] Figure 2 CEEMDAN decomposition diagram after segmentation under normal conditions

[0037] Figure 3 Exploded view of CEEMDAN after segmentation in case of closing refusal

[0038] Figure 4 CEEMDAN decomposition diagram of the segmented connecting rod falling off

[0039] Figure 5 Exploded view of CEEMDAN after segmentation with loose base

[0040] Figure 6 Fault diagnosis recognition rate diagram of this method DETAILED DESCRIPTION

[0041] Taking the opening process of a ZN63-12 high-voltage circuit breaker as an example, the specific implementation of the present invention is as follows:

[0042] Step 1: Install the AD50S piezoelectric acceleration sensor on the circuit breaker housing. Collect vibration signals of the circuit breaker in normal, closing refusal, connecting rod falling off and base loose conditions. Collect 50 sets of signals for each condition, for a total of 200 sets of data.

[0043] Step 2: Segment the standard signal in the time domain. As a standard, the vibration signal can be divided into the following parts according to the physical action process of the circuit breaker: closing contactor action, closing coil core hitting the tripping device, the process from contact to full contact between the moving and static contacts, and the movement of the connecting rod mechanism. According to the vibration signal in the time domain, it can be seen that from one action process to the next, there is a process from signal attenuation to sudden increase. Therefore, the trough is used as the segmentation point. If the peak is used as the segmentation point, the process is not completed at the peak, resulting in unreasonable segmentation. Therefore, we use the time domain signal of the normal operation of the circuit breaker ( Figure 1 ) Take the signal trough of each actual action interval as the discontinuity point and determine the segmentation points 275, 615, 1005, 8005. The total process can be divided into the following processes: 1. Closing contactor action. 2. Closing coil core hits the tripping device. 3. The process from moving and static contacts to full contact. 4. Linkage mechanism movement.

[0044] In the third step, MATLAB is used to perform CEEMDAN decomposition on the vibration signal x(t) of the time segment divided in the second step, decompose it into several modal components, and calculate the power spectrum of each IMF component. The main peak frequency of the power spectrum of IMF1~IMF4 with larger amplitudes is greater than 1kHz, which is the main component of the original signal frequency structure. The power spectrum frequency and amplitude after IMF5 drop sharply, which is due to the significant attenuation of the vibration level after the signal propagates through the medium and the transmission of the operating mechanism. Therefore, the first four modal components are taken as the research objects to extract the kurtosis.

[0045] Step 4: Take the first four components of the modal components obtained in each segment as the object and calculate the kurtosis as the eigenvalue. Kurtosis describes whether the peak of the distribution composed of samples is abrupt or flat. Calculate the kurtosis of the time series x, which is used to measure the deviation of x from a certain distribution.

[0046] For a data with a capacity of n, the calculation method of sample kurtosis is:

[0047]

[0048] Due to space limitations, only one set of data is taken here for each of the normal, closing refusal, connecting rod detachment and base looseness to calculate the skewness of the first four modal components of each signal segment.

[0049] Table 1 Kurtosis of modal components under four operating conditions

[0050]

[0051] In step 5, the cross-validation method was used to set the test set labels to 1-5 in sequence. Ten sets of samples (40 sets in total) were selected from the 200 sets of data, including normal, closing refusal, connecting rod detachment, and base looseness, as the training set to train the SVM classifier. The remaining 160 sets of data were used as the test set to test the recognition effect of the classifier. The training set and the test set were swapped and the recognition effect test was repeated. The SVM recognition algorithm uses the radial basis kernel function to analyze the multi-classification problem with a one-to-many strategy, and finally determines that the penalty factor c and the RBF kernel function parameter g are 1.568 and 1.885 respectively. The SVM classifier can greatly improve the prediction efficiency while ensuring the prediction accuracy.

[0052] In step 6, 200 sets of data including normal, closing refusal, connecting rod detachment and base loosening signals are used to train the weights of each segment. With each weight change, the weight of the segment with high accuracy will increase faster than the segment with low accuracy. Finally, the voting mechanism is used to determine which fault type the overall signal belongs to.

[0053] Step 7: The prediction accuracy of the fault diagnosis method is as follows: Figure 6As shown in the figure, the accuracy rate can reach 99% in the case of closing refusal, and the accuracy rates of the other three types of normal, connecting rod falling off and base loosening are 98.4%, 97.6% and 96.8% respectively. It has practical significance.

