GIS basin-type insulator ultrasonic flaw detection risk quantification method and device and storage medium

Abstract

The invention discloses a GIS basin-type insulator ultrasonic flaw detection risk quantification method, equipment and a storage medium, belongs to the technical field of power equipment detection, and solves the problem of how to improve the ultrasonic flaw detection precision of insulator defects. According to the invention, multi-array element linear array parameters and a dynamic focusing rule are optimized; performing phase compensation superposition on multi-position echo signals by using a synthetic aperture focusing technology, and suppressing interference by combining phase offset and time-space filtering; constructing a convolutional neural network to realize hierarchical extraction of defect features, capturing defect global features by adopting Transform through sine position coding and a multi-head self-attention mechanism, and completing feature fusion through a feedforward neural network; a defect contour is extracted based on Otsu method dynamic threshold segmentation, and size and depth measurement and risk level evaluation are realized through pixel quantization and echo time calculation; the method is suitable for high-precision detection of hidden defects in the insulator, and full-process intelligentization of insulator defect detection, feature extraction and risk quantification is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment detection, and in particular relates to a method, equipment and storage medium for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators. Background Art

[0002] With the deepening of smart grid construction, GIS (gas-insulated switchgear) has become the core equipment of the power system due to its high reliability. However, its key component, the basin insulator, is prone to hidden defects such as microcracks, air gaps, and metal impurities due to manufacturing process defects and long-term electro-thermal-mechanical stress, which seriously threatens the safe operation of the power grid.

[0003] Traditional ultrasonic testing technology is limited by the number of physical array elements and manual interpretation modes, and has three major bottlenecks: 1) Insufficient resolution: Conventional phased array probes have low sensitivity for detecting microcracks smaller than 0.5mm, and interference from metal inserts can easily cause echo confusion, making it difficult to identify early defects in complex structures (such as the junction between the flange and the insulation body); 2) Feature extraction relies on experience: Defect type discrimination (such as distinguishing cracks from air gaps) is highly dependent on the inspector's subjective analysis of the grayscale and texture of the ultrasonic image, and the misjudgment rate can reach more than 20% in complex scenarios; 3) Fuzzy risk assessment: The lack of a quantitative assessment model can only qualitatively describe the severity of the defect, and cannot combine the defect size, depth and type for multi-dimensional risk grading, making it difficult to support the formulation of differentiated maintenance strategies.

[0004] Synthetic Aperture Focusing (SAFT) can improve resolution by virtually expanding the number of array elements, but single-image imaging still faces the challenge of noise suppression. While convolutional neural networks (CNNs) can automatically extract local defect features, they are inadequate for capturing global correlations, such as long cracks and distributed impurities. Traditional risk assessment methods often rely on threshold-based judgments, failing to integrate defect geometry and type weights, resulting in low confidence in assessment results. The integration of multi-source data and the development of a complete technology chain from high-precision imaging to intelligent identification and quantitative assessment are key challenges in the current field of GIS pot-type insulator inspection. Summary of the Invention

[0005] The technical solution of the present invention is used to solve the problem of how to improve the accuracy of ultrasonic flaw detection of GIS pot-type insulator defects.

[0006] The present invention solves the above technical problems through the following technical solutions: The present invention provides a method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators, comprising: The S1 uses a multi-element linear ultrasonic array, performs focusing law calculations on the array, optimizes array excitation parameters, and plans the multi-element linear ultrasonic array flaw detection scanning path to achieve precise coverage and anti-interference acquisition of the ultrasonic beam. S2 uses synthetic aperture focusing technology to compensate for the phase superposition of echo signals at multiple locations, combining phase cancellation and time-space filtering to suppress interference and improve the detection image resolution; S3 builds a convolutional neural network to achieve hierarchical extraction of defect features and introduces Dropout to enhance generalization capabilities; S4 uses the Transformer model to capture global defect features through sinusoidal position encoding and a multi-head self-attention mechanism, and completes feature fusion through a feedforward neural network. S5 extracts defect contours based on Otsu's dynamic threshold segmentation method, and achieves size and depth measurement and risk level assessment through pixel quantization and echo time calculation.

