A method for classification and quantitative identification of multiple defects on metal component surfaces based on motional eddy currents

Abstract

This invention discloses a method for classifying and quantitatively identifying multiple defects on the surface of metal components based on motional eddy currents, belonging to the field of quantitative defect detection. The method first performs electromagnetic imaging on the electromagnetic field data acquired through motional eddy current detection. The electromagnetic image is then identified to determine the type of each defect and locate the defect based on the image. The magnetic field and detection signal position curves surrounding the defect are then extracted, and characteristic signals are extracted to quantitatively identify the depth and width of the defect.

Description

Technical Field

[0001] The present invention designs a quantitative detection method based on motional eddy currents. The method images the detection results, classifies the defects, and then quantitatively inverts the sizes of the defects according to signal characteristics, which relates to the field of quantitative defect detection. Background Art

[0002] In industrial production, the quality of finished products is easily compromised by shortcomings and limitations in existing technologies, working conditions, and other factors. Surface defects are the most obvious indicator of compromised product quality. Therefore, to ensure acceptable yields and reliable quality, surface defect detection is essential. A "defect" can generally be understood as a missing, missing, or larger area compared to a normal sample. Defects are generally categorized into two types: cracks and corrosion holes. Defect detection involves detecting surface defects on a sample, thereby obtaining information such as its type, location, and size. Manual defect detection was once the mainstream method, but this approach is inefficient; its results are susceptible to subjective influences and cannot meet the requirements of real-time detection. Furthermore, manual defect detection cannot quantitatively measure defects, leading to its gradual replacement by other methods. Currently, researchers have used image recognition technology to identify, locate, and classify defects using electromagnetic imaging images from sensor detection results. However, many issues remain to be explored, including the quantitative relationship between the detection results and sizes of different defects; the impact of electromagnetic imaging methods on defect classification; and the impact of different image recognition technologies on image recognition accuracy. Therefore, the present invention is of great significance in the rapid detection of surface defects of metal components, and can be used in the online quantitative detection of surface defects of important metal components such as high-speed rails, large-diameter pipelines, and engine blades.

[0003] The prior art, "A Method and System for Defect Detection of Steel Structures Based on Image Recognition" (patent application number CN202311250084.X), proposes a method and system for defect detection of steel structures based on image recognition. This method involves obtaining a real-time workpiece image; dividing the real-time workpiece image into N image blocks, performing local binary conversion, and obtaining a first binary code; determining whether the first binary code is identical to a first predetermined binary code; if not, adding the first image block to the sequence of components to be inspected; matching the first predetermined image blocks; obtaining a first comparison feature set; and obtaining a defect detection result set, thereby improving detection efficiency and accuracy. This technology can only identify the presence of defects, but cannot classify or quantify them.

[0004] "A Sewage Pipeline Defect Image Recognition Method, System, Terminal, and Storage Medium" (Application No. CN202311131855.3) proposes a sewage pipe defect image recognition method, system, terminal, and storage medium. The method includes: acquiring real-time video images of the sewage pipe; inputting the real-time video images into a preset defect learning model; determining whether the sewage pipe has defects based on the real-time video images and the defect learning model; if the sewage pipe has defects, determining the defect type based on the real-time video images; determining the defect grade based on the defect type; and generating a defect response plan based on the defect type and grade. This technology only grades defects and cannot perform quantitative detection.

[0005] The paper "Research on LabVIEW-Based Pipeline Defect Detection Design" proposes a pipeline defect detection system based on LabVIEW as the host computer. This system uses VISA functions to configure connection ports and a program structure to construct a robot motion control window, thereby controlling the pipeline robot and collecting internal pipeline image information. This image of the pipeline interior is transmitted back to the host computer. LabVIEW software connects to MATLAB via a MATLAB Script node to preprocess the image using bilateral filtering and image enhancement algorithms, followed by image recognition using the Canny operator algorithm. This technology only identifies defects and cannot classify or quantify them.

