LIBS-Based Classification Method and System for Insulators with Different Formulations

Abstract

The present invention discloses a method and system for classifying insulators with different formulations based on LIBS. The method includes the following steps: S1. Acquisition and analysis of insulator standard spectral lines. LIBS is used to analyze insulators from different manufacturers to obtain spectral data, and the spectral data is preprocessed; S2. Build a neural network and use the preprocessed spectral data for training; S3. Use the trained neural network for the classification of actual insulator samples. The present invention utilizes laser-induced breakdown spectroscopy technology. By generating a laser pulse with an extremely high power density, a plasma is induced on the surface of the insulator sample, and the plasma is collected by an optical fiber to obtain spectral information, so as to realize the classification of insulator formulations from different manufacturers.

Description

Technical Field

[0001] The present invention belongs to the technical field of insulator classification, and particularly relates to a method and system for classifying insulators with different formulations based on LIBS. Background Art

[0002] Since the mid - to late 1990s, composite insulators made of high - temperature vulcanized silicone rubber, fittings, and epoxy glass core rods have been widely used in high - voltage transmission lines. Affected by environmental factors (humidity, pollution, lightning strikes, ultraviolet radiation), with the increase of operation time, composite insulators will show aging phenomena such as cracking, powdering, and reduced hydrophobicity. The number of damaged insulators also gradually increases, threatening the safe operation of power transmission and transformation equipment. Line maintenance personnel will comprehensively evaluate insulators from multiple angles such as the operation years, pollution level, powdering degree, and hydrophobicity of composite insulators to guide the operation and retirement of insulators.

[0003] For composite insulators that show phenomena such as seal failure, insulation degradation, and flashover of unknown cause, in addition to exploring the accident causes and aging mechanisms, it is also necessary to count the manufacturers to which the accident - prone insulators belong. When the failure probability of a certain batch of insulators of a certain manufacturer is relatively high, it indicates that there may be defects in the formulation or production process of the insulator silicone rubber. This information has important reference value for users to select manufacturers and manufacturers to improve product quality. In recent years, accurate records of the manufacturers of insulators put into operation have been available, but the records of the manufacturers of insulators on operating lines in the early days were usually incomplete.

[0004] Laser - Induced Breakdown Spectroscopy (LIBS) is to ablate a sample with a high - energy pulsed laser, generate a plasma on the surface of the sample, and the component information of the ablated sample is contained in the plasma emission spectrum collected by an optical fiber. Summary of the Invention

[0005] The main purpose of the present invention is to overcome the disadvantages and deficiencies of the prior art, and propose a method and system for classifying insulators with different formulations based on LIBS. By using laser - induced breakdown spectroscopy, a laser pulse with an extremely high power density is generated to induce the generation of plasma on the surface of the insulator sample, and the optical fiber is used to collect the plasma to obtain spectral information, so as to realize the classification of insulators from different manufacturers.

[0006] In order to achieve the above - mentioned purpose, the present invention adopts the following technical solutions:

[0007] A method for classifying insulators with different formulations based on LIBS includes the following steps:

[0008] S1. Acquisition and analysis of insulator standard spectra: Analyze insulators from different manufacturers using LIBS to obtain spectral data and preprocess the spectral data;

[0009] S2. Build a neural network and train it using the preprocessed spectral data;

[0010] S3. Use the trained neural network for the classification of actual insulator samples.

[0011] Furthermore, before obtaining the insulator standard spectra, it also includes building a LIBS system. The LIBS system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer; the optical path system includes a reflector and a convex lens;

[0012] The process of obtaining the standard spectra through the LIBS system is as follows:

[0013] The laser generates pulsed laser, which is focused on the surface of the sample insulator through a reflecting lens and a lens. The sample is ablated, excited, evaporated, and dissociated to form a plasma with high temperature and high electron density on the surface. During the cooling process of the high-temperature and high-density plasma, bremsstrahlung and recombination radiation generate ionization lines of each element to form a continuous background spectrum. After the continuous radiation significantly decays, a large number of atomic and ionic line spectra start to be radiated. Electrons in excited atoms and molecules transition between discrete bound energy levels and emit line spectra corresponding to the wavelengths, that is, atomic emission spectra. The spectral information is collected by an optical fiber and sent to the spectrometer for spectral splitting; the ICCD detector coupled with the spectrometer completes photoelectric conversion and transmission, and finally the computer completes data acquisition and storage.

