GIS (Geographic Information System) equipment mechanical defect diagnosis method based on Grubrum angle field and dual-channel PCNN-Attention neural network

Abstract

The invention relates to a GIS (Gas Insulated Switchgear) equipment mechanical defect diagnosis method based on a Gramb angle field and a dual-channel PCNN-Attention neural network, and belongs to the technical field of gas insulated switchgear mechanical vibration defect diagnosis. The method solves the problems that traditional diagnosis depends on artificial feature extraction, so that subjectivity is high, information mining is insufficient, and defect severity evaluation is missing. According to the technical scheme, the method comprises the steps that a one-dimensional vibration signal is converted into a GASF two-dimensional image and a GADF two-dimensional image through a GASF field so as to completely reserve time sequence topological features; carrying out data enhancement by adopting an image geometric transformation technology so as to improve the generalization ability of the model; and a dual-channel PCNN-Attention model is constructed, and synchronous intelligent identification of defect types and severity is realized through parallel feature extraction and dynamic weight optimization of an attention mechanism. According to the method, the diagnosis accuracy, reliability and adaptive capacity are improved, and support is provided for equipment state operation and maintenance.

Description

Technical Field

[0001] This invention belongs to the field of mechanical vibration defect diagnosis technology for gas-insulated switchgear, and relates to a method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network. Background Technology

[0002] Gas-insulated switchgear (GIS), as a core component of modern power transmission and transformation networks, is widely used in large-scale power engineering projects due to its advantages such as small footprint, high reliability, and strong environmental compatibility. With the continuous growth in the number of GIS devices put into operation and the increasing demand, it has become a key infrastructure for ensuring the stable operation of the power grid. However, in the long-term operation process, cases of abnormal noise and vibration caused by mechanical defects and sudden failures in GIS equipment have been increasing year by year, posing a serious threat to the safety of the power system.

[0003] Domestic and international researchers have conducted extensive studies on the diagnosis of mechanical defects in GIS equipment. For example, domestic research institutions have proposed methods to characterize the state of mechanical defects using vibration spectrum characteristic parameters based on experimental simulations and vibration data analysis, such as identifying busbar loosening defects by the amplitude ratio of specific frequency components. Other studies have used Fourier transform, coherence function, or energy-average S-transform to extract frequency domain features of vibration signals, and combined them with classification algorithms such as support vector machines to achieve preliminary identification of mechanical defects. These methods can obtain some physical features, but they mainly rely on manual extraction of a limited number of vibration characteristic parameters, exhibiting strong subjective selectivity and insufficient depth in exploring defect features, lacking an effective strategy for assessing the severity of defects.

[0004] Existing diagnostic methods are mostly based on manually designed feature indicators, which make it difficult to fully preserve the global temporal information and local dynamic characteristics of vibration signals, resulting in poor robustness under complex operating conditions. Furthermore, traditional methods do not adequately consider the quantitative assessment of defect severity, limiting their accurate application in condition-based maintenance. Therefore, there is an urgent need for a new method that can adaptively extract features, fully preserve signal information, and simultaneously identify defect types and assess severity.

[0005] To address the aforementioned shortcomings, this invention proposes a diagnostic method based on Gram angle field and dual-channel pulse-coupled neural network-attention mechanism. By using signal-to-image conversion and deep learning technology, it overcomes the limitations of traditional methods, such as feature loss and strong subjectivity, and provides a reliable technical path for the early and accurate diagnosis of mechanical defects in GIS equipment. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network includes the following steps: The one-dimensional time-series signal of mechanical vibration of GIS equipment is converted into two-dimensional images of Gramian Angular Summation Field (GASF) and Gramian Angular Difference Field (GADF) through Gramian Angular Field (GAF). Data augmentation processing is performed on the GASF and GADF images by image geometric transformation fusion; The enhanced image is input into a dual-channel pulse-coupled neural network-attention (PCNN-Attention) model for training and classification, and the model outputs diagnostic results on defect type and severity.

[0008] Furthermore, the conversion of the one-dimensional time-series signal into two-dimensional GASF and GADF images includes: Denoising the vibration signal; The amplitude of the denoised signal is normalized to the [-1, 1] interval; The normalized signal is converted into polar angle and polar radius through polar coordinate mapping; The GASF and GADF images are calculated based on the polar angle.

