Carbon brush wear detection method and system

Abstract

The invention relates to the technical field of equipment maintenance, and discloses a carbon brush wear detection method and system. According to the method, real-time carbon brush surface image data and carbon brush working parameters are collected and subjected to data preprocessing, then a carbon brush surface abrasion detection model and a carbon brush parameter detection model are used for comprehensive evaluation, and finally whether the carbon brush needs to be replaced or not is judged according to the evaluation result. Wherein the carbon brush surface wear detection model is trained by adopting an improved RealNet algorithm, and the carbon brush parameter detection model adopts a random forest algorithm. The method improves the accuracy and efficiency of detection, provides a powerful guarantee for the safe operation of the power equipment, has the advantages of high detection efficiency, small error, wide application range and the like, and is of great significance for improving the maintenance level of the power equipment and reducing the maintenance cost.

Description

Technical Field

[0001] The invention relates to the technical field of equipment maintenance, and in particular to a carbon brush wear detection method and a system thereof. Background Art

[0002] As an important component in electrical equipment, the wear state of carbon brushes directly affects the operating efficiency and safety of the equipment. Traditional detection methods mainly rely on manual inspections, which have problems such as low detection efficiency and large errors. With the development of image processing technology and machine learning, carbon brush wear detection methods based on image recognition have emerged, but such methods often only focus on the image features of the carbon brush surface and ignore the working parameters of the carbon brush, resulting in incomplete detection results. At the same time, the existing detection methods also have deficiencies in data preprocessing and model training, which affects the accuracy and efficiency of detection. Therefore, there is an urgent need for a carbon brush wear detection method that can comprehensively consider the carbon brush surface image and working parameters to achieve accurate detection and timely replacement judgment. Summary of the invention

[0003] In view of the above-mentioned technical deficiencies, an object of the present invention is to provide a carbon brush wear detection method and system thereof, so as to solve the problems of low accuracy and efficiency of carbon brush wear detection in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a method for detecting carbon brush wear, the method comprising:

[0006] Step S100: collecting real-time carbon brush surface image data and carbon brush operating parameters, and performing data preprocessing on the carbon brush surface image data and the carbon brush operating parameters;

[0007] Step S200: inputting the carbon brush surface image data after data preprocessing into the carbon brush surface wear detection model, and outputting the probability of defects on the carbon brush surface;

[0008] Step S300: inputting the carbon brush working parameters after data preprocessing into the carbon brush parameter detection model, and outputting the carbon brush performance evaluation result;

[0009] Step S400: If the probability of defects on the surface of the carbon brush is greater than a preset threshold or the carbon brush performance evaluation result is unqualified, it is determined that the carbon brush needs to be replaced.

[0010] Preferably, in a possible implementation manner of the first aspect, the data preprocessing of the carbon brush surface image data includes image denoising and size normalization processing.

[0011] Preferably, in a possible implementation manner of the first aspect, the carbon brush operating parameters include carbon brush resistivity, carbon brush operating current and carbon brush operating voltage.

[0012] Preferably, in a possible implementation manner of the first aspect, the data preprocessing of the carbon brush operating parameters includes data cleaning, outlier processing and fast Fourier transform.

[0013] Preferably, in a possible implementation of the first aspect, the carbon brush surface wear detection model is trained by an improved RealNet algorithm, using a carbon brush surface image dataset, wherein the carbon brush surface image dataset is composed of historical carbon brush surface image data and abnormal images generated by an SDAS method based on the historical carbon brush surface image data.

[0014] Preferably, in a possible implementation of the first aspect, the improved RealNet algorithm training process includes:

[0015] Input historical carbon brush surface image data and use SDAS to generate synthetic anomaly samples with controllable anomaly intensity;

[0016] The pre-trained image features are selected through the AFS mechanism, and the reconstruction residual selection module and the attention mechanism are used to select the reconstruction residual of abnormal information;

[0017] The selected image feature subset is input into the RealNet reconstruction module for reconstruction to obtain the reconstruction error, which is iteratively adjusted to minimize the reconstruction error in the normal area of ​​the carbon brush surface image and maximize the reconstruction error in the abnormal area of ​​the carbon brush surface image.