[0014] Furthermore, in step S1, use the LIBS system to analyze insulators from multiple different manufacturers respectively to obtain spectral data. Specifically:

[0015] Use the LIBS system to analyze insulators from multiple different manufacturers respectively. Use a press to cut a 5-cm-diameter disc at the skirt part of the insulator, and perform LIBS detection at 10 different positions on the surface of the sample insulator, and take the average value of the spectral data at 10 points.

[0016] Furthermore, the preprocessing of the spectral data specifically includes:

[0017] Baseline correction: Remove the interference of the background spectrum and find the characteristic element spectra corresponding to the fillers used in the insulator firing process. Specifically:

[0018] Adopt the baseline correction method, divide the spectrum into N parts on average, select a series of representative baseline characteristic points from them, perform linear, polynomial, or spline function interpolation on these points to construct a baseline, and subtract the data point set from the baseline to correct the baseline to y = 0;

[0019] Wavelet transform filtering is used for noise reduction. The original spectrum is decomposed by a wavelet function to obtain a low-frequency signal containing spectral feature information and a high-frequency signal containing noise information. A threshold is set to remove the high-frequency signal and retain the low-frequency signal, thereby denoising the original spectrum.

[0020] With reference to the information of the standard spectral lines in the atomic spectral database and considering the resolution of the spectrometer, the characteristic element spectral lines of the insulator are determined.

[0021] Furthermore, the preprocessing of the spectral data also includes:

[0022] Based on the recursive feature elimination algorithm, spectral data feature selection is performed, and the linear discriminant analysis algorithm is used to reduce the dimensionality of the feature spectral data. Specifically:

[0023] All the initial data points are used as features to form a feature set, which is input into the classification model. The correlation of each feature is calculated to obtain a sorted list of feature correlation scores. According to this list, some features with poor correlation are removed, and this iterative process is repeated to select the optimal feature subset and reduce the amount of spectral information data.

[0024] By calculating the within-class scatter matrix S w and the between-class scatter matrix S b , calculate the matrix S w -1 S b , perform eigen decomposition on the matrix S w -1 S b and calculate the eigenvectors corresponding to the largest d eigenvalues to form W. Through the formula Y = W T X, the high-dimensional original features are transformed into low-dimensional new features, reducing the data dimensionality.

[0025] Furthermore, when performing spectral data feature selection, the feature selection algorithm includes but is not limited to REF;

[0026] When performing dimensionality reduction, the algorithms used include but are not limited to LDA.

[0027] Furthermore, in step S2, building the neural network is specifically as follows:

[0028] Build a BP neural network. The BP neural network includes three layers, namely the input layer, the hidden layer, and the output layer. Each layer includes multiple neurons, and the neurons are interconnected;

[0029] The number of neurons in the input layer is equal to the number of input variables in the data to be processed, and the number of neurons in the output layer is equal to the number of outputs associated with each input.

[0030] Further, in step S2, the preprocessed spectral data is used as a training sample, and the training sample is input into the built neural network for training and recognition. The trained neural network is used for the classification and recognition of insulators from different manufacturers.

[0031] The training process of the BP neural network is specifically as follows:

[0032] The BP neural network is a supervised learning. During its training process, the output results are propagated forward, and the input of each layer of neurons only receives the output of the previous layer of neurons.

[0033] Error backpropagation is performed to feedback the deviation of each intermediate layer. Using the gradient descent method, the weights change in the direction of the negative gradient. The above steps are iteratively repeated. When the error between the predicted value and the true value is less than the set threshold, the training of the BP neural network is completed.

