Microwave detection system and visualized quantitative evaluation method for internal defects of composite insulator

Abstract

The application discloses a kind of composite insulator internal defect microwave detection system and visual quantitative evaluation method, belong to microwave nondestructive testing technical field, detection system includes vector network analyzer, computer, test piece, coaxial cable, rectangular waveguide and fixture, three-axis scanning table, scanning table controller;Evaluation method includes: using detection system to carry out cylindrical surface scanning to test piece to obtain the reflection signal of each scanning point on test piece;Reflection signal is adaptively direct wave suppression processing to obtain defect image after direct wave suppression;Defect image is processed to obtain defect image with suppressed background noise and enhanced defect region contrast;Based on high-contrast defect image, its gray histogram is used to determine threshold, image segmentation is carried out, and binary defect image is obtained;Based on binary defect image, defect position and area evaluation are carried out.The application can effectively carry out high-precision visual detection and accurate quantitative characterization to the internal defect of composite insulator, and has important application value for ensuring the smooth and safe operation of power system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of microwave non-destructive testing, and particularly relates to a microwave detection system for internal defects of composite insulators and a visual quantitative evaluation method. BACKGROUND

[0002] The structure of a typical composite insulator comprises a silicon rubber sleeve, a glass steel core rod and metal connections at both ends. Due to complex process conditions in the manufacturing process and harsh service environments, such as high voltage and harsh weather conditions, typical damages such as air gaps between the silicon rubber sleeve and the glass steel core rod, carbonization defects in the glass steel core rod and cracks caused by stress corrosion cracking are prone to occur in the composite insulator. These defects are hidden in the interior of the composite insulator and are not easy to detect, which poses a serious threat to the integrity of the composite insulator and the safety of the power line. Therefore, it is essential to use effective non-destructive testing techniques to non-destructively detect, image and quantitatively evaluate the internal defects of the composite insulator.

[0003] So far, researchers in the field of microwave non-destructive testing at home and abroad have mostly focused on the detection and imaging of internal defects of flat plate-shaped test pieces such as glass fiber reinforced polymers and carbon fiber reinforced polymers. However, they have not touched on the internal defects of actual engineering structures such as composite insulators which have a complex shape and contain multiple layers of dielectric materials. Therefore, there is still a gap in the research on high-precision defect imaging and quantitative evaluation of internal defects of composite insulators using microwave non-destructive testing. In view of this, it is necessary to develop a microwave detection system for internal defects of composite insulators and a visual quantitative evaluation method. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application aims to provide a microwave detection system for internal defects of composite insulators and a visual quantitative evaluation method, which fills the gap in the domestic visual quantitative detection of internal defects of composite insulators using microwave non-destructive testing technology.

[0005] To achieve the above-mentioned application purposes, the technical scheme adopted by the present application is as follows:

[0006] A microwave detection system for internal defects of composite insulators, comprising a vector network analyzer, a computer, a three-axis scanning table controller, a three-axis scanning table, a clamp, a rectangular waveguide, a composite insulator test piece and a coaxial cable.

[0007] The vector network analyzer is used to transmit microwave signals of a specific frequency band and receive reflected signals from the composite insulator test piece, and to calculate the reflection coefficient S 11 and transmit the reflection coefficient to the computer.

[0008] The vector network analyzer is electrically connected with a computer, the computer controls parameters of a microwave signal transmitted by the vector network analyzer through LABVIEW software, including a working frequency band, a frequency interval, a reflection coefficient data format, power, simultaneously receives data transmitted by the vector network analyzer and stores the data;

[0009] The vector network analyzer is electrically connected with a rectangular waveguide through a coaxial cable to transmit the transmitted signal and the reflected signal, the coaxial cable is a transmission line, and the rectangular waveguide is a scanning probe, which transmits a TE 10 wave;

[0010] The three-axis scanning table includes a rotating shaft and two linear shafts, the rotating shaft drives the composite insulator sample to rotate stably through a clamping fixing device, and the two linear shafts are used for axial linear scanning and control of the composite insulator sample, and together with the rotating shaft, the cylindrical surface of the composite insulator sample is scanned;

[0011] The lifting shaft in the linear shaft is mechanically connected with the rectangular waveguide through a clamp, so that the rectangular waveguide is scanned by the three-axis scanning table to scan the composite insulator sample;

[0012] The three-axis scanning table controller is electrically connected with the computer, the rotating shaft and the linear shaft, and the computer controls movement of the three-axis scanning table through the three-axis scanning table controller.

[0013] Further, the microwave detection system obtains information of internal defects of the composite insulator sample by analyzing near-field reflection signals from the composite insulator sample, i.e., a near-field microwave reflection method.

[0014] Further, the rotating shaft, the linear shaft and the clamping fixing device are combined to realize stable scanning of a cylindrical surface of a trunk part of the composite insulator sample.

[0015] Further, by controlling a sampling frequency of the vector network analyzer, a moving speed of the rectangular waveguide and a rotating speed of the rotating shaft, uniform spatial sampling of two-dimensional cylindrical surface data of the composite insulator sample is ensured.