[0007] Furthermore, the number of elements of the multi-element linear ultrasonic array in step S1 is , array element spacing satisfy ,in, is the ultrasonic wavelength, satisfying conditions, is the ultrasonic propagation speed, and the center frequency is , thus avoiding signal aliasing, the effective detection width of the probe is , thereby covering the GIS pot insulator flange area.

[0008] Furthermore, the method for performing focal law calculation on the multi-element linear ultrasonic array and optimizing array excitation parameters described in step S1 is as follows: Let the focus point coordinates be , No. The coordinates of the array elements are , the linear array along Axis distribution, the array element excitation delay time is:

[0009] Through delayed excitation, the sound waves of each array element are superimposed in phase at the focal point to form a high-intensity sound beam.

[0010] Furthermore, the method described in step S2 of using synthetic aperture focusing technology to compensate and superimpose phases of multi-position echo signals, combining phase cancellation and spatiotemporal filtering to suppress interference, and improving the detection image resolution is specifically as follows: The S21 ultrasonic array probe stops at multiple locations one by one. The number of location points is , each position triggers all Each array element transmits or receives ultrasonic waves to form an echo matrix ,in, is the scan position index, , is the array element index, ; S22 performs phase compensation on the signals of the array elements at each position and then adds them together; S23 calculation element in The sound waves emitted from the point reach the measured point And return the distance value ; According to the propagation speed of ultrasonic waves , and the corresponding ultrasonic signal propagation time is obtained ; The effective length of the synthetic aperture is defined as the length corresponding to the farthest position of the array element that can radiate the sound field to that point, and is related to the half-power angle of the array element radiation:

[0011] in, is the effective length of the synthetic aperture, is the sound wave length, is the distance to the target point, is the array element spacing; Within the effective length, the measured point The relative sound pressure intensity is the superposition amplitude of the sound fields radiated by each array element at this location:

[0012] Combined with the Rayleigh criterion, the lateral resolution of synthetic aperture focusing is:

[0013] in, The effective length of the synthetic aperture is 100 nm, and the resolution is significantly improved; S24 uses phase cancellation algorithm to suppress metal insert echo; S25 uses spatiotemporal filtering to suppress scattered noise.

[0014] Furthermore, the method of suppressing the metal insert echo using the phase cancellation algorithm is as follows: Preset metal insert positions , calculate its echo phase as:

[0015] in, is the distance from the insert surface to the probe; Superimpose inverted signals during signal synthesis , cancel the fixed interference source echo, invert the signal The calculation formula is as follows:

[0016] in, is the metal insert echo signal, is an imaginary factor.

[0017] Furthermore, the method of suppressing scattered noise by using spatiotemporal filtering is as follows: Perform spatiotemporal filtering on each array element signal. The spatiotemporal filtering formula is as follows:

[0018] in, is the filter window coefficient, , is the filter scale factor, parameter and are the horizontal and vertical coordinate values ​​of the image respectively; and are all counting factors, of The value is An integer, take It can effectively suppress random scattering noise while retaining defect echo details.

[0019] Furthermore, the method of constructing a convolutional neural network in step S3 to achieve hierarchical extraction of defect features and introduce Dropout to enhance generalization capability is as follows: S31 multiplies the weight value of the convolution kernel with the corresponding element value of the feature map within the receptive field point by point and adds them together. The resulting matrix is ​​then added with a bias matrix to form the output feature map of the convolution layer. S32 adds an activation function between two connected neurons to enhance the expressive power of the network; S33 samples the image through the pooling layer, reducing the height and width of the input feature map and outputting the feature map as follows:

[0020] in, Is the input feature map, which is a two-dimensional matrix; the pooling window size is , the step size is , is the height of the feature map, is the width of the feature map, is the height of the feature map after the pooling operation, is the width of the feature map after the pooling operation; S34 uses sub-pixel convolution to sample pixels on a channel, using The convolution operation changes the number of channels of the original feature map from Expand to And maintain and unchanged, among which, is the upsampling rate, and then the shape of the feature map is transformed from Transformed into , to achieve the purpose of expanding the resolution of the feature map; S35 converts intermediate features into final output through a fully connected layer, classifies information, and uses the Dropout algorithm to randomly discard information from certain nodes during information transmission, thereby reducing the correlation between nodes and improving the network's generalization ability.

[0021] Furthermore, in step S4, the Transformer model is used to capture the global features of the defect through sinusoidal position encoding and multi-head self-attention mechanism, and feature fusion is completed through a feedforward neural network; S41 adds a position encoding layer before the Transformer, using sinusoidal fixed position encoding to assign a unique value to each position through a combination of sine and cosine functions; S42 uses a self-attention mechanism to calculate the attention coefficient between each embedding vector and other vectors to model global dependencies; S43 In Transformer, the feedforward neural network consists of two fully connected layers and a ReLU activation function to perform mapping and nonlinear changes, thereby extracting richer defect features and completing feature fusion.

[0022] The present invention also provides a device including a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above-mentioned GIS pot-type insulator ultrasonic flaw detection risk quantification method, and the processor is configured to execute the program stored in the memory.

[0023] The present invention also provides a storage medium having a computer program stored thereon. When the computer program is run by a processor, the steps of the above-mentioned method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators are executed.

[0024] The beneficial effects of the present invention are as follows: The present invention optimizes the parameters of multi-element linear arrays and dynamic focusing rules to achieve precise coverage and anti-interference acquisition of ultrasonic beams; utilizes synthetic aperture focusing technology (SAFT) to compensate and superimpose phases of echo signals at multiple locations, combines phase cancellation and spatiotemporal filtering to suppress interference and improve detection resolution; constructs a convolutional neural network (CNN) to achieve hierarchical extraction of defect features through operations such as convolution, activation, and pooling, and introduces Dropout to enhance generalization capability; adopts the Transformer model to capture global defect features through sinusoidal position encoding and multi-head self-attention mechanism, and completes feature fusion through a feedforward neural network; based on the Otsu method dynamic threshold The defect contour is segmented and extracted, and the size, depth measurement and risk level assessment are achieved through pixel quantization and echo time calculation. The method of the present invention combines synthetic aperture focusing technology (SAFT), convolutional neural network (CNN) and Transformer model to achieve hierarchical extraction of defect features and global dependency analysis. At the same time, dynamic threshold segmentation and multi-dimensional quantization algorithm are used to complete the defect size, depth and risk level assessment, realizing the intelligentization of the entire process of GIS pot-type insulator defect detection, feature extraction and risk quantification. It is suitable for high-precision detection of hidden defects such as microcracks, air gaps, metal impurities, etc. in pot-type insulators in gas-insulated switchgear. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of a method for quantifying ultrasonic flaw detection risks of GIS pot-type insulators according to an embodiment of the present invention; Figure 2 This is an ultrasonic echo signal diagram after being processed by the synthetic aperture focusing technology of the ultrasonic flaw detection risk quantification method for GIS pot-type insulators according to an embodiment of the present invention; Figure 3 This is a diagram showing the CNN defect feature extraction results of the ultrasonic flaw detection risk quantification method for GIS pot-type insulators according to an embodiment of the present invention; Figure 4 This is a diagram showing the Transformer global feature enhancement results of the GIS basin insulator ultrasonic flaw detection risk quantification method according to an embodiment of the present invention; Figure 5 is the threshold segmentation result of the Otsu method in the GIS pot-type insulator ultrasonic flaw detection risk quantification method according to an embodiment of the present invention; Figure 6 This is the defect segmentation and risk level assessment result of the GIS pot-type insulator ultrasonic flaw detection risk quantification method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0027] The technical solution of the present invention is further described below with reference to the accompanying drawings and specific embodiments: Example 1 like Figure 1 As shown, a method for quantifying ultrasonic flaw detection risks of GIS pot-type insulators according to an embodiment of the present invention includes the following steps: 1. Use a multi-element linear ultrasonic array, calculate the focusing law of the multi-element linear ultrasonic array, optimize the array excitation parameters, plan the multi-element linear ultrasonic array flaw detection scanning path, and achieve accurate coverage and anti-interference acquisition of the ultrasonic beam.

