Circuit breaker fault detection method and system

Abstract

The invention discloses a circuit breaker fault detection method and system, and relates to the technical field of circuit breaker fault detection, and the method comprises the steps: carrying out the data collection of a circuit breaker, and carrying out the first enhancement processing; encoding the data, and performing first feature optimization; and optimizing characteristic attributes, and identifying a fault mode. According to the circuit breaker fault detection method provided by the invention, the sound signal processing process of human ears is simulated, the auditory characteristic of the sound signal of the power equipment is extracted, and meanwhile, the signal information can be reflected more comprehensively by combining the auditory characteristic with the time-frequency characteristic in consideration of the limited signal information reflecting capability of a single characteristic. All the processes adopt intelligent automatic processes. The intelligent and automatic fault detection method can effectively reduce the labor cost and time, greatly improves the operation and maintenance efficiency, and improves the fault response speed and the maintenance accuracy at the same time. The method has better effects in the aspects of operation and maintenance efficiency, cost and comprehensiveness.

Description

Technical Field

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

[0002] 10kV circuit breakers usually operate in harsh environments with high humidity, high temperature and high voltage, which can easily lead to premature aging and mechanical failure of the equipment. Due to the working environment and operating requirements of 10kV circuit breakers, accurately identifying mechanical failures of circuit breakers in power distribution equipment is a comprehensive problem.

[0003] Currently, most detection methods in the market use displacement sensors to detect characteristic parameters such as the opening and closing time, total stroke, overtravel and bounce of the circuit breaker mechanism, and evaluate the health of the circuit breaker through the detection data. This detection method can only determine the health status of the second half of the stroke of the circuit breaker mechanism, and cannot warn of front-end faults such as opening and closing coil failure and mechanism transmission jamming.

[0004] Voiceprint recognition technology can achieve all-round monitoring of circuit breaker faults by extracting and identifying the voiceprint signals generated by the opening and closing coil current, circuit breaker transmission mechanism and auxiliary contacts during operation, and comparing them with standard values. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is that the existing circuit breaker fault detection method cannot determine the state of the front half of the short-circuit mechanism, cannot warn of the opening and closing coil fault, the mechanism transmission is stuck, and how to perform voiceprint recognition.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a circuit breaker fault detection method, comprising collecting data from the circuit breaker and performing a first enhancement process; encoding the data and performing a first feature optimization; optimizing feature attributes and identifying fault modes.

[0008] As a preferred solution of the circuit breaker fault detection method of the present invention, wherein: the data collection of the circuit breaker includes using a sensor to collect the circuit breaker operation sound signal.

[0009] As a preferred solution of the circuit breaker fault detection method of the present invention, the first enhancement processing includes optimizing the quality of the sound signal.

[0010] As a preferred solution of the circuit breaker fault detection method of the present invention, wherein: the encoding of data includes encoding of sound signals.

[0011] As a preferred solution of the circuit breaker fault detection method of the present invention, the first feature optimization includes extracting features from the sound signal.

[0012] As a preferred solution of the circuit breaker fault detection method of the present invention, wherein: the optimizing feature attributes comprises using linear discriminant analysis to perform preliminary dimension reduction on the joint cepstral coefficients, retaining the main category distinguishing features;

[0013] Kernel principal component analysis was used to further reduce the dimensionality of the nonlinear structure of the data;

[0014] The reduced feature vector is input into the MobileNet_YOLOv4 neural network model for training to generate a fault recognition model.

[0015] As a preferred solution of the circuit breaker fault detection method of the present invention, wherein: the identification of the fault mode includes constructing a MobileNet_YOLOv4 neural network structure under the PyTorch framework, and performing model training based on the circuit breaker operation data set;

[0016] The initial learning rate and number of iterations are set to optimize the model parameters, the accuracy and recall of the model are evaluated using the validation set and test set, and the recognition performance is improved by optimizing the algorithm.