[0034] Further, step S3 is specifically as follows:

[0035] Actual operation measurements are carried out under the same instrument parameters. The LIBS system is used to analyze the actual sample insulators to obtain spectral data. After data preprocessing, the spectral data is input into the trained neural network to obtain the manufacturer attribution of the insulators.

[0036] The present invention also includes a classification system for insulators with different formulations based on LIBS. The system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer. The optical path system includes a reflector and a convex lens. The system uses the method provided by the present invention to classify insulators with different formulations.

[0037] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0038] 1. Based on LIBS spectral data, the NIST database, and machine learning algorithms, the present invention can quickly distinguish the differences in the types of elements and component contents of insulator samples from different manufacturers, and further construct a classification and evaluation method for insulator formulations from different manufacturers, providing a new basis for evaluating the operating performance of insulators from different manufacturers and providing a basis for the power grid material department to purchase insulators.

[0039] 2. When distinguishing the differences in the types of elements and component contents of samples, TGA has the defect of being unable to subdivide the content and types of non-thermally decomposable components. Although the EDS and XPS test methods can obtain the elemental composition and content of samples, the samples need to be sent to the laboratory for analysis. However, the LIBS detection makes up for these defects, and the LIBS device has portability and can perform on-site in-situ detection, reducing the test cycle.

[0040] 3. The present invention is based on LIBS. LIBS detection is based on the plasma generated by ablating the surface of the sample. According to the scanning electron microscope test, the laser ablation radius and depth are in the micron order, which is almost non-destructive.

[0041] 4. The present invention is based on LIBS. The LIBS detection device has the ability of on-site remote survey and can adapt to the detection under different environments. Description of the Drawings

[0042] Figure 1 is the flow chart of the method of the present invention;

[0043] Figure 2 is the schematic diagram of the LIBS system;

[0044] Figure 3 is the element spectral line diagram of an insulator of a certain manufacturer in the embodiment;

[0045] Figure 4 is the insulator classification result in the embodiment;

[0046] Figure 5 is the schematic structural diagram of the BP neural network in the embodiment. Detailed Embodiment

[0047] The present invention will be further described in detail below in conjunction with the embodiments and the drawings, but the embodiments of the present invention are not limited thereto.

[0048] Embodiment

[0049] When the LIBS experimental parameters remain unchanged, due to the raw material materials, formula designs, and manufacturing processes of insulators from different manufacturers being completely different, the spectral signals obtained by laser ablating insulators from different manufacturers are different. According to the characteristic spectral line intensities of the corresponding elements of the fillers used in the insulator firing process, the manufacturer attribution of some insulators can be preliminarily judged; the spectral data is classified through an algorithm, and then the classification of the formulas of insulators from different manufacturers is realized.

[0050] As Figure 1 shown, the method for classifying insulators with different formulas based on LIBS of the present invention includes the following steps:

[0051] S1. Build a LIBS system. As Figure 2 shown, the LIBS system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer; the optical path system includes a reflector and a convex lens;

[0052] The process of obtaining the standard spectral line through the LIBS system is as follows:

[0053] The laser generates pulsed laser, which is focused on the surface of the sample insulator through a reflecting lens and a lens. The sample is ablated, excited, evaporated and dissociated, forming a plasma with high temperature and high electron density on the surface. During the cooling process of the high-temperature and high-density plasma, Bremsstrahlung and recombination radiation generate ionization lines of various elements to form a continuous background spectrum. After the continuous radiation significantly attenuates, a large number of atomic and ionic line spectra start to be radiated. The electrons in the excited atoms and molecules transition between discrete bound energy levels and emit line spectra corresponding to the wavelengths, that is, atomic emission spectra. The spectral information is collected through an optical fiber and transported to a spectrometer for spectral splitting; the ICCD detector coupled with the spectrometer completes photoelectric conversion and transmission, and finally the computer completes data acquisition and storage; the purpose of delay control is to adjust the delay time between the emission of the laser pulse and the detection of the optical signal to avoid the strong continuous background spectrum in the recombination radiation and Bremsstrahlung stages, so as to obtain atomic emission spectra with high signal-to-noise ratio and signal-to-background ratio.