[0009] Furthermore, the calculation formula for the normalization process is as follows: ,in The denoised signal i The amplitude of each sampling point and These are the minimum and maximum values ​​of the denoised signal, respectively. This is the normalized signal amplitude.

[0010] Furthermore, in the polar coordinate mapping, the polar angle The calculation formula is: Polar diameter The calculation formula is: ,in i For sampling point index,N This represents the total number of sampling points.

[0011] Furthermore, the formula for calculating the GASF image is: The formula for calculating the GADF image is as follows: ,in i and j This is the index for the sampling points.

[0012] Furthermore, the data augmentation processing of the image geometric transformation fusion includes at least one of random rotation, random flipping, random translation, and random shearing.

[0013] Furthermore, the random rotation angle range is [-15°, 15°] with a step size of 1°; the random flip includes horizontal flip and vertical flip, with a flip probability of 0.5; the random translation offset range is [-10, 10] pixels; and the random cropping angle range is [-5°, 5°].

[0014] Furthermore, the dual-channel PCNN-Attention model includes a symmetrical dual-channel structure, which is used to process GASF images and GADF images respectively. Each channel includes a convolutional layer, a pooling layer and an attention layer. The attention layer is used to dynamically allocate feature weights to enhance key defect information.

[0015] Furthermore, the feature weights of the attention layer are calculated as follows: ,in For the first k Attention weights for each feature channel, For the first k Importance score of each feature channel M This represents the total number of feature channels.

[0016] Furthermore, the method also includes: classifying test set images based on a trained dual-channel PCNN-Attention model, outputting a probability distribution of defect categories, with the category corresponding to the maximum probability being the defect type and severity diagnosis result.

[0017] The beneficial effects of this invention are as follows: (1) This invention achieves multiple technological breakthroughs in the field of mechanical defect diagnosis of gas-insulated switchgear by innovatively combining Gram angle field conversion with the attention mechanism of a dual-channel pulse-coupled neural network. This method can completely preserve the temporal topology and dynamic characteristics of vibration signals, transforming one-dimensional signals into two-dimensional image representations containing rich information, fundamentally solving the problem of incomplete information caused by manual feature extraction in traditional methods.

[0018] (2) Through the design of a dual-channel network architecture, this invention achieves simultaneous capture and deep fusion of global amplitude distribution features and local mutation features. This parallel processing mechanism ensures comprehensive coverage of features of different types of mechanical defects, while the introduction of the attention mechanism further strengthens the weight allocation of key defect information and effectively improves the model's ability to distinguish subtle defect features.

[0019] (3) At the data processing level, the image geometric transformation enhancement strategy adopted in this invention significantly improves the generalization performance of the model under limited sample conditions. This method can not only adapt to small differences in sensor installation position and angle, but also effectively cope with various operating condition changes during equipment operation, demonstrating strong robustness.

[0020] (4) This invention enables the simultaneous completion of defect type identification and severity assessment, providing a comprehensive decision-making basis for equipment condition-based operation and maintenance. This method overcomes the limitations of traditional diagnostic methods, significantly improves the accuracy and reliability of diagnostic results, and provides a new technical path for intelligent operation and maintenance of power equipment.

[0021] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the present invention; Figure 2 Flowchart for constructing a dual-channel PCNN-Attention neural network GIS equipment mechanical defect diagnosis model; Figure 3 This is the confusion matrix for the test set. Detailed Implementation

[0023] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0024] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0025] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present 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, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0026] I. Implementation Process of the Invention like Figure 1 As shown, this invention provides a method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network, which mainly includes the following three steps: Step 1: Image Conversion of Vibration Signals from Mechanical Defects in GIS Based on Gram Angle Field. Using Gram angle field as the core technology, the one-dimensional vibration time-series signals of mechanical defects in GIS equipment are converted into two types of two-dimensional images: Gram Angle Sum Field (GASF) and Gram Angle Difference Field (GADF). This constructs an image dataset covering normal conditions, conductor contact loosening (mild / moderate / severe), and disconnector base loosening (mild / moderate / severe) under different load currents (600A, 1200A, 1800A, 2400A).

