Power fitting defect detection method and system based on visual intelligence

Abstract

The application belongs to the technical field of data processing, and specifically discloses a power hardware defect detection method and system based on visual intelligence, wherein the method comprises a power hardware defect detection method based on visual intelligence, which comprises the following steps: obtaining an initial image data set containing a target power equipment surface magnetic field distribution image according to the magnetic field imaging data of a target power equipment component collected in a target environment; performing feature extraction on the initial image data set to obtain a feature vector set, wherein the feature vector set represents a magnetic field anomaly mode in the image; and comparing the feature vector set with a preset deviation standard, and if the comparison result shows that the deviation exceeds the standard, it is judged that the environment interference affects, and an interference correction parameter is obtained, wherein the interference correction parameter is used for adjusting the imaging setting; and the application aims to solve the problems of the influence of complex environment interference on the accuracy of magnetic field imaging data and the accurate classification of defect types in the prior art.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, specifically to a method and system for detecting defects in power fittings based on visual intelligence. Background Technology

[0002] In the complex operating environment of power equipment, magnetic field imaging technology is used to detect the magnetic field distribution on the equipment surface to identify potential defects and anomalies. However, accurately acquiring and analyzing magnetic field imaging data in a variable and interference-filled environment to achieve precise identification of equipment defects has become a core and urgent technical challenge. This major problem stems from the fact that magnetic field imaging data in complex environments is easily affected by external interference, such as electromagnetic noise, temperature changes, and surrounding metal structures. This leads to blurred or distorted magnetic field distribution features in the initially acquired image data, making it difficult to accurately reflect the true state of the equipment and thus affecting the reliability of subsequent defect identification. Further, in practical applications, for example, in the inspection of high-voltage transmission equipment, strong electromagnetic interference or mechanical vibration may exist around the equipment during operation. The acquired magnetic field images may exhibit abnormal pattern shifts, directly causing the inability to correctly distinguish between magnetic field changes caused by defects in the equipment itself and artifacts caused by environmental interference during feature extraction. This results in significant deviations in subsequent classification and defect type determination.

[0003] In addition, some related minor issues have arisen around this core problem, such as how to dynamically correct interference effects in data processing to improve image quality, how to ensure that the classification model's ability to distinguish different defect types adapts to various devices and scenarios, and how to optimize the recognition logic when defect features do not completely match the standard template.

[0004] These seemingly minor issues all essentially serve to address the core problem: ensuring the accuracy of magnetic field imaging data and the reliability of defect identification in complex environments. This is particularly crucial and unique in actual power equipment inspection scenarios, where the challenges of diverse equipment types and varying operating environments make solving this problem especially important. Summary of the Invention

[0005] This invention provides a method and system for detecting defects in power fittings based on visual intelligence, aiming to solve the problems of the impact of complex environmental interference on the accuracy of magnetic field imaging data and the accurate classification of defect types in the prior art.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A visual intelligence-based method for detecting defects in power fittings includes: obtaining an initial image dataset containing images of the magnetic field distribution on the surface of the target power equipment based on magnetic field imaging data of the target power equipment components in a target environment; extracting features from the initial image dataset to obtain a set of feature vectors, which represent abnormal magnetic field patterns in the images; comparing the set of feature vectors with a preset deviation standard, and if the comparison result shows that the deviation exceeds the standard, it is determined to be due to environmental interference, and interference correction parameters are obtained, which are used to adjust imaging settings; adjusting the magnetic field imaging protocol through the interference correction parameters to obtain a corrected image dataset; classifying the corrected image dataset to obtain a distribution of defect types covering common equipment problems; if the matching degree between the defect type distribution and the standardized design template is lower than the standard, updating the parameter precision matching model to obtain optimized defect feature vectors; and generating an automated detection process sequence based on the optimized defect feature vectors to obtain the final defect identification result.

[0007] In one aspect of this disclosure, obtaining an initial image dataset containing an image of the magnetic field distribution on the surface of the target power equipment based on the collected magnetic field imaging data of the target power equipment components in the target environment includes: The magnetic field data of the target power equipment in the target environment are collected by a magnetic field imaging device to construct an initial image dataset containing the magnetic field distribution on the surface of the equipment. The initial image dataset is denoised to obtain the first set of processed images. Based on the magnetic field distribution images in the first image set, magnetic field data features of the device surface are extracted to obtain a feature dataset; The feature dataset is classified to determine the magnetic field distribution pattern of different device components in complex environments, and the classification result is obtained. If the magnetic field distribution of one or more device components in the classification results exceeds the preset threshold range, then a depth analysis is performed on the images in the first image set corresponding to that component to obtain detailed distribution information of the abnormal area; Based on the distribution information of the abnormal areas, and combined with environmental impact data, a correction process is performed to obtain a corrected second image set. Based on the second image set, a magnetic field distribution adjustment model of the target device in a complex environment is generated to obtain the final optimized magnetic field distribution parameters.

[0008] In one aspect of this disclosure, the step of extracting features from the initial image dataset to obtain a set of feature vectors, the set of feature vectors representing magnetic field anomaly patterns in the image, includes: A pre-built feature extraction network is used to process the initial image dataset to generate a set of feature vectors, which are used to reflect the magnetic field anomaly pattern. By performing cluster analysis on the feature vector set and using the K-means algorithm to group the vectors, preliminary abnormal pattern classification results are obtained. Based on the preliminary abnormal pattern classification results, the feature vector distribution range within each classification group is obtained. If the vector distribution within a certain group exceeds the preset threshold, it is marked as a potential abnormal pattern group, and a high-risk area is obtained. Based on the marked potential abnormal pattern group, representative feature vectors are extracted from it to generate an abnormal pattern subset and obtain a refined feature representation. Based on the refined feature representation, the support vector machine algorithm is used to further classify the subset of abnormal patterns to determine whether there are significant magnetic field anomaly patterns. By spatially mapping the classified magnetic field anomaly patterns, a spatial distribution map of the anomaly patterns is generated, and a visualized anomaly distribution result is obtained. The abnormal concentrated areas in the spatial distribution map are obtained, and the abnormal patterns are verified in combination with image data to determine whether the abnormal patterns are consistent with the actual image features, so as to obtain the final abnormal pattern confirmation result.

[0009] In one aspect of this disclosure, the step of comparing the feature vector set with a preset deviation standard, and determining that environmental interference affects the result if the comparison result shows that the deviation exceeds the standard, and obtaining interference correction parameters, wherein the interference correction parameters are used to adjust the imaging settings, includes: The feature vector set data is acquired, and relevant information is extracted from the imaging device through a pre-established acquisition module. If the integrity of the extracted data meets the preset threshold requirements, it is determined to be valid data, and a preliminary vector set is obtained. The preliminary vector set is compared with the preset deviation standard, and the difference between the two is calculated using statistical analysis methods. If the difference exceeds the standard range, it is determined that there is environmental interference, and a preliminary indicator of the interference effect is obtained. Based on the preliminary indicators of the interference impact, interference patterns in historical data records are obtained. By comparing the similarity between the current indicators and historical patterns, potential interference types are determined, and the classification results of interference types are obtained. Based on the classification results of the interference type, the pre-established correction parameter database is called to obtain the correction parameters that match the current interference type, and the specific values ​​of the correction parameters are obtained. Based on the specific values ​​of the calibration parameters, the relevant configuration of the imaging settings is adjusted. Through the automated setting adjustment module, the parameter values ​​are applied to the imaging device to obtain the adjusted imaging configuration. The operating status of the adjusted imaging configuration is obtained, and the operating data of the target device is collected. If the operating data is consistent with the expected output, the adjustment process is determined to be complete, and the final imaging effect optimization status is obtained. Based on the final optimized imaging effect, the complete process data of this interference correction is recorded, and the data is archived into the historical record library through the storage module to obtain reference information that can be called later.

[0010] In one aspect of this disclosure, the step of adjusting the magnetic field imaging protocol using the interference correction parameters to obtain the corrected image dataset includes: Interference features are extracted from the raw magnetic field imaging data to obtain preliminary interference distribution information; Based on the preliminary interference distribution information, the imaging protocol is adjusted to generate an optimized imaging parameter configuration. By re-executing the magnetic field imaging process using the optimized imaging parameter configuration, a corrected preliminary image dataset is obtained. If the reduction of interference in the initial image dataset does not reach the preset threshold, the protocol parameters are iteratively adjusted to regenerate the corrected image until the preset conditions are met, and the adjusted image dataset is obtained. Based on the adjusted image dataset, an image quality assessment tool is used to determine whether the image quality meets the imaging accuracy requirements. The image quality assessment results are used to determine whether the imaging accuracy is up to standard. If the image quality is up to standard, the image is then optimized in a second step using data processing methods to obtain the final image dataset of the corrected image. The final corrected image dataset is used to save the data using a storage tool, ensuring data integrity and completing the processing flow.