[0054] S2. Acquisition and analysis of the standard spectral lines of the insulator. Use LIBS to analyze insulators from different manufacturers to obtain spectral data and preprocess the spectral data; specifically in this embodiment:

[0055] Use the LIBS system to analyze insulators from multiple different manufacturers respectively. For insulators from different manufacturers, use a molding machine to take circular pieces with a diameter of 5 cm at the skirt part of the insulator, and perform LIBS detection at 10 different positions respectively to obtain spectral data of insulators from different manufacturers.

[0056] The preprocessing of the spectral data includes:

[0057] S21. Baseline correction, removing the interference of the background spectrum and finding the characteristic element spectral lines corresponding to the fillers used in the insulator firing process; specifically:

[0058] During the LIBS measurement process, due to the influence of sample properties, environmental differences and the performance of hardware equipment, the original spectra collected by LIBS usually have a relatively high baseline and noise, which is not conducive to subsequent spectral data analysis; the relatively high baseline is due to the existence of Bremsstrahlung and recombination radiation. Relying solely on delay control cannot eliminate the influence of Bremsstrahlung and recombination radiation. Using the baseline correction method, the spectrum is evenly divided into N parts, and a series of representative baseline characteristic points are selected from them. Linear, polynomial or spline function interpolation is performed on these points to construct the baseline, and subtracting the data point set from the baseline can correct the baseline to y = 0;

[0059] Wavelet transform filtering is used for noise reduction. The original spectrum is decomposed through wavelet functions to obtain a low-frequency signal containing spectral characteristic information and a high-frequency signal containing noise information. A threshold is set to remove the high-frequency signal and retain the low-frequency signal to denoise the original spectrum;

[0060] Referring to the information of standard spectral lines in the atomic spectral database of the National Institute of Standards and Technology (NIST) (including the types of elements, wavelengths, spectral line transition probabilities, upper and lower energy level differences, etc.), and comprehensively considering the resolution of the spectrometer (there is a drift in the central wavelength of the spectral line), the characteristic element spectral lines of the insulator can be roughly determined.

[0061] S22. Perform spectral data feature selection based on the recursive feature elimination algorithm, and use the linear discriminant analysis algorithm to reduce the dimension of the feature spectral data; specifically:

[0062] The amount of spectral information data collected by LIBS is extremely large. If all the data is directly input into the model for qualitative or quantitative analysis, it is likely to cause problems such as long model training time, difficult convergence, and low model accuracy. The recursive feature elimination algorithm is a wrapper-type feature selection algorithm that uses the prediction performance of the classification model as the evaluation criterion. Its search starting point is the complete set. All the initial data points are used as features to form a feature set, which is input into the classification model. The correlation of each feature is calculated to obtain a feature correlation score ranking table. According to this table, some features with poor correlation are eliminated, and this iterative process is repeated to select the optimal feature subset, greatly reducing the amount of spectral information data;

[0063] The feature subset still has a relatively high dimension, and the dimension of the features can be reduced. The original features disappear after dimensionality reduction, and the new features contain the information of the original features. Linear discriminant analysis (LDA) is a supervised linear pattern recognition algorithm. For a classification problem, it is considered to map the data to a low-dimensional space so that the data points of the same category are as compact as possible, and the data points of different categories are as far apart as possible. By calculating the within-class scatter matrix S w and the between-class scatter matrix S b , calculate the matrix S w -1 S b , perform eigen-decomposition on the matrix S w -1 S b , and calculate the eigenvectors corresponding to the largest d eigenvalues to form W. Through the formula Y = W T X, the high-dimensional original features are transformed into low-dimensional new features, reducing the data dimension.

[0064] In this embodiment, when performing spectral data feature selection, the feature selection algorithm includes but is not limited to REF;

[0065] When performing dimensionality reduction processing, the algorithms used include but are not limited to LDA.

[0066] S3. Build a BP neural network and train it using the preprocessed spectral data. Specifically, use the preprocessed spectral data as training samples, input the training samples into the built neural network for training and recognition, and use the trained neural network for the classification and recognition of insulators from different manufacturers.

