A method for fast crack location based on eddy current convolution kernel
Patent Information
- Application Number
- CN202311368773.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-20
- Publication Date
- 2026-09-25
- Estimated Expiration
- 2043-10-20
AI Technical Summary
[0004]本发明提供了一种基于涡流卷积核的裂纹快速定位方法，解决了裂纹定位基于深度学习的缺陷信息检索需要花费大量时间以及分形涡流传感器难以评估裂纹信息的问题
[0012]本发明的有益效果是：通过涡流场数据准备三值化卷积核，对C扫描信号进行处理，得到单独目标信号，然后使用三值化卷积核对单独目标信号进行卷积，能够完成对裂纹定量识别的快速定位，并且本发明不需要和机器学习一样需要训练集，可以处理单个C扫描信号，能够提高裂纹定位的速度，并且通过涡流分布矩阵构建卷积核，解决了裂纹定位基于深度学习的缺陷信息检索需要花费大量时间以及分形涡流传感器难以评估裂纹信息的问题。
Smart Images
Figure CN117517449B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of crack location technology, specifically relating to a rapid crack location method based on eddy current convolution kernel. Background Technology
[0002] Quantitative defect assessment is a key indicator for evaluating the condition of components using eddy current testing results. Information in quantitative defect assessment typically includes defect location, geometry, and physical parameters. Therefore, defect information is multi-dimensional, and traditional physics-based optimization algorithms require significant computational resources and time. However, deep learning-based defect information retrieval methods require large amounts of training data. Current technologies do not utilize the known physical field information of the detection system, resulting in low inversion efficiency and failing to reduce the time and hardware resources required for quantitative defect assessment.
[0003] Fractal eddy current sensors exhibit high consistency and low false negative rate in detecting short cracks in all directions. However, the region where the C-scan signal of such sensors exceeds the noise threshold is much larger than the crack size. Therefore, traditional methods struggle to assess crack information. Eddy current detection, on the other hand, is based on the interaction between eddy currents and cracks. Consequently, the shape of the C-scan signal from a fractal eddy current sensor is directly related to the crack morphology and the distribution of the eddy current field. Summary of the Invention
[0004] This invention provides a method for rapid crack localization based on eddy current convolution kernels, which solves the problems that deep learning-based defect information retrieval for crack localization requires a lot of time and that fractal eddy current sensors are difficult to evaluate crack information.
[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is: a rapid crack localization method based on eddy current convolution kernel, comprising the following steps:
[0006] S1. Use the finite element method to simulate the sensor detection process of cracks, obtain the eddy current field, and extract the simulated eddy current field data;
[0007] S2. Calculate the eddy current distribution matrix based on the simulated eddy current field data, and construct a ternary convolution kernel using the eddy current distribution matrix;
[0008] S3. Obtain the C-scan signal through a fractal eddy current sensor, divide the C-scan signal into regions of interest, obtain individual target signals, and process the individual target signals through an interpolation algorithm to make the individual target signals have the same spatial resolution as the ternary convolution kernel;
[0009] S4. Use a ternary convolution kernel to perform convolution operations on individual target signals;
[0010] S5. Perform a power-law transformation on the convolutional data to change the size distribution of the data and obtain the signal center of the individual target signal.
[0011] S6. Calculate the coordinates of the signal center and offset the coordinates using spatial resolution to obtain the crack location, thus completing the rapid location of the crack.
[0012] The beneficial effects of this invention are as follows: by preparing a ternary convolution kernel using eddy current field data, processing the C-scan signal to obtain a single target signal, and then using the ternary convolution kernel to convolve the single target signal, rapid location of crack quantitative identification can be achieved. Furthermore, this invention does not require a training set like machine learning, can process a single C-scan signal, and can improve the speed of crack location. Moreover, by constructing the convolution kernel using the eddy current distribution matrix, it solves the problems of the time-consuming defect information retrieval based on deep learning for crack location and the difficulty of evaluating crack information by fractal eddy current sensors.