[0018] Preferably, in a possible implementation manner of the first aspect, the carbon brush parameter detection model is trained using a random forest algorithm, specifically comprising:

[0019] Collect historical carbon brush operating parameters, including historical carbon brush resistivity, historical carbon brush operating current, historical carbon brush operating voltage, etc., as well as carbon brush performance evaluation results;

[0020] Performing data preprocessing on historical carbon brush working parameters to form a historical carbon brush working parameter data set as a training sample;

[0021] Features are randomly selected from the samples to construct a decision tree. Multiple decision trees form a random forest. Each tree makes independent predictions and the evaluation results are obtained by voting.

[0022] Preferably, in a possible implementation manner of the first aspect, step S400 specifically includes:

[0023] Determine whether the probability of defects on the surface of the carbon brush is greater than a preset threshold;

[0024] If it is greater than the preset threshold, it is determined that the carbon brush needs to be replaced;

[0025] If it is not greater than the preset threshold, it is determined whether the carbon brush performance evaluation result is unqualified;

[0026] If the carbon brush performance evaluation result is unqualified, it is determined that the carbon brush needs to be replaced.

[0027] In a second aspect, the present invention provides a carbon brush wear detection system, the system comprising:

[0028] An image acquisition module, used for acquiring image data of the carbon brush surface;

[0029] The parameter acquisition module is used to collect the carbon brush working parameters, including the carbon brush resistivity, carbon brush working current and carbon brush working voltage;

[0030] A preprocessing module is used to perform data preprocessing on the collected carbon brush surface image data and carbon brush working parameters, wherein the image data preprocessing includes image denoising and size normalization processing, and the working parameter preprocessing includes data cleaning, outlier processing and fast Fourier transform;

[0031] The model detection module includes a carbon brush surface wear detection model and a carbon brush parameter detection model. The carbon brush surface wear detection model is used to receive the preprocessed carbon brush surface image data and output the probability of defects on the carbon brush surface. The carbon brush parameter detection model is used to receive the preprocessed carbon brush working parameters and output the carbon brush performance evaluation results.

[0032] The result judgment module is used to judge whether the probability of defects on the surface of the carbon brush is greater than a preset threshold or whether the carbon brush performance evaluation result is unqualified. If so, it is judged that the carbon brush needs to be replaced.

[0033] The beneficial effects of the present invention are: by collecting carbon brush surface image data and carbon brush operating parameters and performing data preprocessing, the accuracy and availability of the data are improved. At the same time, by using the carbon brush surface wear detection model and the carbon brush parameter detection model, the carbon brush surface image and operating parameters are comprehensively considered to achieve accurate detection of the carbon brush wear state. In addition, by setting preset thresholds and judgment logic, carbon brushes that need to be replaced can be found in time, providing a strong guarantee for the safe operation of power equipment. The detection method of the present invention has the advantages of high detection efficiency, small error, and a wide range of applications, and is of great significance for improving the maintenance level of power equipment and reducing maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0035] Figure 1 A flow chart of a carbon brush wear detection method is provided for this application.

[0036] Figure 2 A carbon brush wear detection system structure diagram is provided for this application.

[0037] Explanation of the accompanying drawings: 1-image acquisition module, 2-parameter acquisition module, 3-preprocessing module, 4-model detection module, 5-result judgment module. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in 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.

[0039] Embodiment 1: Figure 1 As shown, the present invention provides a carbon brush wear detection method, the method comprising:

[0040] Step S100: collecting real-time carbon brush surface image data and carbon brush operating parameters, and performing data preprocessing on the carbon brush surface image data and the carbon brush operating parameters.