[0067] In this embodiment, the BP neural network consists of three layers, namely the input layer, the hidden layer (middle layer), and the output layer. Each layer includes multiple neurons, and the neurons are interconnected.

[0068] The number of neurons in the input layer is equal to the number of input variables in the data to be processed, and the number of neurons in the output layer is equal to the number of outputs associated with each input.

[0069] The steps for training the BP neural network are as follows:

[0070] The BP neural network is a supervised learning. During its training process, the output results are propagated forward, and the input of each layer of neurons only receives the output of the previous layer of neurons.

[0071] Error backpropagation, feedback the deviation of each intermediate layer, and use the gradient descent method to change the weights in the negative gradient direction. The above steps are iterated cyclically. When the error between the predicted value and the true value is less than the set threshold, the training of the BP neural network model is completed.

[0072] In this embodiment, as Figure 5 shown, I is the input layer, H is the hidden layer, and O is the output layer. W is the weight connecting the neurons of the front and back layers. For example, the input of H1 is the weighted sum (plus the bias) of the outputs of the previous layer of neurons, that is, I1*W1 + I2*W2 + b1*1. The output of H1 is the "activation" of this input. In the model, the activation is embodied in the form of an activation function. The commonly used activation function is the Sigmoid function. Generally, the weights and biases are randomly initialized. Based on the loss function, list the error between the true value and the predicted value. Looking from back to front, find the derivative of the error with respect to the weights and biases, and let the weight and bias parameters change along the negative gradient direction, and keep cycling until the error is less than the set threshold.

[0073] In LIBS detection, the input layer of the neural network receives the spectral feature data of various samples, finds a suitable activation function and adjusts the weights W and biases b connecting the neurons, fits the input features and the target, and outputs the determination of the sample category from the output layer.

[0074] S4. Use the trained BP neural network for the classification of actual insulator samples.

[0075] Under the same instrument parameters, actual operation measurements are carried out. The LIBS system is used to analyze the actual sample insulator to obtain spectral data. After data preprocessing, the spectral data is input into the trained neural network to obtain the manufacturer attribution of the insulator. As Figure 3 and Figure 4 shown, it is the element spectral line diagram of an insulator of a certain manufacturer in this embodiment and the insulator classification result in the embodiment.

[0076] In another embodiment, a classification system for insulators with different formulations based on LIBS is also provided. The system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer; the optical path system includes a reflector and a convex lens; the system classifies insulators with different formulations by using the classification method of the above embodiment.

[0077] It should also be noted that in this specification, terms such as "including", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without more limitations, an element defined by the statement "including one..." does not exclude the existence of another identical element in the process, method, article or device including the said element.