[0016] The visualization and quantitative evaluation method of the microwave detection system of internal defects of the composite insulator includes the following steps:

[0017] S1, composite insulator microwave reflection signal acquisition

[0018] The microwave detection system is used to scan a cylindrical surface of the composite insulator sample to obtain reflection signals of each scanning point on the composite insulator sample;

[0019] The working parameters of the microwave detection system are determined, including the working frequency band, the frequency interval, the lift-off, the scanning range and the scanning step, and then the reflection signal of the composite insulator test piece is collected under the determined working parameters; the working frequency band with optimal microwave penetration ability and resolution is determined according to the relative dielectric constant of each layer of the composite insulator test piece, the maximum step interval of the frequency is determined according to the distance from the rectangular waveguide to the composite insulator test piece, the optimal lift-off is determined according to the experimental verification, the scanning range is determined according to the size of the composite insulator test piece, and the optimal scanning step is determined according to the required defect size accuracy, the shortest wavelength in the working frequency band and the aperture size of the rectangular waveguide; based on the working parameters, the microwave detection system realizes the two-dimensional cylindrical surface sampling of the composite insulator test piece by controlling the vector network analyzer and the three-axis scanning table through the computer, and completes the reflection signal collection of the composite insulator test piece;

[0020] S2, microwave imaging visualization of internal defects of the composite insulator

[0021] The reflection signal is subjected to adaptive direct wave suppression processing to obtain a defect imaging diagram after direct wave suppression;

[0022] The detected reflection wave includes reflection wave carrying defect information and direct wave not carrying defect information; the direct wave is the echo from the interface between the rectangular waveguide and the coaxial cable, and the echo from each hetero-interface in the composite insulator test piece except the defect interface, which does not carry any defect information and is a common component of the reflection wave of each defect point and the reflection wave of non-defect point, and at the same time accounts for a large proportion in the reflection wave, seriously affecting the distinction between the defect signal and the non-defect signal and the contrast between the defect region and the non-defect region in the subsequent defect imaging; therefore, in view of the characteristics of the direct wave, an adaptive direct wave suppression algorithm based on matrix singular value decomposition is proposed to process the collected reflection signal; signal features are extracted from the reconstructed reflection signal for defect imaging;

[0023] The implementation process of the adaptive direct wave suppression algorithm based on matrix singular value decomposition is as follows: the reflection signal is reconstructed into a matrix A, the matrix A is subjected to singular value decomposition to obtain a diagonal matrix S, at this time, the first loop is entered, the first k singular values in the diagonal matrix S are set to 0, for the first iteration, k = 1, then the diagonal matrix S' after the first k singular values are set to 0, i.e. after the direct wave suppression, is brought into the singular value decomposition formula again, the matrix A is inversely reconstructed to obtain the reflection coefficient matrix A' after the elimination of the direct wave; the time domain signal area SA is extracted from the reflection coefficient matrix A' after the elimination of the direct wave as a signal feature for defect imaging; based on the obtained defect image, the Tenengrad function is used to calculate the sharpness; the condition for terminating the iteration is that the difference between the sharpness of the defect image obtained in the current iteration and that in the last iteration is less than 10 -3If the termination condition cannot be met, update k, k' = k + 1, enter the next loop, if the termination condition is met, stop iteration and jump out of the loop; through iteration, obtain the function of the definition about the iteration step length, select the first peak point of the defect image corresponding to the first peak point of the defect image as the optimal defect image, and output; in this way, the direct wave component in the diagonal matrix S is determined, and the best defect imaging result is obtained;

[0024] The specific implementation process is as follows:

[0025] S2.1 Perform matrix singular value decomposition on the reflection signal data set, eliminate the direct wave, and reconstruct the reflection signal after eliminating the direct wave

[0026] The reflection signal obtained by two-dimensional cylindrical scanning of the composite insulator test piece is reconstructed into a matrix A in the following way:

[0027]

[0028] Wherein, S is the reflection coefficient of the scanning point at the coordinate point (x, y) when the frequency is l 11 , and l' is the last frequency point in the working frequency band; perform singular value decomposition on A, and express it as:

[0029] A = USV T

[0030] Wherein, U is a left singular matrix, V is a right singular matrix, and S is a diagonal matrix storing the eigenvalues of A, i.e., singular values, the form of the diagonal matrix S is:

[0031]

[0032] Wherein, the diagonal elements of S are i , i.e., the singular values of the matrix A, wherein 1 ≦ i ≦ r, r is the rank of A, and i ≥ σ2 ≥ … ≥ σ r , the singular values reflect the characteristics of the reflection signal, and some singular values with the largest values represent the direct waves with the highest proportion in the reflection wave; first, set the largest singular value in the diagonal matrix S to 0 to obtain the diagonal matrix S' after direct wave suppression:

[0033]

[0034] The left singular matrix and the right singular matrix remain unchanged, and S' is brought back to the matrix singular value decomposition formula to inverse reconstruct the reflection signal:

[0035] US'V T = A'

[0036] Wherein, A' is the reflection signal matrix after eliminating direct wave;

[0037] S2.2 Extracting signal features in the reconstructed reflection signal, and defect imaging

[0038] The reflection coefficient matrix A' after eliminating direct wave is split into reflection signal data at each spatial scanning point, and the frequency domain signal after reconstruction of each scanning point is converted into time domain signal through inverse Fourier transform:

[0039] y = ifft(b)

[0040] Wherein b is the reflection signal at each spatial scanning point in A', and y is the converted time domain signal; After obtaining the time domain signal of each scanning point, the area surrounded by the time domain signal and the time axis is extracted as the signal feature:

[0041] SA = ∫ydt

[0042] Wherein, SA is the area surrounded by the time domain signal and the time axis, which is called time domain signal area, and t is the horizontal coordinate time of the time domain signal; Extracting the time domain signal area SA of each scanning point for defect imaging, thus realizing the visualization of defects;

[0043] For two-dimensional cylindrical surface scanning, the scanning plane is a cylindrical surface, which is converted into a plane by using the angle-arc length conversion formula during imaging:

[0044]

[0045] Wherein, arc is the arc length corresponding to one rotation angle of the rotation axis, θ is the rotation angle step of the composite insulator test piece in the rotation process, unit is degree, and cir is the circumference of the cylindrical surface where the defect is located; Thus, the scanned cylindrical surface is converted into the imaging plane, and in the imaging plane, the horizontal axis direction is the axial direction of the test piece in the scanning process, and the vertical axis direction corresponds to the circumferential direction in the scanning process;

[0046] S2.3 Calculating the image definition in S2.2 by using Tenengrad function model

[0047] After obtaining the defect image, Tenengrad function is introduced to evaluate the definition of the obtained defect image; In the establishment process of Tenengrad function, Sobel convolution kernel is used, and Sobel convolution kernel is expressed as:

[0048]

[0049] Wherein G x And G y Represent the horizontal and vertical convolution kernels respectively, and the sliding window for gradient calculation at coordinates (x, y) in the defect image is expressed as:

[0050]

[0051] where f(x, y) is the pixel value of the defect image at coordinate (x, y), the horizontal gradient f x and the vertical gradient f y The gradient g at coordinate (x, y) is obtained by convolution of the kernel and the sliding window:

[0052]

[0053] The gradient g at coordinate (x, y) is obtained by convolution of the kernel and the sliding window:

[0054]

[0055] The 95% screening condition is introduced to screen the gradient of each coordinate point in the defect image: in order to reduce the influence of background noise on the clarity of the defect image, the gradients of all coordinates are arranged from the minimum value to the maximum value, and the gradient values of the 95% percentile of all coordinates are extracted for further calculation, denoted as g'(x', y'), and the final image clarity γ is represented as:

[0056]

[0057] where num is the number of coordinate points with gradient values greater than 95% of the percentile of all coordinate points.

[0058] S2.4. Adjusting the singular value sequence value adaptively to realize accurate removal of direct wave signal based on image clarity

[0059] The matrix singular value decomposition and Tenengrad function model are used to determine the direct wave component, and the defect image after maximum suppression of the direct wave is obtained.

[0060] After obtaining the clarity of the defect image obtained by setting the first, i.e. the largest singular value in the diagonal matrix S to 0, return to the diagonal matrix S, set the first two singular values to 0, and reconstruct the reflection signal again. Repeat the above process: extract the time domain signal area SA in the reconstructed reflection signal for defect imaging, and calculate the image clarity based on the defect image using the Tenengrad function model; the difference between the image clarity calculated this time and the image clarity calculated last time is calculated, if the difference is less than 10 -3 , then exit the loop, if the difference is greater than or equal to 10 -3Then, the diagonal matrix S is returned again, the first three singular values are set to 0, and the defective image is generated and the image sharpness is calculated; by analogy, the iteration termination condition is finally met to jump out of the loop, and a group of sharpness vectors γ(k) corresponding to the number of singular values set to 0 in the diagonal matrix is obtained, in which the defective image corresponding to the first peak point is selected as the defective image after the direct wave is suppressed to the greatest extent and is output;

[0061] S3, using a Gaussian high-pass filter-contrast limited adaptive histogram equalization composite algorithm to enhance the defect imaging result in S2

[0062] First, the defective image is processed by a Gaussian high-pass filter. The Gaussian high-pass filter suppresses the low-frequency part of the frequency domain image to achieve the effect of image sharpening. Its effect in the defect imaging image is to suppress the background noise in the image. The Gaussian high-pass filter is expressed as:

[0063]

[0064] Where H(i,j) is the frequency domain form of the image, h is the coefficient of the Gaussian filter, Gau(i,j) is the image after the Gaussian filter transformation in the frequency domain, row is the number of rows of the image, col is the number of columns of the matrix, and D0 is the transformation parameter.

[0065] Based on the image processed by the Gaussian high-pass filter, the contrast limited adaptive histogram equalization processing is continued. This processing enhances the contrast of the defect area in the image without enhancing the background noise. By sequentially performing Gaussian high-pass filter processing and contrast limited adaptive histogram equalization processing on the defective image, the purpose of further suppressing background noise and enhancing contrast is achieved.

[0066] S4, using a gray scale histogram to determine the threshold of image segmentation, segmenting the image in S3 to obtain a binary image

[0067] After obtaining the defect image with high contrast through signal processing and image processing, a gray level histogram of the defect image is drawn; envelope processing is performed to obtain the envelope line of the gray level histogram of each defect image; in the envelope line of the gray level histogram, two wave crests and a wave trough in the middle will appear, a wave band centered on the first wave crest represents an aggregation area of pixel points of the background region, and a wave band centered on the second wave crest represents an aggregation area of pixel points of the defect region, the gray level value corresponding to the number of pixel points of the wave trough is determined by finding the minimum value between the two wave crests; at the same time, the gray mean value of the defect-free imaging image is calculated by taking the defect-free imaging image as a reference; the gray mean value is taken as a compensation standard, and the gray threshold value is obtained by adding the gray mean value of the defect-free imaging image to the gray level value corresponding to the wave trough of the envelope line of the gray level histogram of each defect image; the gray threshold value is used for image segmentation of each defect image to obtain a binary defect image;

[0068] S5, defect position and area evaluation is performed on the binary defect image

[0069] The binary defect image is in 0-1 distribution, the pixel point with a value of 1 is a defect point, and the pixel point with a value of 0 is a background point; for defect position evaluation, the defect region is a single connected region, and the defect position is obtained by finding the centroid of the region; for defect area evaluation, the number of defect points in the binary defect image is counted, and the size corresponding to the pixel point in the defect image is obtained by using a scanning step, and the planar size of the defect is obtained by multiplying the size by the number of defect points.

[0070] The present application has the following advantages:

[0071] The present application provides a microwave detection system and a visual quantitative evaluation method for internal defects of a composite insulator, which can realize visual detection and quantitative evaluation of internal defects of the composite insulator by using a microwave reflection method, and can provide real-time defect information, thereby ensuring the structural integrity and service safety of the composite insulator.

[0072] 1. The microwave detection system of the present application can perform smooth and uniform two-dimensional cylindrical surface scanning on the main part of the composite insulator, and measure the reflection signal.

[0073] 2. The adaptive direct wave suppression algorithm used in the present application can adaptively adjust the singular value sequence, effectively suppress the direct wave component in the reflection signal, and significantly enhance the contrast between the defect region and the background region in the defect imaging result.

[0074] 3. The Gaussian high-pass filtering-contrast limited adaptive histogram equalization composite algorithm used in the present application can further suppress the background noise of the defect imaging image, enhance the defect region, improve the contrast of the defect imaging image, and improve the imaging quality.