[0011] In one aspect of this disclosure, the step of classifying the corrected image dataset to obtain a distribution of defect types covering common device problems includes: The corrected image dataset is processed to identify the types of defects it contains and obtain preliminary defect classification results; By analyzing the preliminary defect classification results, the distribution of defect types is summarized, and the frequency and main categories of each type of defect are obtained. If the confidence level of a certain type of defect in the preliminary classification results is lower than the preset threshold, then the image content of that part will be re-corrected to obtain optimized image data. Based on the optimized image data, a secondary classification process is performed using the same classification model to determine the accuracy of the defect type and obtain the updated classification result. Based on the updated classification results, and combined with the common categories of equipment problems, the types are divided to obtain the correspondence between defect types and equipment problems; Based on the statistical analysis of the correspondence between the classified defect types and equipment problems, the correlation pattern between defect distribution and equipment problems is obtained, and the main sources of problems are determined. Based on the aforementioned association pattern, a support vector machine model is used to further verify the relationship between defect distribution and equipment problems, thereby obtaining the final mapping result between defect type distribution and problem coverage.

[0012] In one aspect of this disclosure, the step of updating the parameter precise matching model to obtain the optimized defect feature vector if the matching degree between the defect type distribution and the standardized design template is lower than the standard includes: By analyzing the defect types and distribution, initial matching data is obtained, and a comparison benchmark with the standardized template is determined. If the comparison between the initial matching data and the standardized template shows a low matching degree, a preset optimization mechanism is used to screen the parameter adjustment schemes and obtain a set of parameters after preliminary adjustment. Based on the initially adjusted parameter set, and in response to the requirements for building a precise model, the internal weights of the model are updated to determine whether the design matching requirements are met. If the updated model weights still do not meet the design matching standard, the defect features are reclassified using the support vector machine algorithm through iterative calculation to obtain the optimized feature vectors. Based on the optimized feature vector and the distribution of defect types, the output logic of the model is adjusted to obtain the final defect feature representation. By using the final defect feature representation, the matching degree with the standardized template is verified, evaluation data of the optimization results is obtained, and it is determined whether they meet the expected standards. If the evaluation data indicates that the optimization results still need improvement, the model parameters are readjusted based on the correlation between the distribution and the defect characteristics to obtain a more accurate matching output.

[0013] In one aspect of this disclosure, the step of generating an automated detection process sequence based on the optimized defect feature vector to obtain the final defect identification result includes: By using OpenCV to extract and analyze defect features, an initial feature dataset is constructed to obtain preliminary defect description information; Based on the preliminary defect description information, load data as input, determine the classification boundary between vectors, and obtain the optimized feature representation; Based on the optimized feature representation, key control points are extracted. If the key control points meet the preset threshold range, the sequence is determined to be valid, and an executable process scheme is obtained. Based on the executable process scheme, the defect identification status of the intermediate stage is obtained; Based on the defect identification status at the intermediate stage, numerical features are extracted and Pearson correlation coefficients are calculated with the preset final conclusion values. The Pearson correlation coefficients are configured as follows:

[0014] Where Xi is the numerical feature of the defect identification status; X represents the mean of the numerical features of the defect identification status; Yi represents the final numerical value; Y represents the mean of the final numerical values; If the correlation coefficient r is higher than the preset standard, the final identification result is confirmed.

[0015] 9. A method and system for detecting defects in power fittings based on visual intelligence, characterized in that it includes: The data acquisition module is configured to obtain an initial image dataset containing an image of the magnetic field distribution on the surface of the target power equipment based on the acquired magnetic field imaging data of the target power equipment component in the target environment. The feature extraction module is configured to perform feature extraction on the initial image dataset to obtain a set of feature vectors, wherein the set of feature vectors represents the magnetic field anomaly pattern in the image; The interference judgment module is configured to compare the feature vector set with a preset deviation standard. If the comparison result shows that the deviation exceeds the standard, it is judged to be affected by environmental interference, and interference correction parameters are obtained. The interference correction parameters are used to adjust the imaging settings. The imaging correction module is configured to perform the step of adjusting the magnetic field imaging protocol through the interference correction parameters and obtaining the corrected image dataset; The defect classification module is configured to perform classification processing on the corrected image dataset to obtain a distribution of defect types covering common equipment problems; The model optimization module is configured to update the parameters to accurately match the model and obtain the optimized defect feature vector if the matching degree between the defect type distribution and the standardized design template is lower than the standard. The result generation module is configured to generate an automated detection process sequence based on the optimized defect feature vector to obtain the final defect identification result.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention constructs an initial image dataset, extracts magnetic field distribution feature vectors, compares them with a preset deviation standard, identifies environmental interference, generates correction parameters, and then adjusts the imaging protocol to obtain a corrected image dataset with reduced interference. Subsequently, a classification model is used to analyze the defect type distribution of the corrected data. If the matching degree is insufficient, an optimization mechanism is used to update the parameters, generating accurate defect feature vectors, ultimately forming an automated detection process and outputting defect identification results. The core innovation of this invention lies in its dual mechanism of interference correction and parameter optimization, which significantly improves the accuracy of magnetic field imaging and the reliability of defect identification in complex environments, providing efficient technical support for the safe operation of target power equipment. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the defect detection method for power fittings in this invention.

[0019] Figure 2 This is a schematic diagram of the defect detection method for power fittings in this invention. Detailed Implementation

[0020] The present invention will be further described below with reference to embodiments. These embodiments are merely some, not all, of the embodiments described. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the protection scope of the present invention.

[0021] Please see Figure 1 as well as Figure 2 As shown in the figure, this embodiment discloses a method and system for detecting defects in power fittings based on visual intelligence, which may specifically include: Step 1 focuses on acquiring data from the power equipment and its components using a specialized magnetic field imaging device to obtain an initial image dataset containing the magnetic field distribution on the equipment surface. It should be noted that complex environments may include conditions such as high temperature, high humidity, and strong electromagnetic interference, which can significantly affect the magnetic field distribution. Therefore, the acquisition equipment must possess high sensitivity and anti-interference capabilities. During the acquisition process, magnetic field sensors can be deployed at multiple points to ensure coverage of key areas on the equipment surface, such as transformer casings and wire connection points. The acquired data is stored in image form, forming the initial image dataset, laying the foundation for subsequent processing.

[0022] In one embodiment, the acquisition parameters for the initial image dataset can be adjusted based on the device type and environmental characteristics. For example, the magnetic field distribution of a transformer operating in a high-temperature environment may be affected by thermal effects. In this case, the acquisition frequency can be increased and the single acquisition time shortened to reduce the interference of thermal noise on the data. The acquisition device can be equipped with a temperature compensation module to perform preliminary correction on the acquired magnetic field data by monitoring the ambient temperature in real time, ensuring the accuracy of the initial image dataset. This approach can effectively improve data quality and provide reliable support for subsequent analysis.

[0023] In steps 1-2, image preprocessing techniques are used to denoise the initial image dataset to obtain the first set of processed images.

[0024] Specifically, noise reduction primarily targets noise introduced during the acquisition process, such as environmental electromagnetic interference or image blurring caused by equipment vibration. Filtering techniques can be used to smooth image data, removing high-frequency noise while preserving the main characteristics of the magnetic field distribution.

[0025] It should be noted that during the filtering process, an appropriate filter window size must be selected based on the characteristics of the image data to avoid excessive smoothing that could lead to the loss of details in the magnetic field distribution. The processed first image set will serve as the basis for subsequent feature extraction.

[0026] For example, when processing images of the magnetic field distribution on the surface of a transformer, if obvious point noise is found in the image, the noise points can be removed by median filtering. Median filtering replaces noise points by calculating the median value of the area surrounding the pixel, which can effectively protect the edge information of the image and prevent the features of key areas of magnetic field distribution from being smoothed out. This processing method can reduce the interference of noise on subsequent analysis while ensuring image clarity.

[0027] In steps 1-3, magnetic field data features on the surface of the device are extracted based on the magnetic field distribution images in the first image set to form a feature dataset.