[0027] Step 2: Data augmentation processing involving image geometric transformation fusion. Image geometric transformation techniques combining random rotation, flipping, translation, and shearing are used to augment the GASF and GADF images. Based on the importance of feature representation according to different defect types and severity, enhancement strategies are employed to strengthen key defect features and reconstruct the enhanced image dataset.

[0028] Step 3: Construct a GIS equipment mechanical defect diagnosis model using a dual-channel PCNN-Attention neural network. Enhanced image data is labeled with feature tags based on defect type and severity, and divided into training and test sets. The training set is input into the dual-channel PCNN-Attention neural network for training. During model training, features are extracted in parallel through dual channels, and key information weights are optimized using an attention mechanism to construct the GIS equipment mechanical defect diagnosis model, enabling intelligent identification of defect states and severity assessment.

[0029] 1. Image-based conversion of vibration signals from mechanical defects in GIS based on Gram angle field The Gram angle field is chosen as the core algorithm for signal conversion primarily because of its unique advantage in preserving the complete temporal structure and physical meaning of the signal. This method transforms the time series into an image with clear geometric meaning through polar coordinate mapping. The Gram angle and field focus on the overall energy distribution and steady-state characteristics of the signal, while the Gram angle difference field is more sensitive to local changes and transient abrupt changes in the signal. This dual-modal representation can simultaneously capture the vibration energy changes caused by mechanical defects and the waveform features generated by dynamic processes such as impact and friction, thus providing complementary and rich visual evidence for distinguishing different types of defects and their severity. Furthermore, the images generated by the Gram angle field have a fixed size and regular matrix form, which can naturally adapt to deep learning models such as convolutional neural networks, realizing a paradigm shift from complex signal processing to end-to-end intelligent image recognition.

[0030] One-dimensional time-series signals of mechanical defect vibration in GIS equipment (covering normal state, conductor contact looseness - mild / moderate / severe, and disconnector base looseness - mild / moderate / severe) were extracted at a sampling rate of 125kHz and a sampling duration of 0.01s and converted into two-dimensional Gram angle field images. The specific conversion method is as follows: set up The time-domain sequence of the vibration signal is given, n = 1, 2, ..., N; where N is the number of sample points. The time-domain sequence is then subjected to Gaussian threshold filtering to remove background noise, yielding the denoised time-domain signal. .

[0031] (1) is defined as the mapping operation between the amplitude of the denoised signal and the extreme value range of the signal, which normalizes the signal amplitude to the range of [-1,1] to eliminate the influence of dimensions. Its calculation formula is Equation (1): (1) in, The denoised signal The amplitude of each sampling point , These are the minimum and maximum values ​​of the denoised signal, respectively. This is the normalized signal amplitude.

[0032] (2) Polar coordinate transformation: defined as the mapping of a normalized signal to a polar coordinate system, converting the signal amplitude to polar angles and the time index to polar radii. The calculation formulas are (2) and (3) respectively: (2) (3) in, For the first The polar angle corresponding to each sampling point (mapped to) (interval) For the first The polar radius corresponding to each sampling point (mapped to) (interval) This is the index of the signal sampling points.

[0033] (3) Gram angle sum field (GASF): defined as the cosine operation of the polar angle sum, used to characterize the global amplitude distribution and steady-state characteristics of the signal, and its calculation formula is Equation (4): (4) in, For GASF image number Line number Column pixel values, For the first The polar angle corresponding to each sampling point.

[0034] (4) Gram angle difference field (GADF): defined as the sine operation of the polar angle difference, used to capture the dynamic changes and abrupt changes of the signal, and its calculation formula is Equation (5): (5) In the formula, For GADF image number Line number The pixel values ​​of the column.

[0035] (5) Feature vector set construction: The generated N×N dimension GASF and GADF images are classified according to defect type and severity to construct an image dataset. The number of images of normal state and various types of defects (including different degrees of severity) is consistent to ensure balanced sample distribution.