[0075] 4. The image segmentation method based on gray histogram adopted by the application can accurately determine the threshold of the defect area in the defect imaging image, obtain a binary image, and combine the centroid positioning and area calculation to accurately quantitatively evaluate the position and planar area of the internal defect of the composite insulator. BRIEF DESCRIPTION OF DRAWINGS

[0076] Figure 1 A microwave detection system for internal defects of a composite insulator.

[0077] Wherein, 1, a vector network analyzer; 2, a computer; 3, a scanning table controller; 4, a rotating shaft; 5, a straight shaft; 6, a clamp; 7, a rectangular waveguide; 8, a composite insulator sample; 9, a coaxial cable.

[0078] Figure 2 A visualization quantitative evaluation method based on the detection system shown in the figure. Figure 1

[0079] Figure 3 A schematic diagram of the position and size of the air gap, carbonized defect and crack inside the composite insulator sample.

[0080] Figure 4 A schematic diagram of the scanning plane and the imaging plane.

[0081] Figure 5 A flowchart of the adaptive direct wave suppression algorithm.

[0082] Figure 6 A preliminary imaging result of extracting the time domain signal area SA as a signal feature.

[0083] Figure 7 An imaging result after signal processing using the adaptive direct wave suppression algorithm.

[0084] Figure 8 A defect imaging result after image processing using the composite algorithm of Gaussian high-pass filtering-contrast-limited adaptive histogram equalization.

[0085] Figure 9 A schematic diagram of determining the threshold value using the gray histogram.

[0086] Figure 10 A binary defect imaging result obtained by image segmentation using the determined threshold value.

[0087] Figure 11 A defect area evaluation result bar chart. DETAILED DESCRIPTION

[0088] ​The specific embodiments of the present application are described below to facilitate the understanding of the present application for those skilled in the art, but it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, it is obvious that various changes are within the spirit and scope of the present application defined and determined by the appended claims, and all the inventions utilizing the concept of the present application are within the scope of protection.

[0089] As shown in the specific embodiments of the present application, a microwave detection system for internal defects of composite insulators is provided, which comprises a vector network analyzer 1, a computer 2, a scanning table controller 3, a three-axis scanning table, a clamp 6, a rectangular waveguide 7, a composite insulator test piece 8, and a coaxial cable 9. Figure 1

[0090] The vector network analyzer 1 is used to transmit microwave signals of a specific frequency band and receive reflected signals from the composite insulator test piece 8, and to calculate the reflection coefficient S 11 and transmit the reflection coefficient to the computer 2; the vector network analyzer 1 is electrically connected to the computer 2, the computer 2 controls the parameters of the microwave signals transmitted by the vector network analyzer 1 through LABVIEW software, including the working frequency band, the frequency interval, the reflection coefficient data format, and the power, and simultaneously receives and stores the data transmitted by the vector network analyzer 1; the vector network analyzer is electrically connected to the rectangular waveguide through the coaxial cable, so as to transmit the input and reflected signals; the coaxial cable is a transmission line for transmitting TEM waves, and the rectangular waveguide 7 is a scanning probe for transmitting TE 10 waves; the three-axis scanning table comprises a rotating shaft 4 and two linear shafts 5, the rotating shaft 4 is used to drive the composite insulator test piece 8 to rotate, and the rotating shaft 4 and the other end are both provided with clamping and fixing devices, so as to realize the stable rotation of the composite insulator test piece 8, and the linear shafts 5 are used for axial linear scanning and control lifting, and together with the rotating shaft 4, realize the cylindrical surface scanning of the composite insulator test piece 8; the lifting shaft in the linear shaft 5 is mechanically connected to the rectangular waveguide 7 through the clamp 6, so as to realize the scanning of the composite insulator test piece 8 by the three-axis scanning table driving the rectangular waveguide 7; the three-axis scanning table controller 3 is electrically connected to the computer 2, and the computer 2 controls the movement of the three-axis scanning table through the three-axis scanning table controller 3; the scanning table controller 3 is electrically connected to the rotating shaft 4 and the linear shaft 5.

[0091] The microwave detection system uses the near-field microwave reflection method to obtain the information of the internal defects of the composite insulator test piece 8.

[0092] Specifically, the detection system realizes the stable cylindrical surface scanning of the cylindrical main part of the composite insulator test piece through the rotating shaft 4, the linear shaft 5, and the clamping and fixing devices, and ensures the uniform spatial sampling of the two-dimensional cylindrical surface data of the composite insulator test piece by controlling the sampling frequency of the vector network analyzer 1, the moving speed of the rectangular waveguide 7, and the rotating speed of the rotating shaft 4.​

[0093] The application also provides a visual quantitative evaluation method of a microwave detection system based on internal defects of a composite insulator, and the overall process is as shown in Figure 2 The method comprises the following steps:

[0094] S1, collecting microwave reflection signals of the composite insulator

[0095] The microwave detection system is used to perform cylindrical surface scanning on the composite insulator sample 8 to obtain reflection signals of each scanning point on the composite insulator sample 8

[0096] The working parameters of the microwave detection system are determined, including the working frequency band, the frequency interval, the lift-off, the scanning range and the scanning step, and then the reflection signals of the composite insulator sample 8 are collected by using the microwave detection system under the determined working parameters; in this example, according to the penetration ability and resolution of microwaves to the silicone rubber sleeve and the glass steel core rod in the composite insulator sample 8, the working frequency band is selected as 26.5-40 GHz, the frequency interval is selected as 0.01 GHz, the best lift-off is determined as 1 mm according to experimental verification, the scanning range is determined as 40 mm x 360° according to the length of the cylindrical main part of the composite insulator sample 8, and the best scanning step is determined as 0.5 mm x 2° according to the minimum linear size 2 mm of the internal defects of the composite insulator, the shortest wavelength 0.65 mm propagating in the composite insulator and the minimum linear size 3.56 mm of the rectangular waveguide 7; based on the working parameters, the microwave detection system controls the vector network analyzer 1 and the three-axis scanning table through the computer 2 to realize two-dimensional cylindrical surface sampling of the composite insulator sample 8, and completes the reflection collection of the composite insulator sample 8; the defect type, size and position in the composite insulator sample 8 are as shown in Figure 3 .