[0006] The study "Research on Polyethylene Gas Pipeline Defect Detection Based on CNN and Image Processing" designed an image acquisition machine capable of operating on a DN110 pipeline. The overall control system is based on a Raspberry Pi. Through a designed expansion board, Mecanum wheels, and a reduced body, the entire image acquisition device is miniaturized and adaptable. It uses a selected CMOS industrial camera to capture images of PE gas pipeline defects, which are recorded in the machine via an SD card and then uploaded to a host computer for identification and analysis. Secondly, a convolutional neural network is used to classify pipeline defect images. Multiple sets of data enhancement are performed on the acquired images to increase the data volume. A PD-VGG model was established and trained to identify pipeline defects. After multiple experimental comparisons, the most suitable model parameters were selected and optimized, improving the training model's accuracy. This technology identifies and classifies defects, as well as quantifies them. However, its detection targets are large defects, making it difficult to detect small defects.

[0007] "In-pipe inspection robotic system for defect detection and identification using image processing" utilizes image processing (IP) and machine vision (MV) technologies to create a pipeline inspection robot to detect and identify various pipeline anomalies, including blockages, internal holes, cracks, and corrosion on the pipeline's inner surface. The robot's control is fully based on the Internet of Things (IoT), and real-time visual inspection is implemented for effective analysis. The image processing through simulation focuses on generating and displaying the output of various image processing transformations, such as image blurring, image smoothing, image conversion to binary, fine image formation, logarithmic transformation, and image contouring. This technology only images defects and does not perform qualitative or quantitative analysis of them.

[0008] The paper "Detection algorithm of defects on polyethylene gas pipe using image recognition" proposes a defect detection algorithm based on image recognition. First, the captured image is preliminarily screened to identify defective images. A gamma correction algorithm is then used to enhance the image, followed by a double filtering algorithm to eliminate noise interference with image recognition. A modified Sobel edge detection algorithm is then used to extract defect edges. An adaptive thresholding method is then used to segment the image defects, and an open operation is performed to effectively extract the defect outline. Finally, defect feature parameters are extracted and fed into an optimized support vector machine for defect recognition. This technique only images and classifies defects, not quantifies them.

[0009] At present, most pipeline defect detection technologies remain at the level of neural network recognition and classification, and there are few quantitative methods. To address this problem, this paper proposes a defect classification and quantitative detection method based on neural network image recognition and combined with a motion-induced eddy current detection probe. Summary of the Invention

[0010] With the development of detection requirements and process technology, although motional eddy current detection probes can detect metal surface defects in high-speed environments, their data processing methods still have certain problems. The present invention solves the problem that the current processing of motional eddy current signals can only be used for defect identification and classification, but cannot quantify the defect size.

[0011] The present invention performs electromagnetic imaging on the magnetic field data results output by the motional eddy current probe, and then uses image recognition to classify the types of surface defects of metal components and locate cracks based on the electromagnetic image, and then quantitatively identify the defect size by extracting the characteristic values ​​of the detection results corresponding to the defect position. In actual detection, defects in metal components appear randomly, and are mostly divided into cracks and corrosion holes. Different types of defects appear randomly, so it is necessary to identify cracks and corrosion holes. At the same time, for cracks and corrosion holes, since the spatial structures of the two are different, the defect size and quantitative relationship are also different. Different mapping relationships are needed to characterize cracks and corrosion holes. For this reason, this paper proposes a new quantitative detection method based on motional eddy currents, such as Figure 1 As shown. According to this method, the magnetic field data obtained by the motional eddy current probe is first collected. Then, electromagnetic imaging is used to convert the data into an image, and a two-dimensional image of the time and electromagnetic signal is obtained, so that the defects in the image can be identified in the subsequent process. If the identification result is a crack, the position of the corresponding defect curve is determined according to the identification position, and then the crack quantitative inversion mode is entered. The Δt1 and ΔB1 of the corresponding characteristic curve are extracted by the method described above. By comparing Δt1 with the crack width w d The crack width w is inverted by the quantitative relationship of d , and then according to the ΔB1 and crack depth d under this crack width d The crack depth d is inverted by the quantitative relationship d , and get the final crack size; if the identification result is a corrosion hole, then enter the corrosion hole quantitative inversion mode, extract the Δt2 and ΔB2 of the corresponding characteristic curve, and calculate the final crack size by comparing Δt2 with the corrosion hole diameter D p The quantitative relationship between the corrosion hole diameter D p , and then according to the ΔB2 under the corrosion hole diameter and the corrosion hole depth d p The crack depth d is inverted by the quantitative relationship p The final corrosion hole size is obtained. This method for classifying and quantitatively identifying multiple defects on the surface of metal components based on motional eddy currents can complete the characterization process including signal detection, electromagnetic imaging, defect data extraction, and defect size quantification.