[0028] It should be noted that the magnetic field data features mainly include information such as magnetic field strength, distribution gradient, and local anomalies. These features can reflect the operating status of the equipment components. During the extraction process, image segmentation techniques can be used to divide the magnetic field distribution image into multiple regions, and the feature values ​​of each region can be calculated to form a feature dataset. Such a dataset can provide data support for subsequent classification and anomaly detection.

[0029] In one possible implementation, different feature extraction strategies can be set for different parts of the device components.

[0030] For example, for the transformer core area, the mean and variance of the magnetic field strength are extracted to determine whether there is magnetic saturation; while for the connecting wire area, the trend of magnetic field gradient changes is focused on to detect whether there is a risk of local overheating or breakage. By extracting features by region, the magnetic field distribution pattern on the surface of the equipment can be captured more accurately, and the representativeness of the feature dataset can be improved.

[0031] In steps 1-4, the feature dataset is classified using the support vector machine algorithm to determine the magnetic field distribution pattern of different device components in complex environments and to determine the classification result.

[0032] Specifically, the support vector machine algorithm constructs a classification hyperplane to divide the feature dataset into different categories. Each category corresponds to a magnetic field distribution pattern. During the classification process, classification labels need to be pre-set according to the type of equipment components and the operating environment, such as normal state, minor abnormality, and severe abnormality. The classification results can intuitively reflect whether the magnetic field distribution of each component meets expectations, providing a basis for subsequent anomaly analysis.

[0033] For example, when classifying transformer components, the mean magnetic field strength and gradient changes in the feature dataset can be used as the primary classification criteria. If the mean magnetic field strength of a component is significantly higher than the normal range, and the gradient changes exhibit irregular fluctuations, it may be classified as a minor anomaly. This classification method can help quickly identify potentially problematic components, reduce the workload of subsequent analysis, and improve detection efficiency.

[0034] In steps 1-5, if the magnetic field distribution of a certain device component in the classification results exceeds a preset threshold range, depth analysis is performed on the images in the first image set corresponding to that component to obtain detailed distribution information of the abnormal region. It should be noted that depth analysis mainly uses image enhancement and region segmentation techniques to amplify the detailed features of the abnormal region, extracting its boundary range and intensity distribution. During the analysis, historical data can be combined to make preliminary inferences about the causes of the abnormal region's formation, such as whether it is caused by environmental interference or equipment aging. The final abnormal region distribution information will provide a precise basis for subsequent correction.

[0035] In one embodiment, depth analysis of anomalous regions can be achieved using multi-scale image processing techniques.

[0036] For example, for anomaly regions of magnetic field on the surface of a transformer casing, the approximate range of the anomaly can be determined first through low-resolution analysis, and then the boundary details of the anomaly region can be extracted through high-resolution analysis. This multi-scale analysis method can effectively avoid the loss of details and improve the accuracy of anomaly region location, providing more reliable data support for subsequent processing.

[0037] In steps S1-6, the distribution information of the abnormal areas is combined with environmental impact data for correction processing to obtain the corrected second image set.

[0038] Specifically, environmental impact data includes parameters such as temperature, humidity, and electromagnetic interference intensity, which can directly affect the magnetic field distribution. During the calibration process, an environmental impact model can be constructed to correlate environmental parameters with magnetic field anomalies, calculate calibration coefficients, and adjust the image data for the anomaly areas. The calibrated second image set can more realistically reflect the actual magnetic field distribution of the equipment components.

[0039] For example, when correcting the magnetic field distribution image on the surface of a transformer, if an abnormal area is found to be highly correlated with a high-temperature environment, the image data can be adjusted using a temperature correction model. The model calculates the corresponding correction coefficient based on the degree of influence of temperature rise on magnetic field strength and corrects the magnetic field strength value in the abnormal area. This correction method can effectively eliminate the influence of environmental interference, making the image data closer to the actual state of the equipment.

[0040] In steps S1-7, a magnetic field distribution adjustment model of the device components in a complex environment is generated based on the second image set, and the final magnetic field distribution optimization parameters are determined.

[0041] It should be noted that the adjustment model generates a set of optimized parameters by modeling the magnetic field distribution characteristics in the second image set. These parameters guide the adjustment of equipment operation or the optimization of imaging equipment parameters. During the modeling process, machine learning methods can be combined to perform regression analysis on the image data, predict the magnetic field distribution trend under different environmental conditions, and determine the optimal parameter combination.

[0042] In one possible implementation, different modeling strategies can be set according to the equipment type and environmental conditions when constructing the magnetic field distribution adjustment model.

[0043] For example, for power equipment operating in high-humidity environments for extended periods, the impact of humidity on magnetic field distribution can be analyzed, and an adjustment model with humidity as the primary variable can be constructed. Conversely, for equipment operating in environments with strong electromagnetic interference, electromagnetic interference intensity needs to be incorporated as a key variable. Through targeted modeling, optimized parameters that better meet actual needs can be generated, improving the stability of equipment operation.

[0044] In step 2, a feature extraction network is used to extract features from the initial image dataset to obtain a set of feature vectors, which represent the magnetic field anomaly patterns in the image.

[0045] Specifically, the feature extraction network extracts feature vectors that reflect magnetic field anomaly patterns by performing multi-layer feature abstraction on the initial image dataset. The network structure can include multiple convolutional layers and pooling layers to capture local and global features in the image. The extracted feature vector set will serve as the basis for subsequent anomaly pattern analysis.

[0046] For example, when processing transformer magnetic field imaging data, the feature extraction network can first extract the local variation features of the magnetic field strength in the image through convolutional layers, and then perform dimensionality reduction on the features through pooling layers to generate feature vectors that can summarize the overall distribution trend. This extraction method can effectively capture abnormal magnetic field patterns, such as local areas of excessively strong or weak magnetic fields, providing data support for subsequent classification and analysis.

[0047] In step 2-1, cluster analysis is performed on the feature vector set, and the K-means algorithm is used to group the vectors to obtain preliminary abnormal pattern classification results.

[0048] It's important to note that the K-means algorithm groups similar vectors together by calculating the distance between them, with each group corresponding to a possible magnetic field anomaly pattern. During clustering, the number of groups needs to be pre-set based on the data distribution characteristics to ensure the classification results are discriminative. Preliminary classification results can help quickly identify potential anomaly patterns.

[0049] In one embodiment, clustering analysis of a set of feature vectors can be performed by adjusting clustering parameters according to the operating environment of the device components.

[0050] For example, in environments with strong electromagnetic interference, the anomaly patterns of magnetic fields may be more complex. In such cases, the number of clusters can be increased to capture more subtle differences. For devices in stable environments, the number of clusters can be reduced to simplify the classification process. This adjustment can improve the relevance of the clustering results and provide more accurate classification basis for subsequent analysis.

[0051] In step 2-2, based on the preliminary abnormal pattern classification results, the feature vector distribution range within each classification group is obtained. If the vector distribution within a certain group exceeds a preset threshold, it is marked as a potential abnormal pattern group to identify high-risk areas. Specifically, the feature vector distribution range is determined by calculating the statistical characteristics of the vectors within the group, such as mean and variance. If the variance of a certain group is significantly higher than that of other groups, it indicates that the feature vectors within that group have large differences and may contain abnormal patterns. Therefore, it is marked as a potential abnormal pattern group and requires further analysis.

[0052] For example, when analyzing abnormal patterns in the magnetic field of a transformer, if the variance of the eigenvectors in a certain classification group is much higher than that in other groups, and the magnetic field strength values ​​of some vectors exceed the normal range, it can be marked as a potential abnormal pattern group. This marking method can quickly identify high-risk areas, reduce the blindness of subsequent analysis, and improve detection efficiency.

[0053] In steps 2-3, representative feature vectors are extracted from the marked potential abnormal pattern groups to generate an abnormal pattern subset, resulting in a refined feature representation.

[0054] It should be noted that representative feature vectors can be determined by calculating the center point or typical value of vectors within a group. The extracted vectors can summarize the main features of the group, and the generated subset of abnormal patterns will be used for further classification and verification, reducing data redundancy and improving analysis accuracy.

[0055] In one possible implementation, the extraction of representative feature vectors can be achieved by combining the distribution density of vectors within the group for filtering.

