Porcelain insulator live detection method and system based on mechanical pulse excitation
Through the porcelain insulator live detection method based on mechanical pulse excitation, vibration response simulation and deep learning model are used for detection, the traditional detection method requires power outage and high cost are solved, and the online, fast and accurate detection of porcelain insulators is achieved.
Patent Information
- Application Number
- CN202411803567.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-04-08
AI Technical Summary
Traditional porcelain insulator detection methods require a power outage, which affects the normal operation of the power system, and has high detection cost and time.
The porcelain insulator live detection method based on mechanical pulse excitation is adopted, and the detection result database is established through vibration response simulation, and the voiceprint characteristics of insulators are extracted using multiple excitation methods and advanced signal processing technologies, and feature fusion and state judgment are combined with deep learning models.
It realizes online, fast and accurate detection of porcelain insulators, reduces inspection costs and time, and can grasp the operating conditions in a timely manner to avoid power outages.
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of porcelain insulator detection in an electric power system, and relates to a porcelain insulator live detection method and system based on mechanical pulse excitation. Background Art
[0002] As a key insulating component of the power system, porcelain post insulators are easily affected by various factors such as production process, material stress, and operating environment. They are prone to defects such as entrapment, pores, and microcracks, which lead to fractures, power outages, and secondary power disasters, causing immeasurable losses. Most traditional porcelain insulator inspection methods need to be carried out under power outage conditions, which not only affects the normal operation of the power system, but also increases the cost and time of inspection. In recent years, with the development of technology, live inspection technology has gradually been applied to the inspection of porcelain insulators. The vibroacoustic method is a relatively new non-destructive testing method in recent years. Compared with other methods for mechanical inspection of post porcelain insulators, it has the advantages of short inspection time, live operation, and no impact on the corona voltage on the top of the insulator. Cracks near the upper and lower flanges can be accurately detected. Summary of the invention
[0003] The present invention proposes a method and system for detecting live porcelain insulators based on mechanical pulse excitation, aiming to realize online, rapid and accurate detection of porcelain insulators, while reducing detection cost and time, so as to timely grasp the operation status and avoid power outage accidents.
[0004] The technical solution of the present invention is achieved in this way:
[0005] Technical Theme 1
[0006] The present invention provides a method for detecting live porcelain insulators based on mechanical pulse excitation, comprising:
[0007] S1: Establishing insulator test result database through vibration response simulation:
[0008] The lower flange side of the insulator is stimulated by knocking and continuous white noise excitation. The response signal of the insulator is detected by the receiver probe to obtain the vibration signal data of the insulator. In the continuous white noise excitation detection, the probes of the white noise signal transmitter and the vibration signal receiver are both pressed against the flange at the bottom of the insulator, and the instrument sends continuous white noise to the insulator through the transmitter probe for excitation.
[0009] The Mel Frequency Cepstral Coefficient (MFCC) algorithm is used to extract the voiceprint features of the vibration signal;
[0010] The extracted features are fused using convolutional neural network (CNN) to form feature vectors and stored in the database;
[0011] S2: Inspection of insulators to be inspected:
[0012] Perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data;
[0013] S3: Comparative analysis:
[0014] The extracted feature values are input into a pre-trained deep BP neural network model, which adds at least one hidden layer to the standard BP neural network and uses the ReLU activation function and dropout strategy for improvement;
[0015] The detection results are output by the deep BP neural network model to determine the state of the insulator.
[0016] Preferably, the step S1 further comprises: performing multi-scale decomposition on the signal using wavelet transform to extract features of different frequency bands;
[0017] The Hilbert-Huang transform (HHT) is used to perform time-frequency analysis on nonlinear and non-stationary signals and extract the instantaneous frequency and amplitude information.
[0018] Preferably, the S1 step further includes:
[0019] The long short-term memory network (LSTM) is introduced to fuse the extracted features to form a feature vector;
[0020] Combined with the Transformer model, it enhances the focus on key features;
[0021] Add noise of different types and intensities to the training data to improve the robustness of the model;
[0022] Generate more diverse training samples through generative adversarial networks (GANs).
[0023] Preferably, the S2 step further includes:
[0024] Use wireless sensor networks for real-time data collection to achieve remote and real-time data collection;
[0025] The collected data is transmitted to the cloud in real time for centralized processing and analysis.
[0026] Preferably, the S2 step further includes:
[0027] Multi-sensor fusion technology is used to combine vibration sensors, temperature sensors and humidity sensors to detect the vibration signals, surface temperature and ambient humidity of the insulators.