[0012] Therefore, the technical solution of the present invention is a method for classifying and quantitatively identifying multiple defects on the surface of metal components based on motional eddy currents, comprising the following steps:

[0013] S1. Signal collection: Collecting magnetic field signal data returned by the array-type motional eddy current detection probe when detecting the surface of the metal component;

[0014] S2. Electromagnetic imaging: Use binary difference to visualize the detection results and obtain a two-dimensional image of the magnetic field intensity and time on the surface of the metal component;

[0015] S3. Image recognition: A convolutional neural network is used to classify and identify defects in the visualized electromagnetic image, classifying the defects into cracks and corrosion holes, and locating the defects based on the electromagnetic image.

[0016] S4. Extract characteristic signals: Extract the curve of magnetic field intensity B and time t before and after the defect location x mm, and extract the peak value B of the curve tB. a , and then extract the corresponding trough value B b , we get the peak-to-peak value ΔB=|B a -B b |; Extract curve tB from peak time t a , trough time t b , we get the peak-to-peak time difference Δt=|t a -t b |;x has little influence on the extraction results, so 10mm is selected.

[0017] S5. Calculate the crack width or corrosion hole diameter based on the defect type, peak-to-peak value ΔB, and peak-to-peak time difference Δt;

[0018] S51. Establish the relationship between Δt and crack width: set up m groups of n experiments according to actual needs, where the i-th experiment in the j-th group is to set the crack width on the metal component to be w di , i=1,2,…,n, j=1,2,…,m,w di Each group has the same correspondence and each group is different, the depth is d dj Defects, d dj The same within the group and different in different groups, measured Δt, ΔB; the depth of the collection of the jth group is d dj Defect data, with Δt as the horizontal axis, w d The width-Δt equation w is obtained by linear fitting of the vertical coordinate d =α1*Δt+β1, where α1 and β1 are constant coefficients determined by the motional eddy current detection probe, w d is the crack width;

[0019] S52. Establish the relationship between Δt and the diameter of the corrosion hole: set up m groups of n experiments, where the i-th experiment in the j-th group is to set the diameter of the corrosion hole D on the metal component. pi , D pi Each group has the same correspondence and each group is different, the depth is d pj Defects, d pj The same within the group and different in different groups, measure Δt, ΔB; first collect the depth of group j, all of which are d j Defect data, with Δt as the horizontal axis, D p The diameter-Δt equation D is obtained by linear fitting of the vertical coordinate p=α2*Δt+β2, where α2 and β2 are constant coefficients, determined by the motional eddy current detection probe, D p Indicates the diameter of the corrosion hole;

[0020] S53, extract the characteristic signal Δt of the unknown defect, and substitute Δt into the obtained width-Δt or diameter-Δt equation according to the defect type to obtain the crack width w d Or the diameter of the corrosion hole D p ;

[0021] S6. Calculate the crack depth or corrosion hole depth based on the defect type, peak-to-peak value ΔB, and peak-to-peak time difference Δt;

[0022] S61, collect the same width w of each experimental group in S51 di Defect data, with ΔB as the horizontal axis and depth d d The depth-ΔB equation d is obtained by linear fitting of the vertical axis d =α3*ΔB+β3, where α3 and β3 are constant coefficients, which are determined by the motional eddy current detection probe and width; when the probe geometry and electrical parameters remain unchanged, α3 and β3 are respectively d Fitting, get coefficient -w d Fitting equation α3=σ1*w d +λ1 and β3=σ2*w d +λ2, where the four coefficients are determined by the motional eddy current detection probe; σ1 and λ1 represent α3 and w d The coefficients of the linear relationship between σ2 and λ2 represent β3 and w d The coefficients of the linear relationship between them, because α3, β3 must be determined by w d To determine, different d The lower α3 and β3 are different.

