A method for in-depth defect analysis of eddy current signals based on machine learning

Abstract

The present invention specifically relates to a defect depth analysis method of an eddy current signal based on machine learning, comprising the following steps: (1) constructing a known defect matrix X; (2) constructing a known defect phase row vector Y; (3) generating a radial basis network Net according to the known defect matrix X and the known defect phase row vector Y; (4) constructing an unknown defect signal vector x; (5) inputting the radial basis network Net and the unknown defect signal vector x into a simulation function sim(), and outputting a phase angle θ corresponding to the unknown defect signal vector x; (6) writing a phase angle θ and a depth correspondence table into a vector D with a length of 180; (7) according to the vector D, finding out the corresponding defect depth D(θ)% of the phase angle θ obtained in step (5). The defect depth analysis method of an eddy current signal based on machine learning of the present invention uses the artificial neural network to predict the phase angle of an unknown defect, and uses the phase angle and depth correspondence table to give the defect depth.

Description

Technical Field

[0001] The present invention relates to the technical field of nondestructive testing signal processing, and in particular to a method for analyzing eddy current signal defect depth based on machine learning. Background Art

[0002] In eddy current signal processing, the measurement of defect depth is obtained by calculating the phase angle of the defect signal. In "Eddy Current Detection" published by Machinery Industry Press in 2006, Xu Kebei and Zhou Junhua defined the phase angle of the defect signal as follows: take two points with the maximum impedance of the response signal, and define the upper side of the vertical line as the positive direction, connect the two points with a straight line, and the angle between the straight line and the negative direction of the horizontal direction. The Chinese patent with publication number CN 111351842A discloses a method for measuring the defect phase angle based on eddy current signal differential technology. So far, eddy current detection personnel and detection computers have done a lot of work in defect detection. How to use these data and results to predict the depth of unknown defects is a problem we need to consider. Summary of the invention

[0003] Based on this, it is necessary to provide a method for defect depth analysis of eddy current signals based on machine learning to address the problem of how to predict the defect depth based on the high-frequency differential signal of the defect and the phase angle of the defect. This method uses the high-frequency differential signal of the defect obtained in eddy current detection and the phase angle of the defect calculated based on the high-frequency differential signal of the defect to train an artificial neural network based on RBF, and uses the artificial neural network to predict the phase angle of an unknown defect, and uses a phase angle and depth correspondence table to give the defect depth.

[0004] In order to achieve the above object, the present invention provides the following technical solutions:

[0005] A method for analyzing defect depth of eddy current signals based on machine learning comprises the following steps: (1) constructing a known defect matrix X; (2) constructing a known defect phase row vector Y; (3) generating a radial basis network Net according to the known defect matrix X and the known defect phase row vector Y; (4) constructing an unknown defect signal vector x; (5) inputting the radial basis network Net and the unknown defect signal vector x into a simulation function sim(), and outputting a phase angle θ corresponding to the unknown defect signal vector x; (6) writing a correspondence table between the phase angle θ and the depth into a vector D with a length of 180; (7) according to the vector D, finding out the defect depth D(θ)% corresponding to the phase angle θ obtained in step (5).

[0006] Furthermore, in step (1), a horizontal component h and a vertical component v of length M are cut from the high-frequency differential signal of each known defect, and a column vector of length 2M is used It means that N known defects are arranged in sequence to form a 2M×N known defect matrix X.

[0007] Furthermore, in step (2), the N known defect phase angles in step (1) are rounded to integers and arranged in sequence to form a known defect phase row vector Y of length N.

[0008] Furthermore, in step (3), select newrbe() in NewNetwork Functions in the Neural Network Toolbox of MATLAB, input the known defect matrix X and the known defect phase row vector Y to generate a radial basis network Net, that is, Net = newrbe(X, Y).

[0009] Furthermore, in step (4), the high-frequency differential signal of the unknown defect to be detected is intercepted with a horizontal component hx and a vertical component vx of length M, and a defect signal vector of length 2M is used. express.

[0010] Furthermore, in step (7), a defect of θ<40° is an internal injury, θ>40° is an external injury, and θ=40° is a through hole.