[0028] 1.1 Multi-element linear ultrasonic array design A multi-element linear ultrasonic array is used, with the number of elements being , array element spacing satisfy ,in, is the ultrasonic wavelength, satisfying conditions, is the ultrasonic propagation speed, and the center frequency is , thus avoiding signal aliasing, the effective detection width of the probe is , thereby covering the GIS pot insulator flange area.

[0029] 1.2. Calculate the focal law of multi-element linear ultrasonic arrays and optimize array excitation parameters Let the focus point coordinates be , No. The coordinates of the array elements are , the linear array along Axis distribution, the array element excitation delay time is: (1) Through delayed excitation, the sound waves of each array element are superimposed in phase at the focal point to form a high-intensity sound beam.

[0030] 1.3. Planning the multi-element linear ultrasonic array flaw detection scanning path 1.3.1. The probe rotates along the circumference of the insulator and steps along the axial direction at the same time. , forming a spiral scanning track to ensure that there is no blind spot in detection.

[0031] 1.3.2. For the junction between the flange and the insulation body, which is the high-incidence area of ​​defects, the step spacing is reduced to , improving local resolution.

[0032] By optimizing the array element layout and excitation parameters, flexible focusing and scanning of the ultrasonic beam can be achieved, thereby improving the detection coverage capability of complex structures of pot-type insulators (such as flanges and inserts).

[0033] Step 2: Use synthetic aperture focusing (SAFT) technology, which collects echo signals from multiple positions by moving the ultrasonic array probe to simulate the effect of a large aperture array and break through the limitation of the number of physical array elements. The synthetic aperture focusing technology compensates and superimposes the phases of the echo signals from multiple positions, combines phase cancellation and spatiotemporal filtering to suppress interference and improve the resolution of the detected image.

[0034] 2.1. The ultrasonic array probe stops at multiple locations one by one. The number of location points is , each position triggers all Each array element transmits or receives ultrasonic waves to form an echo matrix ,in, is the scan position index, , is the array element index, .

[0035] 2.2. Phase compensation and superposition of the array element signals at each position: (2) in, Synthesize the signal for the focus point, is the coordinate of the measuring point, For the Position No. The time delay corresponding to the sound path difference from each array element to the focal point.

[0036] 2.3、Array element The sound waves emitted from the point reach the measured point And return the distance value The calculation formula is as follows: (3) According to the propagation speed of ultrasonic waves , and the corresponding ultrasonic signal propagation time is obtained : (4) The effective length of the synthetic aperture is defined as the length corresponding to the farthest position of the array element that can radiate the sound field to that point, and is related to the half-power angle of the array element radiation: (5) in, is the effective length of the synthetic aperture, is the wavelength of the sound wave, is the distance to the target point, is the array element spacing.

[0037] Within the effective length, the measured point The relative sound pressure intensity is the superposition amplitude of the sound fields radiated by each array element at this location: (6) Combined with the Rayleigh criterion, the lateral resolution of synthetic aperture focusing is: (7) in, The effective length of the synthetic aperture is 200 nm, and the resolution is significantly improved.

[0038] 2.4. Use phase cancellation algorithm to suppress metal insert echo Preset metal insert positions , calculate its echo phase as: (8) in, is the distance from the insert surface to the probe.