[0023] S62, collect the same diameter D of each experimental group in S52 pi Defect data, with ΔB as the horizontal axis and depth d p The depth-ΔB equation d is obtained by linear fitting of the vertical axis p =α4*ΔB+β4, where α4 and β4 are constant coefficients, which are determined by the motional eddy current detection probe and width. When the probe remains unchanged, α4 and β4 are respectively p Fitting, get coefficient -D p Fitting equation α4=σ3*D p +λ3 and β4=σ4*D p +λ4, where the four coefficients are determined by the motional eddy current probe;

[0024] S63, extract the characteristic signal ΔB of the unknown defect, first calculate the known crack width w dOr the diameter of the corrosion hole is substituted into the coefficient equation to obtain the depth-ΔB equation, and then substituted ΔB into the depth-ΔB equation to obtain the depth d of the defect d or d p .

[0025] This patent uses the technical solutions of motional eddy current detection and neural network image recognition, which enables it to achieve excellent technical results in classifying and quantitatively detecting tiny defects in a high-speed environment. The sizes of tiny defects in the patent are: crack width 0.1-1mm, depth 18mm; corrosion hole diameter 18mm, corrosion hole depth 2-10mm. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Flowchart of defect quantitative detection method.

[0027] Figure 2 The defect distribution of the turntable; (a) is the model diagram; (b) is the actual diagram.

[0028] Figure 3 Schematic diagram of detection location distribution.

[0029] Figure 4 Schematic diagram of electromagnetic imaging results.

[0030] Figure 5 Schematic diagram of the model structure used in this article; (a) is the structural flow chart; (b) is the structural parameter diagram.

[0031] Figure 6 Schematic diagram of Δt and ΔB characteristic values. DETAILED DESCRIPTION

[0032] The present invention proposes a method for classifying and quantitatively identifying multiple defects on the surface of metal components based on motional eddy currents. This method has been implemented in the quantitative detection of multiple types of surface defects on high-speed rotating metal components. The specific implementation is as follows:

[0033] (1) Electromagnetic imaging

[0034] During the implementation process, in order to simulate the high-speed motion state of the kinematic eddy current test, a turntable specimen with target defects was designed. The specimen material was 45# steel and had the following defects:

[0035] ① The crack opening widths are 0.1mm, 0.2mm, 0.3mm, 0.5mm, 0.8mm, and 1.0mm, respectively, and the defect depths are 10%, 20%, 40%, 50%, and 80% of the specimen thickness;

[0036] ②The diameters of the corrosion holes are 1mm, 2mm, 3mm, 4mm, 5mm and 8mm respectively; the depths are 20%, 40%, 50%, 60% and 100% of the specimen thickness.

[0037] According to the above requirements, five turntables are designed with a diameter of 300mm and a thickness of 10mm. Six cracks of the same depth and six holes of the same depth are set on each turntable. The defect distribution on the five turntables is as follows: Figure 2 Taking the first turntable as an example, its geometric dimensions are shown in Table 1. The defect distribution of the following four turntables does not change, and the depths of cracks and corrosion holes increase in sequence, as shown in Table 2.

[0038] Table 1 Turntable specific dimensions

[0039]

[0040] Table 2 Changes in turntable defect depth

[0041] Turntable number Crack depth (mm) Corrosion hole depth (mm) 1 1 2 2 2 4 3 4 5 4 5 6 5 8 10

[0042] The turntable is mounted on a high-speed turntable and the defects on the turntable are detected using a motion-induced eddy current detection probe at the same linear speed of 5m / s. The detection position is as follows: Figure 3 shown.