[0013] Furthermore, the specific steps of step S2 are as follows:
[0014] S21. Based on the simulated eddy current field data, calculate the eddy current distribution using finite element simulation and calculate the eddy current distribution matrix;
[0015] S22. Normalize and binarize the eddy current distribution matrix in sequence to obtain the binarized convolution kernel of the eddy current field.
[0016] S23. The binary convolution kernel of the eddy current field is ternary to obtain the ternary convolution kernel.
[0017] Furthermore, the expression for the eddy current distribution matrix in step S21 is:
[0018]
[0019] Where J represents the eddy current distribution matrix, R represents the real part of the eddy current distribution signal, and I represents the imaginary part of the eddy current distribution signal;
[0020] The expression for the binarized convolution kernel of the eddy current field in step S22 is as follows:
[0021]
[0022] Among them, J b [m,n] represents the element in the m-th row and n-th column of the binary convolution kernel, J[m,n] represents the element in the m-th row and n-th column of the normalized eddy current distribution matrix, and max(J) represents the maximum value of the eddy current distribution matrix J;
[0023] The expression for the ternary convolution kernel in step S23 is:
[0024]
[0025] Among them, J z [m,n] represents the element in the m-th row and n-th column of the ternary convolution kernel, and M represents the ternary convolution kernel J. z The number of rows and columns of a matrix.
[0026] The beneficial effects of the above-mentioned further scheme are as follows: the ternary convolution kernel constructed by the present invention has an antisymmetric structure, and the positive and negative halves of the ternary convolution kernel can cancel random noise. The crack signal also has an approximately antisymmetric structure. When the ternary convolution kernel and the crack signal overlap and their positive and negative parts overlap respectively, the signal can be enhanced.
[0027] Furthermore, the specific steps of step S3 are as follows:
[0028] S31. Perform detrending processing on the C scan signal matrix along the scanning direction to obtain the detrending signal;
[0029] S32. Rotate the coordinates of the detrending signal;
[0030] S33. Divide the rotated signal into regions of interest to obtain individual target signals;
[0031] S34. Process individual target signals through interpolation algorithms to obtain the actual data distribution of individual target signals, so that individual target signals and ternary convolution kernels have the same spatial resolution.
[0032] The beneficial effects of the above-mentioned further scheme are as follows: by using interpolation methods, the distance between C-scan signal data is supplemented to the real scene, and the actual data distribution of individual target signals is obtained, so that individual target signals and ternary convolution kernels have the same spatial resolution, so as to facilitate the location of targets by using convolution operations.
[0033] Furthermore, the formula for coordinate rotation in step S32 is:
[0034] S r [p,q]=S R,d [p,q]cosθ+S I,d [p,q]sinθ
[0035] Among them, S r [p,q] represents the element in the p-th row and q-th column of the rotated signal, S R,d [p,q] represents the element in the p-th row and q-th column of the real part of the detrended signal, S I,d [p,q] represents the p-th row and q-th column element of the detrended signal's imaginary part, and θ represents the rotation angle.
[0036] The beneficial effects of the above-mentioned further scheme are: to perform detrending processing on the C-scan signal, and then to perform coordinate rotation on the real and imaginary parts of the signal, so as to enhance the amplitude of the signal and highlight the signal characteristics.
[0037] Furthermore, the specific steps of step S33 are as follows:
[0038] A1. Use the gradient function to calculate the gradient in the vertical direction of the signal after rotation;
[0039] A2. Normalize the gradient and transform it through a threshold to obtain a binary image;
[0040] A3. Perform dilation operations on the binary image in both directions;
[0041] A4. Extract the boundaries of the dilated image, and use the outer rectangle of each boundary as the region of interest to segment out individual target signals.
[0042] The beneficial effects of the above-mentioned further scheme are as follows: using the gradient function to calculate the gradient in the vertical direction can reveal the changing trend characteristics of the target signal. Then, by converting through a threshold, noise can be reduced and the target signal can be highlighted. Performing dilation operation can fill the gaps in the target signal, making the target more continuous and complete and strengthening the edges. The image boundary is extracted, and a box is defined according to the minimum and maximum coordinates of each signal boundary. The box is used to segment each individual target signal in the C-scan signal, thereby achieving data segmentation.