[0017] Another object of the present invention is to provide a circuit breaker fault detection system, which can extract the auditory characteristics of the sound signal of the power equipment by imitating the processing process of the human ear on the sound signal, thereby solving the problem that the current circuit breaker fault detection method contains the inability to perform voiceprint detection.

[0018] As a preferred solution of the circuit breaker fault detection system of the present invention, it includes:

[0019] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of a circuit breaker fault detection method.

[0020] A computer-readable storage medium stores a computer program, which implements the steps of a circuit breaker fault detection method when executed by a processor.

[0021] Beneficial effects of the present invention: The circuit breaker fault detection method provided by the present invention extracts the auditory features of the sound signals of power equipment by imitating the human ear's processing of sound signals. At the same time, considering that a single feature has limited ability to reflect signal information, combining auditory features with time-frequency features can more comprehensively reflect signal information. All intelligent automated processes are adopted. This intelligent automated fault detection method can effectively reduce labor costs and time, greatly improve operation and maintenance efficiency, and at the same time improve the speed of fault response and the accuracy of maintenance. The present invention achieves better results in terms of operation and maintenance efficiency, cost, and comprehensiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 This is an overall flow chart of a circuit breaker fault detection method provided by the first embodiment of the present invention.

[0024] Figure 2 A GFCC coefficient diagram of a closing sound signal of a circuit breaker fault detection method provided by the first embodiment of the present invention.

[0025] Figure 3 A sound signal combined with cepstral coefficient diagram of a circuit breaker fault detection method provided by the first embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0027] Example 1, reference Figure 1-Figure 3 , as an embodiment of the present invention, provides a circuit breaker fault detection method, comprising:

[0028] S1: Collect data from the circuit breaker and perform a first enhancement process.

[0029] Furthermore, collecting data from the circuit breaker includes using a sensor to collect an operating sound signal of the circuit breaker.

[0030] It should be noted that performing the first enhancement processing includes optimizing the quality of the sound signal.

[0031] The first enhancement may be performed by various methods, such as voiceprint signal denoising based on smoothing filtering or voiceprint signal denoising based on modal decomposition.

[0032] As a preferred embodiment, in this embodiment, the first enhancement processing adopts voiceprint signal denoising based on smoothing filtering, and its principle steps are as follows:

[0033] (1) Moving filtering: The continuously collected voiceprint data is regarded as a set of sequences with a fixed length of N. After re-measurement, the first bit of the above sequence is removed, and the remaining N-1 data are moved forward in sequence. The second collected data is re-inserted into the end of the new sequence. After arithmetic operations are performed on the sequence, the result is the value required for this calculation.

[0034] When processing data, moving average filtering is often used for denoising. Assuming the input is x and the output is y, the calculation formula of the Moving filtering algorithm is as follows:

[0035]

[0036] Among them, y(n) is the output value at time n, which represents the result after filtering. x(n) is the input data at time n, usually the latest sampling point. N is the window size, which represents the number of sampling points used for averaging.

[0037] (2) Savitzky-Golay filtering can be used for polynomial fitting based on the least squares principle. It is a filtering method that uses the least squares method to fit in the time domain. The average value obtained after processing can be expressed as:

[0038]

[0039] Among them, x ksmooth is the value after smoothing at position k, i.e., the filtering result. ω is the half width of the smoothing window, which determines how many adjacent data points are included. 2ω+1 is the total width of the window, including the current point and its adjacent points on the left and right. k+i is the data value near position k. i is the index offset in the range -ω to +ω.

[0040] (3) Lowess filtering algorithm: Since the relationship between different variables is subtle, the equation alone cannot achieve the desired effect. The Lowess filtering algorithm uses a fitted trend line as the basic line, and the data that deviates from this basic line are the abnormal data points that need to be filtered out.