[0043] After testing 10 detection positions in the experiment, the detection results were collected. Then, the contourf function in Matlab was used to perform electromagnetic imaging. Taking the No. 4 turntable as an example, the electromagnetic imaging results are as follows: Figure 4 shown.

[0044] (2) Convolutional neural network structure design for defect classification

[0045] The embodiment realizes the classification of cracks and corrosion holes through convolutional neural network, and its convolutional neural network model is built based on the VGG16 model. Its convolutional neural network structure is as follows: Figure 5 shown.

[0046] (3) Positive relationship between crack geometry information and electromagnetic characteristic signals

[0047] All dimensional defects in the five turntable specimens were detected, and the Δt and ΔB characteristic values ​​of all cracks were extracted according to the above, as shown in the following example: Figure 6 shown.

[0048] First, the Δt of the same crack depth and the ΔB of the same crack width are extracted and compared with the crack width w. d and crack depth d d The fitting results are shown in Table 3.

[0049] Table 3 Δt~w d and ΔB~d d Linear fitting formula

[0050] <![CDATA[d d (mm)]]> <![CDATA[Δt~w d Linear Fit Formula]]> Goodness of fit R-Square 8 <![CDATA[w d =-14.65*Δt+41.38]]> 0.95 5 <![CDATA[w d =-14.72*Δt+39.62]]> 0.96 4 <![CDATA[w d =-15.23*Δt+40.17]]> 0.95 2 <![CDATA[w d =-14.3*Δt+43.08]]> 0.97 1 <![CDATA[w d =-14.35*Δt+43.27]]> 0.96 <![CDATA[w d (mm)]]> <![CDATA[ΔB~d d Linear Fit Formula]]> Goodness of fit R-Square 1 <![CDATA[d d =10.89*ΔB+7.693]]> 0.96 0.8 <![CDATA[d d =7.683*ΔB+5.567]]> 0.97 0.5 <![CDATA[d d =6.543*ΔB+3.487]]> 0.97 0.3 <![CDATA[d d =5.113*ΔB+3.047]]> 0.99 0.2 <![CDATA[d d =3.703*ΔB+2.867]]> 0.98 0.1 <![CDATA[d d =2.673*ΔB+1.987]]> 0.94

[0051] Thus w d and ΔB~d d The results of fitting α3 and β3 in the linear fitting formula are shown in Table 4.

[0052] Table 4w d Fitting results with α3, β3

[0053] <![CDATA[d d ~ΔB formula coefficient]]> <![CDATA[With w d Linear fitting formula]]> R-Square <![CDATA[α3]]> <![CDATA[α3=8.208*w d +2.134]]> 0.97 <![CDATA[β3]]> <![CDATA[β3=5.83*w d +1.29]]> 0.96

[0054] (4) Positive relationship between corrosion hole geometry and electromagnetic characteristic signals

[0055] The same operation as that for cracks is performed on corrosion hole defects to obtain Tables 5 and 6.

[0056] Table 5D p ~Δt and d p ~ΔB linear fitting formula

[0057] <![CDATA[d p (mm)]]> <![CDATA[D p ~Δt linear fitting formula]]> Goodness of fit R-Square 10 <![CDATA[D p =2.397*Δt+16.37]]> 0.96 6 <![CDATA[D p =1.995*Δt+16.85]]> 0.96 5 <![CDATA[D p =2.486*Δt+12.14]]> 0.95 4 <![CDATA[D p =2.557*Δt+12.03]]> 0.99 2 <![CDATA[D p =2.351*Δt+12.49]]> 0.98 <![CDATA[D p (mm)]]> <![CDATA[d p ~ΔB linear fitting formula]]> Goodness of fit R-Square 8 <![CDATA[d p =11.59*ΔB+18.04]]> 0.96 5 <![CDATA[d p =3.739*ΔB+11.01]]> 0.98 4 <![CDATA[d p =2.881*ΔB+7.244]]> 0.96 3 <![CDATA[d p =1.523*ΔB+1.977]]> 0.97 2 <![CDATA[d p =1.112*ΔB+0.501]]> 0.98 1 <![CDATA[d p =0.679*ΔB+0.127]]> 0.97