[0043] Furthermore, the expression for the gradient in step A1 is:
[0044]
[0045] Where G[p,q] represents the gradient of the element in the p-th row and q-th column of the rotated signal matrix, |·| represents taking the absolute value, and S r [p, q+1] represents the element in the p-th row and q+1-th column of the rotated signal matrix, S r [p,q-1] represents the element in the p-th row and q-1-th column of the rotated signal matrix, and Q' represents the data width of the rotated signal;
[0046] The formula for the expansion operation in step A3 is:
[0047]
[0048] Among them, G e G represents the dilated image. b Represents a binary image. Indicates the expansion operation, SE y and SE x Both represent linear structuring elements of the expansion operation.
[0049] Furthermore, the formula for the convolution operation in step S4 is:
[0050] C b [r,s]=|S I,i *J z |
[0051] r = 1, 2, ..., R, s = 1, 2, ..., S
[0052]
[0053] Among them, C b [r,s] represents the element in the r-th row and s-th column of the convolutional data matrix, where S... I,i J represents a single target signal in the i-th region of interest. z The ternary convolution kernel is represented by *, the convolution operation is represented by |·|, the absolute value is represented by R, the width of the convolutional data matrix is represented by S, and W is the height of the convolutional data matrix. i H represents the width of the individual target signal in the i-th region of interest. i J represents the height of a single target signal in the i-th region of interest, and M represents the ternary convolution kernel J. z The number of rows and columns of a matrix.
[0054] The beneficial effect of the above-mentioned further scheme is that by using a ternary convolution kernel to perform convolution operations on the segmented individual target signals, target localization can be achieved.
[0055] Furthermore, the formula for the power-law transformation in step S5 is:
[0056] C d [p,q]=C p [p,q] τ
[0057] Among them, C d [p,q] represents the element in the p-th row and q-th column of the signal center matrix, C p [p,q] represents the response signal of the power-law transform, and τ represents the coefficients of the power-law transform.
[0058] The beneficial effect of the above-mentioned further scheme is that by performing a power-law transformation on the convolutional data, the size distribution of the data can be changed, highlighting the signal center.
[0059] The expression for the crack location in step S6 is:
[0060] (P,Q)=argmax(C d [p,q])
[0061]
[0062] Where (P,Q) represents the position coordinates when the signal center amplitude is at its maximum, and argmax(·) represents the function to find the maximum position coordinates. s represents the location of the crack in the i-th region of interest. d The spatial resolution of a single target signal, s k R represents the spatial resolution of the ternary convolution kernel. i T represents the right part of the i-th region of interest. i This represents the top of the i-th region of interest.
[0063] The beneficial effect of the above-mentioned further scheme is that: the coordinates of the signal center of the power law transform are calculated, and then offset to obtain the coordinates of the crack, thus completing the crack location. Attached Figure Description
[0064] Figure 1 This is a flowchart of the crack rapid localization method based on eddy current convolution kernel of the present invention.
[0065] Figure 2 This is a schematic diagram illustrating the preparation of the convolution kernel for this invention.
[0066] Figure 3 The region of interest is defined for this invention.
[0067] Figure 4 This is a schematic diagram of crack location in the present invention.
[0068] Figure 5 This refers to the location and error of the crack in this invention. Detailed Implementation
[0069] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.
[0070] Example
[0071] like Figure 1 As shown, this invention provides a method for rapid crack localization based on eddy current convolution kernels, comprising the following steps:
[0072] S1. Use the finite element method to simulate the sensor detection process of cracks, obtain the eddy current field, and extract the simulated eddy current field data;
[0073] S2. Calculate the eddy current distribution matrix based on the simulated eddy current field data, and construct a ternary convolution kernel using the eddy current distribution matrix;
[0074] S3. Obtain the C-scan signal through a fractal eddy current sensor, divide the C-scan signal into regions of interest, obtain individual target signals, and process the individual target signals through an interpolation algorithm to make the individual target signals have the same spatial resolution as the ternary convolution kernel;
[0075] S4. Use a ternary convolution kernel to perform convolution operations on individual target signals;
[0076] S5. Perform a power-law transformation on the convolutional data to change the size distribution of the data and obtain the signal center of the individual target signal.