[0041] Specifically, the carbon brush surface image data is acquired by photographing with an industrial camera, and the captured carbon brush surface image data is transmitted to the control center through the network. The carbon brush surface image data is subjected to data preprocessing including image denoising and size normalization. The carbon brush working parameters include carbon brush resistivity, carbon brush working current and carbon brush working voltage. The carbon brush resistivity is acquired through a resistivity tester, the carbon brush working current is acquired through a Hall current sensor, and the carbon brush working voltage is acquired through a voltage divider voltage sensor. The acquired carbon brush working parameters are transmitted to the data center through a data interface. The data preprocessing of the carbon brush working parameters includes data cleaning, outlier processing and fast Fourier transform.

[0042] In this embodiment, the resistivity tester adopts TS500GHD carbon material resistance tester, the Hall current sensor adopts LTS25-NP Hall current sensor, and the voltage divider voltage sensor adopts LMV358 voltage divider voltage sensor.

[0043] For the carbon brush surface image data, data preprocessing is performed. First, the OpenCV library is used to read and process the carbon brush surface image. The path of the image file is passed in through the image_path parameter. The read image is denoised and the Gaussian blur algorithm is used to remove the noise in the image. Gaussian blur is a commonly used image processing technique that smoothes the image by weighted averaging the neighborhood of each pixel in the image. The cv2.GaussianBlur() function is used to perform Gaussian blur processing on the image, where the blur kernel size is set to (5,5) and the standard deviation is set to 0. After the denoising process is completed, the image is normalized and the denoised image is resized using the cv2.resize() function. The target size is passed in through the target_size parameter, which is a tuple representing the width and height of the adjusted image. The target size is set to (256,256). Finally, the processed image is returned.

[0044] For the data preprocessing of carbon brush working parameters, first use the Pandas library to process the working parameter data of carbon brushes. The data is stored in the form of DataFrame and contains three columns: 'resistivity', 'working current' and 'working voltage'. Next, clean the data. The main purpose of cleaning is to delete the null values ​​(NaN) in the DataFrame. Use the dropna() method of DataFrame to delete all rows containing null values. After completing the data cleaning, outlier processing is performed. Outliers refer to values ​​that are significantly different from most data points in the data. Using the 3σ principle, first calculate the mean (mean) and standard deviation (std) of each data column, then find the data points whose absolute value of the difference from the mean is greater than 3 times the standard deviation, and regard these points as outliers. Finally, delete these outliers from the DataFrame. After outlier processing, fast Fourier transform is performed on the carbon brush working current and carbon brush working voltage. The fft function in the NumPy library is used to perform fast Fourier transform on the current and voltage data. The input of the fft function is time series data, and the output is data represented in the frequency domain. Finally, return the processed DataFrame.

[0045] Step S200: inputting the carbon brush surface image data after data preprocessing into the carbon brush surface wear detection model, and outputting the probability of defects on the carbon brush surface.

[0046] Specifically, the carbon brush surface wear detection model is trained by the improved RealNet algorithm and adopts the carbon brush surface image dataset, which consists of historical carbon brush surface image data and abnormal images generated based on the historical carbon brush surface image data by the SDAS method.

[0047] The improved RealNet algorithm training process includes: inputting historical carbon brush surface image data and using SDAS to generate synthetic abnormal samples with controllable abnormal intensity; selecting pre-trained image features through the AFS mechanism, and applying the reconstruction residual selection module and the attention mechanism to select the reconstruction residual of the abnormal information; inputting the selected image feature subset into the RealNet reconstruction module for reconstruction, obtaining the reconstruction error, and continuously iteratively adjusting to minimize the reconstruction error of the normal area of ​​the carbon brush surface image and maximize the reconstruction error of the abnormal area of ​​the carbon brush surface image.

[0048] In this embodiment, the carbon brush surface image data after data preprocessing is input into the carbon brush surface wear detection model, and the carbon brush surface wear detection model is trained using the improved RealNet algorithm.

