An overhead insulated cable production quality on-line monitoring method

Abstract

The application discloses an overhead insulated cable production quality online monitoring method and relates to the technical field of online monitoring.The steps of the method comprise the following steps: collecting real-time data of an insulation layer of an overhead insulated cable and preprocessing the real-time data to obtain preprocessed real-time data of the insulation layer; acquiring a surface image of an ultraviolet-resistant coating, extracting features by using a multi-scale local binary method, and obtaining feature data of the ultraviolet-resistant coating; acquiring qualified sample data of the overhead insulated cable; calculating the similarity between the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer by using a Hamming distance algorithm to obtain an insulation layer similarity score; and calculating the similarity between the feature data of the ultraviolet-resistant coating and the qualified sample feature data by using a Hausdorff distance to obtain an ultraviolet-resistant coating similarity score.The overall quality score is calculated according to the insulation layer similarity score and the ultraviolet-resistant coating similarity score, and the online monitoring of the production quality of the overhead insulated cable is realized.

Description

Technical Field

[0001] This invention relates to the field of online monitoring engineering, specifically to an online monitoring method for the production quality of overhead insulated cables. Background Technology

[0002] Overhead insulated cables, as the core carrier of electrical energy transmission in power systems, are widely used in urban distribution networks, industrial parks, and power supply in remote areas. The integrity and uniformity of the insulation layer, as well as the surface condition of the UV-resistant coating, directly determine the cable's insulation performance, weather resistance, and service life, and are crucial to the safety and stability of power transmission. With the power industry's increasing demands for transmission reliability, online monitoring of cable quality during production and timely control of the insulation layer and UV-resistant coating quality have become key requirements for ensuring cable product qualification and reducing subsequent maintenance costs.

[0003] Traditional methods for monitoring the quality of overhead insulated cables often involve identifying surface defects in images of the insulation wire surface. This method acquires optical images of the insulation wire surface and uses image processing techniques such as edge detection and threshold segmentation to identify surface defects such as scratches, breaks, and dents. Based on parameters such as the number and size of the defects, it determines whether the appearance quality of the cable insulation layer is up to standard.

[0004] However, this method of identifying defects on the surface of the insulation wire is limited to monitoring visible defects in the insulation layer. It cannot cover the quality monitoring needs of the UV-resistant coating of overhead insulated cables. As a result, the problem of uniform thickness of the UV-resistant coating is difficult to detect effectively. Ultimately, some cable products with potential failure risks of UV-resistant coatings enter the market, affecting their long-term service life in complex outdoor environments. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an online monitoring method for the production quality of overhead insulated cables, solving the problem of not being able to cover the quality monitoring requirements of the UV-resistant coating on overhead insulated cables.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an online monitoring method for the production quality of overhead insulated cables, comprising the following steps:

[0007] Step S1: Collect real-time data of the insulation layer of the overhead insulated cable and perform data preprocessing to obtain preprocessed real-time data of the insulation layer;

[0008] Step S2: Obtain an image of the UV-resistant coating surface of the overhead insulated cable, and use the multi-scale local binary method to extract features from the UV-resistant coating surface image to obtain UV-resistant coating feature data;

[0009] Step S3: Obtain qualified sample data of overhead insulated cables, including qualified sample data of insulation layer and characteristic data of qualified sample data of UV-resistant coating;

[0010] Step S4: Calculate the similarity between the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer to obtain the insulation layer similarity score;

[0011] Step S5: Use Hausdorff distance to calculate the similarity between the UV-resistant coating feature data and the UV-resistant coating qualified sample feature data to obtain the UV-resistant coating similarity score;

[0012] Step S6: Based on the insulation layer similarity score and the UV-resistant coating similarity score, calculate the overall quality score, and realize online monitoring of the production quality of overhead insulated cables based on the overall quality score.

[0013] Preferably, real-time data of the insulation layer of the overhead insulated cable is collected and preprocessed to obtain preprocessed real-time data of the insulation layer, including:

[0014] During the data acquisition phase, cameras deployed on the production line periodically capture images of the insulation layer surface of the insulated cable. Simultaneously, a laser thickness gauge is used to measure the thickness of the insulation layer in real time. The laser thickness gauge is based on the principle of laser triangulation, which calculates the thickness by emitting a laser beam and receiving the reflected signal, and outputs continuous time series data.

[0015] For the preprocessing of the surface image of the insulation layer of the insulated cable, the acquired image is first compressed and grayscaled to reduce computational complexity. After the image preprocessing is completed, feature extraction and encoding are performed. Discrete cosine transform (DCT) is performed on the grayscale image to convert the image from the spatial domain to the frequency domain to capture its frequency features. Next, the low-frequency domain matrix is ​​encoded to generate an image fingerprint.

[0016] The preprocessing method for the insulation layer thickness of overhead insulated cables is as follows. Since the thickness data is a continuously collected numerical sequence, data cleaning and normalization are required. First, the original insulation layer thickness data is subjected to moving average filtering to smooth random noise. Then, the filtered thickness data is normalized and scaled to the [0,1] interval to avoid the influence of dimensions.

[0017] After preprocessing the image and thickness data, the two are integrated into preprocessed real-time insulation layer data. .

[0018] Preferably, obtaining images of the UV-resistant coating surface of the overhead insulated cable includes:

[0019] The surface of the UV-resistant coating of the overhead insulated cable is scanned using a near-infrared spectral sensor. The sensor captures an image of the UV-resistant coating surface of the overhead insulated cable at a preset resolution. This surface image is a grayscale image, and the grayscale value of each pixel reflects the reflection characteristics of the coating to near-infrared light. During scanning, the ambient lighting conditions need to be controlled to avoid external interference and ensure image quality.

[0020] Preferably, the anti-UV coating surface image is extracted using a multi-scale local binary method to obtain anti-UV coating feature data, including:

[0021] The basic Local Binary Mode (LBP) application process is as follows: On the preprocessed image of the UV-resistant coating surface, LBP is applied. For each pixel, a circular neighborhood is defined centered on it, and multiple sets of scale parameters are set. For example, the radius R is defined as 1, 2, and 3, and the number of sampling points P is 8, 16, and 24 respectively. The gray values ​​of the neighboring pixels are then calculated. With the gray value of the center pixel The difference is binarized using the sign function s(x), as shown in the following formula:

[0022]

[0023] in, ; It is the gray value of the p-th pixel in the neighborhood, starting from... Obtain from a circular neighborhood centered at R with radius R; is the gray value of the center pixel, directly obtained from the preprocessed near-infrared grayscale image. 'c' explicitly indicates that this gray value belongs to the calculated center pixel and is the benchmark point for the entire LBP feature calculation process. The gray values ​​of all neighboring pixels are compared to it. Subsequently, the mathematical formula for generating the code is as follows:

[0024]

[0025] Where P is the number of sampling points, which is a preset constant; weights Based on fixed-location assignment, this step extracts local symbolic features, but it is sensitive to noise and rotational changes.

[0026] To enhance robustness, a multi-scale extension and improvement method is integrated. The SMDCP method is employed, combining sign and magnitude information. First, the mean value of local neighboring pixels is calculated using the following formula:

[0027]

[0028] in, It is the average value of local neighborhood pixels, used as a threshold substitute. This enhances the algorithm's robustness to noise because the mean is more stable than individual pixel values, and the sign pattern... The definition is as follows:

[0029]

[0030] in, The subscript S in the code represents the name of the mathematical function, explicitly indicating that the key to this improvement lies in using a sign function to generate binary code, thereby achieving stronger noise resistance; this is an improvement on LBP, which uses neighborhood mean. Non-center pixel value As a threshold, a symbol pattern is generated. This improvement effectively suppresses noise interference within a uniform region and enhances feature stability. Subsequently, amplitude calculation is performed to extract differential amplitude information, as shown in the following formula:

[0031]

[0032] in, The amplitude represents the absolute value of the grayscale difference. This amplitude information reflects the strength of the contrast and is an important feature of the texture. Combining amplitude information with symbolic information can more comprehensively describe the texture. At the same time, a directional contour mode is introduced to capture the spatial orientation of the texture. First, the orientation angle of the neighboring points relative to the center point is calculated, as follows:

[0033]

[0034] in, It is the direction angle; σ is a small regularization constant; then the continuous direction angles are quantized into finite discrete direction numbers, as shown in the following formula:

[0035]

[0036] in, B is the quantized direction number, and B is the preset total number of direction partitions. Combining amplitude information with sign and direction information can more comprehensively describe the texture. At the same time, the LBPmr method is introduced to improve the weighting problem, that is, to adjust the amplitude... Perform ascending sorting to obtain the dynamic weighted index. The final integrated coding formula is:

[0037]

[0038] Among them, s It is a symbolic value; It is the amplitude Sort index obtained by ascending order; sort index by amplitude Replaces fixed weights This makes the encoding no longer dependent on the fixed geometric position of the neighboring points, but on the relative strength of their gray-level contrast, thus making the features invariant to the rotation of the image and enhancing robustness. It is a directional weighting function that makes features more discriminative of texture patterns in specific directions; in addition, the GNP operator is applied to utilize absolute grayscale information, as shown in the following formula:

[0039]

[0040] in, It represents the number of discrete intervals and is a preset constant. It is an interval matcher that, through a judgment operation, matches consecutive center pixel gray values. This is converted into a discrete number GNP value representing the grayscale range to which it belongs; The nth threshold is adaptively determined by analyzing the global histogram of historical qualified sample images and using image segmentation algorithms such as the maximum inter-class variance method to better separate different gray-level distributions. Multi-threshold segmentation is used to separate the gray-level values ​​of the center pixels. Mapping to discrete intervals;

[0041] Multi-scale and multi-feature fusion is employed. At various set scales, SMDCP, LBPmr, and GNP operators are applied to generate multiple feature images. A grayscale histogram is calculated for each feature image. All histogram feature vectors are concatenated to generate a compact feature vector, which serves as the feature data for the UV-resistant coating. .

[0042] Preferably, obtaining qualified sample data for overhead insulated cables includes:

[0043] The data collection phase involves a comprehensive scan of certified overhead insulated cables using a high-precision sensor system. Specifically, near-infrared spectral sensors and accompanying imaging equipment are used to simultaneously capture images of the insulation layer surface and the UV-resistant coating surface. Insulation layer data acquisition focuses on surface texture and thickness information, using a camera to capture high-resolution images and combining this with a laser thickness gauge to record thickness values ​​in real time. UV-resistant coating data acquisition utilizes a near-infrared spectral sensor to scan the coating surface and obtain grayscale images. The grayscale value of each pixel reflects the coating's reflectivity to near-infrared light. The scanning process is conducted under strictly controlled environmental conditions, such as maintaining constant temperature and stable lighting, to minimize external interference. All raw data comes directly from the sensor output to ensure representativeness and authenticity, and is stored in matrix form for easy subsequent processing.

[0044] Preferably, the qualified sample data includes qualified sample data for the insulation layer and characteristic data for qualified samples for the UV-resistant coating, including:

[0045] The surface image of the insulation layer is compressed and converted to grayscale. Average pooling is used to reduce the image size, and a weighted average method is applied to convert it to grayscale to highlight brightness features. Feature extraction is then performed, including discrete cosine transform to capture frequency information and generate a compact image fingerprint. For thickness data, moving average filtering is performed to smooth noise, and normalization is applied to scale it to a standard range. Finally, the image and thickness features are integrated into a unified multidimensional feature vector, representing the qualified sample data of the insulation layer. ;

[0046] For the UV-resistant coating sample data in the qualified sample data of overhead insulated cables, the processing method is exactly the same as in step S2. First, an adaptive median filter is used to smooth the image and reduce noise interference. Then, texture features are extracted based on multi-scale local binary mode, including sign, amplitude, and absolute grayscale information. A compact feature representation, i.e., the UV-resistant coating qualified sample feature data, is generated through multi-scale analysis and feature dimensionality reduction. .

[0047] Preferably, the similarity score for the insulation layer is obtained by calculating the similarity between the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer, including:

[0048] For calculating the similarity of the surface texture of the insulating layer, the method is the Hamming distance algorithm based on image fingerprints. The Hamming distance is defined as the number of different characters at corresponding positions in two strings of equal length, and its mathematical expression is:

[0049]

[0050]

[0051] in, F is the Hamming distance, representing the number of different characters between two strings; F is the sample data image fingerprint, a hexadecimal string of length 16, obtained by processing qualified sample cables in step S3; G is the real-time data image fingerprint, a hexadecimal string of length 16, obtained by processing cables produced online in step S1. and It is the i-th character in the string; δ is an indicator function, when... = When δ=0, otherwise δ=1. Among the parameters, the string length is 16, which is obtained by DCT transformation and binary sequence conversion. The characters come from hexadecimal encoding and represent the low-frequency domain features of the image.

[0052] Thickness similarity can be directly obtained by calculating the complement of the absolute difference, as shown in the following formula:

[0053]

[0054] in, The sample thickness data is obtained by preprocessing the thickness data of qualified sample cables in step S3. It is real-time thickness data, which is obtained by preprocessing the thickness data of the online-produced cable in step S1;

[0055] By combining the similarity results of surface texture and thickness, the similarity score of the insulating layer is obtained. Since the two parts have different dimensions, they need to be normalized first. The Hamming distance H of the surface texture can be converted into a similarity score using the following formula:

[0056]

[0057] in, This is the image similarity score; 16 represents the string length and thickness similarity. The result has been normalized. Then, a weighted average is calculated based on actual needs, as shown in the following formula:

[0058]

[0059] in, It is an insulation layer similarity score; and It is a weighting coefficient, and the initial value can be set according to the importance of the insulation layer texture and thickness to the overall quality.

[0060] Preferably, the similarity score of the UV-resistant coating is obtained by using Hausdorff distance to calculate the similarity between the feature data of the UV-resistant coating and the feature data of qualified UV-resistant coating samples, including:

[0061] The Hausdorff distance is used to measure the similarity between two point sets J and K, where J represents the sample data point set and K represents the real-time data point set. The formula is as follows:

[0062]

[0063] in, It is a bidirectional maximum-minimum distance, the ultimate measure of difference between two point sets J and K; it takes the maximum of the directed distances in two directions, ensuring that it captures the maximum degree of mismatch between the two point sets in any direction. The smaller the value, the more similar the two point sets are; h(J,K) represents the directed Hausdorff distance from point set J to point set K, and h(K,J) represents the directed Hausdorff distance from point set K to point set J.

[0064] The Hausdorff distance value is then converted into an intuitive similarity score, calculated using the following formula:

[0065]

[0066] in, The UV-resistant coating similarity score is a standardized value ranging from 0 to 1, used to represent the degree of similarity between the UV-resistant coatings. The closer the score is to 1, the more similar the sample data is to the real-time data; H(J,K) is the actual distance value calculated using the aforementioned Hausdorff distance, representing the degree of difference between point set J and point set K; the denominator... The design ensures the mathematical stability of the formula when H(J,K)=0. 1; when As the score increases, it gradually decreases, but never becomes negative or exceeds 1.

[0067] Preferably, based on the insulation layer similarity score and the UV-resistant coating similarity score, the overall quality score is calculated as follows:

[0068] The overall quality score of overhead insulated cables is obtained by combining the similarity scores of the insulation layer and the coating layer using a weighted average method. The weighted average formula is as follows:

[0069]

[0070] in, It is the overall quality score, a value between 0 and 1, with a higher value indicating better overall quality; The insulation layer similarity score is derived from the output of step S4. This score is calculated based on the similarity of the insulation layer data and is typically in the range of 0 to 1, where 1 represents complete similarity and 0 represents complete dissimilarity. The UV-resistant coating similarity score, derived from the output of step S5, is calculated based on the similarity of the coating data. A higher score indicates more consistent coating quality. and These are weighting coefficients, representing the relative importance of the insulation layer and coating layer in the overall quality assessment, satisfying... + =1, these weights are usually set based on domain knowledge or historical data to ensure that the score reflects the actual quality priority.

[0071] Preferably, online monitoring of the production quality of overhead insulated cables, based on an overall quality score, includes:

[0072] After the overall quality score is calculated, it needs to be compared with a preset threshold to generate the monitoring results. The threshold comparison formula is as follows:

[0073]

[0074] The determination of the preset threshold T should be based on the principles of statistical process control. This involves collecting no fewer than 500 historical qualified samples, calculating their quality scores, and initializing T to the P5 or P10 quantile of that score set. Subsequently, the threshold should be validated and optimized using a validation set containing samples with known defects, focusing on the false negative rate to ensure it is below an acceptable risk level of 0.5%. The final T value is typically between 0.85 and 0.95 and should be fine-tuned according to specific product specifications and safety levels. The monitoring result is a binary output used to guide the production process. If the score is greater than or equal to the threshold, the cable is considered qualified and can proceed to the next stage; otherwise, it is considered unqualified and may require further inspection or adjustment.