[0041] In another feasible embodiment, the first enhancement adopts the voiceprint signal denoising based on modal decomposition, and its principle steps are as follows: filtering and purifying the collected circuit breaker operation sound signal includes adopting the EMD denoising method based on multi-scale analysis, and performing high-frequency zero padding on the noise-dominated high-frequency IMFs to change its resampling rate, so that the EMD can perform the next-level decomposition of the IMFs that are mainly noise components in the first-level decomposition to obtain the corresponding IMFs in the second-level decomposition. Then distinguish the noise-dominated IMFs and information-dominated IMFs in the second-level decomposition.

[0042] Specifically, the traditional denoising method CMSE based on EMD is first used to divide the IMFs in the first-level decomposition into two groups, one group is low-frequency IMFs considered to be information components, and the other group is high-frequency IMFs considered to be noise components. The EMD denoising method based on multi-scale analysis performs high-frequency zero padding on the noise-dominated high-frequency IMFs to change their resampling rate, so that EMD can perform the next-level decomposition of these IMFs that are mainly noise components in the first-level decomposition to obtain the corresponding IMFs in the second-level decomposition. Then, the noise-dominated IMFs and information-dominated IMFs in the second-level decomposition are distinguished. The scaling index is calculated using DFA to study the statistical relationship between IMFs in the second-level decomposition. The larger the signal scaling index, the smoother its oscillation trend, so more components in the signal can be regarded as information components. On the other hand, the smaller the signal scaling index, the higher the signal oscillation frequency, so more components in the signal can be regarded as noise components. In order to further extract effective information from the noise-dominated IMFs in the second-level decomposition, the threshold processing strategy will be used in combination with multi-scale analysis based on EMD.

[0043] S2: Encode the data and perform first feature optimization.

[0044] Furthermore, encoding data includes encoding a sound signal.

[0045] It should be noted that performing the first feature optimization includes extracting features from the sound signal.

[0046] The first feature optimization may be performed by various methods, such as methods based on Mel Filter Cepstral Coefficients (MFCC) or Gamma Filter Cepstral Coefficients (GFCC).

[0047] As a preferred embodiment, in this embodiment, the first feature optimization is based on Mel-frequency cepstral coefficients (MFCC), and its principle steps are as follows: the GFCC coefficient of the sound signal can realize the encoding and compression of the sound signal information by imitating the perception characteristics of the human ear to the sound signal, and the GFCC coefficient is applied to the feature extraction of the circuit breaker closing sound signal, which can effectively extract the spectrum envelope and details of the closing sound signal in different mechanical states, and then reflect the frequency domain distribution characteristics of the closing sound signal.

[0048] The Gammatone filter bank performs an equivalent to the impulse response characteristics of the basilar membrane, thereby achieving auditory filtering. The center frequencies of the filters are equidistant from the positions of the basilar membrane, and the step response function Bm(t) of the mth filter is shown in the following formula.

[0049]

[0050] Where V is the filter gain. β is the filter order. f m is the center frequency of the mth filter. m is the initial phase. θ(t) is the step function. c m is the attenuation factor, which is related to f m The relationship is:

[0051]

[0052] The result of blind separation of sound signal during the closing time of the circuit breaker As a frame. To extract the i-th sound signal component As an example, the GFCC coefficient of the sound signal component Perform windowing and then use fast Fourier transform (FFT) to get its spectrum sequence X i (k), X i (k) The output of the Gammatone filter bank is exponentially compressed. The compression factor is set to 0.2 based on experience. The mth filter performs exponential compression on the signal. The filtering and compression process can be expressed as

[0053]

[0054] in, is the compression result. n is X i (k) is the number of FFT points. m (k) is the mth filter in X i (k) The gain of the kth point corresponding to the frequency.

[0055] The compressed output of each filter is decorrelated by discrete cosine transform, and the result is the sound component The GFCC cepstrum coefficient, discrete cosine transform formula is

[0056]

[0057] Among them, d i (j) is the j-th dimension cepstrum coefficient. I is the number of filters.