[0056] For example, for a group of anomalous patterns that are relatively concentrated, the center point vector can be selected as the representative; while for a group that is relatively dispersed, multiple typical vectors can be selected to comprehensively reflect the characteristics within the group. Through this extraction method, a more representative subset of anomalous patterns can be generated, providing high-quality data for subsequent processing.

[0057] In steps 2-4, based on the refined feature representation, the support vector machine algorithm is used to further classify the subset of abnormal patterns to determine whether there are significant magnetic field anomalies. Specifically, the support vector machine algorithm divides the subset of abnormal patterns into two categories, significant anomalies and non-significant anomalies, by constructing classification boundaries. During the classification process, historical data and expert experience can be combined to set classification thresholds to ensure the reliability of the classification results.

[0058] For example, when classifying abnormal magnetic field patterns in transformers, if the feature vectors in a subset show a persistently high magnetic field strength, and the distribution area is concentrated in critical parts of the equipment, it can be classified as a significant abnormal pattern. This classification method can help accurately identify abnormal patterns that have a significant impact on equipment operation, providing a basis for subsequent correction and optimization.

[0059] In steps 2-5, spatial mapping is performed on the classified magnetic field anomaly patterns to generate a spatial distribution map of the anomaly patterns, thus obtaining a visualized anomaly distribution result.

[0060] It should be noted that spatial mapping projects abnormal patterns onto specific areas of the device surface by associating feature vectors with image spatial locations, forming an intuitive spatial distribution map. This distribution map clearly shows the location, size, and intensity of abnormal regions, providing visual support for subsequent verification.

[0061] In one embodiment, the mapping method can be adjusted according to the characteristics of the device structure for generating the spatial distribution map.

[0062] For example, for transformer equipment, the spatial distribution map can be combined with the equipment's 3D model to mark the specific location of abnormal areas on the equipment surface, such as the top or side connection points of the iron core. This mapping method can intuitively reflect the relationship between the abnormal distribution and the equipment structure, which is convenient for subsequent analysis and processing.

[0063] In steps 2-6, the abnormal concentrated areas in the spatial distribution map are obtained, and the abnormal patterns are verified in combination with the image data to determine whether the abnormal patterns are consistent with the actual image features, and to determine the final abnormal pattern confirmation result.

[0064] Specifically, the verification process compares the abnormal areas in the spatial distribution map with the magnetic field distribution characteristics in the original image data to determine whether the two match. If there is a discrepancy, further analysis is needed to determine the cause, such as whether it is caused by image processing errors. The final confirmation result will serve as an important basis for equipment defect detection.

[0065] For example, when verifying anomaly patterns in transformer magnetic fields, if the spatial distribution map shows a certain area as a concentrated area of ​​anomalies, but the magnetic field distribution in that area is normal in the original image data, the image processing steps can be traced back to check for feature extraction or classification errors. This verification method can effectively avoid misjudgment and ensure the accuracy of the anomaly pattern confirmation results.

[0066] Thus far, this embodiment of the invention has completed the preliminary judgment of the magnetic field distribution law of power equipment components in complex environments and the identification of abnormal patterns by acquiring, processing, extracting features and analyzing abnormal patterns of magnetic field imaging data. Subsequent steps will further combine environmental interference correction and defect classification to optimize the detection results and improve the reliability and safety of equipment operation.

[0067] Based on the foregoing content, this invention provides a detailed description of the method for detecting defects in power equipment components based on magnetic field imaging data, focusing on subsequent steps such as environmental interference correction, defect classification, and optimization, in order to further improve the implementation of the technical solution.

[0068] In step 3, the feature vector set is compared with the preset deviation standard. If the comparison result shows that the deviation exceeds the standard, it is determined to be due to environmental interference, and the interference correction parameter is obtained. The interference correction parameter is used to adjust the imaging settings.

[0069] Specifically, the feature vector set is extracted from the initial image dataset and reflects the characteristic information of the magnetic field distribution, while the preset deviation standard is a benchmark value set based on historical data and the normal operating status of the equipment. By comparing the difference between the two, it can be determined whether there is environmental interference, such as the influence of strong electromagnetic fields or temperature fluctuations on the imaging results. If the deviation exceeds the standard range, it is necessary to further analyze the source of interference and generate corresponding correction parameters to optimize the settings of the imaging equipment.

[0070] In step 3-1, feature vector set data is acquired. Relevant information is extracted from the imaging device through a pre-established acquisition module. If the integrity of the extracted data meets the preset threshold requirements, it is determined to be valid data, and a preliminary vector set is obtained.

[0071] It should be noted that the acquisition module connects to the imaging device via a data interface to obtain feature vector data in real time and verifies the integrity of the data, such as checking for missing data or outliers. During the verification process, valid data can be filtered out by setting data integrity thresholds, such as requiring a certain percentage of data coverage. The initial vector set will serve as the basis for subsequent comparisons and analyses.

[0072] In one embodiment, the sampling frequency of the acquisition module can be adjusted according to different device operating environments for the acquisition of feature vector data.

[0073] For example, in environments with strong electromagnetic interference, the sampling frequency can be increased to acquire more data points and improve data integrity; while in more stable environments, the frequency can be appropriately reduced to decrease the data processing burden. This adjustment method ensures that the initial vector set collected has high reliability, providing a solid foundation for subsequent analysis.

[0074] In step 3-2, the preliminary vector set is compared with the preset deviation standard. Statistical analysis is used to calculate the difference between the two. If the difference exceeds the standard range, it is determined that environmental interference exists, thus identifying a preliminary indicator of the interference's impact. Specifically, the statistical analysis method calculates the mean and variance of each feature value in the vector set and compares them item by item with the preset standard to obtain the difference value. The calculation of the difference value can be based on multiple dimensions of the feature vectors, such as magnetic field strength and distribution gradient. If the difference value exceeds the preset range, it indicates that environmental interference has a significant impact on the imaging data, and further analysis of the specific type of interference is required.

[0075] For example, when analyzing transformer magnetic field imaging data, if the mean magnetic field strength in the initial vector set is significantly higher than a preset standard, and the distribution gradient exhibits irregular fluctuations, it can be preliminarily determined that environmental interference exists. This comparison method can quickly identify preliminary signs of interference, providing a basis for determining the type of interference in subsequent analyses.

[0076] In step 3-3, based on the preliminary indicators of interference impact, interference patterns in historical data records are obtained. By comparing the similarity between the current indicators and historical patterns, potential interference types are determined, and the classification results of interference types are obtained.

[0077] It should be noted that the historical data records store various environmental interference patterns, such as magnetic field shifts caused by high temperatures and signal attenuation due to humidity. By calculating the similarity between the current interference pattern and historical patterns, for example through distance calculation of feature vectors, the most likely type of interference can be determined. The classification results will guide the selection of subsequent correction parameters.

[0078] In one possible implementation, the determination of the type of interference can be aided by combining environmental monitoring data.

[0079] For example, if the current ambient temperature is significantly higher than the historical average and the interference indicator shows an abnormal increase in magnetic field strength, it can be inferred that the interference type is high temperature. By combining historical data and real-time environmental information, the type of interference can be classified more accurately, providing a reliable basis for generating correction parameters.

[0080] In steps 3-4, based on the classification results of the interference type, the pre-established correction parameter database is called to obtain the correction parameters that match the current interference type and determine the specific values ​​of the correction parameters.

[0081] Specifically, the calibration parameter database stores calibration schemes for different types of interference, such as magnetic field strength correction coefficients for high-temperature interference and signal gain adjustment values ​​for humidity interference. By matching the current interference type, the corresponding calibration parameter values ​​can be quickly obtained, providing guidance for subsequent imaging settings adjustments.

[0082] For example, when dealing with high-temperature interference, if the classification results indicate that the current interference type is high-temperature influence, correction parameters for high temperatures can be retrieved from the database. For instance, the signal gain of the imaging device can be reduced to decrease the amplification effect of high temperatures on the magnetic field data. This parameter selection method can effectively adjust for specific interference types, improving the specificity of the correction.

[0083] In steps 3-5, the imaging settings are adjusted according to the specific values ​​of the calibration parameters. An automated setting adjustment module applies these parameter values ​​to the imaging device, resulting in the adjusted imaging configuration. It should be noted that the setting adjustment module connects to the imaging device's control system and automatically applies the calibration parameters to the device settings, such as adjusting sensor sensitivity or signal processing algorithm parameters. The adjusted imaging configuration will be used for subsequent data acquisition to reduce the impact of environmental interference.

[0084] In one embodiment, a multi-level adjustment mechanism can be set for adjusting imaging settings.