[0075] This invention provides an online monitoring method for the production quality of overhead insulated cables, involving machine learning and deep learning technologies, which has the following beneficial effects:

[0076] (1) This online monitoring method for the production quality of overhead insulated cables extracts features from the surface image of the UV-resistant coating of overhead insulated cables using a multi-scale local binary mode. Compared with traditional single-scale feature extraction methods, it can comprehensively capture multi-dimensional feature information of the UV-resistant coating surface of overhead insulated cables, from micro-texture to macro-contour, by setting various different combinations of neighborhood radii and sampling points. Actual production line testing verified that this method achieves a detection rate of over 98.5% for minor coating defects, an improvement of over 13 percentage points compared to the 85.2% of the traditional single-scale LBP method. It preserves the detailed texture features of the coating surface while also considering the overall structural characteristics, enabling the extracted features to more comprehensively reflect the true quality state of the coating and providing richer and more accurate feature support for subsequent similarity calculations.

[0077] (2) This online monitoring method for the production quality of overhead insulated cables improves the local binary pattern (LBP) based on amplitude ranking, replacing the fixed geometric position weights with the relative strength of grayscale contrast, thus making the features rotationally invariant. Simultaneously, it incorporates the global N-value pattern (GNP), fully utilizing the absolute grayscale information of the coating surface, achieving deep fusion of local partial sign features, amplitude features, and global grayscale features. Experimental comparisons show that, under simulated illumination changes and image rotation interference, the similarity stability of the feature vectors extracted by this method is far superior to that of traditional LBP features. This effectively overcomes the limitation that traditional single features can only reflect part of the coating characteristics, significantly improving the discriminative ability and stability of the feature vectors.

[0078] (3) An online monitoring method for the production quality of overhead insulated cables is proposed. This method uses the Hausdorff distance algorithm to calculate the similarity between UV-resistant coating sample data and real-time data. Compared to traditional point-to-point similarity calculation methods, this method comprehensively captures the global distribution differences of coating features by measuring the overall maximum and minimum distance between two feature point sets. In experiments comparing the results with those of human experts, the correlation coefficient between the similarity score based on Hausdorff distance and the expert evaluation results reached 0.95, while the traditional method was only 0.78. This avoids the misjudgment of similarity caused by local feature point matching deviations in traditional methods, making the calculation results more objectively reflect the consistency of the overall coating quality and improving the accuracy of quality assessment.

[0079] (4) The online monitoring method for the production quality of overhead insulated cables introduces a dual-queue intermediate storage structure consisting of a primary priority queue (MPQ) and a secondary priority queue (SPQ), and adopts an incremental access node traversal method. This allows for dynamic screening of effective feature points and filtering of noise interference points during the calculation process. On test images with artificially added 5% salt-and-pepper noise, the similarity score output by the INC-HD algorithm of this method fluctuates within ±2%, while the fluctuation range of the standard Hausdorff distance algorithm exceeds ±15%. This avoids the influence of noise points on the distance calculation results, significantly improves the algorithm's anti-noise capability, and ensures that even with slight noise interference on the coating surface, a stable and reliable similarity score can still be output, guaranteeing the robustness of the online monitoring results. Attached Figure Description

[0080] Figure 1 This is a flowchart of an online monitoring method for the production quality of overhead insulated cables proposed in this invention.

[0081] Figure 2 This invention provides a hierarchical diagram of insulation layer similarity scores obtained from an online monitoring method for the production quality of overhead insulated cables.

[0082] Figure 3 This is a hierarchical diagram illustrating the online monitoring of the production quality of overhead insulated cables in the online monitoring method for overhead insulated cables proposed in this invention. Detailed Implementation

[0083] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0084] Please see Figure 1-3This invention provides a technical solution: an online monitoring method for the production quality of overhead insulated cables. Specifically, the following online monitoring method for the production quality of overhead insulated cables is provided; please refer to [link / reference]. Figure 1 The method includes the following steps:

[0085] Step S1: Collect real-time data of the insulation layer of the overhead insulated cable and perform data preprocessing to obtain preprocessed real-time data of the insulation layer.

[0086] This step involves acquiring real-time data on the insulation layer of overhead insulated cables and performing data preprocessing to obtain preprocessed real-time insulation layer data. The insulation layer data consists of two parts: first, images of the insulation layer surface captured using a camera; and second, measurements of the insulation layer thickness using a laser thickness gauge. Then, the insulation layer surface images and insulation layer thickness are preprocessed and feature extracted. Finally, the processed image features and thickness data are integrated into unified preprocessed real-time insulation layer data for subsequent similarity calculations.

[0087] During the data acquisition phase, cameras deployed on the production line periodically capture images of the insulation layer surface of the insulated cables, while a laser thickness gauge measures the insulation layer thickness in real time. The laser thickness gauge, based on the principle of laser triangulation, calculates the thickness by emitting a laser beam and receiving the reflected signal, outputting continuous time-series data. The acquired raw images and thickness data may contain noise or outliers, therefore preprocessing is necessary to ensure data quality.

[0088] For the preprocessing of the surface image of the insulation layer of the insulated cable, the acquired image is first compressed and converted to grayscale to reduce computational complexity. Image compression uses average pooling, dividing the high-resolution image into multiple regions, and averaging the pixel values ​​of each region to generate a compressed image with a resolution of 32x32 pixels. Average pooling preserves overall features by reducing image size while improving processing speed. Next, the compressed image is converted to grayscale using a weighted average method to convert the color image to a grayscale image P. The weighted average method assigns different weights to the red, green, and blue channels based on the human eye's sensitivity to different colors, calculating the grayscale value of each pixel. The mathematical formula is as follows:

[0089]

[0090] Here, Grayscale represents the calculated pixel grayscale value; R, G, and B represent the pixel values ​​of the red, green, and blue channels in the input image, respectively, obtained from the raw data directly captured by the camera. The coefficients 0.299, 0.587, and 0.114 are fixed weights set according to the characteristics of human vision. This process eliminates color information, highlights brightness features, and facilitates subsequent analysis.

[0091] After image preprocessing, feature extraction and encoding are performed. A Discrete Cosine Transform (DCT) is applied to the grayscale image to transform it from the spatial domain to the frequency domain, capturing its frequency features. The mathematical formula for the Discrete Cosine Transform is:

[0092]

[0093] Where D represents the 32x32 dimensional transformation coefficient matrix obtained after the transformation, P represents the 32x32 pixel grayscale image matrix, and A is the 32x32 dimensional transformation matrix. This is the transpose of matrix A. The elements of the transformation matrix A. Defined by the following formula:

[0094]

[0095] Where N is the dimension of the matrix, with a value of 32, determined by the size of the compressed image; i and j are the row and column indices of the matrix, respectively, with values ​​ranging from 0 to 31. After the discrete cosine transform, the low-frequency information of the image is concentrated in the upper left corner of the transform coefficient matrix. Therefore, the 8x8 submatrix in the upper left corner is taken as the low-frequency domain matrix to focus on the overall structural features of the image.

[0096] Next, the low-frequency domain matrix is ​​encoded to generate an image fingerprint. The average value of the low-frequency domain matrix is ​​calculated, and then each element in the matrix is ​​compared to the average value. If the element value is greater than or equal to the average value, it is recorded as 1; otherwise, it is recorded as 0. This generates a 64-bit binary sequence, which is then converted into a hexadecimal string, ultimately resulting in a 16-bit string as the image feature representation. This process encodes image information into a compact hash value, facilitating subsequent similarity calculations.

[0097] The preprocessing method for the insulation layer thickness of overhead insulated cables is as follows. Since the thickness data is a continuously collected numerical sequence, data cleaning and normalization are required. First, a moving average filter is applied to the original insulation layer thickness data to smooth out random noise. The mathematical formula for the moving average filter is:

[0098]

[0099] in, This represents the filtered thickness value, and k is the size of the sliding window. This is the original thickness measurement value at time point t within the window, which is time-series data acquired in real time from a laser thickness gauge. The parameter k is set according to the sampling frequency and noise level, and is set to 8 in this scheme. Next, the filtered thickness data is normalized and scaled to the [0,1] interval to avoid the influence of dimensions. The mathematical formula for subsequent normalization is as follows:

[0100]

[0101] in, This represents the normalized thickness value; This represents the filtered thickness value; and These represent the minimum and maximum thickness values, respectively, obtained by statistically analyzing a large amount of historically verified and qualified cable insulation thickness data. Finally, the normalized average thickness value is combined with the image feature string to form the preprocessed real-time insulation data.

[0102] After preprocessing the image and thickness data, the two are integrated into preprocessed real-time insulation layer data. Image features are represented as hexadecimal strings, and thickness features are represented as normalized values ​​or statistical vectors. During integration, the image strings and thickness values ​​are concatenated into a single feature vector. This real-time data encapsulates the surface condition and thickness information of the insulation layer, providing a foundation for subsequent quality assessment. The entire preprocessing process ensures the reliability, consistency, and real-time nature of the data, meeting the requirements of online monitoring.