[0028] Preferably, the S2 step further includes:
[0029] Perform preliminary data processing and feature extraction on sensor nodes to reduce the amount of transmitted data;
[0030] The preprocessed data is transmitted to the cloud for inference and final judgment of the deep learning model.
[0031] Preferably, the S3 step further includes:
[0032] Add multiple hidden layers to the deep BP neural network model to improve the model's expressiveness;
[0033] Use random forest and gradient boosting machine (GBM) for ensemble learning to improve classification accuracy;
[0034] Dynamically adjust the judgment threshold to improve the sensitivity and accuracy of detection;
[0035] Introduce anomaly detection algorithms to identify and process abnormal data points.
[0036] Through the visual interface, the detection results and status of the insulator are displayed in real time;
[0037] Automatically generate inspection reports including inspection data, analysis results and recommended actions.
[0038] Technical Theme 2
[0039] The present invention also provides a device using the above-mentioned porcelain insulator live detection method based on mechanical pulse excitation, comprising:
[0040] Excitation module, used for continuous excitation of knocking and white noise on the flange side of the lower end of the insulator;
[0041] A receiving module, used for acquiring vibration signal data of the insulator;
[0042] A feature extraction module is used to extract the voiceprint features of the vibration signal;
[0043] Feature fusion module, used to fuse the extracted features to form a feature vector;
[0044] A database module is used to store feature vectors and establish an insulator detection result database;
[0045] The detection module is used to perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data;
[0046] The analysis module is used to analyze the extracted characteristic values, output the detection results, and determine the status of the insulator.
[0047] Technical Theme 3
[0048] The present invention also provides a porcelain insulator live detection system based on mechanical pulse excitation, the system comprising:
[0049] one or more memories for storing instructions; and
[0050] One or more processors are used to call and run the instructions from the memory to execute the method described in the above detection method.
[0051] Technical Theme 4
[0052] The present invention also provides a computer-readable storage medium, the computer-readable storage medium comprising:
[0053] The program, when the program is executed by a processor, the method described in the above detection method is executed.
[0054] The working principle and beneficial effects of the present invention are:
[0055] 1. The present invention provides a method for detecting live porcelain insulators based on mechanical pulse excitation. The method extracts the voiceprint features of insulators through multiple excitation methods and advanced signal processing technology, and uses a deep learning model for feature fusion and state judgment. The device has innovative features such as multi-modal excitation, advanced signal processing, multi-sensor fusion, edge computing and adaptive threshold, which can improve the comprehensiveness, accuracy and real-time performance of detection.
[0056] Among them, multimodal excitation combines multiple excitation methods to improve the comprehensiveness and accuracy of detection;
[0057] Advanced signal processing introduces a variety of signal processing techniques to extract richer feature information;
[0058] Deep learning and self-attention mechanism combine multiple deep learning models to enhance the expressiveness and robustness of the model;
[0059] Multi-sensor fusion and edge computing enable real-time collection and processing of multi-sensor data, improving the real-time and reliability of detection;
[0060] Dynamic adjustment and anomaly detection introduces adaptive thresholds and anomaly detection algorithms to improve detection sensitivity and accuracy.
[0061] 2. The present invention also provides a porcelain insulator live detection device and system based on mechanical pulse excitation. Through practical application analysis, this technology can accurately detect defects in porcelain post insulators, including cracks, fractures, etc., effectively avoiding safety accidents and economic losses caused by insulator failures. It has broad application prospects and can provide more reliable protection for the safe and stable operation of power systems. DETAILED DESCRIPTION
[0062] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0063] Example 1
[0064] A method for detecting live porcelain insulators based on mechanical pulse excitation, comprising:
[0065] S1: Establishing insulator test result database through vibration response simulation:
[0066] The lower flange side of the insulator is stimulated by knocking and continuous white noise excitation. The response signal of the insulator is detected by the receiver probe to obtain the vibration signal data of the insulator. In the continuous white noise excitation detection, the probes of the white noise signal transmitter and the vibration signal receiver are both pressed against the flange at the bottom of the insulator, and the instrument sends continuous white noise to the insulator through the transmitter probe for excitation.
[0067] The Mel Frequency Cepstral Coefficient (MFCC) algorithm is used to extract the voiceprint features of the vibration signal; the specific steps are as follows:
[0068] (1) Data preprocessing
[0069] 1. Signal acquisition
[0070] 2. Pre-filtering: Pre-filter the collected vibration signal to remove high-frequency noise and low-frequency interference.