[0039] Superimpose inverted signals during signal synthesis , as shown in formula (9), cancels the echo of the fixed interference source.

[0040] (9) in, is the metal insert echo signal, is an imaginary factor.

[0041] 2.5. Suppressing Scattering Noise Using Space-Time Filtering Perform spatiotemporal filtering on each array element signal, as shown in formula (10): (10) in, is the filter window coefficient, , is the filter scale factor, parameter and are the horizontal and vertical coordinate values ​​of the image respectively; and are all counting factors, of The value is An integer, take It can effectively suppress random scattering noise while retaining defect echo details.

[0042] Step 3: Construct a convolutional neural network (CNN) to achieve hierarchical extraction of defect features through operations such as convolution, activation, and pooling, and introduce Dropout to enhance generalization capabilities.

[0043] 3.1. Multiply the weight value of the convolution kernel with the corresponding element value of the feature map within the receptive field range point by point and add them together. The resulting matrix plus a bias matrix can form the output feature map of the convolution layer. The specific process of a convolution calculation is shown in formula (11): (11) in, The output feature map Rank The value of the column, , are the width and height of the convolution kernel respectively, is the weight value of the convolution kernel, is the element value at the corresponding position of the feature map, For the Rank The offset value of the column, and are all counting factors, The value of integer, The value of An integer.

[0044] 3.2. Add activation functions such as Sigmoid (Formula (12)), Tanh (Formula (13)), ReLU (Formula (14)), and Leaky ReLU (Formula (15)) between two connected neurons to enhance the network's expressive power and better fit various curves.

[0045] (12) (13) (14) (15) in, is the gradient of the Leaky ReLU function on negative input.

[0046] 3.3. Sampling the image through the pooling layer simulates the process of dimensionality reduction and abstraction of the objects seen by the human visual system, and highly summarizes the input feature map, reducing the height of the input feature map. and width , output feature map : (16) in, Is the input feature map, which is a two-dimensional matrix; the pooling window size is , the step size is , is the height of the feature map, is the width of the feature map, is the height of the feature map after the pooling operation, is the width of the feature map after pooling operation; .

[0047] 3.4. Use sub-pixel convolution to sample the pixels on a channel, using The convolution operation changes the number of channels of the original feature map from Expand to And maintain and unchanged, among which is the upsampling rate, and then the shape of the feature map is transformed from Transformed into , to achieve the purpose of expanding the resolution of the feature map.

[0048] 3.5. The intermediate features are converted into the final output through the fully connected layer to classify the information. When the number of layers is large and the original data is small, overfitting is prone to occur. The Dropout algorithm is used to randomly discard the information of certain nodes during the information transmission process, thereby reducing the correlation between nodes and improving the generalization ability of the network.

[0049] Step 4: Use the Transformer model to capture the global features of the defect through sinusoidal position encoding and multi-head self-attention mechanism, and complete feature fusion through the feedforward neural network.

[0050] 4.1. Adding a position encoding layer before the Transformer can add position information to the vector. Sine fixed position encoding is used to assign a unique value to each position through a combination of sine and cosine functions. The calculation process is shown in formula (17): (17) in, For the Patch dimensional position encoding, The dimension of the model embedding vector.

[0051] 4.2. The self-attention mechanism is used to calculate the attention coefficient between each model embedding vector and other vectors to model global dependencies. Multi-head self-attention is mainly used. Multi-head self-attention is based on scaled dot product self-attention. The model is divided into multiple heads. Each head learns a different attention score matrix in parallel on a different subspace. The number of heads is , the multi-head self-attention will input vector The three linear changes are Each head ,in , They are The query vector, key vector and value vector of each head, and The shape is , The shape is , They are The dimension of the vector. For each Calculate their scaled dot product attention respectively, and finally The scaled dot product attention outputs of the heads are concatenated along the channels and linearly transformed Get the final multi-head self-attention output. Although there is heads, but due to the corresponding reduction of each head The dimension of , therefore, does not increase the computational complexity of the network. The calculation process of multi-head self-attention is shown in formula (18) and formula (19).