Claims

1. A detection method for circuit breaker fault diagnosis, characterized in that: Includes steps: (1) Install the circuit breaker vibration sensor device, adjust the sensor parameters, collect the circuit breaker vibration signal, and start the test and calculation analysis process; (2) According to the physical process of circuit breaker operation, the waveform in the time domain is observed, and the horizontal axis of the waveform in the time domain is converted into the horizontal axis of the sampling points, and the trough of each physical moment is taken as the segmentation point to divide it into four segments. (3) Perform CEEMDAN decomposition on each physical time segment. (4) The kurtosis of the decomposed signal is calculated and the statistical significance of the kurtosis is used as the characteristic value for pattern recognition, which can further improve the fault recognition rate during the operation of the circuit breaker and can be specific to the action type. (5) Use the SVM algorithm for pattern recognition and compare the kurtosis of each segment with the standard library of normal, closing refusal, base looseness, and connecting rod detachment to identify whether the measured signal segment is normal or has some kind of fault. (6) After determining the fault condition of each signal segment, a voting mechanism is introduced to determine the fault condition of the overall signal. (7) The method is verified using the collected data samples to obtain the accuracy of fault diagnosis of the circuit breaker, which serves as a sign of the practicality of the method.

2. The method for segmenting the collected circuit breaker vibration signal according to claim 1, characterized in that: According to the actual physical process of the circuit breaker, it is divided into the following four processes: the closing contactor action is completed, the closing coil core hits the tripping device action is completed, the moving and static contacts are completed from contact to full contact, and the connecting rod mechanism movement is completed. The vibration signal is segmented according to these physical processes.

3. The method for segmenting the collected circuit breaker vibration signal according to claim 1, characterized in that: The number of sampling points at the trough of each physical moment plus 5% of the sampling points from the trough to the peak are taken as segmentation points to divide the signal into four segments. In the time domain diagram, there is a common time from the end of the previous physical process of the circuit breaker to the beginning of the next physical process. Therefore, a period of time after the trough moment (5% of the sampling points from the trough to the peak) is taken to reduce the waveform aliasing between physical processes and improve the accuracy of segmentation.

4. According to claim 1, the CEEMDAN decomposition of the segmented circuit breaker in the conditions of normal, closing refusal, connecting rod falling off and base loosening is performed, characterized in that: Calculate the power spectrum of each IMF component. The main peak frequencies of the power spectra of IMF1 to IMF4 with larger amplitudes are all greater than 1kHz, which are the main components of the original signal frequency structure. The power spectrum frequency and amplitude drop sharply after IMF5, which is due to the significant attenuation of the vibration level after the signal propagates through the medium and the transmission of the operating mechanism. Therefore, the first four modal components of each segment are taken as the research objects.

5. The method of using the kurtosis of the vibration signal extracted and segmented and then subjected to CEEMDAN decomposition as a feature quantity according to claim 1, characterized in that: Kurtosis is a description of whether the peak value of the distribution of the sample is abrupt or flat. Using the kurtosis of the modal component of the vibration signal as a characteristic quantity of the vibration signal when analyzing the circuit breaker fault can better achieve the purpose of accurate identification.

6. The method for circuit breaker fault diagnosis based on SVM according to claim 1, characterized in that: The cross-validation method is used, and the test set labels are set to 1-5 in sequence. Ten sets of samples (40 sets in total) of the 200 sets of data are used as training sets to train the SVM classifier, and the remaining 160 sets of data are used as test sets to test the recognition effect of the classifier. The training set and the test set are swapped and the recognition effect test is repeated. The SVM recognition algorithm uses the radial basis kernel function to analyze multi-classification problems with a one-to-many strategy. The SVM classifier can greatly improve the prediction efficiency while ensuring the prediction accuracy.

7. The method for determining which fault condition the overall signal belongs to by introducing a voting mechanism according to claim 1, characterized in that: An initial weight is given to each segment. When training with data samples, the accuracy is calculated every fifty trials. The weight is changed for each segment, and the accuracy of the fifty groups of data is approximated as the probability of correct diagnosis. The changed weight is the previous weight plus the probability of correct diagnosis multiplied by the previous weight. After training, high-weighted segments indicate that the segment has a higher accuracy rate for fault classification and has a higher decision power for the fault type of the final overall signal. Determine what type of fault the overall signal belongs to. The weight update formula is: oh (i) =ω (i-1) +Pω (i-1) 。

Images