[0058] Table 6D p Fitting results with α4, β4

[0059] <![CDATA[D p ~ΔB formula coefficient]]> <![CDATA[With D p Linear fitting formula]]> R-Square <![CDATA[α4]]> <![CDATA[α4=1.532*D p -2.252]]> 0.95 <![CDATA[β4]]> <![CDATA[β4=3.094*D p -5.987]]> 0.98

[0060] (5) Quantitative detection method for various surface defects of high-speed rotating metal components

[0061] Six cracks and six corrosion holes were randomly selected for inspection, and their defect information is shown in Table 7.

[0062] Table 7 Size and location of defects detected

[0063] Crack size (mm) 0.1×4 0.2×5 0.3×8 0.5×5 0.8×2 1.0×1 Distance detection starting point position (mm) 106 228 531 382 903 789 Corrosion hole size (mm) 1×5 2×6 3×10 4×6 5×4 8×2 Distance detection starting point position (mm) 925 540 255 213 545 103

[0064] 5.1 Obtaining electromagnetic images and classifying defects

[0065] The inspection speed was 5 m / s. After completing the inspection of 12 defects, the corresponding data was subjected to electromagnetic imaging and the images were input into a CNN convolutional neural network for image recognition. The recognition results are shown in Table 8. When the neural network judgment bit K is 0, the defect is judged to be a corrosion hole, and when the judgment bit K is 1, the defect is judged to be a crack. All 12 defects in the table were accurately identified. Because the input image for image recognition is pixel-stepping, the time of defect occurrence can be determined based on the input image number. The defect curve is then intercepted by intercepting the curve near the defect, and the magnetic induction intensity curve at the defect location is extracted based on the image sequence number.

[0066] Table 8 Defect type recognition results

[0067]

[0068] 5.2 Defect Location

[0069] At the current speed, using this imaging method, the number of pixels per 1 meter of probe movement is 3300. Therefore, multiplying the sequence number by 0.303 yields the millimeter distance of the defect location. This also provides the defect curve based on the pixels. The inversion accuracy of the defect location is shown in Table 9.

[0070] Table 9 Defect position inversion results

[0071]

[0072]

[0073] 5.3 Quantitative Defect Identification

[0074] After obtaining the magnetic induction intensity curves at the defects, the Δt and ΔB eigenvalues ​​of the 8th inspection position curve, which passes through the center of all corrosion holes, were extracted for the corrosion holes. Cracks were randomly selected from the 3rd to 8th inspection position curves, and the Δt and ΔB eigenvalues ​​were obtained based on the selected curves. Based on the quantitative method and quantitative relationship curves obtained above, the 12 defect sizes were inverted. The inversion results are shown in Tables 10 and 11.

[0075] Table 10 Crack size inversion results

[0076] Crack size (mm) 0.1×4 0.2×5 0.3×8 0.5×5 0.8×2 1.0×1 Δt(mm) 38.72 36.85 34.13 30.60 25.67 24.65 Inversion width (mm) 0.108 2.012 0.291 0.478 0.812 0.951 Width inversion error (mm) +0.008 +0.012 -0.009 -0.022 +0.012 -0.049 Width relative error (%) 8 6 3 4 1.5 5.4 ΔB (Gauss) 15.89 21.6 40.2 35.7 27.8 12.2 Inversion depth (mm) 4.14 5.37 8.8 5.03 2.18 0.89 Depth inversion error (mm) +0.14 +0.37 +0.8 +0.03 +0.18 -0.11 Depth relative error (%) 3.5 7.4 10 0.6 9 11