[0052] (18) (19) in, For the The calculation results of individual attention, To scale the dot product attention, and Is the size of The matrix, Is the size of The matrix, is the overall calculation output of multi-head attention, is the splicing along the channel dimension, Is the size of The matrix, is the dimension of the final output vector.

[0053] 4.3. In the Transformer, a feedforward neural network (FFN) is typically used after the self-attention mechanism to perform mapping and nonlinear transformations, thereby extracting richer defect features. The FFN consists of two fully connected layers (FC) and a ReLU activation function. The calculation process is shown in Equation (20).

[0054] (20) in, is the input vector, is a bias vector of the same size as the input vector.

[0055] Step 5: Extract the defect contour based on the dynamic threshold segmentation of the Otsu method, and realize the size, depth measurement and risk level assessment through pixel quantization and echo time calculation.

[0056] 5.1. Use the Otsu method to traverse the entire grayscale interval so that the grayscale value variance between parts C1 and C2 is maximized under the threshold, and the image is divided into the background area and the target area, thereby performing dynamic threshold segmentation. The calculation formula of the Otsu algorithm is: (twenty one) (twenty two) (twenty three) (twenty four) (25) (26) Substituting formula (25) into formula (26) yields: (27) By traversing the entire grayscale interval, we can obtain the inter-class variance Maximum threshold , this threshold is the threshold required for image segmentation; Indicates the proportion of target image pixels in the total number of image pixels. represents the average grayscale of the target image pixels, Indicates the proportion of background image pixels in the total number of pixels in the entire image. represents the average grayscale of the background image pixels, Indicates calculating the total average grayscale of the pixels in the entire image. Represents the between-class variance, and the size of the image is , Indicates that the image grayscale value is less than the threshold The number of pixels, Indicates that the image grayscale value is greater than the threshold The number of pixels.

[0057] 5.2. Quantify defect size, measure depth, and conduct risk assessment Calculate the length and width of the defect size , ,in , is the number of defective pixels, , The defect depth is determined by ultrasonic echo signal. , is the echo time, is the ultrasonic propagation speed. According to different weights , calculate the risk level ,in, is the weight of defect length, is the weight of the defect width, The weight of the defect depth; conduct risk assessment, divide it into high, medium and low risk levels, and number its location to increase visibility.

[0058] Experimental testing The experimental data for this study was collected from GIS equipment at a power supply bureau. These equipment had varying ages, ranging from newly commissioned equipment to equipment over 10 years old. Ultrasonic testing technology was used to inspect the basin insulators in the GIS equipment, generating a large amount of ultrasonic echo signal data.

[0059] In the experiment, representative samples were selected from this actual inspection data. To more comprehensively validate the algorithm's performance, various defect types and sizes were tested. The simulated defects included microcracks, air gaps, and metal inclusions. Defect sizes ranged from tiny defects (such as an air gap with an 8-pixel radius) to larger defects (such as a crack with a major axis of approximately 40 pixels). These simulated defect data were combined with the actual inspection data to form the dataset for this experiment.

[0060] Use professional ultrasonic testing equipment to test the pot insulators in GIS equipment. During the test, multiple test points are arranged on the surface of the pot insulator, and a multi-element linear array scanning method is used to obtain ultrasonic echo signals at different positions, such as Figure 2 As shown, the synthetic aperture focusing technology is used to superimpose the signals after time delay compensation, so that the signals from the same target are enhanced in phase. By processing the signals at different positions, focused imaging of the entire detection area is achieved, thereby improving the resolution.

[0061] Then, CNN defect feature extraction is performed. The Canny edge detection algorithm is used. The closed operation structural element is a disk with a radius of 1. The defect contour is located by edge detection. The closed operation fills small gaps to improve edge continuity. The CNN defect feature extraction result is obtained, as shown in the figure. Figure 3 shown.