[0011] Beneficial technical effects of the present invention:

[0012] The eddy current signal defect depth analysis method based on machine learning of the present invention utilizes high-frequency differential signals of known defects and defect phase angle information to train an artificial neural network, uses the network to predict the phase angle of unknown defects, and gives the defect depth; utilizes prior data, and can continuously add new training data to continuously improve network performance and improve the accuracy of detection and identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is the phase and depth correspondence diagram. DETAILED DESCRIPTION

[0014] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] The present invention provides a method for analyzing defect depth of eddy current signals based on machine learning, comprising the following steps: (1) constructing a known defect matrix X; (2) constructing a known defect phase row vector Y; (3) generating a radial basis network Net according to the known defect matrix X and the known defect phase row vector Y; (4) constructing an unknown defect signal vector x; (5) inputting the radial basis network Net and the unknown defect signal vector x into a simulation function sim(), and outputting a phase angle θ corresponding to the unknown defect signal vector x; (6) writing a phase angle θ and depth correspondence table into a vector D with a length of 180; (7) according to the vector D, finding out the defect depth D(θ)% corresponding to the phase angle θ obtained in step (5).

[0016] Furthermore, in step (1), a horizontal component h and a vertical component v of length M are cut from the high-frequency differential signal of each known defect, and a column vector of length 2M is used It means that N known defects are arranged in sequence to form a 2M×N known defect matrix X.

[0017] Furthermore, in step (2), the N known defect phase angles in step (1) are rounded to integers and arranged in sequence to form a known defect phase row vector Y of length N.

[0018] Furthermore, in step (3), select newrbe() in NewNetwork Functions in the Neural Network Toolbox of MATLAB, input the known defect matrix X and the known defect phase row vector Y to generate a radial basis network Net, that is, Net = newrbe(X, Y).

[0019] Furthermore, in step (4), the high-frequency differential signal of the unknown defect to be detected is intercepted with a horizontal component hx and a vertical component vx of length M, and a defect signal vector of length 2M is used. express.

[0020] Furthermore, in step (7), a defect of θ<40° is an internal injury, θ>40° is an external injury, and θ=40° is a through hole.

[0021] The algorithm verification of the eddy current signal defect depth analysis method based on machine learning of the present invention includes the following steps:

[0022] 1. Create a known defect matrix X and a known defect phase row vector Y

[0023] Download the data obtained by six measurements of the calibration tube used in the test heat transfer tube of a nuclear power plant. After determining the defect center point, intercept length M and phase angle of each defect signal in each data file, generate the known defect matrix X and the known defect phase row vector Y according to steps (1) and (2).

[0024] 2. Create the test defect matrix test_X and the test defect phase row vector test_Y

[0025] Here, the test defect matrix test_X is randomly formed by adding Gaussian white noise to 1 / 3 of the above signals with a signal-to-noise ratio SNR of 40 dB, and the test defect phase row vector test_Y keeps the category unchanged.

[0026] 3. Use steps (3) and (5) to solve the phase of test_X, and compare the obtained results with , and the accuracy of the test results is 91.5278%. According to the application background characteristics, errors below 2° are acceptable. In fact, the computer classification results show a difference of plus or minus 1°, which shows that this method can be used to achieve angle detection when the data volume is large enough, that is, the training data includes almost all kinds of defects.

[0027] The defect depth analysis method of eddy current signal based on machine learning of the present invention is performed to solve the defect depth, including the following steps:

[0028] 1. Given phase and depth mapping relationship diagram

[0029] See also Figure 1 The data can be downloaded from the CEddy software. The phase and depth mapping relationship of the data of the nuclear power plant can be seen in DepthTable. The position index represents the phase from 0 to 179°, and the component value represents the depth value (%) corresponding to the position index. The first 40 components represent internal injuries, the 41st component is a through hole, and the following components represent external injuries.