[0077] S6. Calculate the coordinates of the signal center and offset the coordinates using spatial resolution to obtain the crack location, thus completing the rapid location of the crack.
[0078] The specific steps of step S2 are as follows:
[0079] S21. Based on the simulated eddy current field data, calculate the eddy current distribution using finite element simulation and calculate the eddy current distribution matrix;
[0080] S22. Normalize and binarize the eddy current distribution matrix in sequence to obtain the binarized convolution kernel of the eddy current field.
[0081] S23. The binary convolution kernel of the eddy current field is ternary to obtain the ternary convolution kernel.
[0082] In this embodiment, the eddy current distribution is calculated using finite element simulation. The calculated eddy current distribution is a matrix composed of complex numbers. The eddy current distribution matrix is normalized to the range of 0 to 1 to obtain the normalized eddy current distribution matrix. Then, the normalized eddy current distribution matrix is binarized to obtain the eddy current field binarized convolution kernel. Finally, the eddy current field binarized convolution kernel is ternary.
[0083] The expression for the eddy current distribution matrix in step S21 is:
[0084]
[0085] Where J represents the eddy current distribution matrix, R represents the real part of the eddy current distribution signal, and I represents the imaginary part of the eddy current distribution signal;
[0086] The expression for the binarized convolution kernel of the eddy current field in step S22 is as follows:
[0087]
[0088] Among them, J b[m,n] represents the element in the m-th row and n-th column of the binary convolution kernel, J[m,n] represents the element in the m-th row and n-th column of the normalized eddy current distribution matrix, and max(J) represents the maximum value of the eddy current distribution matrix J;
[0089] The expression for the ternary convolution kernel in step S23 is:
[0090]
[0091] Among them, J z [m,n] represents the element in the m-th row and n-th column of the ternary convolution kernel, and M represents the ternary convolution kernel J. z The number of rows and columns of a matrix.
[0092] In this embodiment, the ternary convolution kernel J z The matrix is M×M in size and is a square matrix.
[0093] The specific steps of step S3 are as follows:
[0094] S31. Perform detrending processing on the C scan signal matrix along the scanning direction to obtain the detrending signal;
[0095] S32. Rotate the coordinates of the detrending signal;
[0096] S33. Divide the rotated signal into regions of interest to obtain individual target signals;
[0097] S34. Process individual target signals through interpolation algorithms to obtain the actual data distribution of individual target signals, so that individual target signals and ternary convolution kernels have the same spatial resolution.
[0098] In this embodiment, the real and imaginary parts of the detrending signal are rotated from -180° to 180° by a rotation angle θ in 180° steps. In all cases, we use the maximum value of the peak-to-peak value of the real part of the detrending signal.
[0099] The formula for coordinate rotation in step S32 is:
[0100] S r [p,q]=S R,d [p,q]cosθ+S I,d [p,q]sinθ
[0101] Among them, S r [p,q] represents the element in the p-th row and q-th column of the rotated signal, S R,d [p,q] represents the element in the p-th row and q-th column of the real part of the detrended signal, S I,d [p,q] represents the p-th row and q-th column element of the detrended signal's imaginary part, and θ represents the rotation angle.
[0102] The specific steps of step S33 are as follows:
[0103] A1. Use the gradient function to calculate the gradient in the vertical direction of the signal after rotation;
[0104] A2. Normalize the gradient and transform it through a threshold to obtain a binary image;
[0105] A3. Perform dilation operations on the binary image in both directions;
[0106] A4. Extract the boundaries of the dilated image, and use the outer rectangle of each boundary as the region of interest to segment out individual target signals.