[0097] S2, microwave imaging visualization of internal defects of the composite insulator

[0098] The reflection signals are subjected to adaptive direct wave suppression processing to obtain a defect imaging graph after direct wave suppression;

[0099] The specific implementation process is as follows:

[0100] S2.1, the reflection signal data set is subjected to matrix singular value decomposition, direct wave elimination is performed, and the reflection signal after elimination of the direct wave is reconstructed

[0101] The reflection signals obtained by two-dimensional cylindrical surface scanning on the composite insulator sample 8 are reconstructed into a matrix A in the following manner:

[0102]

[0103] wherein, S is the reflection coefficient of the scanning point located at coordinate point (x, y) when the frequency is l 11 , l' is the last frequency point in the working frequency band; A is singular value decomposed, which can be expressed as:

[0104] A=USV T

[0105] Wherein, U is the left singular matrix, V is the right singular matrix, and S is the diagonal matrix storing the eigenvalues of A, that is, singular values. The form of the diagonal matrix S is:

[0106]

[0107] Wherein, the diagonal elements σ i (1≦i≦r, r is the rank of A) of S are singular values of matrix A, and σ1≥σ2≥…≥σ i ≥…≥σ r The singular values reflect the characteristics of the reflected signal, wherein some singular values with the largest values represent the direct wave with the highest proportion in the reflected wave. First, the largest singular value in the diagonal matrix S is set to 0 to obtain the diagonal matrix S' after suppressing the direct wave:

[0108]

[0109] The left singular matrix and the right singular matrix remain unchanged, and S' is brought back to the matrix singular value decomposition formula to reconstruct the reflected signal:

[0110] US'V T =A'

[0111] Wherein, A' is the reflected signal matrix after eliminating the direct wave;

[0112] S2.2 extracts signal features from the reconstructed reflected signal to perform defect imaging

[0113] The reflected coefficient matrix A' after eliminating the direct wave is split into reflected signal data at each spatial scanning point, and then the frequency domain signal after reconstruction of each scanning point is converted into a time domain signal through inverse Fourier transform:

[0114] y=ifft(b)

[0115] Wherein b is the reflected signal at each spatial scanning point in A', and y is the converted time domain signal; after obtaining the time domain signal of each scanning point, the area surrounded by the time domain signal and the time axis is extracted as the signal feature:

[0116] SA=∫ydt

[0117] Wherein, SA is the area surrounded by the time domain signal and the horizontal axis, namely the time axis, denoted as the time domain signal area, and t is the horizontal coordinate time of the time domain signal; the time domain signal area SA of each scanning point is extracted for defect imaging, thereby realizing the visualization of the defect;

[0118] For two-dimensional cylindrical surface scanning, the scanning plane is a cylindrical surface, and the cylindrical surface is converted into a plane by using an angle-arc length conversion formula during the imaging process:

[0119]

[0120] Wherein, arc is the arc length corresponding to the rotation angle of the rotation axis 4, θ is the rotation angle step of the composite insulator test piece 8 during the rotation process, and cir is the circumference of the cylindrical surface where the defect is located; thus, the scanned cylindrical surface is converted into the plane used for imaging, and a schematic diagram of the conversion of the scanning plane into the imaging cylindrical surface is shown in Figure 4 In the imaging plane, the horizontal axis direction is the axial direction of the test piece during the scanning process, and the vertical axis direction corresponds to the circumferential direction during the scanning process;

[0121] S2.3 Calculate the image definition in S2.2 by using the Tenengrad function model

[0122] After obtaining the defect image, the Tenengrad function is introduced to evaluate the definition of the obtained defect image; in the establishment process of the Tenengrad function, the Sobel convolution kernel is used, and the Sobel convolution kernel can be expressed as:

[0123]

[0124] Wherein, G x and G y represent the horizontal and vertical convolution kernels, respectively, and the sliding window for gradient calculation at the coordinate (x, y) in the defect image can be expressed as:

[0125]

[0126] Wherein, f(x, y) is the pixel point value of the defect image at the coordinate (x, y), the horizontal gradient f x and the vertical gradient f y at the coordinate (x, y) can be obtained by convolution of the convolution kernel and the sliding window:

[0127]

[0128] Then, the gradient g at the coordinate (x, y) can be expressed as:

[0129]

[0130] Compared with the general Tenengrad model, the application introduces a 95% screening condition to screen the gradient of each coordinate point in the defect image; in order to reduce the influence of background noise on the clarity of the defect image, the gradient of all coordinates is arranged from the minimum value to the maximum value, and the gradient value of the 95th percentile of all coordinates is extracted for further calculation, denoted as g'(x', y'), and the reason for selecting 95% is that the gradient value of the 95th percentile is considered as a defect edge point based on the 95% confidence interval, so as to effectively reduce the influence of background noise on the calculation of image clarity; the final image clarity γ can be represented as:

[0131]

[0132] Wherein, num is the number of coordinate points with gradient values greater than 95th percentile in all coordinate points;

[0133] S2.4. Adjusting the singular value sequence value adaptively to realize accurate removal of direct wave signal based on image clarity

[0134] The matrix singular value decomposition and the Tenengrad function model are used to determine the direct wave component, and the defect image after the maximum suppression of the direct wave is obtained;