[0078] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for classifying insulators with different formulations based on LIBS, characterized in that, It includes the following steps: S1. Acquisition and analysis of the standard spectrum of insulators. Use LIBS to analyze insulators from different manufacturers to obtain spectral data and preprocess the spectral data; The preprocessing of the spectral data specifically includes: Baseline correction. Remove the interference of the background spectrum and find the characteristic element spectral lines corresponding to the fillers used in the insulator firing process. Specifically: Adopt the baseline correction method. Divide the spectrum into N parts on average, select a series of representative baseline characteristic points from them, perform linear, polynomial or spline function interpolation on these points to construct the baseline, and subtract the data point set from the baseline to correct the baseline to y = 0; Use wavelet transform filtering for noise reduction. Decompose the original spectrum through the wavelet function to obtain the low-frequency signal containing spectral characteristic information and the high-frequency signal containing noise information. Set the threshold to remove the high-frequency signal and retain the low-frequency signal to denoise the original spectrum; Refer to the information of the standard spectral lines in the atomic spectral database and consider the resolution of the spectrometer to determine the characteristic element spectral lines of the insulator; The preprocessing of the spectral data also includes: Based on the recursive feature elimination algorithm for spectral data feature selection, use the linear discriminant analysis algorithm to perform dimensionality reduction processing on the characteristic spectral data. Specifically: Take all the initial data points as features to form a feature set, input it into the classification model, calculate the correlation of each feature, obtain the feature correlation score ranking table, and eliminate some features with poor correlation according to this table. Repeat this iterative process to select the optimal feature subset and reduce the spectral information data volume; By calculating the within-class scatter matrix S w and the between-class scatter matrix S b , calculate the matrix S w -1 S b , for the matrix S w -1 S b perform eigen-decomposition and calculate the eigenvectors corresponding to the largest d eigenvalues to form W. Through the formula Y = W T X, transform the high-dimensional original features into low-dimensional new features to reduce the data dimension; S2. Build a neural network and use the preprocessed spectral data for training; In step S2, use the preprocessed spectral data as the training sample, input the training sample into the built neural network for training and recognition, and use the trained neural network for the classification and recognition of insulators from different manufacturers; The specific process of BP neural network training is: The BP neural network is a supervised learning. During its training process, the output result is propagated forward, and the input of each layer of neurons only accepts the output of the previous layer of neurons; Error backpropagation. Feedback the deviation of each intermediate layer, and use the gradient descent method to make the weight change in the negative gradient direction. The above steps are iterated cyclically. When the error between the predicted value and the true value is less than the set threshold, the BP neural network training is completed; S3. Use the trained neural network for the classification of actual insulator samples. Specifically: Perform actual operation measurement under the same instrument parameters. Use the LIBS system to analyze the actual sample insulators to obtain spectral data, perform data preprocessing, input the spectral data into the trained neural network, and obtain the manufacturer attribution of the insulators.

2. The method for classifying different-formula insulators based on LIBS according to claim 1, wherein Before obtaining the standard spectrum of the insulator, it also includes building a LIBS system. The LIBS system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer; The optical path system includes a reflector and a convex lens; The process of obtaining the standard spectrum through the LIBS system is: The laser generates pulsed laser light, which is focused on the surface of the sample insulator through a reflecting lens and a lens. The sample is ablated, excited, evaporated, and dissociated, forming a plasma with high temperature and high electron density on the surface. During the cooling process of the high-temperature and high-density plasma, bremsstrahlung and recombination radiation generate ionization lines of various elements to form a continuous background spectrum. After the continuous radiation significantly decays, a large number of atomic and ionic line spectra begin to be radiated. The electrons in the excited atoms and molecules transition between discrete bound energy levels and emit line spectra corresponding to the wavelengths, that is, atomic emission spectra. The spectral information is collected by an optical fiber and transmitted to a spectrometer for spectral splitting; the photoelectric conversion and transmission are completed by an ICCD detector coupled with the spectrometer, and finally, data acquisition and storage are completed by a computer.

3. The method for classifying insulators with different formulations based on LIBS according to claim 1, characterized in that, In step S1, the LIBS system is used to analyze insulators from multiple different manufacturers respectively to obtain spectral data, specifically: The LIBS system is used to analyze insulators from multiple different manufacturers respectively. A circular piece with a diameter of 5 cm is taken at the skirt part of the insulator by a molding machine, and LIBS detection is performed at 10 different positions on the surface of the sample insulator, and the average value of the spectral data at 10 points is taken.

4. The method for classifying insulators with different formulations based on LIBS according to claim 1, characterized in that, When performing spectral data feature selection, the feature selection algorithm includes but is not limited to REF; When performing dimensionality reduction processing, the algorithms used include but are not limited to LDA.

5. The method for classifying different formulation insulators based on LIBS according to claim 1, characterized in that, Step S2, building a neural network specifically: Build a BP neural network. The BP neural network includes three layers, namely the input layer, the hidden layer, and the output layer. Each layer includes multiple neurons, and the neurons are interconnected; The number of neurons in the input layer is equal to the number of input variables in the data to be processed, and the number of neurons in the output layer is equal to the number of outputs associated with each input.

6. The insulator classification system based on LIBS is characterized in that The system includes a laser, an optical path system, a controller, a spectrometer, a delay controller, and a computer; the optical path system includes a reflector and a convex lens; the system classifies insulators with different formulations by using the method described in any one of claims 1-5.

Images