[0071] 3. Framing: The pre-processed signal is divided into multiple short-time frames. The length of each frame is 20-40 milliseconds, and the frame shift (overlap between adjacent frames) is 10-20 milliseconds.
[0072] (2) Mel-frequency cepstral coefficient (MFCC) extraction
[0073] 1. Fourier transform: Perform a fast Fourier transform (FFT) on each frame to obtain a frequency domain representation.
[0074] 2. Mel filter bank: The frequency domain signal is passed through a set of Mel filters. The design of the Mel filter bank is based on the auditory characteristics of the human ear and can better capture the characteristics of the audio signal. The number of Mel filters is 20-40.
[0075] 3. Logarithmic energy calculation: Calculate the logarithmic energy of each filter output.
[0076] 4. Discrete Cosine Transform (DCT): Perform discrete cosine transform on the logarithmic energy to obtain Mel Frequency Cepstral Coefficients (MFCC). Take the first 12-13 MFCC coefficients as the feature vector.
[0077] 5. Mean Normalization: In order to eliminate DC offset and reduce the impact of background noise, the MFCC coefficients of each frame are mean normalized.
[0078] (3) Feature selection: 12-20 coefficients.
[0079] The extracted features are fused using convolutional neural network (CNN) to form feature vectors and stored in the database. The specific steps are as follows:
[0080] 1. Data preparation
[0081] Input data: extracted audio features.
[0082] Label: If you need to train the model, you also need to prepare corresponding label data.
[0083] 2. Build a CNN model
[0084] Input layer: defines the shape of the input, for example (None, 13, num_frames), where num_frames is the number of time frames.
[0085] Convolutional layers: Add multiple convolutional layers, each of which is usually followed by an activation function (such as ReLU) and a pooling layer (such as max pooling).
[0086] Fully connected layers: The output of the convolutional layer is flattened and then passed through one or more fully connected layers.
[0087] Output layer: Depending on the type of task (classification or regression), define the structure of the output layer. For example, for classification tasks, you can use the Softmax activation function.
[0088] 3. Training the model
[0089] Loss Function: Select a loss function, such as cross entropy loss (for classification tasks).
[0090] Optimizer: Select an optimizer, such as Adam or SGD.
[0091] Training process: Train the model using the training dataset and periodically evaluate the performance on the validation set.
[0092] 4. Extract feature vectors
[0093] Feature extraction: After training is complete, the intermediate layers of the model (usually the last convolutional layer or fully connected layer) can be used to extract feature vectors.
[0094] Shape of feature vector: The shape of the feature vector depends on the chosen layer and is usually a vector of fixed length.
[0095] 5. Store in database
[0096] Database preparation: Select a database system (MySQL) and create a table or collection to store feature vectors.
[0097] Data format: Convert the feature vector to a format suitable for storage, such as JSON or binary data.
[0098] Insert data: Insert the feature vector and its related information (such as file name, label, etc.) into the database.
[0099] S2: Inspection of insulators to be inspected:
[0100] Perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data;
[0101] S3: Comparative analysis:
[0102] The extracted feature values are input into the pre-trained deep BP neural network model, which adds at least one hidden layer on the basis of the standard BP neural network and uses the ReLU activation function and dropout strategy for improvement; the specific steps are as follows: 1. Data preparation
[0103] Input data: Extracted eigenvalues of insulators.
[0104] Label: If you need to train the model, you also need to prepare corresponding label data (for example, normal, damaged, etc.).
[0105] 2. Build a deep BP neural network model
[0106] Input layer: defines the shape of the input, for example (None, num_features), where num_features is the length of the feature vector.
[0107] Hidden layers: Add multiple hidden layers, each of which is usually followed by an activation function (such as ReLU) and a dropout layer.
[0108] Output layer: Depending on the task type (classification or regression), define the structure of the output layer. For example, for a binary classification task, you can use the Sigmoid activation function.
[0109] 3. Load the pre-trained model
[0110] Load model: Load a model that has been pre-trained on a similar task.
[0111] Fine-tuning the model: Some layers can be fine-tuned according to the needs of specific tasks.
[0112] 4. Input eigenvalues
[0113] Eigenvector: The extracted eigenvalues are input into the model.
[0114] Standardization: Ensure that the scale of input feature values is consistent, usually normalized. Use StandardScaler to normalize the feature values to ensure that the scale of input data is consistent.
[0115] 5. Output test results
[0116] Forward propagation: The output is calculated by forward propagation of the model.