[0103] This step, as the initial stage of the quality monitoring method of this invention, plays a crucial role in constructing a standardized real-time data source for the insulation layer, providing a reliable data foundation for subsequent similarity calculations and quality assessments. This step systematically collects surface images and thickness data of the insulation layer of overhead insulated cables, and performs image feature encoding and thickness signal preprocessing respectively. This transforms the heterogeneous raw data into a unified feature representation, ensuring not only data quality and consistency but also achieving a quantitative characterization of the appearance and structural properties of the insulation layer. This data preparation work directly supports the similarity calculation in the subsequent step S4, enabling effective comparison between the real-time insulation layer data and sample data, thus laying a vital data foundation for the entire quality monitoring process.

[0104] Step S2: Obtain an image of the UV-resistant coating surface of the overhead insulated cable, and use the multi-scale local binary method to extract features from the UV-resistant coating surface image to obtain UV-resistant coating feature data.

[0105] In the data collection phase, this step runs parallel to step S1, forming an independent data acquisition and processing channel. First, a near-infrared spectral sensor is used to scan the surface of the UV-resistant coating on the overhead insulated cable. The sensor captures an image of the UV-resistant coating surface at a preset resolution (e.g., 200×200 pixels). This surface image is a grayscale image, where the grayscale value of each pixel reflects the coating's reflectivity to near-infrared light. Ambient lighting conditions must be controlled during scanning to avoid external interference and ensure image quality. The captured image is input into the preprocessing system as raw data, and the raw pixel grayscale values ​​are directly used for subsequent processing.

[0106] In the image preprocessing stage, the original coating image may be affected by noise (such as Gaussian noise or salt-and-pepper noise), requiring preprocessing to enhance the robustness of feature extraction. An adaptive median filter is used to smooth the anti-UV coating surface image. This filter dynamically adjusts the window size (e.g., an initial window of 3×3, expanding to a maximum of 7×7) through the following steps: First, the median of pixels within the local window is calculated. If the median is within the grayscale range and not a noise point, the center pixel is replaced with the median; otherwise, the window is expanded and recalculated until the condition is met. After preprocessing, the image retains texture details while reducing noise interference, laying the foundation for feature extraction.

[0107] In the feature extraction stage, feature extraction is based on multi-scale local binary patterns, combining symmetric mean and orientation contour patterns (SMDCP) with improved methods, namely, amplitude-ranked local binary pattern (LBPmr) and global N-value pattern (GNP). The specific steps are as follows:

[0108] The basic Local Binary Mode (LBP) application process is as follows: Local Binary Mode is applied to the preprocessed image of the UV-resistant coating surface. For each pixel, a circular neighborhood is defined centered on it, with multiple sets of scale parameters set. For example, the radius R is defined as 1, 2, and 3, corresponding to 8, 16, and 24 sampling points P. The grayscale values ​​of the neighboring pixels are then calculated. With the gray value of the center pixel The difference is binarized using the sign function s(x), as shown in the following formula:

[0109]

[0110] in, ; It is the gray value of the p-th pixel in the neighborhood, starting from... Obtain from a circular neighborhood centered at R with radius R; The gray value of the center pixel is directly obtained from the preprocessed near-infrared grayscale image. 'c' explicitly indicates that this gray value belongs to the calculated center pixel and serves as the reference point for the entire LBP feature calculation process; the gray values ​​of all neighboring pixels are compared to it. Subsequently, the mathematical formula for generating the code is as follows:

[0111]

[0112] Where P is the number of sampling points, which is a preset constant; weights This method is based on fixed-location assignment. It extracts localized symbolic features, but is sensitive to noise and rotational variations.

[0113] To enhance robustness, a multi-scale extension and improvement method is integrated, employing the SMDCP method that combines sign and magnitude information. First, the mean value of local neighborhood pixels is calculated using the following formula:

[0114]

[0115] in, It is the average value of local neighborhood pixels, used as a threshold substitute. This enhances the algorithm's robustness to noise because the mean is more stable than individual pixel values. (Symbol pattern) The definition is as follows:

[0116]

[0117] in, The subscript S in the code represents the name of the mathematical function, explicitly indicating that the key to this improvement lies in using a sign function to generate binary code, thereby achieving stronger noise resistance. This is an improvement on LBP. It uses neighborhood mean. Non-center pixel value As a threshold, a symbol pattern is generated. This improvement effectively suppresses noise interference within uniform regions and enhances feature stability. Subsequently, amplitude calculation is performed to extract differential amplitude information, as shown in the following formula:

[0118]

[0119] in, The amplitude represents the absolute value of the grayscale difference. This amplitude information reflects the strength of the contrast and is an important feature of the texture. Combining amplitude information with symbolic information allows for a more comprehensive description of the texture. Simultaneously, a directional contour mode is introduced to capture the spatial orientation of the texture. First, the orientation angle of the neighboring points relative to the center point is calculated, using the following formula:

[0120]

[0121] in, It is the direction angle; σ is a small regularization constant; then the continuous direction angles are quantized into finite discrete direction numbers, as shown in the following formula:

[0122]

[0123] in, B is the quantized direction number, and B is the preset total number of direction partitions. Combining amplitude information with sign and direction information can more comprehensively describe the texture. At the same time, the LBPmr method is introduced to improve the weighting problem, that is, to adjust the amplitude... Perform ascending sorting to obtain the dynamic weighted index. The final integrated coding formula is:

[0124]

[0125] Among them, s It is a symbolic value; It is the amplitude The sorting index obtained by ascending sorting. The sorting index based on amplitude. Replaces fixed weights This makes the encoding no longer dependent on the fixed geometric position of the neighboring points, but on the relative strength of their gray-level contrast, thus making the features invariant to the rotation of the image and enhancing robustness. This is a directional weighting function that makes features more discriminative of texture patterns in specific directions. Furthermore, the GNP operator is applied using absolute grayscale information, as shown in the following formula:

[0126]

[0127] in, It represents the number of discrete intervals and is a preset constant. It is an interval matcher that, through a judgment operation, matches consecutive center pixel gray values. This is converted into a discrete number GNP value representing the grayscale range to which it belongs; The nth threshold is adaptively determined using image segmentation algorithms such as the maximum inter-class variance method by analyzing the global histogram of historical qualified sample images to better separate different gray-level distributions. Multi-threshold segmentation (e.g., N=2) is used to separate the gray-level values ​​of the center pixels. Mapped to discrete intervals.

[0128] Multi-scale and multi-feature fusion is employed. At various set scales (e.g., R=1, 2, 3), SMDCP, LBPmr, and GNP operators are applied to generate multiple feature images. For each feature image, its grayscale histogram is calculated, and all histogram feature vectors are concatenated to form a high-dimensional joint feature vector. To reduce redundancy, an improved dimensionality reduction algorithm is used. Based on dominant pattern statistics, dominant patterns whose cumulative probability in their histogram statistics reaches a set threshold are retained, while the remaining low-probability patterns are merged into their nearest neighbor dominant patterns. The specific steps are as follows:

[0129] The probabilistic processing is as follows: Input the concatenated high-dimensional joint feature histogram vector H, whose dimension is D (i.e., the total number of bins in the histogram); normalize the histogram vector H to convert it into a probability distribution vector P, as shown in the following formula:

[0130]

[0131] Where P(i) represents the probability of feature pattern i appearing in the entire feature image, i=1,2,...,D.

[0132] Then, dominant patterns are selected by sorting all patterns in the probability distribution vector P according to their probability values ​​P(i) from largest to smallest; a cumulative probability threshold θ is set. This threshold is the core parameter controlling the degree of information preservation after dimensionality reduction; starting with the pattern with the highest probability, their probabilities are accumulated sequentially until the cumulative probability first reaches or exceeds the preset threshold θ; these selected patterns for accumulation constitute the dominant pattern set. These patterns represent the most prominent and frequently occurring texture features in the image.

[0133] Then, low-probability patterns are merged, and all those not selected are... Low-probability patterns need to be merged to reduce dimensionality. The merging strategy is to merge each low-probability pattern into its nearest dominant pattern, where nearest means the closest Euclidean distance in the histogram's index space or eigenvalue space. Specifically, this involves: traversing each low-probability pattern j; calculating the distance from j to the set of dominant patterns. Find the distance d(j,k) of each pattern k; find the dominant pattern with the smallest distance. The count value of the low-probability mode j, i.e., H(j), is accumulated into the dominant mode. On the count value.