[0058] According to Shannon sampling theorem, the bandwidth of the Gammatone filter group is set to 22.05kHz. In a noise-free environment, a closing sound signal is one frame, the signal frame length is 3s, the number of filters is 64 based on experience, the order is 4, and the GFCC coefficient corresponding to the sound signal is as follows: Figure 2 shown.

[0059] In another feasible embodiment, the first feature optimization adopts the gamma filter cepstral coefficient (GFCC), and its principle steps are as follows: Similar to the Gammatone filter bank, the Mel filter bank is a triangular pass filter bank equivalent to the nonlinear characteristics of human ear pitch perception, and each filter bandwidth is equal under the Mel scale. The conversion relationship between the sound frequency f and the Mel scale frequency Mel (f) is shown in the following formula:

[0060]

[0061] The frequency response of the mth Mel filter is H m (k) is shown in the following formula:

[0062]

[0063] Where k is the frequency and f(m) is the center frequency of the mth filter.

[0064] Similar to GFCC, the spectrum signal passing through the Mel filter is exponentially compressed, and the formula is:

[0065]

[0066] in, is the compression result. n is X i (k) is the number of FFT points. m (k) is the mth filter in X i (k) The gain of the kth point corresponding to the frequency.

[0067] The compressed output of each filter is decorrelated by discrete cosine transform, and the result is the sound component The MFCC cepstrum coefficients, discrete cosine transform formula is

[0068]

[0069] Among them, d i (j) is the j-th dimension cepstrum coefficient. I is the number of filters.

[0070] The exponentially compressed MFCC and GFCC coefficients of different types of sound signals are extracted using a filter bank with similar auditory effects. The number of Mel filter banks and Gammatone filter banks is selected from empirical values. The number of Mel filter banks is 26, the number of Gammatone filter banks is 64, and the order is 4. According to Shannon sampling theorem, the filter bandwidth is set to 22.05kHz. The joint cepstrum coefficients of the sound signal are as follows: Figure 3 shown.

[0071] S3: Optimize feature attributes and identify failure modes.

[0072] Furthermore, optimizing feature properties involves performing preliminary dimensionality reduction on the joint cepstral coefficients using linear discriminant analysis, retaining the main class-distinguishing features.

[0073] Kernel principal component analysis is used to further reduce the dimensionality of the nonlinear structure of the data.

[0074] The reduced eigenvectors are input into the MobileNet_YOLOv4 neural network model for training to generate a fault recognition model. The feature extraction method is used to simplify the feature attributes, including linear dimensionality reduction and nonlinear dimensionality reduction. The extracted joint cepstral coefficients are weighted by LDA and then KPCA dimensionality reduction is performed. The non-zero eigenvalues ​​of the kernel matrix are arranged from large to small. The cumulative contribution rate trend is shown in the figure. The joint cepstral coefficients are weighted by LDA and then KPCA dimensionality reduction is performed. The non-zero eigenvalues ​​of the kernel matrix are arranged from large to small. The cumulative contribution rate trend is shown in the figure. According to the criterion that the cumulative contribution rate is greater than 85%, the eigenvectors corresponding to the top 5 eigenvalues ​​are selected to form the dimensionality reduction matrix G, and the eigenvector is reduced from 62 dimensions to 5 dimensions. In order to reflect the linear separability of the eigenvector in the feature space after dimensionality reduction, the feature data of the first 3 dimensions are selected for visualization. The results are shown in Figure 3 As shown, it can be seen that the feature vector after optimization and dimensionality reduction has good separability in three-dimensional space, which is conducive to subsequent pattern recognition.

[0075] It should be noted that identifying the failure mode includes building a MobileNet_YOLOv4 neural network structure under the PyTorch framework and training the model based on the circuit breaker operation dataset.