[0107] In this embodiment, edges are calculated and adjacent points are connected to form a closed edge contour to obtain the boundary of each signal. By finding the boundary of each non-zero region, the outer rectangle of each boundary is taken as the region of interest (ROI). The ROI consists of four positions: the left, right, bottom, and top positions. Each ROI contains a single target signal; the single target signal in the i-th ROI is W. i ×H i The matrix, where W i =R i -L i +1,H i =T i -B i +1, W i H represents the width of the individual target signal in the i-th region of interest. i L represents the high level of a single target signal in the i-th region of interest. i R represents the left part of the i-th region of interest. i B represents the right part of the i-th region of interest. i T represents the bottom of the i-th region of interest. i This represents the top of the i-th region of interest.
[0108] The expression for the gradient in step A1 is:
[0109]
[0110] Where G[p,q] represents the gradient of the element in the p-th row and q-th column of the rotated signal matrix, |·| represents taking the absolute value, and S r [p, q+1] represents the element in the p-th row and q+1-th column of the rotated signal matrix, S r [p,q-1] represents the element in the p-th row and q-1-th column of the rotated signal matrix, and Q' represents the data width of the rotated signal;
[0111] The formula for the expansion operation in step A3 is:
[0112]
[0113] Among them, G e G represents the dilated image. b Represents a binary image. Indicates the expansion operation, SE y and SE x Both represent linear structuring elements of the expansion operation.
[0114] The formula for the convolution operation in step S4 is:
[0115] C b [r,s]=|S I,i *J z |
[0116] r = 1, 2, ..., R, s = 1, 2, ..., S
[0117]
[0118] Among them, C b [r,s] represents the element in the r-th row and s-th column of the convolutional data matrix, where S... I,i J represents a single target signal in the i-th region of interest. z The ternary convolution kernel is represented by *, the convolution operation is represented by |·|, the absolute value is represented by R, the width of the convolutional data matrix is represented by S, and W is the height of the convolutional data matrix. i H represents the width of the individual target signal in the i-th region of interest. i J represents the height of a single target signal in the i-th region of interest, and M represents the ternary convolution kernel J. z The number of rows and columns of a matrix.
[0119] The formula for the power-law transformation in step S5 is:
[0120] C d [p,q]=C p [p,q] τ
[0121] Among them, C d [p,q] represents the element in the p-th row and q-th column of the signal center matrix, C p [p,q] represents the response signal of the power-law transform, and τ represents the coefficients of the power-law transform.
[0122] In this embodiment, power-law transformation is used to improve the signal-to-noise ratio (SNR) of the signal, and the signal center C at the defect location is used. dThe amplitude of the signal is higher than that of the noise. In the power-law transform, the higher the signal strength, the greater the signal amplification factor when the power-law transform coefficient τ>1. Therefore, the signal-to-noise ratio (SNR) will be improved through the power-law transform, and the SNR is increased by a factor of τ after this transform.
[0123] The expression for the crack location in step S6 is:
[0124] (P,Q)=argmax(C d [p,q])
[0125]
[0126] Where (P,Q) represents the position coordinates when the signal center amplitude is at its maximum, and argmax(·) represents the function to find the maximum position coordinates. s represents the location of the crack in the i-th region of interest. d The spatial resolution of a single target signal, s k R represents the spatial resolution of the ternary convolution kernel. i T represents the right part of the i-th region of interest. i This represents the top of the i-th region of interest.
[0127] In this embodiment, taking the real part of the 100kHz excitation frequency detection signal as an example, the performance of the sensor is verified by adding random noise to the signal. Then, the relative error of the sensor in locating the crack in the test block C scanning signal at different excitation frequencies is processed and analyzed.
[0128] The eddy current field data on the surface of the test block were obtained using the finite element simulation software COMSOL. For example... Figure 2 As shown in (a), the eddy current field data in both directions range from -7.5 mm to 7.5 mm, and the distance between two adjacent points is 0.1 mm. Figure 2 (b) and Figure 2 (c) shows the convolution kernel after binarization and ternaryization.
[0129] Figure 3 Twenty-eight regions of interest were established, namely the real part, imaginary part, rotated real part, normalized absolute gradient, binary gradient, dilated gradient, and region of interest (ROI) of the C-scan signal.