[0135] The overall process of the adaptive direct wave suppression algorithm is shown in Figure 5 The reflection signal is reconstructed into a matrix A, and the diagonal matrix S is obtained by singular value decomposition of A, and then the loop is entered, the first k singular values (k = 1, 2, 3, …, 1000) in S are set to 0, and then the diagonal matrix S' after setting the first k singular values to 0 is brought into the singular value decomposition formula again, and the reflection coefficient matrix A' after eliminating the direct wave is obtained by inverse reconstruction of the matrix A; the time domain signal area SA is extracted as a signal feature based on the reflection coefficient matrix A' after eliminating the direct wave, and the defect imaging is performed; based on the obtained defect imaging image, the clarity is calculated by using the Tenengrad function model; the difference between the clarity obtained by the current iteration and the clarity obtained by the last iteration is less than 10 -3 , if the termination condition cannot be met, update k, k' = k + 1, and enter the next loop, if the termination condition is met, stop iteration and jump out of the loop; through iteration, the function γ(k) of clarity about iteration step length is obtained, the defect image corresponding to the first peak point of γ(k) is selected as the optimal defect image, and the output is obtained; in this way, the direct wave component in the matrix S is determined, and the best defect imaging result is obtained;

[0136] The defect imaging is performed on the reflection signal without processing by the adaptive direct wave suppression algorithm, i.e. the original reflection signal, and the result is shown in Figure 6 Figure 6 ​Air gap defects and carbonization defects can be observed, but cracks are not observed, and there is significant background noise. Figure 7 The image in the middle is a defect image obtained after processing with the proposed adaptive direct wave suppression algorithm. As can be seen from the image, compared to… Figure 6 , Figure 7 The contrast between the defective area and the background area is significantly improved, and... Figure 7 The middle was successfully reflected Figure 6 Cracks not observed in the sample.

[0137] S3. The defect imaging results in S2 are enhanced using a composite algorithm of Gaussian high-pass filtering and contrast-limited adaptive histogram equalization.

[0138] In this example, for Figure 7 First, bicubic interpolation is performed to smooth the image, followed by Gaussian high-pass filtering and contrast-limited adaptive histogram equalization to obtain a defect image with further enhanced contrast, such as... Figure 8 As shown.

[0139] S4. Determine the image segmentation threshold using the gray-level histogram, and segment the image from S3 to obtain a binarized image.

[0140] Draw them separately Figure 8 The grayscale histograms of various defect images are shown below. The grayscale histogram of the air gap defect is shown below. Figure 9 As shown; in Figure 9 Two peaks and a trough can be observed in the envelope. The band centered on the first peak represents the clustering area of ​​background pixels, and the band centered on the second peak represents the clustering area of ​​air gap defect pixels. The gray value corresponding to the number of pixels in the trough is determined by finding the minimum value between the two peaks. The gray value of the defect-free image is calculated and used as a compensation standard. Figure 9 The threshold of the air gap defect image can be obtained by adding the gray value corresponding to the valley of the middle envelope to the gray mean of the defect-free image. Following this method, thresholds for carbonization defects and cracks can be obtained separately. Image segmentation is then performed based on these gray value thresholds to obtain binary defect images of the three typical defects, such as... Figure 10 As shown.

[0141] S5. Evaluate the defect location and area of ​​the binarized defect image.

[0142] Search Figure 10The center of mass of each defect region can obtain the defect position, wherein the evaluation positions of the air gap defect, the carbonization defect and the crack are (58.4mm, 20mm), (38.6mm, 11.2mm) and (46.4mm, 19.2mm) respectively, and the actual positions thereof are (59mm, 20mm), (38mm, 12mm) and (46.4mm, 19.2mm) respectively, and the evaluation error is controlled within 0.8mm; the number of pixel points corresponding to each defect is counted Figure 10 The pixel point size after bicubic interpolation is 0.5mm*0.3mm, the plane size of the defect can be obtained by multiplying the size with the counted number of defect points, and the evaluation result of the defect size is shown in Table 2. Figure 11 The average error is 5.1%; the obtained high-precision defect imaging result, the evaluation result of the defect position and the defect area all show the effectiveness of the microwave detection system and the visualized quantitative evaluation method for the internal defects of the composite insulator.