[0085] For example, after applying the calibration parameters, a set of data can be collected through a small-scale test to check whether the adjustment effect meets expectations; if the effect is not good, the parameters can be further fine-tuned until the optimal configuration is achieved. This multi-level adjustment method can ensure the accuracy of the imaging configuration and avoid over-calibration caused by a one-time adjustment.

[0086] In steps 3-6, the adjusted imaging configuration running status is obtained. The device running data is collected through the real-time monitoring module. If the running data is consistent with the expected output, the adjustment process is judged to be complete, and the final imaging effect optimization status is determined.

[0087] Specifically, the real-time monitoring module collects key indicators during the operation of the imaging equipment, such as signal-to-noise ratio and data stability, to determine whether the adjusted configuration has achieved the expected results. If the operating data is consistent with the preset target, it indicates that the correction parameters have been successfully applied and the imaging effect has been optimized.

[0088] For example, when monitoring a transformer magnetic field imaging device, if the adjusted configuration's operating data shows a significant reduction in the signal-to-noise ratio and an improvement in the clarity of the acquired magnetic field distribution image, it can be determined that the adjustment process is complete. This monitoring method can provide timely feedback on the adjustment effect, ensuring the imaging device operates stably in complex environments.

[0089] In steps 3-7, based on the final optimized imaging effect, the complete process data of this interference correction is recorded. This data is then archived into the historical record library via the storage module, providing reference information for later retrieval. It should be noted that the storage module organizes key data from the correction process, such as interference type, correction parameter values, and comparisons of effects before and after adjustment, into structured records and archives them in the historical record library. These records can provide a reference for handling similar interference in the future, improving correction efficiency.

[0090] In one possible implementation, process data storage can utilize categorization tags, such as archiving by interference type or equipment type. For instance, high-temperature interference correction process data can be categorized under the high-temperature interference tag for easy and rapid retrieval later. This storage method allows for the construction of a rich reference database, providing data support for future interference correction.

[0091] In step 4, the magnetic field imaging protocol is adjusted by interference correction parameters to obtain a corrected image dataset, which reduces the impact of interference.

[0092] Specifically, the magnetic field imaging protocol refers to a series of rules and parameter settings followed by imaging equipment during data acquisition and processing. By applying interference correction parameters and adjusting key parameters in the protocol, such as signal gain and filter strength, the impact of environmental interference on imaging results can be effectively reduced, and a corrected image dataset can be generated.

[0093] In step 4-1, interference features are extracted from the original magnetic field imaging data using a preset interference correction model to obtain preliminary interference distribution information.

[0094] It should be noted that the interference correction model extracts features related to environmental interference, such as noise distribution and signal shift, by performing feature decomposition on the original data. During the extraction process, environmental monitoring data can be combined to determine the main sources of interference features, forming preliminary interference distribution information and providing a basis for subsequent adjustments.

[0095] In one embodiment, the focus of the model's analysis can be adjusted according to different device operating environments for the extraction of interference features.

[0096] For example, in high humidity environments, we can focus on extracting humidity-related interference features, such as signal attenuation amplitude; while in strong electromagnetic interference environments, we can focus on the frequency characteristics of noise distribution. Through targeted extraction, we can more accurately describe the interference distribution and improve the correction effect.

[0097] In step 4-2, based on the preliminary interference distribution information, the imaging protocol is adjusted using a parameter optimization method to generate an optimized imaging parameter configuration.

[0098] Specifically, the parameter optimization method analyzes interference distribution information to calculate the most suitable imaging parameter values ​​for the current environment. For example, it adjusts the signal gain to compensate for signal loss caused by interference, or adjusts filtering parameters to remove noise at specific frequencies. The optimized parameter configuration will then be directly applied to the imaging protocol.

[0099] For example, when processing transformer magnetic field imaging data, if preliminary interference distribution information indicates that the main interference originates from high-frequency noise, the filtering parameters can be adjusted using optimization methods to enhance the suppression of high-frequency signals. This adjustment method can effectively reduce the impact of interference on the imaging results and improve the quality of the image data.

[0100] In step 4-3, the magnetic field imaging process is re-executed using the optimized imaging parameter configuration to obtain the corrected preliminary image dataset.

[0101] It should be noted that when the imaging process is re-executed, the device will acquire data according to the optimized parameter configuration, such as adjusting the sensor sensitivity or signal processing method. The acquired preliminary image dataset will reflect the magnetic field distribution after interference correction, providing a basis for subsequent quality assessment.

[0102] In one possible implementation, a multi-round acquisition mechanism can be set up for the re-imaging process. For example, after the initial acquisition, if interference is found in some areas of the data, parameters can be adjusted specifically, and data for those areas can be reacquired. Through multiple rounds of acquisition, the quality of the initial image dataset can be ensured, and residual interference can be reduced.

[0103] In step 4-4, if the degree of interference reduction in the initial image dataset does not reach the preset threshold, the protocol parameters are iteratively adjusted to regenerate the corrected image until the preset conditions are met, thus obtaining the adjusted image dataset.

[0104] Specifically, the degree of interference reduction can be evaluated by calculating the noise level or sharpness index of the image data. If the preset threshold is not reached, it indicates that the current parameter configuration still needs to be optimized. The parameters can be gradually adjusted through iterative methods, such as gradually reducing the signal gain or increasing the filtering intensity, until the image quality meets the requirements.

[0105] For example, when processing transformer magnetic field imaging data, if the noise level of the initial image dataset is still higher than a preset threshold, the filtering parameters can be iteratively adjusted to gradually enhance the noise suppression capability until the noise level is reduced to an acceptable range. This iterative approach ensures the quality of the image dataset and provides reliable data for subsequent analysis.

[0106] In steps 4-5, an image quality assessment tool is used to determine whether the image quality meets the imaging accuracy requirements for the adjusted image dataset.

[0107] It should be noted that image quality assessment tools comprehensively evaluate whether an image meets accuracy requirements by calculating indicators such as image sharpness, contrast, and noise ratio. During the assessment process, different accuracy standards can be set according to the equipment's testing needs; for example, higher sharpness is required for critical areas.

[0108] In one embodiment, for image quality assessment, different assessment weights can be set according to the importance of device components.

[0109] For example, for images of transformer core areas, the weight of the sharpness index can be increased to ensure the imaging quality of critical areas; while for non-critical areas, the standards can be appropriately relaxed. This evaluation method ensures that the image dataset meets the needs of practical applications.

[0110] In steps 4-6, based on the image quality assessment results, if the imaging accuracy does not meet the standard, the corrected image is further optimized using data processing methods to obtain the final corrected image dataset. Specifically, secondary optimization can be achieved through image enhancement techniques, such as adjusting image contrast or enhancing local details to improve image quality. The optimized image dataset will be used as the final result for subsequent defect detection.

[0111] For example, when processing transformer magnetic field imaging images, if the evaluation results show insufficient clarity in key areas, local enhancement techniques can be used to process these areas, improving the visibility of details in the magnetic field distribution. This secondary optimization method can further improve image quality and ensure the accuracy of detection results.

[0112] In steps 4-7, the final corrected image dataset is used to save the data using a storage tool, ensuring data integrity and completing the processing flow.

[0113] It should be noted that the storage tool saves the corrected image data in a structured format and verifies the integrity of the data, such as checking for data loss or corruption. The saved data will serve as the basis for subsequent defect classification and analysis.

[0114] In one possible implementation, a multi-level backup mechanism can be set up for storing image data. For example, data can be stored simultaneously on local devices and remote servers to ensure data security; at the same time, timestamps and device identifiers can be added to the data to facilitate subsequent traceability. This storage method can effectively protect the image dataset of the correction effect and provide a guarantee for subsequent processing.

[0115] In step 5, a classification model is used to classify the corrected image dataset to determine the defect type distribution, which covers common equipment problems.

[0116] Specifically, the classification model performs feature analysis on the corrected image data, associating the magnetic field distribution characteristics in the image with equipment defect types to identify possible defect categories, such as overheating risk caused by excessively strong magnetic fields or structural damage caused by uneven magnetic fields. The classification results will reflect the distribution of defect types, providing guidance for equipment maintenance.

[0117] In step 5-1, a pre-trained classification model is used to process the corrected image dataset, identify the types of defects contained therein, and obtain preliminary defect classification results.