[0134] Finally, a compact feature vector, i.e., the UV-resistant coating feature data, is generated. The specific steps are as follows: After merging, the dimension of the feature vector is reduced from the original high dimension D to the number of dominant patterns. This new vector Composed of the merged dominant pattern statistics, this represents the dimensionality-reduced, compact data on the UV-resistant coating characteristics. .

[0135] This step, a crucial component of the UV-resistant coating quality monitoring process for overhead insulated cables, involves acquiring images of the UV-resistant coating surface using a near-infrared spectral sensor and extracting texture features using a multi-scale local binary mode (LDM) method. This transforms the raw optical signal into structured data. By combining basic LDM with an improved algorithm, this step effectively overcomes the sensitivity of traditional methods to noise, rotation, and illumination variations. It not only extracts the microscopic texture features of the coating surface but also enhances the data's discriminative power and robustness through multi-scale analysis and feature dimensionality reduction. This process provides a highly discriminative feature vector foundation for subsequent steps, acting as a bridge between them—converting physical coating characteristics into computable digital features and providing high-quality input for subsequent machine learning models or statistical analyses, thus supporting the reliability and accuracy of the entire monitoring system.

[0136] Step S3: Obtain qualified sample data of overhead insulated cables, including qualified sample data of insulation layer and characteristic data of qualified sample data of UV-resistant coating.

[0137] This step plays a crucial role in the quality monitoring process of overhead insulated cables. Its core objective is to collect data on qualified overhead insulated cables produced and to generate reliable sample data of qualified overhead insulated cables through standardized preprocessing, providing a foundation for subsequent model training or evaluation. This step ensures data consistency and comparability. By applying the mature processing methods of steps S1 and S2, it efficiently extracts features of the insulation layer and UV-resistant coating, thereby supporting the accuracy and robustness of the entire monitoring system. The following section details the data collection and preprocessing process:

[0138] The data collection phase involves a comprehensive scan of certified overhead insulated cables using a high-precision sensor system. Specifically, near-infrared spectroscopy sensors and accompanying imaging equipment simultaneously capture images of the insulation layer and the UV-resistant coating. Insulation layer data acquisition focuses on surface texture and thickness information, using a camera to capture high-resolution images (e.g., 500×500 pixels) and combining this with a laser thickness gauge to record thickness values ​​in real time. UV-resistant coating data acquisition utilizes a near-infrared spectroscopy sensor to scan the coating surface, obtaining grayscale images (e.g., 200×200 pixels), where the grayscale value of each pixel reflects the coating's reflectivity to near-infrared light. The scanning process is conducted under strictly controlled environmental conditions, such as maintaining a constant temperature (20-25°C) and stable lighting, to minimize external interference. All raw data comes directly from the sensor output, ensuring representativeness and accuracy, and is stored in matrix form for easy subsequent processing.

[0139] In the preprocessing stage, existing methods are applied to the insulation layer and UV-resistant coating separately to ensure data quality and enhance feature consistency. To ensure complete consistency between the feature representations of the sample data and the real-time data, thereby guaranteeing the effectiveness of subsequent similarity calculations, the insulation layer sample data in the qualified sample data of overhead insulated cables is processed using the exact same method as in step S1. First, the surface image of the insulation layer is compressed and converted to grayscale, average pooling is used to reduce the image size, and a weighted average method is used to convert it into a grayscale image to highlight brightness features. Subsequently, feature extraction is performed, including discrete cosine transform (DCT) to capture frequency information and generate a compact image fingerprint. For the thickness data, moving average filtering is performed to smooth noise, and normalization is performed to scale it to a standard range. Finally, the image and thickness features are integrated into a unified multi-dimensional feature vector, which is the qualified sample data of the insulation layer. For the UV-resistant coating sample data in the qualified sample data of overhead insulated cables, the processing method is exactly the same as in step S2. First, an adaptive median filter is used to smooth the image and reduce noise interference. Then, texture features, including sign, amplitude, and absolute grayscale information, are extracted based on multi-scale local binary mode (such as LBP and its improved algorithm). A compact feature representation, i.e., the UV-resistant coating qualified sample feature data, is generated through multi-scale analysis and feature dimensionality reduction. The entire preprocessing process was carried out in parallel, ensuring high quality and consistency for both types of data.

[0140] After preprocessing, the qualified sample data of the insulation layer were named The characteristic data of qualified samples with UV-resistant coatings were named These naming conventions facilitate data management and subsequent citations, ensuring the continuity of the process. Step S3, through systematic data collection and standardized processing, provides high-quality sample input for the quality monitoring of overhead insulated cables, playing a crucial role in both transforming physical characteristics into calculable features and laying a reliable foundation for subsequent similarity calculations and model applications.

[0141] Step S4: Calculate the similarity between the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer to obtain the insulation layer similarity score.

[0142] The Hamming distance algorithm was used to analyze the qualified sample data of the insulation layer. Real-time data of pre-processed insulation layer Similarity calculations are performed to obtain an insulation layer similarity score. This step is a core component of the quality monitoring process, aiming to quantitatively compare qualified insulation layer sample data. Real-time data with pre-treated insulation layer The differences between the data are used to assess the degree to which the current state of the insulation layer deviates from the acceptable standard. Insulation layer data includes both surface texture and thickness; therefore, similarity calculations must be performed separately. The surface texture data uses image fingerprinting and Hamming distance methods, while the thickness data requires similarity calculations using numerical data. Finally, a comprehensive score is obtained by combining the results of both parts, providing a basis for quality judgment.

[0143] First, the similarity calculation for the surface texture of the insulating layer is performed using the Hamming distance algorithm based on image fingerprints. The image fingerprint is a 16-character hexadecimal string generated from qualified sample data or real-time data through image processing (as described in steps S1 and S3). The Hamming distance is defined as the number of different characters at corresponding positions between two strings of equal length, and its mathematical expression is:

[0144]

[0145]

[0146] in, F is the Hamming distance, representing the number of different characters between two strings; F is the sample data image fingerprint, a hexadecimal string of length 16, obtained by processing qualified sample cables in step S3; G is the real-time data image fingerprint, a hexadecimal string of length 16, obtained by processing cables produced online in step S1. and It is the i-th character in the string; δ is an indicator function, when... = δ=0 if true, otherwise δ=1. The string length in the parameters is 16, obtained through DCT transformation and binary sequence conversion. The characters are derived from hexadecimal encoding, representing low-frequency features of the image. A smaller Hamming distance value H indicates more similar textures; if used directly, it needs to be converted into a score through normalization or thresholding.

[0147] In terms of connection, surface texture similarity reflects appearance features, but insulation layer thickness, as a structural parameter, also requires similarity calculation. The thickness data is a preprocessed numerical value (such as a scalar or vector obtained through moving average filtering and normalization in step S1), therefore, thickness similarity calculation is based on absolute difference. Assuming the sample thickness data is... The real-time thickness data comes from a qualified database. If the thickness similarity is derived from the current measurement, it can be directly obtained by calculating the complement of the absolute difference, as shown in the following formula:

[0148]

[0149] in, The sample thickness data is obtained by preprocessing the thickness data of qualified sample cables in step S3. The thickness data is real-time, obtained by preprocessing the thickness data of the online-produced cables in step S1. This formula directly provides a similarity score within the range [0,1], with values ​​closer to 1 indicating greater thickness similarity. If the thickness data is a multi-dimensional vector (e.g., containing mean and variance), Euclidean distance can be used and normalized. This processing method is consistent with the setting of statistical features for extracting thickness data in step S1, ensuring the scalability of the solution.

[0150] Finally, the similarity results of surface texture and thickness are combined to obtain the insulation layer similarity score. Since the two components have different dimensions, normalization is required first. The Hamming distance H of the surface texture can be converted into a similarity score using the following formula:

[0151]

[0152] in, This is the image similarity score; 16 is the string length. Thickness similarity. Normalized. Then, a weighted average is calculated based on actual needs, using the following formula:

[0153]

[0154] in, It is an insulation layer similarity score; and These are weighting coefficients, and their initial values ​​can be set based on the importance of the insulation layer texture and thickness to the overall quality (for example, when more attention is paid to surface defects, the initial value can be set to...). =0.6, =0.4), and its optimal value can be accurately calibrated using optimization methods such as grid search on historical datasets; Insulation layer similarity score The closer the similarity score is to 1, the better the overall quality of the insulation layer. The entire calculation process ensures the comprehensiveness and interpretability of the similarity score, laying the foundation for the quality determination in step S6.