Claims

1. A visualization and quantitative evaluation method for a microwave detection system for internal defects in composite insulators, the system comprising a vector network analyzer (1), a computer (2), a three-axis scanning stage controller (3), a three-axis scanning stage, a fixture (6), a rectangular waveguide (7), a composite insulator specimen (8), and a coaxial cable (9); The vector network analyzer (1) is used to transmit microwave signals in a specific frequency band and receive reflected signals from the composite insulator specimen (8), while simultaneously calculating the reflection coefficient S. 11 And transmit the reflection coefficient to the computer (2); The vector network analyzer (1) is electrically connected to the computer (2). The computer (2) controls the parameters of the microwave signal emitted by the vector network analyzer (1) through LABVIEW software, including the working frequency band, frequency interval, reflection coefficient data format, and power. At the same time, it receives and stores the data transmitted by the vector network analyzer (1). The vector network analyzer (1) is electrically connected to the rectangular waveguide (7) via a coaxial cable (9) to transmit input and reflected signals. The coaxial cable serves as the transmission line, propagating TEM waves, while the rectangular waveguide (7) acts as the scanning probe, propagating TEM waves. 10 Wave; The triaxial scanning stage includes a rotating axis (4) and two linear axes (5); the rotating axis (4) drives the composite insulator specimen (8) to rotate smoothly through the clamping and fixing device; the two linear axes (5) are used to perform axial linear scanning and control lifting of the composite insulator specimen (8) respectively, and together with the rotating axis (4) realize the cylindrical surface scanning of the composite insulator specimen (8); The lift-off axis in the linear axis (5) is mechanically connected to the rectangular waveguide (7) through the clamp (6), thereby enabling the three-axis scanning stage to drive the rectangular waveguide (7) to scan the composite insulator specimen (8); The three-axis scanning stage controller (3) is electrically connected to the computer (2), the rotary axis (4) and the linear axis (5). The computer (2) controls the movement of the three-axis scanning stage through the three-axis scanning stage controller (3). Its features are, The visualization and quantitative evaluation method includes the following steps: S1. Microwave Reflection Signal Acquisition of Composite Insulator A microwave detection system was used to scan the cylindrical surface of the composite insulator specimen (8) to obtain the reflection signals of each scanning point on the composite insulator specimen (8); The working parameters of the microwave detection system are determined, including the working frequency band, frequency interval, lift-off, scanning range, and scanning step size. Then, the reflected signal of the composite insulator specimen (8) is collected using the microwave detection system under the determined working parameters. The working frequency band with optimal microwave penetration capability and resolution is determined based on the relative permittivity of each layer of material in the composite insulator specimen (8). The maximum frequency step interval is determined based on the distance from the rectangular waveguide (7) to the composite insulator specimen (8). The optimal lift-off is determined based on experimental verification. The scanning range is determined based on the size of the composite insulator specimen (8). The optimal scanning step size is determined based on the required defect size accuracy, the shortest wavelength in the working frequency band, and the aperture size of the rectangular waveguide (7). Based on the working parameters, the microwave detection system controls the vector network analyzer (1) and the three-axis scanning stage through the computer (2) to realize two-dimensional cylindrical surface sampling of the composite insulator specimen (8) and complete the acquisition of the reflected signal of the composite insulator specimen (8). S2. Microwave imaging visualization of internal defects in composite insulators Adaptive direct wave suppression processing is applied to the reflected signal to obtain a defect image after direct wave suppression; The detected reflected waves include reflected waves carrying defect information and direct waves not carrying defect information. The direct waves are the echoes from the interface between the rectangular waveguide (7) and the coaxial cable (9) and the echoes from all heterogeneous interfaces in the composite insulator specimen (8) except for the defect interface. These echoes do not carry any defect information and are a common component of the reflected waves from each defect point and the reflected waves from non-defect points. They also account for a large proportion of the reflected waves, which seriously affects the distinction between defect signals and non-defect signals and the contrast between defect areas and non-defect areas in subsequent defect imaging. In view of this, an adaptive direct wave suppression algorithm based on matrix singular value decomposition is proposed to process the collected reflected signals, based on the characteristics of the direct waves. Signal features are extracted from the reconstructed reflected signals to perform defect imaging. The implementation process of the adaptive direct wave suppression algorithm based on matrix singular value decomposition is as follows: The reflected signal is reconstructed into matrix A. Singular value decomposition is performed on matrix A to obtain a diagonal matrix S. Then, in the loop, the first k singular values ​​in the diagonal matrix S are set to 0. For the first iteration, k = 1. Subsequently, the diagonal matrix S' after setting the first k singular values ​​to 0, i.e., the direct wave suppressed matrix, is substituted back into the singular value decomposition formula to perform inverse reconstruction of matrix A, obtaining the reflection coefficient matrix A' after eliminating the direct wave. The time-domain signal area SA is extracted as a signal feature from the reflection coefficient matrix A' after eliminating the direct wave, and defect imaging is performed. Based on the obtained defect image, the sharpness is calculated using the Tenengrad function. The iteration terminates when the difference in sharpness between the defect image obtained in the current iteration and the previous iteration is less than 10. -3 If the termination condition is not met, then update k, k' = k + 1, and enter the next loop. If the termination condition is met, then stop the iteration and exit the loop. The sharpness function γ(k) with respect to the iteration step size is obtained through iteration. The defect image corresponding to the first peak point of γ(k) is selected as the optimal defect image and output. In this way, the direct wave component in the diagonal matrix S is determined and the best defect imaging result is obtained. S3. The defect imaging results in S2 are enhanced using a composite algorithm of Gaussian high-pass filtering and contrast-limited adaptive histogram equalization. First, the defect image is processed by a Gaussian high-pass filter. The Gaussian high-pass filter achieves image sharpening by suppressing the low-frequency components of the frequency domain image. In the defect image, its effect is to suppress background noise. The Gaussian high-pass filter is expressed as: Where H(i,j) is the frequency domain form of the image, h is the coefficient of the Gaussian filter, Gau(i,j) is the image after Gaussian filter transformation in the frequency domain, row is the number of rows of the image, col is the number of columns of the matrix, and D0 is the transformation parameter. The image processed by the Gaussian high-pass filter is further subjected to contrast-limited adaptive histogram equalization, which enhances the contrast of the defective region in the image without increasing the background noise. By sequentially performing Gaussian high-pass filtering and contrast-limited adaptive histogram equalization on the defective image, the purpose of further suppressing background noise and enhancing contrast is achieved. S4. Determine the image segmentation threshold using the gray-level histogram, and segment the image from S3 to obtain a binarized image. After obtaining high-contrast defect images through signal and image processing, a grayscale histogram of the defect images is plotted. Envelope processing is then performed to obtain the grayscale histogram envelope of each defect image. Within the grayscale histogram envelope, two peaks and a trough appear. The band centered on the first peak represents the clustering area of ​​background pixels, and the band centered on the second peak represents the clustering area of ​​defect pixels. The grayscale value corresponding to the number of pixels in the trough is determined by finding the minimum value between the two peaks. Simultaneously, using a defect-free image as a reference, the grayscale mean of the defect-free image is calculated. This grayscale mean is used as a compensation standard. The grayscale threshold is obtained by adding the grayscale value corresponding to the trough of the grayscale histogram envelope of each defect image to the grayscale mean of the defect-free image. This grayscale threshold is then used to segment each defect image, resulting in a binarized defect image. S5. Evaluate the defect location and area of ​​the binarized defect image. The binarized defect image is distributed as 0-1, with pixels of value 1 being defect points and pixels of value 0 being background points. For defect location assessment, the defect region is a simply connected region, and the defect location is obtained by finding the centroid of the region. For defect area assessment, the number of defect points is counted in the binarized defect image, and the size corresponding to the pixel in the defect image is obtained by using the scanning step size. Multiplying this size by the number of defect points gives the planar size of the defect.