[0130] In a binary gradient process, the threshold is 2% of the maximum gradient value. For example... Figure 3 As shown, the binary gradients around each crack location are not connected, so a dilation operation from an image algorithm is used to connect the regions of each crack. Besides... Figure 3Outside the region of interest in the upper right corner, each crack signal is framed within its respective region of interest. For the region of interest in the upper right corner, the crack length is 0.5 mm, and the signal is affected by noise, causing its region of interest to be smaller than the actual size.
[0131] The results of crack location are as follows Figure 4 As shown, the signal exhibits an antisymmetric distribution, with the upper and lower halves displaying positive and negative values, respectively. After convolution, the signal center is enhanced. Then, through a power-law transformation, the signal at the peak position is significantly enhanced. Finally, the location of the crack is pinpointed at the peak position.
[0132] Figure 5 The results of the C-scan signal localization are shown; the black dot represents the crack center, located at the center of the crack signal. To evaluate the algorithm's accuracy, positional error and relative error were calculated.
[0133] Position error e L Defined as:
[0134]
[0135]
[0136]
[0137] relative error δ L Defined as:
[0138]
[0139] Where L represents the distance between the a-th actual crack center location and the first actual crack center location. x represents the distance between the a-th estimated crack center location and the first estimated crack center location, |·| represents taking the absolute value. a The x-coordinate of the position of the a-th actual crack center is represented by y. a Let x1 represent the ordinate of the a-th actual crack center location, x1 represent the a-th actual crack center location, and y1 represent the ordinate of the first actual crack center location. This represents the x-coordinate of the estimated crack center location (a-th). This represents the ordinate of the a-th estimated crack center location. This represents the x-coordinate of the first estimated crack center location. This represents the ordinate of the first estimated crack center location.
[0140] like Figure 5 As shown, for crack location, the maximum positional error is 1.95 mm, and the maximum relative error is less than 2.61%.
[0141] This invention constructs a ternary convolution kernel using an eddy current distribution matrix. Unlike many other machine learning algorithms that require a lot of training data and are black-box algorithms, the algorithm proposed in this invention does not require a training set, can process single C-scan data, and can quickly obtain eddy current distribution without spending a lot of time retrieving defect information.
[0142] Cracks often have multiple information such as location, shape, orientation, and physical property parameters that need to be inverted. Inversion based on optimization algorithms usually requires a lot of operation time. This invention can quickly reduce the time and hardware resources required for quantitative defect assessment and improve inversion efficiency to a certain extent.
[0143] The region where the C-scan signal of the fractal eddy current sensor exceeds the noise threshold is much larger than the crack size; therefore, a ternary convolution kernel is constructed. Unlike many convolution algorithms that derive their kernels from digital image processing methods, the ternary convolution kernel of this invention is constructed from an eddy current distribution based on the interaction between eddies and cracks. Therefore, this invention solves the problems of time-consuming defect information retrieval based on deep learning for crack localization and the difficulty of evaluating crack information using fractal eddy current sensors.
Claims
1. A method for rapid crack localization based on eddy current convolution kernels, characterized in that, Includes the following steps: S1. Use the finite element method to simulate the sensor detection process of cracks, obtain the eddy current field, and extract the simulated eddy current field data; S2. Calculate the eddy current distribution matrix based on the simulated eddy current field data, and construct a ternary convolution kernel using the eddy current distribution matrix; S3. Obtain the C-scan signal through a fractal eddy current sensor, divide the C-scan signal into regions of interest, obtain individual target signals, and process the individual target signals through an interpolation algorithm to make the individual target signals have the same spatial resolution as the ternary convolution kernel; S4. Use a ternary convolution kernel to perform convolution operations on individual target signals; S5. Perform a power-law transformation on the convolutional data to change the size distribution of the data and obtain the signal center of the individual target signal. S6. Calculate the coordinates of the signal center and offset the coordinates by spatial resolution to obtain the crack location and complete the rapid location of the crack. The specific steps of step S2 are as follows: S21. Based on the simulated eddy current field data, calculate the eddy current distribution using finite element simulation and calculate the eddy current distribution matrix; S22. Normalize and binarize the eddy current distribution matrix in sequence to obtain the binarized convolution kernel of the eddy current field. S23. The binary convolution kernel of the eddy current field is ternary to obtain the ternary convolution kernel; The specific steps of step S3 are as follows: S31. Perform detrending processing on the C scan signal matrix along the scanning direction to obtain the detrending signal; S32. Rotate the coordinates of the detrending signal; S33. Divide the rotated signal into regions of interest to obtain individual target signals; S34. Process individual target signals through interpolation algorithms to obtain the actual data distribution of individual target signals, so that individual target signals and ternary convolution kernels have the same spatial resolution. The specific steps of step S33 are as follows: A1. Use the gradient function to calculate the gradient in the vertical direction of the signal after rotation; A2. Normalize the gradient and transform it through a threshold to obtain a binary image; A3. Perform dilation operations on the binary image in both directions; A4. Extract the boundaries of the dilated image, and use the outer rectangle of each boundary as the region of interest to segment out individual target signals.