[0155] This step, a crucial link in the overhead insulated cable quality monitoring process, plays a core role in quantifying the differences between qualified insulation sample data and pre-processed real-time insulation data through similarity calculation using the Hamming distance algorithm and supplementary numerical values. This generates an insulation similarity score, thereby achieving an objective assessment of the current state of the insulation layer. This step integrates surface texture and thickness features. Surface texture is captured by Hamming distance comparison based on image fingerprints, capturing microscopic texture changes; thickness data is calculated using normalized absolute difference to reflect structural parameter deviations. Finally, a standardized score is output by weighted averaging the two results, providing direct and reliable input for the subsequent quality judgment step S6. This serves as a bridge between the preceding and following steps—transforming data comparison into an operational scoring indicator and laying the data foundation for the entire monitoring system's decision-making process, ensuring the comprehensiveness and accuracy of the quality assessment.

[0156] Step S5: Use Hausdorff distance to calculate the similarity between the UV-resistant coating feature data and the UV-resistant coating qualified sample feature data to obtain a UV-resistant coating similarity score.

[0157] This step uses the Hausdorff distance to calculate the similarity between the feature data of qualified anti-UV coating samples and the feature data of anti-UV coatings, in order to obtain the anti-UV coating similarity score.

[0158] Hausdorff distance is used to measure the similarity between two point sets J and K, where J represents the sample data point set (such as coating texture feature points) and K represents the real-time data point set. The formula is as follows:

[0159]

[0160] in, It is the bidirectional maximum-minimum distance, the ultimate measure of difference between two point sets J and K. It takes the maximum of the directed distances in both directions, ensuring that it captures the maximum degree of mismatch between the two point sets in any direction. The smaller the value, the more similar the two point sets are; h(J,K) represents the directed Hausdorff distance from point set J to point set K, and h(K,J) represents the directed Hausdorff distance from point set K to point set J, defined as:

[0161]

[0162]

[0163] in, d(a,b) is a feature vector in the sample data point set J and the real-time data point set K; d(a,b) is the distance function between point a and point b, usually using Euclidean distance, and the calculation formula is as follows:

[0164]

[0165] in, and These are the eigenvalues ​​of eigenvectors a and b in the k-th dimension; This represents the dimension of the feature vector. A smaller value for H(J,K) indicates greater similarity between point sets J and K. This method is sensitive to noise, which can lead to inaccurate similarity calculations. To overcome the noise sensitivity of the basic Hausdorff distance, upper and lower bound estimation methods are proposed to improve computational efficiency and robustness. These methods are calculated based on the minimum bounding rectangle (MBR) of the point set, thus reducing the overhead of directly traversing all points. The lower bound of the Hausdorff distance is given by: [Formula omitted for brevity]. Defined as:

[0166]

[0167] Among them, MBR J and MBR K are the minimum bounding rectangles MBR of point sets J and K, which are automatically generated by R-tree indexing and are used to approximately represent the spatial extent of the entire point set. MINDIST represents the set of all boundary faces of the minimum bounding rectangle MBRJ; ) is noodles The minimum distance to MBRK is obtained by calculating the distance between the face and the nearest point in K. The parameter i iterates through all faces of J, ensuring the lower bound is the maximum distance among all possible faces. This is derived from spatial indexing techniques for fast similarity estimation. Similarly, the upper bound of the Hausdorff distance... (J,K) is defined as:

[0168]

[0169] Among them, MAXMINDIST ( ) is noodles The maximum minimum distance to MBR K is calculated by considering the distance between the face and the farthest point of K. These upper and lower bound parameters are automatically generated from the point set using R-tree indexing.

[0170] The Incremental Hausdorff Distance Similarity Metric (INC-HD) is an improved method that aims to accurately calculate the Hausdorff distance by dynamically traversing the R-tree nodes of a point set, while improving computational efficiency and robustness to noise. This algorithm emphasizes process description; its core lies in using a primary priority queue (MPQ) and a secondary priority queue (SPQ) to manage the node traversal order, thereby avoiding unnecessary computational overhead. The basic process of the INC-HD algorithm is described below:

[0171] The INC-HD algorithm begins with the initialization phase. First, it constructs R-tree index structures, RX and RY, for the sample data set X and the real-time data set Y, respectively. Then, it initializes a primary priority queue MPQ, which stores entries in descending order. Each entry contains a node NX (from RX), an upper bound UB (representing the upper bound of the Hausdorff distance from NX to RY), and an associated secondary priority queue SPQ. Simultaneously, it initializes SPQ, which stores entries in ascending order. Each entry contains a node NY (from RY) and a lower bound LB (representing the lower bound of the Hausdorff distance from NX to NY). During initialization, the root node of RX is inserted into MPQ with its upper bound set to infinity, while the root node of RY is inserted into SPQ with its lower bound set to 0, thus laying the foundation for subsequent traversal.

[0172] Next, the algorithm enters the iterative traversal phase. In each iteration, the algorithm retrieves the head node (i.e., the entry with the largest UB) from the MPQ and checks the head node of its associated SPQ (i.e., the entry with the smallest LB). Based on the level depth of nodes NX and NY, the algorithm decides whether to traverse the child nodes of RX or RY: if the leaf level distance of NX is farther than that of NY, the TRA-X function is called to process the subtree of RX; otherwise, the TRA-Y function is called to process the subtree of RY. The TRA-X function calculates new upper and lower bounds for each child node C of NX, updates the SPQ, and inserts the new entry into the MPQ; similarly, the TRA-Y function updates the SPQ and MPQ for the child nodes of NY. This process dynamically adjusts the traversal order through the branch and bound principle, prioritizing nodes that may produce larger Hausdorff distances, thereby gradually narrowing the range of the upper and lower bounds.

[0173] During the traversal, the algorithm continuously compares the distance estimates between nodes. When all nodes in the MPQ and SPQ are leaf nodes (i.e., actual data points), the algorithm terminates and returns the current Hausdorff distance value. This process ensures that the algorithm only visits necessary nodes, reducing redundant computation. Ultimately, the INC-HD algorithm achieves efficient distance calculation through this dynamic queue management. Compared to the basic method, it handles noise interference better and provides a reliable basis for coating similarity scoring. The entire process demonstrates the algorithm's advantages in balancing accuracy and efficiency, making it suitable for practical applications.

[0174] The Hausdorff distance value is then converted into an intuitive similarity score, calculated using the following formula:

[0175]

[0176] in, This is the UV-resistant coating similarity score, a standardized value ranging from 0 to 1, used to represent the degree of similarity between UV-resistant coatings. The closer the score is to 1, the more similar the sample data is to the real-time data (i.e., the smaller the Hausdorff distance); H(J,K) is the actual distance value calculated using the aforementioned Hausdorff distance, representing the degree of difference between point set J (sample data) and point set K (real-time data); the denominator... The design ensures the mathematical stability of the formula when H(J,K)=0 (indicating complete similarity). 1; when As the distance increases, the score gradually decreases, but never becomes negative or exceeds 1. Adding 1 is to avoid the error of dividing by zero and to smooth the score curve, making it suitable for various distance values.

[0177] This step calculates similarity between UV-resistant coating feature data and qualified UV-resistant coating sample feature data using Hausdorff distance. The core function of this step is to quantify the consistency of coating features, thereby generating a UV-resistant coating similarity score. Specifically, it measures the maximum and minimum distance between two point sets based on the basic Hausdorff distance formula, and improves upon this by introducing the INC-HD algorithm. This algorithm utilizes a primary priority queue (MPQ) and a secondary priority queue (SPQ) to dynamically traverse R-tree nodes to calculate upper and lower bounds, thereby enhancing computational efficiency and noise resistance. This step provides crucial similarity indicators for subsequent coating quality assessment, decision support, and further data analysis, ensuring the reliability and robustness of the entire inspection process.

[0178] Step S6: Based on the insulation layer similarity score and the UV-resistant coating similarity score, calculate the overall quality score, and realize online monitoring of the production quality of overhead insulated cables based on the overall quality score.

[0179] Based on the insulation layer similarity score and UV-resistant coating similarity score obtained in the previous steps, the overall quality score of the overhead insulated cable is calculated and compared with a preset threshold, thereby achieving online monitoring of the production quality of the overhead insulated cable. This step is the final decision-making stage of the quality monitoring process, aiming to comprehensively evaluate the overall production quality of the cable and ensure that it meets the standard requirements.