[0036] 2. Data Augmentation Processing for Image Geometric Transformation Fusion Image geometric transformation techniques are used to enhance the features of GASF and GADF images with different defect types. Furthermore, key defect information is strengthened through transformation strategies to construct a multimodal enhanced image dataset. The specific processing methods are as follows: set up The image is the original GASF or GADF image, with a resolution of [resolution value missing]. The image is subjected to multi-dimensional geometric transformations in sequence, and the transformed image is defined as follows:

[0037] (1) Random rotation: defined as the angular transformation of the image around the center point, with the rotation angle range set to Step size is The transformation relationship is given by equation (6) to simulate the difference in sensor installation angle: (6) in, The rotated image, For random rotation angles, These are the pixel coordinates of the image.

[0038] (2) Random flip: Defined as a mirror transformation of the image along the horizontal or vertical axis. The probability of both horizontal and vertical flips is set to 0.5 to enhance the robustness of the signal direction. The horizontal flip transformation relationship is given by equation (7): (7) in, This is the image after being horizontally flipped.

[0039] (3) Random translation: defined as the pixel offset of the image in the horizontal and vertical directions, with the translation range set to [-10, 10] pixels to cover small deviations in the sensor installation position. The horizontal translation transformation relationship is given by equation (8): (8) in, The image after horizontal translation. This represents a random pixel offset in the horizontal direction.

[0040] (4) Random cropping: Defined as cropping the image along a horizontal or vertical angle, with the cropping angle range set to 1. The horizontal shear transformation relationship is given by equation (9) to simulate the difference in vibration displacement of equipment: (9) in, The image after horizontal cropping. This is the horizontal shearing angle.

[0041] (5) Data augmentation: Data augmentation is performed on GASF and GADF images for each type of defect (including different degrees of severity), and training and test sets are divided. The training set is used for model parameter optimization, and the test set is used for diagnostic performance verification to ensure that the augmented data still retains the original defect characteristics.

[0042] 3. A dual-channel PCNN-Attention neural network was used to construct a diagnostic model for mechanical defects in GIS equipment. In the feature extraction and classification stages, a dual-channel PCNN-Attention network structure is employed to process GASF and GADF images in parallel. This design fully considers the different physical meanings carried by the two image modalities, extracting features from their respective channels and then dynamically fusing and enhancing the information most relevant to the defects through an attention mechanism. The structural characteristics of the pulse-coupled neural network help capture spatiotemporally correlated regional features in the image, while the attention mechanism further suppresses irrelevant interference and improves the model's ability to distinguish subtle defect features. Combined with image geometric transformation data augmentation strategies, this method maintains strong generalization ability and robustness even with limited samples, ultimately achieving accurate and reliable diagnosis of the type and severity of mechanical defects in GIS equipment.

[0043] Since the channel feature extraction parameters and attention weight allocation of the dual-channel PCNN-Attention neural network have a significant impact on its diagnostic results, a symmetrical dual-channel structure design is used to strengthen multimodal feature representation. The attention mechanism is then combined to optimize key feature weights, thereby improving the model's accuracy and generalization ability in identifying mechanical defects in GIS equipment. This network inputs GASF and GADF images into the symmetrical channels respectively, extracting global and dynamic features in parallel. The attention mechanism dynamically allocates weights to focus on key defect information, ultimately outputting the defect type and severity assessment results. The network output is defined as the defect category probability distribution, i.e., the optimal solution of the objective function. The feature weights in the attention mechanism are defined as follows: (10) In the formula, For the first Attention weights for each feature channel, For the first Importance score of each feature channel This represents the total number of feature channels. The network adjusts the weight allocation based on feature importance scores to strengthen key defect features and suppress interfering information. (11) In the formula, For the weighted number of Each feature channel outputs, For the original number Each feature channel is output. After determining the optimal weight allocation, the network outputs the defect category probability through a fully connected layer and a Softmax function, completing the diagnostic process.