[0076] The initial learning rate and number of iterations are set to optimize the model parameters. The accuracy and recall of the model are evaluated using the validation set and test set, and the recognition performance is improved by optimizing the algorithm. In order to enable the model to accurately identify and classify the normal mode and various fault types in the voiceprint signal of the circuit breaker, the circuit breaker opening and closing fault recognition model needs to be trained. The program for building the neural network structure of the MobileNet_YOLOv4 neural network model and the program for testing the network performance are written and debugged under the PyTorch 1.8 framework. The operating environment is Linux system Ubuntu 16.04. The operating environment CPU is Intel(R) Core(TM) i7-10750H, the GPU is RTX20606GB, and the entire training process is carried out in a 6GB independent graphics card environment. During the training process, the number of categories to be classified is set to 2, where n7 and n8 represent the labeled data sets of normal samples and the labeled data sets of fault samples, respectively. The validation set is planned to be 123 groups and the test set is 315 groups. The initial learning rate was set to 0.002, and the number of iterations of the entire training process was 10,000. The entire training process took 9 hours.

[0077] Embodiment 2, an embodiment of the present invention, provides a circuit breaker fault detection method. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0078] First, an experiment was conducted to verify the innovation and effectiveness of the intelligent detection process of mechanical faults of circuit breakers. The experiment selected the circuit breaker model "DW45-1000" as the test object, collected the operating sound signals of the circuit breaker in normal operation and fault state, and implemented it in detail based on the three steps in the invention content.

[0079] The experiment simulated the operation of the circuit breaker in a laboratory environment, and used a high-sensitivity sound sensor (model: B&K4189) to collect the operating sound signal of the circuit breaker. The sampling frequency was set to 22.05kHz, and the recording time was 5 seconds. After the acquisition was completed, the quality of the sound signal was optimized, and the environmental noise was removed by combining the moving average filter (Moving filter) and the Savitzky-Golay filter. The moving average filter window was set to 20 points, the Savitzky-Golay filter polynomial order was set to 3, and the smoothing window was 21 points. Finally, a voiceprint signal with a significantly improved signal-to-noise ratio was obtained.

[0080] The feature extraction of the filtered voiceprint signal is performed, and the Mel-frequency cepstral coefficient (MFCC) algorithm is used to simulate the auditory perception of the human ear to the sound signal, and the spectral characteristics of the mechanical state of the circuit breaker are extracted. The specific parameters are 26 filter groups, the frequency band is set to 20Hz to 8kHz, and the window length is 512 points. The extracted features include 13-dimensional cepstral coefficients of each frame signal. Through the visual analysis of the signal features under normal and fault conditions, it is found that the fault state has significant characteristic changes, such as the increase in the mean value and coefficient of variation of the cepstral coefficients.

[0081] The extracted features were reduced in dimension by linear discriminant analysis (LDA) and kernel principal component analysis (KPCA), and the 62-dimensional joint cepstral coefficients were reduced to 5 dimensions. Subsequently, the optimized features were input into the MobileNet_YOLOv4 neural network model for training and testing. The experiment used the PyTorch 1.8 framework, the GPU was RTX 20606GB, the training data set contained 500 groups of normal samples and 500 groups of fault samples, and the validation set and test set contained 100 groups of data respectively. The initial learning rate of the model was set to 0.002, the number of training iterations was 10,000, and the training time was about 9 hours.

[0082] Table 1 Experimental data comparison table

[0083]

[0084]

[0085] In the first enhancement processing stage, through the joint application of moving average filtering and Savitzky-Golay filtering, experimental data show that the signal-to-noise ratio in the normal state is stable between 40.5-41.0dB, while the fault state is 29.5-30.2dB. This shows that the signal in the normal state is purer and the noise component is greatly reduced. Compared with the existing technology where a single filtering method can only increase the signal-to-noise ratio to about 35dB, the present invention can still achieve excellent denoising effect under complex experimental conditions.