[0049] First, based on the abnormal image of the carbon brush surface image data generated by the controllable intensity diffusion anomaly synthesis (SDAS) model, the Perlin noise generator is used to capture various abnormal shapes and binarize them into an abnormal mask M. The normal image is represented as I, the abnormal image generated by SDAS is represented as P, and the image synthesized by image mixing with local abnormalities is represented as A:

[0050]

[0051] in ⊙ represents the element-by-element multiplication operation, and δ is the opacity in the image blending. At the same time, in order to ensure that the generated abnormal area is located in the foreground, a foreground segmentation binarization method based on adaptive threshold is used.

[0052] Next, we select pre-trained image features through the AFS mechanism, apply the reconstruction residual selection module and the attention mechanism to select the reconstruction residual of abnormal information. The AFS mechanism is a self-supervised method for pre-trained feature selection, which reduces the dimension of features and eliminates pre-training bias while managing reconstruction costs. First, we define a set of N triples A n ,I_n, Among them A n ,I_n∈R h×w×3 represents the synthesized abnormal image and the original normal image, Denote the corresponding anomaly mask and denote the pre-trained network as φ_k, φ_k(A_n)∈R h_k×w_k×c_k represents the k-th layer pre-trained features extracted from A_n, where c_k represents the number of channels, and for the i-th feature map, φ k,i (A_n)∈R h_k×w_k , AFS selects m_k feature maps for reconstruction. For the pre-trained features of the k-th layer, the following AFS loss is defined to evaluate the i-th feature map:

[0053]

[0054] where F(·) is a function that performs a normalization operation and converts φ k,i (A n )-φ k,i The resolution of (I_n) is aligned to M n By reconstructing the network G1,...,G k Synthesize abnormal features Reconstruct the original image features at various resolutions The loss function is defined as follows:

[0055]

[0056] The reconstructed residual selection module is represented as E1(A_n),...,E k (A_n), in order to obtain the global reconstruction residual, the low-resolution reconstruction residual is upsampled and then cascaded in a channel manner. The reconstruction residual in E(A_n) is obtained from the pre-trained features of the corresponding reconstruction layer. Features of the same resolution can only effectively capture anomalies within a certain range. Therefore, the reconstruction residual selection module combines the attention mechanism to select only the subset of reconstruction residuals that have the greatest impact on the abnormal information, and generates anomaly scores to achieve the maximum recall rate of abnormal areas.

[0057] Finally, the image feature subset is input into the RealNet reconstruction module for reconstruction to obtain the reconstruction error. After multiple rounds of training and continuous iterative adjustment, the reconstruction error of the normal area of ​​the carbon brush surface image is minimized and the reconstruction error of the abnormal area of ​​the carbon brush surface image is maximized. The maximum value of the abnormal score is used as the image-level abnormal score, and the cross entropy loss is used. To supervise the training of the discriminator. The overall loss function of RealNet is:

[0058]

[0059] After the carbon brush surface wear detection model is trained, the carbon brush surface image data after data preprocessing is input into the carbon brush surface wear detection model. The model outputs a probability value based on the reconstruction error and anomaly score, which is the probability of defects on the carbon brush surface.

[0060] Step S300: inputting the carbon brush operating parameters after data preprocessing into the carbon brush parameter detection model, and outputting the carbon brush performance evaluation result.

[0061] Specifically, the carbon brush parameter detection model is trained using the random forest algorithm, which includes: collecting historical carbon brush working parameters, including historical carbon brush resistivity, historical carbon brush working current, historical carbon brush working voltage, etc., as well as carbon brush performance evaluation results; performing data preprocessing on the historical carbon brush working parameters to form a historical carbon brush working parameter data set as a training sample; randomly selecting features from the samples to construct a decision tree, and multiple decision trees form a random forest. Each tree makes independent predictions, and the evaluation results are obtained through voting.