[0062] Based on the edge information extracted by CNN, the pixels with edge detection results greater than 0.5 are enhanced by 1.5 times to enhance the contrast of the defect area, and the Transformer global feature enhancement result is obtained, as shown in the following figure: Figure 4 shown.

[0063] The Otsu method is used to determine the optimal segmentation threshold according to the bimodal characteristics of the image grayscale histogram, and the area with an area less than 20 pixels is filtered to effectively separate the defect area from the background noise. Figure 5 Shown is the result of Otsu's threshold segmentation.

[0064] Extract the geometric features of the defect (area, aspect ratio, etc.), normalize the area and shape to obtain the normalization factor, calculate the comprehensive risk score based on the features, and divide the risk level, such as Figure 6 The figure shows the defect segmentation and risk level assessment results. In the figure, #1 to #12 represent the defect numbers, green represents low risk, orange represents medium risk, and red represents high risk, which intuitively reflects the degree of defect damage.

[0065] Example 2 A device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the GIS pot-type insulator ultrasonic flaw detection risk quantification method in Example 1, and the processor is configured to execute the program stored in the memory.

[0066] Example 3 A storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for quantifying ultrasonic flaw detection risks of GIS pot-type insulators in embodiment 1.

[0067] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators, characterized in that: include: The S1 uses a multi-element linear ultrasonic array, performs focusing law calculations on the array, optimizes array excitation parameters, and plans the multi-element linear ultrasonic array flaw detection scanning path to achieve precise coverage and anti-interference acquisition of the ultrasonic beam. S2 uses synthetic aperture focusing technology to compensate for the phase superposition of echo signals at multiple locations, combining phase cancellation and time-space filtering to suppress interference and improve the resolution of the detection image; S3 builds a convolutional neural network to achieve hierarchical extraction of defect features and introduces Dropout to enhance generalization capabilities; S4 uses the Transformer model to capture global defect features through sinusoidal position encoding and a multi-head self-attention mechanism, and completes feature fusion through a feedforward neural network. S5 extracts defect contours based on Otsu's dynamic threshold segmentation method, and achieves size and depth measurement and risk level assessment through pixel quantization and echo time calculation.

2. The GIS pot insulator ultrasonic flaw detection risk quantification method according to claim 1 is characterized in that: The number of elements of the multi-element linear ultrasonic array described in step S1 is , array element spacing satisfy ,in, is the ultrasonic wavelength, satisfying conditions, is the ultrasonic propagation speed, and the center frequency is , thus avoiding signal aliasing, the effective detection width of the probe is , thereby covering the GIS pot insulator flange area.

3. The method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators according to claim 2 is characterized in that: The method for performing focal law calculation on the multi-element linear ultrasonic array and optimizing array excitation parameters described in step S1 is as follows: Let the focus point coordinates be , No. The coordinates of the array elements are , the linear array along Axis distribution, the array element excitation delay time is: Through delayed excitation, the sound waves of each array element are superimposed in phase at the focal point to form a high-intensity sound beam.

4. The GIS pot insulator ultrasonic flaw detection risk quantification method according to claim 1 is characterized in that: The method described in step S2 for improving the detection image resolution by using synthetic aperture focusing technology to compensate and superimpose phases of echo signals at multiple locations, combining phase cancellation and spatiotemporal filtering to suppress interference is as follows: The S21 ultrasonic array probe stops at multiple locations one by one. The number of location points is , each position triggers all Each array element transmits or receives ultrasonic waves to form an echo matrix ,in, is the scan position index, , is the array element index, ; S22 performs phase compensation on the array element signals at each position and then adds them together; S23 calculation element in The sound waves emitted from the point reach the measured point And return the distance value ; According to the propagation speed of ultrasonic wave , and the corresponding ultrasonic signal propagation time is obtained ; The effective length of the synthetic aperture is defined as the length corresponding to the farthest position of the array element that can radiate the sound field to that point, and is related to the half-power angle of the array element radiation: in, is the effective length of the synthetic aperture, is the sound wave length, is the distance to the target point, is the array element spacing; Within the effective length, the measured point The relative sound pressure intensity is the superposition amplitude of the sound fields radiated by each array element at this location: Combined with the Rayleigh criterion, the lateral resolution of synthetic aperture focusing is: in, The effective length of the synthetic aperture is 100 nm, and the resolution is significantly improved; The S24 uses a phase cancellation algorithm to suppress metal insert echoes; S25 uses spatiotemporal filtering to suppress scattered noise.