[0086] The experimental results show that the MFCC mean value in the normal state is 12.1-12.3, and the coefficient of variation is only 0.05-0.06, indicating that the signal has stability and consistency in frequency domain distribution. However, the MFCC mean value in the fault state increases significantly to 18.4-18.9, and the coefficient of variation increases significantly to more than 0.20, indicating that the signal spectrum characteristics are obviously abnormal. This difference provides a reliable feature basis for subsequent pattern classification.

[0087] After feature dimensionality reduction, the experiment optimized the 62-dimensional joint cepstrum feature to 5 dimensions, effectively reducing the model calculation complexity while maintaining the integrity of the feature information. The model classification accuracy is over 98% under normal conditions, slightly lower in fault conditions but still maintained at over 94%. Compared with the prior art, the classification accuracy of traditional time domain and frequency domain features mostly hovers around 90%, especially in complex noise environments. The present invention can maintain excellent classification performance under complex experimental conditions by combining feature optimization with a neural network model.

[0088] Existing technologies usually rely on a single feature extraction method, which cannot effectively deal with multiple types of noise or complex spectral distribution, resulting in large fluctuations in classification performance. The present invention comprehensively improves the robustness and recognition ability of signal processing through the combination of multi-stage signal enhancement, feature optimization and deep learning models.

[0089] Embodiment 3, an embodiment of the present invention, provides a circuit breaker fault detection system, including a data acquisition module, a feature extraction module, and a fault analysis module.

[0090] The data acquisition module is used to collect data from the circuit breaker and perform a first enhancement process. The feature extraction module is used to encode the data and perform a first feature optimization. The fault analysis module is used to optimize feature attributes and identify fault modes.

[0091] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0092] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0093] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0094] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiment, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit with a logic gate circuit for implementing a logical function for a data signal, a dedicated integrated circuit with a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc. It should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not limited. Although the present invention is described in detail with reference to the preferred embodiment, it should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, which should be included in the scope of the claims of the present invention.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A circuit breaker fault detection method, characterized in that: include: Collecting data from the circuit breaker and performing a first enhancement process; Encode the data and optimize the first feature; Optimize feature attributes and identify failure modes.

2. The circuit breaker fault detection method according to claim 1, characterized in that: The data collection of the circuit breaker includes collecting the circuit breaker operation sound signal using a sensor.

3. The circuit breaker fault detection method according to claim 2, characterized in that: The performing the first enhancement processing includes optimizing the quality of the sound signal.

4. The circuit breaker fault detection method according to claim 3, characterized in that: The encoding of data includes encoding a sound signal.

5. The circuit breaker fault detection method according to claim 4, characterized in that: The first feature optimization includes extracting features from the sound signal.

6. The circuit breaker fault detection method according to claim 5, characterized in that: The optimizing feature attributes includes performing preliminary dimensionality reduction on the joint cepstral coefficients using linear discriminant analysis to retain the main category distinguishing features; Kernel principal component analysis was used to further reduce the nonlinear structure of the data; The reduced feature vector is input into the MobileNet_YOLOv4 neural network model for training to generate a fault recognition model.

7. The circuit breaker fault detection method according to claim 6, characterized in that: The identifying of the fault mode includes constructing a MobileNet_YOLOv4 neural network structure under a PyTorch framework and performing model training based on a circuit breaker operation data set; The initial learning rate and number of iterations are set to optimize the model parameters, the accuracy and recall of the model are evaluated using the validation set and test set, and the recognition performance is improved by optimizing the algorithm.

8. A system using the circuit breaker fault detection method according to any one of claims 1 to 7, characterized in that: Including data acquisition module, feature extraction module, and fault analysis module; The data acquisition module is used to collect data from the circuit breaker and perform a first enhancement process; The feature extraction module is used to encode the data and perform first feature optimization; The fault analysis module is used to optimize characteristic attributes and identify fault modes.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the circuit breaker fault detection method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the circuit breaker fault detection method according to any one of claims 1 to 7 are implemented.

Images