[0180] First, the overall quality score of overhead insulated cables is obtained by combining the similarity scores of the insulation layer and coating layer using a weighted average method. The weighted average formula is as follows:

[0181]

[0182] in, It is the overall quality score, a value between 0 and 1, with a higher value indicating better overall quality; The insulation layer similarity score is derived from the output of step S4. This score is calculated based on the similarity of the insulation layer data and is typically in the range of 0 to 1, where 1 represents complete similarity and 0 represents complete dissimilarity. The UV-resistant coating similarity score is derived from the output of step S5. This score is calculated based on the similarity of the coating data. The range is similar, and the higher the value, the more consistent the coating quality. and These are weighting coefficients, representing the relative importance of the insulation layer and coating layer in the overall quality assessment, satisfying... + =1, these weights are usually set based on domain knowledge or historical data to ensure that the score reflects the actual quality priority.

[0183] After the overall quality score is calculated, it needs to be compared with a preset threshold to generate the monitoring results. The threshold comparison formula is as follows:

[0184]

[0185] The determination of the preset threshold T should be based on the principles of statistical process control. This involves collecting no fewer than 500 historical qualified samples, calculating their quality scores, and initializing T to the P5 or P10 quantile of that score set. Subsequently, the threshold should be validated and optimized using a validation set containing samples with known defects, focusing on the false negative rate to ensure it is below an acceptable risk level of 0.5%. The final T value is typically between 0.85 and 0.95 and should be fine-tuned according to specific product specifications and safety levels. The monitoring result is a binary output used to guide the production process. If the score is greater than or equal to the threshold, the cable is considered qualified and can proceed to the next stage; otherwise, it is considered unqualified and may require further inspection or adjustment.

[0186] The core function of this step is to calculate the overall quality score of the overhead insulated cable based on the insulation layer similarity score and UV-resistant coating similarity score obtained in the previous steps, using a weighted average formula. The weighting coefficients are set according to domain knowledge, thereby generating a standardized score to quantify the overall quality consistency of the cable. Subsequently, this overall quality score is compared with a preset threshold T, and a binary monitoring result is output, thus realizing online monitoring of the production quality of overhead insulated cables. This step provides the final decision-making basis for the entire online monitoring process, ensuring the objectivity and automation of quality assessment and enhancing the reliability and efficiency of the entire production chain.

[0187] This patent presents an online monitoring method for the production quality of overhead insulated cables, aiming to address the problem that traditional monitoring methods only focus on surface defects of the insulation layer and cannot cover the quality of the UV-resistant coating. By collecting and preprocessing real-time data on the insulation layer—namely, the surface image and thickness—a multi-scale local binary method is used to extract surface image features of the UV-resistant coating, obtaining qualified sample data containing both the insulation layer and the coating. Then, Hamming distance and Hausdorff distance are used to calculate the similarity scores between the insulation layer and the coating, respectively. Finally, the two scores are combined to obtain an overall quality score, enabling online monitoring of cable production quality, ensuring cable insulation performance and weather resistance, and reducing subsequent maintenance costs.

[0188] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0189] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

Claims

1. A method for online monitoring of the production quality of overhead insulated cables, characterized in that, Includes the following steps: Step S1: Collect real-time data of the insulation layer of the overhead insulated cable and perform data preprocessing to obtain preprocessed real-time data of the insulation layer; Step S2: Obtain an image of the UV-resistant coating surface of the overhead insulated cable, and use the multi-scale local binary method to extract features from the UV-resistant coating surface image to obtain UV-resistant coating feature data; Step S3: Obtain qualified sample data of overhead insulated cables, including qualified sample data of insulation layer and characteristic data of qualified sample data of UV-resistant coating; Step S4: Calculate the similarity between the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer to obtain the insulation layer similarity score; Step S5: Use Hausdorff distance to calculate the similarity between the UV-resistant coating feature data and the UV-resistant coating qualified sample feature data to obtain the UV-resistant coating similarity score; Step S6: Based on the insulation layer similarity score and the UV-resistant coating similarity score, calculate the overall quality score, and realize online monitoring of the production quality of overhead insulated cables based on the overall quality score.

2. The method for online monitoring of the production quality of overhead insulated cables according to claim 1, characterized in that, Real-time data of the insulation layer of overhead insulated cables is collected and preprocessed to obtain preprocessed real-time insulation layer data, including: During the data acquisition phase, cameras deployed on the production line periodically capture images of the insulation layer surface of the insulated cable. Simultaneously, a laser thickness gauge is used to measure the thickness of the insulation layer in real time. The laser thickness gauge is based on the principle of laser triangulation, which calculates the thickness by emitting a laser beam and receiving the reflected signal, and outputs continuous time series data. For the preprocessing of the surface image of the insulation layer of the insulated cable, the acquired image is first compressed and grayscaled to reduce computational complexity. After the image preprocessing is completed, feature extraction and encoding are performed. Discrete cosine transform (DCT) is performed on the grayscale image to convert the image from the spatial domain to the frequency domain to capture its frequency features. Next, the low-frequency domain matrix is ​​encoded to generate an image fingerprint. The preprocessing method for the insulation layer thickness of overhead insulated cables is as follows. Since the thickness data is a continuously collected numerical sequence, data cleaning and normalization are required. First, the original insulation layer thickness data is subjected to moving average filtering to smooth random noise. Then, the filtered thickness data is normalized and scaled to the [0,1] interval to avoid the influence of dimensions. After preprocessing the image and thickness data, the two are integrated into preprocessed real-time insulation layer data. .

3. The method for online monitoring of the production quality of overhead insulated cables according to claim 2, characterized in that, Obtain images of the UV-resistant coating surface of the overhead insulated cable, including: The surface of the UV-resistant coating of the overhead insulated cable is scanned using a near-infrared spectral sensor. The sensor captures an image of the UV-resistant coating surface of the overhead insulated cable at a preset resolution. This surface image is a grayscale image, and the grayscale value of each pixel reflects the reflection characteristics of the coating to near-infrared light. During scanning, the ambient lighting conditions need to be controlled to avoid external interference and ensure image quality.

4. The method for online monitoring of the production quality of overhead insulated cables according to claim 3, characterized in that, The surface image of the UV-resistant coating is subjected to feature extraction using a multi-scale local binary method to obtain UV-resistant coating feature data, including: The basic Local Binary Mode (LBP) application process is as follows: On the preprocessed image of the UV-resistant coating surface, LBP is applied. For each pixel, a circular neighborhood is defined centered on it, and multiple sets of scale parameters are set. The radius R is 1, 2, and 3, and the number of sampling points P is 8, 16, and 24 respectively. The gray values ​​of the neighboring pixels are then calculated. With the gray value of the center pixel The difference is binarized using the sign function s(x), as shown in the following formula: ； in, ; It is the gray value of the p-th pixel in the neighborhood, starting from... Obtain from a circular neighborhood centered at R with radius R; is the gray value of the center pixel, directly obtained from the preprocessed near-infrared grayscale image. 'c' explicitly indicates that this gray value belongs to the calculated center pixel and is the benchmark point for the entire LBP feature calculation process. The gray values ​​of all neighboring pixels are compared to it. Subsequently, the mathematical formula for generating the code is as follows: ； Where P is the number of sampling points, which is a preset constant; weights Based on fixed-location assignment, this step extracts local symbolic features, but it is sensitive to noise and rotational changes. To enhance robustness, a multi-scale extension and improvement method is integrated. The SMDCP method is employed, combining sign and magnitude information. First, the mean value of local neighboring pixels is calculated using the following formula: ； in, It is the average value of local neighborhood pixels, used as a threshold substitute. This enhances the algorithm's robustness to noise because the mean is more stable than individual pixel values, and the sign pattern... The definition is as follows: ； in, The subscript S in the code represents the name of the mathematical function, explicitly indicating that the key to this improvement lies in using a sign function to generate binary code, thereby achieving stronger noise resistance; this is an improvement on LBP, which uses neighborhood mean. Non-center pixel value As a threshold, a symbol pattern is generated. This improvement effectively suppresses noise interference within a uniform region and enhances feature stability. Subsequently, amplitude calculation is performed to extract differential amplitude information, as shown in the following formula: ； in, The amplitude represents the absolute value of the grayscale difference. This amplitude information reflects the strength of the contrast and is an important feature of the texture. Combining amplitude information with symbolic information can more comprehensively describe the texture. At the same time, a directional contour mode is introduced to capture the spatial orientation of the texture. First, the orientation angle of the neighboring points relative to the center point is calculated, as follows: ； in, It is the direction angle; σ is a small regularization constant; then the continuous direction angles are quantized into finite discrete direction numbers, as shown in the following formula: ； in, B is the quantized direction number, and B is the preset total number of direction partitions. Combining amplitude information with sign and direction information can more comprehensively describe the texture. At the same time, the LBPmr method is introduced to improve the weighting problem, that is, to adjust the amplitude... Sort in ascending order to obtain a dynamic weighted index. The final integrated coding formula is: ； Among them, s It is a symbolic value; It is the amplitude Sort index obtained by ascending order; sort index by amplitude Replaces fixed weights This makes the encoding no longer dependent on the fixed geometric position of the neighboring points, but on the relative strength of their gray-level contrast, thus making the features invariant to the rotation of the image and enhancing robustness. It is a directional weighting function that makes features more discriminative of texture patterns in specific directions; in addition, the GNP operator is applied to utilize absolute grayscale information, as shown in the following formula: ； in, It is the number of discrete intervals, and is a preset constant; It is an interval matcher that, through a judgment operation, matches consecutive center pixel gray values. This is converted into a discrete number GNP value representing the grayscale range to which it belongs; The nth threshold is adaptively determined by analyzing the global histogram of historical qualified sample images and using image segmentation algorithms such as the maximum inter-class variance method to better separate different gray-level distributions. Multi-threshold segmentation is used to separate the gray-level values ​​of the center pixels. Mapping to discrete intervals; Multi-scale and multi-feature fusion is employed. At various set scales, SMDCP, LBPmr, and GNP operators are applied to generate multiple feature images. A grayscale histogram is calculated for each feature image. All histogram feature vectors are concatenated to generate a compact feature vector, which serves as the feature data for the UV-resistant coating. .