5. The method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators according to claim 4 is characterized in that: The method for suppressing the metal insert echo using the phase cancellation algorithm is as follows: Preset metal insert positions , calculate its echo phase as: in, is the distance from the insert surface to the probe; Superimpose inverted signals during signal synthesis , cancel the fixed interference source echo, invert the signal The calculation formula is as follows: in, is the metal insert echo signal, is an imaginary factor.

6. The method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators according to claim 4 is characterized in that: The method for suppressing scattered noise by using space-time filtering is as follows: Perform spatiotemporal filtering on each array element signal. The spatiotemporal filtering formula is as follows: in, is the filter window coefficient, , is the filter scale factor, parameter and are the horizontal and vertical coordinate values ​​of the image respectively; and are all counting factors, of The value is An integer, take It can effectively suppress random scattering noise while retaining defect echo details.

7. The GIS pot insulator ultrasonic flaw detection risk quantification method according to claim 1 is characterized in that: The method described in step S3 for constructing a convolutional neural network to achieve hierarchical extraction of defect features and introducing Dropout to enhance generalization capability is as follows: S31 multiplies the weight value of the convolution kernel with the corresponding element value of the feature map within the receptive field point by point and adds them together. The resulting matrix is ​​then added with a bias matrix to form the output feature map of the convolution layer. S32 adds an activation function between two connected neurons to enhance the expressive power of the network; S33 samples the image through the pooling layer, reducing the height and width of the input feature map and outputting the feature map as follows: in, Is the input feature map, which is a two-dimensional matrix; the pooling window size is , the step size is , is the height of the feature map, is the width of the feature map, is the height of the feature map after the pooling operation, is the width of the feature map after pooling operation; S34 uses sub-pixel convolution to sample pixels on a channel, using The convolution operation changes the number of channels of the original feature map from Expand to And maintain and unchanged, among which, is the upsampling rate, and then the shape of the feature map is transformed from Transformed into , to achieve the purpose of expanding the resolution of the feature map; S35 converts intermediate features into final output through a fully connected layer, classifies information, and uses the Dropout algorithm to randomly discard information from certain nodes during information transmission, thereby reducing the correlation between nodes and improving the network's generalization ability.

8. The GIS pot insulator ultrasonic flaw detection risk quantification method according to claim 1 is characterized in that: In step S4, the Transformer model is used to capture the global features of the defect through sinusoidal position encoding and multi-head self-attention mechanism, and feature fusion is completed through the feedforward neural network; S41 adds a position encoding layer before the Transformer, using sinusoidal fixed position encoding to assign a unique value to each position through a combination of sine and cosine functions; S42 uses a self-attention mechanism to calculate the attention coefficient between each embedding vector and other vectors to model global dependencies; S43 In Transformer, the feedforward neural network consists of two fully connected layers and a ReLU activation function to perform mapping and nonlinear changes, thereby extracting richer defect features and completing feature fusion.

9. A device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the GIS pot-type insulator ultrasonic flaw detection risk quantification method as described in any one of claims 1 to 8, and the processor is configured to execute the program stored in the memory.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for quantifying the risk of ultrasonic flaw detection of GIS pot-type insulators as described in any one of claims 1 to 8 are executed.

Images