[0044] like Figure 2As shown, this invention constructs a GIS equipment mechanical defect diagnosis model based on a dual-channel PCNN-Attention neural network using a multimodal enhanced image dataset. The specific steps for model construction are as follows: 1) Divide the enhanced images corresponding to the vibration signals into 7 categories of feature labels according to the following conditions: GIS equipment is in normal condition, conductor contact looseness is mild, conductor contact looseness is moderate, conductor contact looseness is severe, disconnector base looseness is mild, disconnector base looseness is moderate, and disconnector base looseness is severe. 2) Extract features from GASF and GADF enhanced images respectively. GASF images focus on global amplitude distribution features, while GADF images focus on dynamic mutation features. A dual-channel parallel structure is used to extract features from the two types of images respectively. 3) Construct a unified dual-channel PCNN-Attention model for defect diagnosis based on feature matrices under different load scenarios. The network input layer uses an image size of 227×227×3. The feature extraction layer consists of 7×7 convolution, 5×5 convolution, 3×3 convolution and max pooling. The attention layer is set with 32 attention heads. The classification layer uses the Softmax function. The Adam optimizer is introduced during training. 4) Using the enhanced image as input to the dual-channel PCNN-Attention model, the test sample set is divided according to the fault category and severity and then input into the diagnostic model. The probability value of each category is output. The category corresponding to the maximum probability is the defect type and severity, thus realizing intelligent diagnosis of mechanical defects of GIS equipment.

[0045] II. Validation 1. Construction of a GIS Equipment Mechanical Defect Signal Acquisition and Detection System To verify the effectiveness of the model, mechanical defect simulation and long-term monitoring tests were conducted based on a full-scale 126kV GIS device. This GIS device includes structures such as busbars, disconnectors, and grounding switches. The test platform consists of a primary main circuit, an induction current-boosting circuit, a grounding protection system, and a display and control system. The GIS main circuit current range is 300A~3000A. The vibration signal detection device comprises sensors, a signal conditioner, a data acquisition card, and a host computer. Specifically, a piezoelectric accelerometer is used to collect vibration signals, with a sensitivity of 1000mV / g, a frequency response range of 10-5000Hz, a measurement range of ±5g, and a signal sampling rate set to 125kHz. This allows for accurate acquisition and storage of defect vibration signals, providing reliable data support for subsequent model verification.

[0046] 2. Division of experimental data The test data covers three types of defects in GIS equipment: loose conductor contacts, loose disconnector base, and normal condition. Measurements were performed at four load currents for each type, and each defect included three severity gradients. The sampling time for a single sample was 0.01 seconds, with a total of 10,640 samples. This included 4,560 samples of loose conductor contacts, 4,560 samples of loose disconnector bases, and 1,520 samples of normal condition. Further breakdown revealed that each severity gradient for each defect had 1,520 samples, and each load current gradient had 380 samples. This balanced data distribution effectively avoided sample bias from interfering with the model validation results.

[0047] 3. Model Training Results and Analysis Figure 3 This is the confusion matrix for the test set. The training set is used for parameter training and optimization of the dual-channel PCNN-Attention model, and the test set is used to verify the model's diagnostic performance. The final test results are presented through the confusion matrix and class accuracy. From the confusion matrix of the test set, we can see that: for severe conductor contact loosening (Category 1), 477 samples were correctly identified with an accuracy of 99.4%; for moderate conductor contact loosening (Category 2), 437 samples were correctly identified with an accuracy of 91.0%; for mild conductor contact loosening (Category 3), 470 samples were correctly identified with an accuracy of 97.9%; for mild disconnector base loosening (Category 4), 439 samples were correctly identified with an accuracy of 91.5%; for severe disconnector base loosening (Category 5), 436 samples were correctly identified with an accuracy of 90.8%; for moderate disconnector base loosening (Category 6), 445 samples were correctly identified with an accuracy of 92.7%; and for normal state (Category 7), all 480 samples were correctly identified with an accuracy of 100%. Based on the confusion matrix calculation, the overall evaluation accuracy of the model reaches 94.8%. It can not only effectively distinguish different types of mechanical defects, but also achieve quantitative assessment of the severity of defects, providing a quantitative basis for the status operation and maintenance of GIS equipment, and fully verifying the effectiveness and practicality of the method of the present invention.