2. The crack rapid localization method based on eddy current convolution kernel according to claim 1, characterized in that, The expression for the eddy current distribution matrix in step S21 is: in, Represents the eddy current distribution matrix. The signal representing the real part of the eddy current distribution. The signal representing the imaginary part of the eddy current distribution; The expression for the binarized convolution kernel of the eddy current field in step S22 is as follows: in, The first convolution kernel is the binarized kernel. Line number Column elements, The first digit represents the normalized eddy current distribution matrix. Line number Column elements, Represents the eddy current distribution matrix The maximum value; The expression for the ternary convolution kernel in step S23 is: in, The ternary convolution kernel represents the first... Line number Column elements, Represents the ternary convolution kernel The number of rows and columns of a matrix.
3. The crack rapid localization method based on eddy current convolution kernel according to claim 1, characterized in that, The formula for coordinate rotation in step S32 is: in, The signal after rotation is the first Line 1 Column elements, The first real part of the detrending signal represents the... Line 1 Column elements, The first imaginary part of the detrending signal represents the... Line 1 Column elements, Indicates the rotation angle.
4. The crack rapid localization method based on eddy current convolution kernel according to claim 1, characterized in that, The expression for the gradient in step A1 is: in, The first element of the rotated signal matrix represents the... Line 1 Gradient of column elements, This indicates taking the absolute value. The first element of the rotated signal matrix represents the... Line 1 Column elements, The first element of the rotated signal matrix represents the... Line 1 Column elements, This indicates the data width of the signal after rotation; The formula for the expansion operation in step A3 is: in, This represents the image after dilation. Represents a binary image. This indicates an expansion operation. and Both represent linear structuring elements of the expansion operation.
5. The crack rapid localization method based on eddy current convolution kernel according to claim 1, characterized in that, The formula for the convolution operation in step S4 is: in, The first element of the data matrix after convolution represents the... Line number Column elements, Indicates the first Individual target signals within a region of interest, Represents the ternary convolution kernel, This represents the convolution operation. This indicates taking the absolute value. This represents the width of the data matrix after convolution. This represents the height of the data matrix after convolution. Indicates the first The width of a single target signal in a region of interest. Indicates the first The high signal intensity of a single target within a region of interest. Represents the ternary convolution kernel The number of rows and columns of a matrix.
6. The crack rapid localization method based on eddy current convolution kernel according to claim 5, characterized in that, The formula for the power-law transformation in step S5 is: in, The first element of the signal center matrix represents the... Line number Column elements, This represents the response signal of the power-law transform. Represents the coefficients of the power-law transformation.
7. The crack rapid localization method based on eddy current convolution kernel according to claim 6, characterized in that, The expression for the crack location in step S6 is: in, This represents the position coordinates where the signal center amplitude is at its maximum. This represents the function to find the maximum position coordinates. Indicates the first The location of the crack in the region of interest. Indicates the spatial resolution of a single target signal. This represents the spatial resolution of the ternary convolution kernel. Indicates the first The right side of the region of interest, Indicates the first The top of the region of interest.