5. The method for online monitoring of the production quality of overhead insulated cables according to claim 4, characterized in that, Obtain qualified sample data of overhead insulated cables, including: The data collection phase involves a comprehensive scan of certified overhead insulated cables using a high-precision sensor system. Specifically, near-infrared spectral sensors and accompanying imaging equipment are used to simultaneously capture images of the insulation layer surface and the UV-resistant coating surface. Insulation layer data acquisition focuses on surface texture and thickness information, using a camera to capture high-resolution images and combining this with a laser thickness gauge to record thickness values ​​in real time. UV-resistant coating data acquisition utilizes a near-infrared spectral sensor to scan the coating surface and obtain grayscale images. The grayscale value of each pixel reflects the coating's reflectivity to near-infrared light. The scanning process is conducted under strictly controlled environmental conditions, maintaining constant temperature and stable lighting to minimize external interference. All raw data comes directly from the sensor output to ensure representativeness and authenticity, and is stored in matrix form for easy subsequent processing.

6. The method for online monitoring of the production quality of overhead insulated cables according to claim 5, characterized in that, The qualified sample data includes qualified sample data for insulation layers and characteristic data for qualified samples for UV-resistant coatings, including: The surface image of the insulation layer is compressed and converted to grayscale. Average pooling is used to reduce the image size, and a weighted average method is applied to convert it to grayscale to highlight brightness features. Feature extraction is then performed, including discrete cosine transform to capture frequency information and generate a compact image fingerprint. For thickness data, moving average filtering is performed to smooth noise, and normalization is applied to scale it to a standard range. Finally, the image and thickness features are integrated into a unified multidimensional feature vector, representing the qualified sample data of the insulation layer. ; For the UV-resistant coating sample data in the qualified sample data of overhead insulated cables, the processing method is exactly the same as in step S2. First, an adaptive median filter is used to smooth the image and reduce noise interference. Then, texture features are extracted based on multi-scale local binary mode, including sign, amplitude, and absolute grayscale information. A compact feature representation, i.e., the UV-resistant coating qualified sample feature data, is generated through multi-scale analysis and feature dimensionality reduction. .

7. The method for online monitoring of the production quality of overhead insulated cables according to claim 6, characterized in that, Similarity calculations are performed on the qualified sample data of the insulation layer and the preprocessed real-time data of the insulation layer to obtain an insulation layer similarity score, including: For calculating the similarity of the surface texture of the insulating layer, the method is the Hamming distance algorithm based on image fingerprints. The Hamming distance is defined as the number of different characters at corresponding positions in two strings of equal length, and its mathematical expression is: ； ； in, F is the Hamming distance, representing the number of different characters between two strings; F is the sample data image fingerprint, a hexadecimal string of length 16, obtained by processing qualified sample cables in step S3; G is the real-time data image fingerprint, a hexadecimal string of length 16, obtained by processing cables produced online in step S1. and It is the i-th character in the string; δ is an indicator function, when... = When δ=0, otherwise δ=1. Among the parameters, the string length is 16, which is obtained by DCT transformation and binary sequence conversion. The characters come from hexadecimal encoding and represent the low-frequency domain features of the image. Thickness similarity can be directly obtained by calculating the complement of the absolute difference, as shown in the following formula: ； in, The sample thickness data is obtained by preprocessing the thickness data of qualified sample cables in step S3. It is real-time thickness data, which is obtained by preprocessing the thickness data of the online-produced cable in step S1; By combining the similarity results of surface texture and thickness, the similarity score of the insulating layer is obtained. Since the two parts have different dimensions, they need to be normalized first. The Hamming distance H of the surface texture can be converted into a similarity score using the following formula: ； in, This is the image similarity score; 16 represents the string length and thickness similarity. The result has been normalized. Then, a weighted average is calculated based on actual needs, as shown in the following formula: ； in, It is an insulation layer similarity score; and It is a weighting coefficient, and the initial value can be set according to the importance of the insulation layer texture and thickness to the overall quality.

8. The method for online monitoring of the production quality of overhead insulated cables according to claim 7, characterized in that, The similarity score of the UV-resistant coating is obtained by calculating the similarity between the feature data of the UV-resistant coating and the feature data of qualified UV-resistant coating samples using Hausdorff distance, including: The Hausdorff distance is used to measure the similarity between two point sets J and K, where J represents the sample data point set and K represents the real-time data point set. The formula is as follows: ； in, It is a bidirectional maximum-minimum distance, the ultimate measure of difference between two point sets J and K; it takes the maximum of the directed distances in two directions, ensuring that it captures the maximum degree of mismatch between the two point sets in any direction. The smaller the value, the more similar the two point sets are; h(J,K) represents the directed Hausdorff distance from point set J to point set K, and h(K,J) represents the directed Hausdorff distance from point set K to point set J. The Hausdorff distance value is then converted into an intuitive similarity score, calculated using the following formula: ； in, The UV-resistant coating similarity score is a standardized value ranging from 0 to 1, used to represent the degree of similarity between the UV-resistant coatings. The closer the score is to 1, the more similar the sample data is to the real-time data; H(J,K) is the actual distance value calculated using the aforementioned Hausdorff distance, representing the degree of difference between point set J and point set K; the denominator... The design ensures the mathematical stability of the formula when H(J,K)=0. 1; when As the score increases, it gradually decreases, but never becomes negative or exceeds 1.

9. The method for online monitoring of the production quality of overhead insulated cables according to claim 8, characterized in that, Based on the similarity scores of the insulation layer and the UV-resistant coating, an overall quality score is calculated, including: The overall quality score of overhead insulated cables is obtained by combining the similarity scores of the insulation layer and the coating layer using a weighted average method. The weighted average formula is as follows: ； in, It is the overall quality score, a value between 0 and 1, with a higher value indicating better overall quality; The insulation layer similarity score is derived from the output of step S4. This score is calculated based on the similarity of the insulation layer data and is typically in the range of 0 to 1, where 1 represents complete similarity and 0 represents complete dissimilarity. The UV-resistant coating similarity score, derived from the output of step S5, is calculated based on the similarity of the coating data. A higher score indicates more consistent coating quality. and These are weighting coefficients, representing the relative importance of the insulation layer and coating layer in the overall quality assessment, satisfying... + =1, these weights are usually set based on domain knowledge or historical data to ensure that the score reflects the actual quality priority.

10. The method for online monitoring of the production quality of overhead insulated cables according to claim 9, characterized in that, Based on an overall quality score, online monitoring of the production quality of overhead insulated cables is achieved, including: After the overall quality score is calculated, it needs to be compared with a preset threshold to generate the monitoring results. The threshold comparison formula is as follows: ； The determination of the preset threshold T should be based on the principles of statistical process control. This involves collecting no fewer than 500 historical qualified samples, calculating their quality scores, and initializing T to the P5 or P10 quantile of that score set. Subsequently, the threshold should be validated and optimized using a validation set containing samples with known defects, focusing on the false negative rate to ensure it is below the acceptable risk level of 0.5%. The final T value is typically between 0.85 and 0.95 and should be fine-tuned according to specific product specifications and safety levels. The monitoring result is a binary output used to guide the production process. If the score is greater than or equal to the threshold, the cable is considered qualified and can proceed to the next stage; otherwise, it is considered unqualified and may require further inspection or adjustment.

Images