[0062] In this embodiment, the carbon brush working parameters after data preprocessing are input into the carbon brush parameter detection model. The carbon brush parameter detection model is trained using a random forest algorithm, and a historical carbon brush working parameter data set is used as a training sample. The historical carbon brush working parameter data set is divided into a training set and a test set according to an 8:2 ratio.

[0063] First, the historical carbon brush working parameter data set includes resistivity, current FFT data (current_fft) and voltage FFT data (voltage_fft). The current FFT data and voltage FFT data are spectrum data after FFT transformation. Then the random forest algorithm is used to train the model. The number of decision trees in the random forest is set to 100, and the random seed is specified to ensure the repeatability of the results. After the model training is completed, the performance of the model is evaluated on the test set. The accuracy of the model is calculated by comparing the labels predicted by the model and the actual labels of the test set.

[0064] After the carbon brush parameter detection model training is completed, the carbon brush working parameters after data preprocessing are input into the carbon brush parameter detection model to determine whether the carbon brush resistivity, carbon brush working current and carbon brush working voltage in the carbon brush working parameters are qualified.

[0065] Step S400: If the probability of defects on the surface of the carbon brush is greater than a preset threshold or the carbon brush performance evaluation result is unqualified, it is determined that the carbon brush needs to be replaced.

[0066] Specifically, determine whether the probability of defects on the carbon brush surface is greater than a preset threshold; if greater than the preset threshold, determine that the carbon brush needs to be replaced; if not greater than the preset threshold, determine whether the carbon brush performance evaluation result is unqualified; if the carbon brush performance evaluation result is unqualified, determine that the carbon brush needs to be replaced.

[0067] In this embodiment, the probability of defects on the surface of the carbon brush is first compared with a preset threshold value, the preset threshold value is 0.6, when the probability of defects on the surface of the carbon brush is greater than the preset threshold value 0.6, it is considered that the degree of wear of the carbon brush has exceeded the acceptable range, and it needs to be replaced to avoid potential failures and safety hazards. If the probability of defects on the surface of the carbon brush is not greater than the preset threshold value, then continue to check the result of the carbon brush performance evaluation in step S300, the carbon brush performance evaluation result is obtained by analyzing the carbon brush operating parameters (including resistivity, operating current and operating voltage) through the carbon brush parameter detection model, if the carbon brush performance evaluation result is unqualified, it means that although the degree of wear on the surface of the carbon brush has not reached the threshold value requiring replacement, its performance can no longer meet the normal working requirements, and it also needs to be replaced.

[0068] Embodiment 2: Figure 2 As shown, the present invention provides a carbon brush wear detection system, the system comprising:

[0069] Image acquisition module 1, used for collecting carbon brush surface image data;

[0070] Parameter collection module 2, used to collect carbon brush operating parameters, including carbon brush resistivity, carbon brush operating current and carbon brush operating voltage;

[0071] The preprocessing module 3 is used to perform data preprocessing on the collected carbon brush surface image data and carbon brush working parameters, wherein the image data preprocessing includes image denoising and size normalization processing, and the working parameter preprocessing includes data cleaning, outlier processing and fast Fourier transform;

[0072] Model detection module 4, including a carbon brush surface wear detection model and a carbon brush parameter detection model, the carbon brush surface wear detection model is used to receive the pre-processed carbon brush surface image data and output the probability of defects on the carbon brush surface, and the carbon brush parameter detection model is used to receive the pre-processed carbon brush working parameters and output the carbon brush performance evaluation results;

[0073] The result judgment module 5 is used to judge whether the probability of defects on the surface of the carbon brush is greater than a preset threshold or whether the carbon brush performance evaluation result is unqualified. If so, it is judged that the carbon brush needs to be replaced.

[0074] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A carbon brush wear detection method, characterized in that: The method comprises: Step S100: collecting real-time carbon brush surface image data and carbon brush operating parameters, and performing data preprocessing on the carbon brush surface image data and the carbon brush operating parameters; Step S200: inputting the carbon brush surface image data after data preprocessing into the carbon brush surface wear detection model, and outputting the probability of defects on the carbon brush surface; Step S300: inputting the carbon brush working parameters after data preprocessing into the carbon brush parameter detection model, and outputting the carbon brush performance evaluation result; Step S400: If the probability of defects on the surface of the carbon brush is greater than a preset threshold or the carbon brush performance evaluation result is unqualified, it is determined that the carbon brush needs to be replaced.