[0117] Determine the status: Determine the status of the insulator based on the output result (for example, an output greater than 0.5 indicates damage, and less than 0.5 indicates normal).
[0118] The detection results are output by the deep BP neural network model to determine the state of the insulator.
[0119] Specifically, the device used in the above detection method includes:
[0120] Excitation module, used for continuous excitation of knocking and white noise on the flange side of the lower end of the insulator;
[0121] A receiving module, used for detecting a response signal of the insulator through a receiver probe to obtain vibration signal data of the insulator;
[0122] A feature extraction module is used to extract the voiceprint features of the relevant vibration signal using a Mel Frequency Cepstral Coefficient (MFCC) algorithm; as another implementation scheme of this embodiment, the feature extraction module also includes: a wavelet transform unit, which is used to perform multi-scale decomposition of the signal and extract features of different frequency bands; a Hilbert-Huang Transform (HHT) unit, which is used to perform time-frequency analysis on nonlinear and non-stationary signals and extract instantaneous frequency and amplitude information;
[0123] A feature fusion module is used to fuse the extracted features using a convolutional neural network (CNN) to form a feature vector; as another implementation scheme of this embodiment, the feature fusion module also includes: a long short-term memory network (LSTM) unit, which is used to fuse the extracted features to form a feature vector; a self-attention mechanism unit, which is used to combine the Transformer model to enhance the focus on key features;
[0124] A database module is used to store feature vectors and establish an insulator detection result database; as another implementation scheme of this embodiment, the database module also includes: a data enhancement unit, which is used to add noise of different types and intensities to the training data to improve the robustness of the model; a data synthesis unit, which is used to generate more diverse training samples through a generative adversarial network (GAN);
[0125] The detection module is used to perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data; as another implementation of this embodiment, the detection module may also include: a wireless sensor network unit, used to perform real-time data collection to achieve remote and real-time data collection; an Internet of Things platform unit, used to transmit the collected data to the cloud in real time for centralized processing and analysis; as another implementation of this embodiment, the detection module also includes: a multi-sensor fusion unit, used to combine vibration sensors, temperature sensors and humidity sensors to detect the vibration signal, surface temperature and ambient humidity of the insulator; as another implementation of this embodiment, the detection module also includes: an edge computing unit, used to perform preliminary data processing and feature extraction on the sensor node to reduce the amount of transmitted data; a cloud analysis unit, used to transmit the preprocessed data to the cloud for reasoning and final judgment of the deep learning model.
[0126] An analysis module is used to input the extracted feature values into a pre-trained deep BP neural network model. The model adds at least one hidden layer on the basis of the standard BP neural network, and is improved using the ReLU activation function and the dropout strategy. The detection results are output by the deep BP neural network model to judge the state of the insulator; as another implementation of this embodiment, the analysis module also includes: a multi-hidden layer structure unit, which is used to add multiple hidden layers to the deep BP neural network model to improve the expression ability of the model; an integrated learning unit, which is used to use random forests and gradient boosting machines (GBM) for integrated learning to improve classification accuracy; as another implementation of this embodiment, the analysis module also includes: an adaptive threshold unit, which is used to dynamically adjust the judgment threshold to improve the sensitivity and accuracy of detection; an abnormality detection unit, which is used to identify and process abnormal data points.
[0127] As another implementation of this embodiment, as another implementation of this embodiment, it also includes:
[0128] Visualization module, used to display the detection results and status of insulators in real time through a visualization interface;
[0129] The report generation module is used to automatically generate test reports, including test data, analysis results and recommended measures.
[0130] Example 2
[0131] A system using the method described in Example 1, the system comprising:
[0132] one or more memories for storing instructions; and
[0133] One or more processors are used to call and run the instructions from the memory to execute the method described in the above detection method.
[0134] The present invention also provides a computer program product, which includes instructions. When the instructions are executed, the terminal device and the network device perform operations of the terminal device and the network device corresponding to the above method.
[0135] The present invention also provides a chip system, which includes a processor for implementing the functions involved in the above content, such as generating, receiving, sending, or processing the data and / or information involved in the above method.
[0136] The chip system may be composed of chips, or may include chips and other discrete devices.
[0137] The processor mentioned in any of the above places can be a CPU, a microprocessor, an ASIC, or one or more integrated circuits for executing programs for controlling the above-mentioned feedback information transmission method.
[0138] In a possible design, the chip system also includes a memory, which is used to store necessary program instructions and data. The processor and the memory can be decoupled and respectively set on different devices, connected by wire or wireless means to support the chip system to implement various functions in the above embodiments. Alternatively, the processor and the memory can also be coupled on the same device.