[0030] (DepthTable=[0 3 5 8 10 13 15 18 20 23 25 28 30 33 35 38 40 43 45 4850 53 55 58 60 63 65 68 70 73 75 78 80 83 85 88 90 93 95 98 100 100 99 99 9898 97 97 97 96 96 95 95 94 94 94 93 93 92 92 91 91 90 90 90 89 88 88 87 87 8686 85 85 84 84 83 83 82 8181 80 80 79 79 78 77 77 76 75 75 74 74 73 72 72 71 70 69 68 67 67 66 65 65 64 63 62 62 61 60 59 58 58 57 56 55 54 53 53 52 51 50 49 48 47 46 45 44 43 42 41 40 39 38 37 36 35 34 33 32 31 30 29 27 26 25 24 23 21 20 19 17 16 15 13 12 11 9 8 6 5 3 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0])

[0031] 2. The high-frequency differential signal of the unknown defect to be detected is made into a column signal that meets the standard according to step (4). Here, the first 10 defect signals are directly extracted from the test set.

[0032] 3. Using the network obtained in Example 1, the 10 defect signals are input into the sim function, and the output phases are 63°, 179°, 6°, 83°, 15°, 153°, 2°, 5°, 19°, and 33°, respectively. Since Gaussian noise is added to the signal, there is a certain error between the phases corresponding to the original signal and the obtained results, which are displayed as 0°, 0°, 0°, 0°, 0°, 0°, 1°, 1°, 1°, and 0°, respectively. Since these results are acceptable, the following results are obtained according to the above depth table:

[0033] Defect 1: Trauma, phase is about 63 degrees, depth is about 90 / 100.

[0034] Defect 2: Trauma, phase is about 179 degrees, depth is about 0 / 100.

[0035] Defect 3: Internal injury, phase is about 6 degrees, depth is about 15 / 100.

[0036] Defect 4: Trauma, phase is about 83 degrees, depth is about 79 / 100.

[0037] Defect 5: Internal injury, phase is about 15 degrees, depth is about 38 / 100.

[0038] Defect 6: Trauma, phase is about 153 degrees, depth is about 17 / 100.

[0039] Defect 7: Internal injury, phase is about 2 degrees, depth is about 5 / 100.

[0040] Defect 8: Internal injury, phase is about 5 degrees, depth is about 13 / 100.

[0041] Defect 9: Internal injury, phase is about 19 degrees, depth is about 48 / 100.

[0042] Defect 10: Internal injury, phase is about 33 degrees, depth is about 83 / 100.

[0043] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for analyzing eddy current signal defects in depth based on machine learning, characterized in that: The method comprises the following steps: (1) constructing a known defect matrix X; (2) constructing a known defect phase row vector Y; (3) generating a radial basis network Net according to the known defect matrix X and the known defect phase row vector Y; (4) constructing an unknown defect signal vector x; (5) inputting the radial basis network Net and the unknown defect signal vector x into a simulation function sim(), and outputting a phase angle θ corresponding to the unknown defect signal vector x; (6) writing a phase angle θ and depth correspondence table into a vector D with a length of 180; (7) according to the vector D, finding out the defect depth D(θ)% corresponding to the phase angle θ obtained in step (5); In step (1), the high-frequency differential signal of each known defect is intercepted with a horizontal component h and a vertical component v of length M, and a column vector of length 2M is used Indicates that N known defects are arranged in sequence to form a 2M×N known defect matrix X; In step (2), the N known defect phase angles in step (1) are rounded to integers and arranged in sequence to form a known defect phase row vector Y with a length of N; In step (4), the high-frequency differential signal of the unknown defect to be detected is intercepted with a horizontal component hx and a vertical component vx of length M, and a defect signal vector of length 2M is used. express.

2. The eddy current signal defect depth analysis method based on machine learning according to claim 1 is characterized in that: In step (3), select newrbe() in New Network Functions in the NeuralNetwork Toolbox of MATLAB, input the known defect matrix X and the known defect phase row vector Y to generate a radial basis network Net, that is, Net = newrbe(X, Y).

3. The eddy current signal defect depth analysis method based on machine learning according to claim 1 is characterized in that: In step (7), a defect of θ<40° is an internal defect, θ>40° is an external defect, and θ=40° is a through hole.

Images