[0118] It should be noted that the classification model learns from the magnetic field distribution characteristics in image data to identify patterns related to defects, such as localized abnormal concentrations or uneven distribution of magnetic fields. The preliminary classification results will list the possible defect types and their locations on the equipment surface.

[0119] In one embodiment, the classification focus of the defect classification model can be adjusted according to the equipment type. For example, for transformer equipment, the model can be trained to identify abnormal magnetic field patterns related to core overheating or conductor breakage; while for other power equipment, the model can be adjusted to identify other common defects. Targeted training can improve the accuracy of the classification results.

[0120] In step 5-2, data analysis is performed on the preliminary defect classification results to summarize the distribution of defect types and determine the frequency and main categories of each type of defect.

[0121] Specifically, data analysis summarizes the main types and concentrated areas of defects by statistically analyzing the frequency and distribution of various types of defects in the classification results. For example, a certain type of defect may be concentrated in a specific part of the equipment, indicating that there is a high risk in that part.

[0122] For example, when analyzing the distribution of transformer defects, if the preliminary classification results show that overheating-related defects occur most frequently in the core area and are concentrated at the top of the core, this area can be inferred to be the main risk point. This analytical approach can help quickly pinpoint the critical problem areas of the equipment, providing direction for subsequent processing.

[0123] In step 5-3, if the confidence level of a certain type of defect in the preliminary classification result is lower than the preset threshold, the image content of that part will be re-corrected to obtain optimized image data.

[0124] It should be noted that the identification confidence score reflects the reliability of the classification model in identifying a certain type of defect. If the confidence score is below the threshold, it indicates that there may be errors in the classification results, and further correction of the relevant image data is required, such as by enhancing image details or adjusting classification parameters to improve data quality.

[0125] In one possible implementation, correction for low-confidence defects can be achieved by combining local image enhancement techniques. For example, if the confidence level for identifying overheating defects in the transformer core area is low, the image of that area can be contrast-enhanced to improve the visibility of details in the magnetic field distribution, and then input into the classification model again for judgment. This correction method can effectively improve the reliability of the classification results.

[0126] In step 5-4, based on the optimized image data, a secondary classification process is performed using the same classification model to determine the accuracy of the defect type and obtain the updated classification result.

[0127] Specifically, secondary classification verifies the accuracy of the initial classification results by re-analyzing the optimized image data. The updated classification results will more closely approximate the actual defects of the equipment, providing a more reliable basis for subsequent analysis.

[0128] For example, when classifying transformer image data in a secondary manner, if the optimized image shows a more pronounced abnormal magnetic field distribution in the core area, it confirms the accuracy of the initial overheating defect assessment and allows for updating the classification results. This secondary classification method effectively reduces misjudgments and improves the accuracy of defect identification.

[0129] In step 5-5, based on the updated classification results and combined with the common categories of equipment problems, the types are divided to determine the correspondence between defect types and equipment problems.

[0130] It should be noted that common categories of equipment problems include overheating, structural damage, and loose connections. By mapping the classification results to these categories, the specific impact of the defect type on equipment operation can be clearly identified. For example, an excessively strong magnetic field may correspond to overheating, while an uneven magnetic field may correspond to structural damage.

[0131] In one embodiment, the correspondence between defect types and equipment problems can be verified by combining historical maintenance records. For example, if historical records show that a certain magnetic field anomaly pattern is often accompanied by overheating problems, this pattern can be associated with the overheating category. This correspondence method allows for a more accurate assessment of the impact of defects on the equipment, providing support for maintenance decisions.

[0132] In steps 5-6, statistical analysis is performed on the correspondence between the classified defect types and equipment problems to obtain the correlation pattern between defect distribution and equipment problems, and to determine the main sources of problems.

[0133] Specifically, statistical analysis identifies the primary sources of problems by calculating the correlation strength between various defects and equipment issues. For example, if a certain type of defect frequently occurs in critical areas of the equipment and is highly correlated with a particular problem, it indicates that the problem is the primary source of risk.

[0134] For example, when analyzing the distribution of transformer defects, if statistical results show that defects with excessively strong magnetic fields in the core area are highly correlated with overheating problems, and these defects are concentrated at the top of the core, it can be determined that the main source of the overheating problem is the top of the core. This analytical approach can help pinpoint the root cause of equipment problems and improve maintenance efficiency.

[0135] In steps 5-7, based on the association pattern, a support vector machine model is used to further verify the relationship between defect distribution and equipment problems, and the final mapping result of defect type distribution and problem coverage is obtained.

[0136] It should be noted that the Support Vector Machine (SVM) model constructs classification boundaries to quantitatively verify the correlation between defect distribution and equipment problems, generating the final mapping result. This mapping result will intuitively reflect the correspondence between various defects and equipment problems, providing comprehensive guidance for equipment maintenance.

[0137] In one possible implementation, the verification of defect distribution and equipment problems can be conducted by combining multi-dimensional features for analysis.

[0138] For example, features such as the location, magnetic field strength, and frequency of occurrence of defects can be input into a support vector machine model to comprehensively assess the correlation strength with equipment problems. This verification method ensures the comprehensiveness and accuracy of the mapping results.

[0139] In step 6, if the defect type distribution matches the standardized design template less than the standard, the parameter matching model is updated through an optimization mechanism to obtain the optimized defect feature vector.

[0140] Specifically, the standardized design template is a baseline model built based on the normal operating conditions of the equipment and design specifications, used to assess whether the defect distribution meets expectations. If the matching degree is lower than the standard, it indicates that there is a deviation in the current defect feature vector, and the model parameters need to be adjusted through optimization mechanisms to improve the matching accuracy.

[0141] In step 6-1, initial matching data is obtained by analyzing the defect types and distribution, and a comparison benchmark with the standardized template is determined.

[0142] It should be noted that the initial matching data includes information such as the distribution area, frequency, and intensity of defect types. By comparing each item with the standardized template, the matching degree value can be calculated. The comparison benchmark will serve as the starting point for subsequent optimizations.

[0143] In one embodiment, different analysis weights can be set according to the key areas of the device for the acquisition of initial matching data.

[0144] For example, the defect distribution in the transformer core area can be weighted more heavily in the matching data, ensuring that the matching degree of key areas receives priority attention. This analytical approach allows for a more accurate determination of the comparison benchmark.

[0145] In step 6-2, if the comparison results between the initial matching data and the standardized template show a low matching degree, a preset optimization mechanism is used to screen the parameter adjustment schemes and obtain the parameter set after preliminary adjustment.

[0146] Specifically, the optimization mechanism analyzes the feature dimensions with low matching degree, filters out parameters that may affect the matching degree, such as feature extraction weights or classification thresholds, and generates an adjustment plan. The initially adjusted parameter set will be used to update the model.

[0147] For example, when dealing with transformer defect distribution, if the low matching degree is due to unreasonable feature weight settings in the core region, the weight of features in that region can be increased through an optimization mechanism to generate a preliminary set of adjustment parameters. This screening method can specifically address the problem of low matching degree and improve model performance.

[0148] In step 6-3, based on the initially adjusted parameter set, the internal weights of the model are updated to meet the requirements for building a precise model, and it is determined whether the design matching requirements are met.

[0149] It should be noted that the internal weights of the model are updated by applying the adjusted parameters to the model training process, such as adjusting the weight distribution of feature vectors or the classification boundary. After the update, the matching degree is recalculated to determine whether the design requirements have been met.

[0150] In one possible implementation, a gradual adjustment strategy can be adopted for updating model weights. For example, the weights of key features can be adjusted within a small range first, and the change in matching degree can be observed; if the effect is not good, the adjustment range can be gradually expanded. This strategy can avoid overfitting due to over-adjustment and ensure a steady improvement in matching degree.

[0151] In step 6-4, if the updated model weights still do not meet the design matching standard, the defect features are reclassified using the support vector machine algorithm through iterative calculation to obtain the optimized feature vectors.

[0152] Specifically, iterative computation optimizes the classification results of defect features by repeatedly adjusting model parameters and classification boundaries, generating feature vectors that better conform to the standardized template. The optimized feature vectors will more accurately reflect the distribution characteristics of defects.

[0153] For example, when optimizing transformer defect features, if the updated model weights still do not improve the matching degree, the feature data of the core region can be reclassified using a support vector machine algorithm, the classification boundary can be adjusted, and optimized feature vectors can be generated. This iterative approach can gradually approach the design standard and improve the quality of the feature vectors.

[0154] In step 6-5, based on the optimized feature vector and the distribution of defect types, the output logic of the model is adjusted to determine the final defect feature representation.