[0139] Optionally, the computer instructions are stored in a memory.
[0140] Optionally, the memory is a storage unit within the chip, such as a register, a cache, etc. The memory can also be a storage unit within the terminal located outside the chip, such as a ROM or other types of static storage devices that can store static information and instructions, RAM, etc.
[0141] It can be understood that the memory in the present invention can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.
[0142] The nonvolatile memory may be a ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), or a flash memory.
[0143] The volatile memory may be a RAM, which is used as an external cache. There are many different types of RAM, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct memory bus RAM.
[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for detecting live porcelain insulators based on mechanical pulse excitation, characterized in that: include: S1: Establishing insulator test result database through vibration response simulation: The lower flange side of the insulator is excited by knocking and white noise continuous excitation, and the response signal of the insulator is detected by the receiver probe to obtain the vibration signal data of the insulator. The Mel frequency cepstral coefficient algorithm is used to extract the voiceprint features of the vibration signal; The extracted features are fused using convolutional neural networks to form feature vectors, which are then stored in the database. S2: Inspection of insulators to be inspected: Perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data; S3: Comparative analysis: The extracted feature values are input into a pre-trained deep BP neural network model, which adds at least one hidden layer to the standard BP neural network and uses the ReLU activation function and dropout strategy for improvement; The detection results are output by the deep BP neural network model to determine the state of the insulator.
2. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: During the white noise continuous excitation detection, the probes of the white noise signal transmitter and the vibration signal receiver are both pressed against the flange side of the lower end of the insulator, and the instrument sends continuous white noise to the insulator through the transmitter probe for excitation.
3. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: The step S1 also includes: performing multi-scale decomposition of the signal using wavelet transform to extract features of different frequency bands; Use Hilbert-Huang transform to perform time-frequency analysis on nonlinear and non-stationary signals and extract instantaneous frequency and amplitude information; The long short-term memory network is introduced to fuse the extracted features to form a feature vector; Combined with the Transformer model, it enhances the focus on key features; Add noise of different types and intensities to the training data to improve the robustness of the model; Generate more diverse training samples through generative adversarial networks.
4. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: The S2 step further includes: Use wireless sensor networks for real-time data collection to achieve remote and real-time data collection; The collected data is transmitted to the cloud in real time for centralized processing and analysis.
5. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: The S2 step further includes: Multi-sensor fusion technology is used to combine vibration sensors, temperature sensors and humidity sensors to detect the vibration signals, surface temperature and ambient humidity of the insulators.
6. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: The S2 step further includes: Perform preliminary data processing and feature extraction on sensor nodes to reduce the amount of transmitted data; The preprocessed data is transmitted to the cloud for inference and final judgment of the deep learning model.
7. The method for detecting live porcelain insulators based on mechanical pulse excitation according to claim 1 is characterized in that: The S3 step further includes: Add multiple hidden layers to the deep BP neural network model to improve the model's expressiveness; Use random forest and gradient boosting machine for ensemble learning to improve classification accuracy; Dynamically adjust the judgment threshold to improve the sensitivity and accuracy of detection; Introduce anomaly detection algorithms to identify and process abnormal data points; Through the visual interface, the detection results and status of the insulator are displayed in real time; Automatically generate inspection reports including inspection data, analysis results and recommended actions.
8. A device using the porcelain insulator live detection method based on mechanical pulse excitation as claimed in claim 1, characterized in that: include: Excitation module, used for continuous excitation of knocking and white noise on the flange side of the lower end of the insulator; A receiving module, used for acquiring vibration signal data of the insulator; A feature extraction module is used to extract the voiceprint features of the vibration signal; Feature fusion module, used to fuse the extracted features to form a feature vector; A database module is used to store feature vectors and establish an insulator detection result database; The detection module is used to perform the same excitation operation on the insulator to be inspected, detect and extract its voiceprint feature data; The analysis module is used to analyze the extracted characteristic values, output the detection results, and determine the status of the insulator.
9. A porcelain insulator live detection system based on mechanical pulse excitation, characterized in that: The system comprises: one or more memories for storing instructions; and One or more processors, configured to call and execute the instructions from the memory to perform the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer readable storage medium comprises: The program, when the program is executed by a processor, the method according to any one of claims 1 to 7 is executed.
Citation Information
Cited By
AI-based composite insulator internal defect ultrasonic detection method
CN121068768A
An AI-based composite insulator internal defect ultrasonic detection method
CN121068768B