2. The visual quantitative assessment method according to claim 1, characterized in that, The specific implementation process of step S2 is as follows: S2.1 Performs matrix singular value decomposition on the reflected signal dataset, performs direct wave cancellation, and reconstructs the reflected signal after direct wave cancellation. The reflected signal obtained by two-dimensional cylindrical scanning of the composite insulator specimen (8) is reconstructed into matrix A in the following manner: in, Let S be the reflection coefficient of the scanning point located at coordinate point (x, y) when the frequency is l. 11 l' is the last frequency point in the operating frequency band; singular value decomposition of A is expressed as: A=USV T Where U is the left singular matrix, V is the right singular matrix, and S is a diagonal matrix storing the eigenvalues, i.e., singular values, of A. The specific form of the diagonal matrix S is as follows: Where, the diagonal element σ of S i These are the singular values ​​of matrix A, where 1 ≦ i ≦ r, r is the rank of A, and σ1 ≥ σ2 ≥ … ≥ σ i ≥…≥σ r These singular values ​​reflect the characteristics of the reflected signal, with the largest singular values ​​representing the direct wave that accounts for the highest proportion in the reflected wave. First, the largest singular value in the diagonal matrix S is set to 0, resulting in the diagonal matrix S' after direct wave suppression. With the left and right singular matrices unchanged, substituting S' back into the matrix singular value decomposition formula, we can inversely reconstruct the reflected signal: US'V T =A' Where A' is the reflected signal matrix after eliminating the direct wave; S2.2 Extract signal features from the reconstructed reflection signal to perform defect imaging. The reflection coefficient matrix A' after eliminating the direct wave is decomposed into reflection signal data at each spatial scan point, and then the reconstructed frequency domain signal at each scan point is converted into a time domain signal through inverse Fourier transform: y = ifft(b) Where b is the reflected signal at each spatial scanning point in A', and y is the converted time-domain signal; after obtaining the time-domain signal at each scanning point, the area enclosed by the time-domain signal and the time axis is extracted as the signal feature: SA=∫ydt Where SA is the area enclosed by the time-domain signal and the horizontal axis, i.e., the time axis, and is denoted as the time-domain signal area; t is the horizontal axis time of the time-domain signal; the time-domain signal area SA of each scanning point is extracted for defect imaging, thereby realizing the visualization of defects; For two-dimensional cylindrical surface scanning, the scanning plane is a cylindrical surface. During the imaging process, the cylindrical surface is converted into a plane using the angle-arc length conversion formula: Where arc is the arc length corresponding to one rotation angle of the rotation axis (4), θ is the rotation angle step of the composite insulator specimen (8) during the rotation process, in degrees, and cir is the circumference of the cylindrical surface where the defect is located; thus, the scanned cylindrical surface is converted into the plane used for imaging. In the imaging plane, the horizontal axis direction is the axial direction of the specimen during the scanning process, and the vertical axis direction corresponds to the circumferential direction during the scanning process. S2.3 uses the Tenengrad function model to calculate the image sharpness in S2.

2. After obtaining the defect image, the Tenengrad function is introduced to evaluate the sharpness of the obtained defect image. In the process of establishing the Tenengrad function, the Sobel convolution kernel is used. The Sobel convolution kernel is expressed as: Among them G x and G y Let represent the horizontal and vertical convolution kernels, respectively. The sliding window expression for gradient calculation at coordinates (x, y) in the defect image is as follows: Where f(x,y) is the pixel value of the defect image at coordinates (x,y), and f is the horizontal gradient at coordinates (x,y). x and vertical gradient f y It is obtained by convolution of the convolution kernel and the sliding window: The gradient g at coordinates (x, y) is expressed as: A 95% screening criterion is introduced to filter the gradients at each coordinate point in the defect image: To reduce the impact of background noise on the image sharpness, the gradients of all coordinates are arranged from minimum to maximum value, and the gradient values ​​with a percentile greater than 95% are extracted for further calculation, denoted as g'(x',y'). The final image sharpness γ is expressed as: Where num is the number of gradient values ​​with a percentile greater than 95% among all coordinate points; S2.4 uses image sharpness as a standard and adaptively adjusts the singular value sequence values ​​to achieve precise removal of direct wave signals. The direct wave component is identified by using matrix singular value decomposition and the Tenengrad function model, resulting in a defect image after suppressing the direct wave to the greatest extent. After obtaining the sharpness of the defect image by setting the first, largest singular value in the diagonal matrix S to 0, return to the diagonal matrix S, set the first two singular values ​​to 0, and reconstruct the reflection signal again; repeat the above process: extract the time-domain signal area SA from the reconstructed reflection signal for defect imaging, and calculate the image sharpness based on the defect image using the Tenengrad function model; calculate the difference between the image sharpness calculated this time and the image sharpness calculated last time; if the difference is less than 10... -3 If the difference is greater than or equal to 10, then exit the loop. -3 Then, we return to the diagonal matrix S, set the first three singular values ​​to 0, and continue to generate defect images and calculate their image sharpness. This process continues until the iteration termination condition is met and the loop is exited, resulting in a sharpness vector γ(k) corresponding to the number of singular values ​​set to 0 in the diagonal matrix. In this sharpness vector, the defect image corresponding to the first peak point is selected as the defect image after suppressing the direct wave to the greatest extent and is output.

3. The visual quantitative evaluation method according to claim 1, characterized in that, The microwave detection system obtains information about internal defects in the composite insulator specimen (8) by analyzing the near-field reflection signal from the composite insulator specimen (8), i.e., the near-field microwave reflection method.

4. The visualization and quantitative evaluation method according to claim 1, characterized in that, By combining the rotating shaft (4), the linear shaft (5), and the clamping and fixing device, a smooth scanning of the cylindrical surface of the main part of the cylindrical shape of the composite insulator specimen was achieved.

5. The visual quantitative assessment method according to claim 1, characterized in that, Uniform spatial sampling of the two-dimensional cylindrical surface data of the composite insulator specimen is ensured by controlling the sampling frequency of the vector network analyzer (1), the moving speed of the rectangular waveguide (7), and the rotation speed of the rotating shaft (4).

Images