[0155] It should be noted that the output logic is adjusted by redefining the mapping relationship between feature vectors and defect types, for example, adjusting the contribution of a certain feature value to the defect type. The final defect feature representation will better reflect the actual defect distribution.

[0156] In one embodiment, adjustments to the output logic can be made in conjunction with the regional characteristics of the defect distribution. For example, if the optimized feature vector shows that the defect feature values ​​in the core region are high, the output logic can be adjusted to increase the contribution of the features in that region to the overheating defect. This adjustment method can generate a more accurate representation of the defect features.

[0157] In step 6-6, the final defect feature representation is used to verify the matching degree with the standardized template, obtain the evaluation data of the optimization results, and determine whether it meets the expected standards.

[0158] Specifically, the verification process evaluates the optimization effect by calculating the matching degree between the final feature representation and the standardized template. If the matching degree reaches the expected standard, the optimization process is complete; if it still does not meet the standard, further adjustments are needed.

[0159] For example, when verifying the feature representation of transformer defects, if the final matching degree value is significantly improved and the feature distribution of the key area is consistent with the standardized template, it can be determined that the optimization result meets expectations. This verification method can ensure the accuracy of defect feature representation and provide a reliable basis for subsequent detection.

[0160] In steps 6-7, if the evaluation data indicates that the optimization results still need improvement, the model parameters are readjusted based on the correlation between the distribution and the defect characteristics to obtain a more accurate matching output.

[0161] It should be noted that adjustments are made to identify shortcomings in the analyzed data, such as a low feature matching degree in a certain area. The relevant parameters are then adjusted accordingly until the matching output reaches the optimal state.

[0162] In one possible implementation, the readjustment of model parameters can be achieved through multiple rounds of iterative optimization. For example, for defect features with low matching degrees in the iron core region, the feature weights and classification thresholds can be gradually adjusted, and the change in matching degree after each adjustment can be observed until the optimal matching output is reached. In this way, the accuracy of the model parameters can be ensured.

[0163] In step 7, an automated detection process sequence is generated based on the optimized defect feature vector to obtain the final defect identification result.

[0164] Specifically, the automated inspection process sequence uses optimized feature vectors as input to construct a series of inspection steps, such as feature extraction, classification, and result output, to automatically complete the defect identification process. The final identification results will intuitively reflect the defect status of equipment components, providing support for maintenance decisions.

[0165] In step 7-1, defect features are extracted and analyzed to construct an initial feature dataset and obtain preliminary defect description information.

[0166] It should be noted that feature extraction involves further decomposing the optimized defect feature vector to extract key feature values, such as anomalies in magnetic field strength and the distribution range. The constructed initial feature dataset will summarize the main characteristics of the defect, forming preliminary descriptive information.

[0167] In one embodiment, different extraction priorities can be set according to the defect type when constructing the feature dataset.

[0168] For example, for overheating-related defects, the focus can be on extracting magnetic field strength and local concentration features; while for structural damage defects, the focus is on the non-uniformity of magnetic field distribution. This extraction method ensures the specificity of the feature dataset.

[0169] In step 7-2, based on the preliminary defect description information, data is loaded as input, the support vector machine algorithm is used to train the model, the classification boundary between vectors is determined, and the optimized feature representation is obtained.

[0170] Specifically, the Support Vector Machine (SVM) algorithm constructs classification boundaries between feature vectors by classifying preliminary defect description information, thereby optimizing the feature representation. The optimized feature representation will more clearly distinguish different defect types.

[0171] For example, when training a transformer defect identification model, if the initial description information shows that the core area has high feature values, a classification boundary can be constructed using a support vector machine algorithm to classify the features of that area as an overheating defect. This training method can improve the discriminative power of the feature representation, providing support for subsequent detection.

[0172] In step 7-3, key control points are extracted from the optimized feature representation. If the key control points meet a preset threshold range, the sequence is considered valid, and an executable workflow is obtained. It should be noted that key control points refer to feature values ​​in the feature representation that have a decisive impact on defect identification, such as the peak magnetic field strength in a certain area. By checking whether the control points meet the threshold range, it can be determined whether the feature representation is suitable for constructing a detection workflow.

[0173] In one possible implementation, the extraction of key control points can be combined with screening of key areas of the equipment. For example, for transformer equipment, the characteristic values ​​of the core area can be used as the main control points to check whether they meet preset thresholds. If they meet the requirements, an executable process plan is generated to ensure the reliability of the detection process.

[0174] In step 7-4, the automated detection module is run through an executable process scheme to obtain the defect identification status of the intermediate stage.

[0175] Specifically, the automated detection module follows the steps of the workflow, sequentially performing operations such as feature extraction, classification, and result output, generating intermediate identification statuses. These intermediate statuses reflect the preliminary results of defect identification, providing a basis for final confirmation.

[0176] For example, during the transformer defect detection process, if an intermediate stage indicates an overheating defect in the core area, this status can be recorded and the process can proceed to subsequent verification steps. This automated detection method can quickly generate preliminary results and improve detection efficiency.

[0177] In step 7-5, based on the defect identification status in the intermediate stage, numerical features are extracted and their correlation with the preset final conclusion value is calculated. If the correlation coefficient is higher than the preset standard, the final identification result is confirmed.

[0178] It should be noted that numerical characteristics include key indicators in the intermediate states, such as the magnetic field strength and distribution range of the defect area. By calculating the correlation with the preset conclusion values, the reliability of the intermediate results can be determined. If the correlation coefficient reaches the standard, it indicates that the final identification result is accurate and reliable.

[0179] In one embodiment, correlation calculation can be performed by combining multi-dimensional features for comprehensive evaluation.

[0180] For example, the correlation between the magnetic field strength characteristics and distribution area characteristics of the intermediate stage and the preset conclusion values ​​is calculated separately, and the reliability of the results is comprehensively judged. This evaluation method ensures the accuracy of the final defect identification results and provides a scientific basis for equipment maintenance.

[0181] In the description of this invention, it should be understood that the terms "coaxial," "bottom," "one end," "top," "middle," "other end," "upper," "side," "top," "inner," "front," "center," "both ends," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0182] Furthermore, the terms “first,” “second,” “third,” and “fourth” are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as “first,” “second,” “third,” or “fourth” may explicitly or implicitly include at least one of those features.

[0183] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "setting," "connection," "fixing," "screw connection," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal connection of two components or the interaction between two components. Unless otherwise explicitly limited, those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0184] 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 equivalents.

Claims

1. A method for detecting defects in power fittings based on visual intelligence, characterized in that, include: Based on the collected magnetic field imaging data of the target power equipment components in the target environment, an initial image dataset containing images of the magnetic field distribution on the surface of the target power equipment is obtained. Feature extraction is performed on the initial image dataset to obtain a set of feature vectors, which represent the magnetic field anomaly patterns in the images; The feature vector set is compared with a preset deviation standard. If the comparison result shows that the deviation exceeds the standard, it is determined to be due to environmental interference. Interference correction parameters are obtained, and the interference correction parameters are used to adjust the imaging settings. The magnetic field imaging protocol is adjusted using the interference correction parameters to obtain the corrected image dataset; The corrected image dataset is classified to obtain a distribution of defect types covering common equipment problems; If the degree of matching between the defect type distribution and the standardized design template is lower than the standard, then update the parameter precise matching model to obtain the optimized defect feature vector; An automated detection process sequence is generated based on the optimized defect feature vector to obtain the final defect identification result.

2. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The initial image dataset obtained based on the collected magnetic field imaging data of the target power equipment components in the target environment, which includes images of the magnetic field distribution on the surface of the target power equipment, includes: The magnetic field data of the target power equipment in the target environment are collected by a magnetic field imaging device to construct an initial image dataset containing the magnetic field distribution on the surface of the equipment. The initial image dataset is denoised to obtain the first set of processed images. Based on the magnetic field distribution images in the first image set, magnetic field data features of the device surface are extracted to obtain a feature dataset; The feature dataset is classified to determine the magnetic field distribution pattern of different device components in complex environments, and the classification result is obtained. If the magnetic field distribution of one or more device components in the classification results exceeds the preset threshold range, then a depth analysis is performed on the images in the first image set corresponding to that component to obtain detailed distribution information of the abnormal area; Based on the distribution information of the abnormal areas, and combined with environmental impact data, a correction process is performed to obtain a corrected second image set. Based on the second image set, a magnetic field distribution adjustment model of the target device in a complex environment is generated to obtain the final optimized magnetic field distribution parameters.

3. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of extracting features from the initial image dataset to obtain a set of feature vectors, wherein the set of feature vectors represents the magnetic field anomaly pattern in the image, includes: A pre-built feature extraction network is used to process the initial image dataset to generate a set of feature vectors, which are used to reflect the magnetic field anomaly pattern. By performing cluster analysis on the feature vector set and using the K-means algorithm to group the vectors, preliminary abnormal pattern classification results are obtained. Based on the preliminary abnormal pattern classification results, the feature vector distribution range within each classification group is obtained. If the vector distribution within a certain group exceeds the preset threshold, it is marked as a potential abnormal pattern group, and a high-risk area is obtained. Based on the marked potential abnormal pattern group, representative feature vectors are extracted from it to generate an abnormal pattern subset and obtain a refined feature representation. Based on the refined feature representation, the support vector machine algorithm is used to further classify the subset of abnormal patterns to determine whether there are significant magnetic field anomaly patterns. By spatially mapping the classified magnetic field anomaly patterns, a spatial distribution map of the anomaly patterns is generated, and a visualized anomaly distribution result is obtained. The abnormal concentrated areas in the spatial distribution map are obtained, and the abnormal patterns are verified in combination with image data to determine whether the abnormal patterns are consistent with the actual image features, so as to obtain the final abnormal pattern confirmation result.

4. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of comparing the feature vector set with a preset deviation standard, and determining that the deviation exceeds the standard if the comparison result indicates environmental interference, and obtaining interference correction parameters, wherein the interference correction parameters are used to adjust the imaging settings, includes: The feature vector set data is acquired, and relevant information is extracted from the imaging device through a pre-established acquisition module. If the integrity of the extracted data meets the preset threshold requirements, it is determined to be valid data, and a preliminary vector set is obtained. The preliminary vector set is compared with the preset deviation standard, and the difference between the two is calculated using statistical analysis methods. If the difference exceeds the standard range, it is determined that there is environmental interference, and a preliminary indicator of the interference effect is obtained. Based on the preliminary indicators of the interference impact, interference patterns in historical data records are obtained. By comparing the similarity between the current indicators and historical patterns, potential interference types are determined, and the classification results of interference types are obtained. Based on the classification results of the interference type, the pre-established correction parameter database is called to obtain the correction parameters that match the current interference type, and the specific values ​​of the correction parameters are obtained. Based on the specific values ​​of the calibration parameters, the relevant configuration of the imaging settings is adjusted. Through the automated setting adjustment module, the parameter values ​​are applied to the imaging device to obtain the adjusted imaging configuration. The operating status of the adjusted imaging configuration is obtained, and the operating data of the target device is collected. If the operating data is consistent with the expected output, the adjustment process is determined to be complete, and the final imaging effect optimization status is obtained. Based on the final optimized imaging effect, the complete process data of this interference correction is recorded, and the data is archived into the historical record library through the storage module to obtain reference information that can be called later.

5. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of adjusting the magnetic field imaging protocol using the interference correction parameters to obtain the corrected image dataset includes: Interference features are extracted from the raw magnetic field imaging data to obtain preliminary interference distribution information; Based on the preliminary interference distribution information, the imaging protocol is adjusted to generate an optimized imaging parameter configuration. By re-executing the magnetic field imaging process using the optimized imaging parameter configuration, a corrected preliminary image dataset is obtained. If the reduction of interference in the initial image dataset does not reach the preset threshold, the protocol parameters are iteratively adjusted to regenerate the corrected image until the preset conditions are met, and the adjusted image dataset is obtained. Based on the adjusted image dataset, an image quality assessment tool is used to determine whether the image quality meets the imaging accuracy requirements. The image quality assessment results are used to determine whether the imaging accuracy is up to standard. If the image quality is up to standard, the image is then optimized in a second step using data processing methods to obtain the final image dataset of the corrected image. The final corrected image dataset is used to save the data using a storage tool, ensuring data integrity and completing the processing flow.

6. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of classifying the corrected image dataset to obtain a distribution of defect types covering common device problems includes: The corrected image dataset is processed to identify the types of defects it contains and obtain preliminary defect classification results; By analyzing the preliminary defect classification results, the distribution of defect types is summarized, and the frequency and main categories of each type of defect are obtained. If the confidence level of a certain type of defect in the preliminary classification results is lower than the preset threshold, then the image content of that part will be re-corrected to obtain optimized image data. Based on the optimized image data, a secondary classification process is performed using the same classification model to determine the accuracy of the defect type and obtain the updated classification result. Based on the updated classification results, and combined with the common categories of equipment problems, the types are divided to obtain the correspondence between defect types and equipment problems; Based on the statistical analysis of the correspondence between the classified defect types and equipment problems, the correlation pattern between defect distribution and equipment problems is obtained, and the main sources of problems are determined. Based on the aforementioned association pattern, a support vector machine model is used to further verify the relationship between defect distribution and equipment problems, thereby obtaining the final mapping result between defect type distribution and problem coverage.

7. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of updating the parameter precise matching model and obtaining the optimized defect feature vector if the matching degree between the defect type distribution and the standardized design template is lower than the standard includes: By analyzing the defect types and distribution, initial matching data is obtained, and a comparison benchmark with the standardized template is determined. If the comparison between the initial matching data and the standardized template shows a low matching degree, a preset optimization mechanism is used to screen the parameter adjustment schemes and obtain a set of parameters after preliminary adjustment. Based on the initially adjusted parameter set, and in response to the requirements for building a precise model, the internal weights of the model are updated to determine whether the design matching requirements are met. If the updated model weights still do not meet the design matching standard, the defect features are reclassified using the support vector machine algorithm through iterative calculation to obtain the optimized feature vectors. Based on the optimized feature vector and the distribution of defect types, the output logic of the model is adjusted to obtain the final defect feature representation. By using the final defect feature representation, the matching degree with the standardized template is verified, evaluation data of the optimization results is obtained, and it is determined whether they meet the expected standards. If the evaluation data indicates that the optimization results still need improvement, the model parameters are readjusted based on the correlation between the distribution and the defect characteristics to obtain a more accurate matching output.

8. The method for detecting defects in power fittings based on visual intelligence according to claim 1, characterized in that: The step of generating an automated detection process sequence based on the optimized defect feature vector to obtain the final defect identification result includes: By using OpenCV to extract and analyze defect features, an initial feature dataset is constructed to obtain preliminary defect description information; Based on the preliminary defect description information, load data as input, determine the classification boundary between vectors, and obtain the optimized feature representation; Based on the optimized feature representation, key control points are extracted. If the key control points meet the preset threshold range, the sequence is determined to be valid, and an executable process scheme is obtained. Based on the executable process scheme, the defect identification status of the intermediate stage is obtained; Based on the defect identification status at the intermediate stage, numerical features are extracted and Pearson correlation coefficients are calculated with the preset final conclusion values. The Pearson correlation coefficients are configured as follows: Where Xi is the numerical feature of the defect identification status; X represents the mean of the numerical features of the defect identification status; Yi represents the final numerical value; Y represents the mean of the final numerical values; If the correlation coefficient r is higher than the preset standard, the final identification result is confirmed.

9. A visual intelligence-based power fitting defect detection system, characterized in that, include: The data acquisition module is configured to obtain an initial image dataset containing an image of the magnetic field distribution on the surface of the target power equipment based on the acquired magnetic field imaging data of the target power equipment component in the target environment. The feature extraction module is configured to perform feature extraction on the initial image dataset to obtain a set of feature vectors, wherein the set of feature vectors represents the magnetic field anomaly pattern in the image; The interference judgment module is configured to compare the feature vector set with a preset deviation standard. If the comparison result shows that the deviation exceeds the standard, it is judged to be affected by environmental interference, and interference correction parameters are obtained. The interference correction parameters are used to adjust the imaging settings. The imaging correction module is configured to perform the step of adjusting the magnetic field imaging protocol through the interference correction parameters and obtaining the corrected image dataset; The defect classification module is configured to perform classification processing on the corrected image dataset to obtain a distribution of defect types covering common equipment problems; The model optimization module is configured to update the parameters to accurately match the model and obtain the optimized defect feature vector if the matching degree between the defect type distribution and the standardized design template is lower than the standard. The result generation module is configured to generate an automated detection process sequence based on the optimized defect feature vector to obtain the final defect identification result.

Images