2. The carbon brush wear detection method according to claim 1, characterized in that: The data preprocessing of the carbon brush surface image data includes image denoising and size normalization processing.

3. The carbon brush wear detection method according to claim 1, characterized in that: The carbon brush operating parameters include carbon brush resistivity, carbon brush operating current and carbon brush operating voltage.

4. The carbon brush wear detection method according to claim 3, characterized in that: The data preprocessing of the carbon brush working parameters includes data cleaning, outlier processing and fast Fourier transform.

5. The carbon brush wear detection method according to claim 1, characterized in that: The carbon brush surface wear detection model is trained by an improved RealNet algorithm and adopts a carbon brush surface image dataset, wherein the carbon brush surface image dataset consists of historical carbon brush surface image data and abnormal images generated based on the historical carbon brush surface image data by an SDAS method.

6. The carbon brush wear detection method according to claim 5, characterized in that: The improved RealNet algorithm training process includes: Input historical carbon brush surface image data and use SDAS to generate synthetic anomaly samples with controllable anomaly intensity; The pre-trained image features are selected through the AFS mechanism, and the reconstruction residual selection module and the attention mechanism are used to select the reconstruction residual of abnormal information; The selected image feature subset is input into the RealNet reconstruction module for reconstruction to obtain the reconstruction error, which is iteratively adjusted to minimize the reconstruction error in the normal area of ​​the carbon brush surface image and maximize the reconstruction error in the abnormal area of ​​the carbon brush surface image.

7. The carbon brush wear detection method according to claim 1, characterized in that: The carbon brush parameter detection model is trained using a random forest algorithm, specifically including: Collect historical carbon brush operating parameters, including historical carbon brush resistivity, historical carbon brush operating current, historical carbon brush operating voltage, etc., as well as carbon brush performance evaluation results; Performing data preprocessing on historical carbon brush working parameters to form a historical carbon brush working parameter data set as a training sample; Features are randomly selected from the samples to construct a decision tree. Multiple decision trees form a random forest. Each tree makes independent predictions and the evaluation results are obtained by voting.

8. The carbon brush wear detection method according to claim 1, characterized in that: The step S400 specifically includes: Determine whether the probability of defects on the surface of the carbon brush is greater than a preset threshold; If it is greater than the preset threshold, it is determined that the carbon brush needs to be replaced; If it is not greater than the preset threshold, it is determined whether the carbon brush performance evaluation result is unqualified; If the carbon brush performance evaluation result is unqualified, it is determined that the carbon brush needs to be replaced.

9. A carbon brush wear detection system, characterized in that: The system comprises: An image acquisition module, used for acquiring carbon brush surface image data; The parameter acquisition module is used to collect the carbon brush working parameters, including the carbon brush resistivity, carbon brush working current and carbon brush working voltage; A preprocessing module is used to perform data preprocessing on the collected carbon brush surface image data and carbon brush working parameters, wherein the image data preprocessing includes image denoising and size normalization processing, and the working parameter preprocessing includes data cleaning, outlier processing and fast Fourier transform; The model detection module includes a carbon brush surface wear detection model and a carbon brush parameter detection model. The carbon brush surface wear detection model is used to receive the preprocessed carbon brush surface image data and output the probability of defects on the carbon brush surface. The carbon brush parameter detection model is used to receive the preprocessed carbon brush working parameters and output the carbon brush performance evaluation results. The result judgment module is used to judge whether the probability of defects on the surface of the carbon brush is greater than a preset threshold or whether the carbon brush performance evaluation result is unqualified. If so, it is judged that the carbon brush needs to be replaced.

Images