[0048] As core equipment in power transmission and transformation networks, the early and accurate diagnosis of mechanical defects in GIS equipment is crucial for ensuring the safe and stable operation of the power grid. However, existing diagnostic methods largely rely on manual extraction of limited vibration feature parameters, which suffers from problems such as strong subjective selectivity, insufficient information mining, lack of defect severity assessment capabilities, and poor robustness under complex operating conditions. This invention proposes a diagnostic method based on Gram angle field and a dual-channel PCNN-Attention neural network. By converting one-dimensional vibration signals into multimodal two-dimensional images of Gram angle sum field (GASF) and Gram angle difference field (GADF) through Gram angle field, it eliminates the need for manual setting of parameters such as frequency domain resolution and time window. This method can completely preserve the global correlation information and dynamic mutation features of the signal, effectively overcoming the subjectivity and feature loss problems of traditional methods. Simultaneously, the symmetrical dual-channel network can focus on the global amplitude distribution of GASF and the local mutation features of GADF respectively, dynamically enhancing key defect information and suppressing interference through an attention mechanism, combined with image geometric transformation data. The enhanced technology addresses the problem of sample scarcity, ultimately achieving simultaneous defect type identification and severity assessment with an overall diagnostic accuracy of 94.8%. It also maintains high robustness under varying load currents and environmental interference. Compared to traditional diagnostic methods relying on single features or models, this approach is not only more adaptable and eliminates the need for manually designed features, but also provides precise quantitative data for GIS equipment status maintenance. This facilitates early defect detection and tiered treatment, reducing the risk of equipment downtime and large-scale power outages caused by defect deterioration. It offers a new and reliable technical path to improve the diagnostic level of mechanical defects in GIS equipment, and has significant practical implications for ensuring the safe and stable operation of the power grid and reducing maintenance costs.

[0049] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network, characterized in that: Includes the following steps: The one-dimensional time-series signal of mechanical vibration of GIS equipment is converted into two-dimensional images of Gram angle sum field (GASF) and Gram angle difference field (GADF) through Gram angle field (GAF); Data augmentation processing is performed on the GASF and GADF images by image geometric transformation fusion; The enhanced image is input into a dual-channel pulse-coupled neural network-attention mechanism (PCNN-Attention model) for training and classification, and the model outputs diagnostic results on defect type and severity.

2. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 1, characterized in that: The process of converting a one-dimensional time-series signal into two-dimensional GASF and GADF images includes: Denoising the vibration signal; The amplitude of the denoised signal is normalized to the [-1, 1] interval; The normalized signal is converted into polar angle and polar radius through polar coordinate mapping; The GASF and GADF images are calculated based on the polar angle.

3. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 2, characterized in that: The calculation formula for the normalization process is as follows: ,in The denoised signal i The amplitude of each sampling point and These are the minimum and maximum values ​​of the denoised signal, respectively. This is the normalized signal amplitude.

4. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 2, characterized in that: In the polar coordinate mapping, the polar angle The calculation formula is: Polar diameter The calculation formula is: ,in i For sampling point index, N This represents the total number of sampling points.

5. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 2, characterized in that: The formula for calculating the GASF image is: The formula for calculating the GADF image is: ,in i and j This is the index for the sampling points.

6. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 1, characterized in that: The image geometric transformation fusion data augmentation process includes at least one of random rotation, random flipping, random translation, and random shearing.

7. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 6, characterized in that: The random rotation angle range is [-15°, 15°] with a step size of 1°; the random flip includes horizontal flip and vertical flip, with a flip probability of 0.5; the random translation offset range is [-10, 10] pixels; the random cropping angle range is [-5°, 5°].

8. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 1, characterized in that: The dual-channel PCNN-Attention model includes a symmetrical dual-channel structure, which is used to process GASF images and GADF images respectively. Each channel includes a convolutional layer, a pooling layer and an attention layer. The attention layer is used to dynamically allocate feature weights to enhance key defect information.

9. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 8, characterized in that: The feature weights of the attention layer are calculated as follows: ,in For the first k Attention weights for each feature channel, For the first k Importance score of each feature channel M This represents the total number of feature channels.

10. The method for diagnosing mechanical defects in GIS equipment based on Gram angle field and dual-channel PCNN-Attention neural network according to claim 1, characterized in that: The method further includes: classifying test set images based on a trained dual-channel PCNN-Attention model, outputting a probability distribution of defect categories, with the category corresponding to the maximum probability being the defect type and severity diagnosis result.

Images