[0077] Table 11 Corrosion hole size inversion results

[0078] Corrosion hole size (mm) 1×5 2×6 3×10 4×6 5×4 8×2 Δt(mm) 16 19 21 25 27 36 Inversion diameter (mm) 1.013 2.053 2.913 4.089 4.825 8.452 Diameter inversion error (mm) +0.013 +0.053 -0.087 +0.089 -0.175 +0.452 Relative error of diameter (%) 1.3 2.6 2.9 2.2 3.5 5.6 ΔB (Gauss) 2 10 16 25 23 39 Inversion depth (mm) 4.589 5.467 10.761 6.387 4.368 2.175 Depth inversion error (mm) -0.411 -0.533 +0.761 +0.387 +0.368 +0.175 Depth relative error (%) 8.2 8.8 7.6 6.4 9.2 8.7

[0079] It can be found that the quantitative sizes of these 12 defects measured using the quantitative detection method based on permanent magnet motional eddy currents do not exceed 10% except for the relative error of 11% in the depth inversion of the crack with a width of 0.1 mm and a depth of 1 mm.

Claims

1. A method for classifying and quantitatively identifying multiple defects on the surface of metal components based on motional eddy currents, comprising the following steps: S1. Signal collection: Collecting magnetic field signal data returned by the array-type motional eddy current detection probe when detecting the surface of the metal component; S2. Electromagnetic imaging: Use binary difference to visualize the detection results and obtain a two-dimensional image of the magnetic field intensity and time on the surface of the metal component; S3. Image recognition: A convolutional neural network is used to classify and identify defects in the visualized electromagnetic image, classifying the defects into cracks and corrosion holes, and locating the defects based on the electromagnetic image. S4. Extract characteristic signals: Extract the curve of magnetic field intensity B and time t before and after the defect location x mm, and extract the peak value B of the curve tB. a , and then extract the corresponding trough value B b , we get the peak-to-peak value ΔB=|B a -B b |; Extract curve tB from peak time t a , trough time t b , we get the peak-to-peak time difference Δt=|t a -t b |; S5. Calculate the crack width or corrosion hole diameter based on the defect type, peak-to-peak value ΔB, and peak-to-peak time difference Δt; S51. Establish the relationship between Δt and crack width: set up m groups of n experiments according to actual needs, where the i-th experiment in the j-th group is to set the crack width on the metal component to be w di , i=1,2,…,n, j=1,2,…,m,w di Each group has the same correspondence and each group is different, the depth is d dj Defects, d dj The same within the group and different in different groups, measured Δt, ΔB; the depth of the collection of the jth group is d dj Defect data, with Δt as the horizontal axis, w d The width-Δt equation w is obtained by linear fitting of the vertical coordinate d =α1*Δt+β1, where α1 and β1 are constant coefficients determined by the motional eddy current detection probe, w d is the crack width; S52. Establish the relationship between Δt and the diameter of the corrosion hole: set up m groups of n experiments, where the i-th experiment in the j-th group is to set the diameter of the corrosion hole on the metal component to be D pi , D pi Each group has the same correspondence and each group is different, the depth is d pj Defects, d pj The same within the group and different in different groups, measure Δt, ΔB; first collect the depth of group j, all of which are d j Defect data, with Δt as the horizontal axis, D p The diameter-Δt equation D is obtained by linear fitting of the vertical coordinate p =α2*Δt+β2, where α2 and β2 are constant coefficients, determined by the motional eddy current detection probe, D p Indicates the diameter of the corrosion hole; S53, extract the characteristic signal Δt of the unknown defect, and substitute Δt into the obtained width-Δt or diameter-Δt equation according to the defect type to obtain the crack width w d Or the diameter of the corrosion hole D p ; S6. Calculate the crack depth or corrosion hole depth based on the defect type, peak-to-peak value ΔB, and peak-to-peak time difference Δt; Step S61: Collect the same width w of each experimental group in S51 di Defect data, with ΔB as the horizontal axis and depth d d The depth-ΔB equation d is obtained by linear fitting of the vertical axis d =α3*ΔB+β3, where α3 and β3 are constant coefficients, which are determined by the motional eddy current detection probe and width; when the probe geometry and electrical parameters remain unchanged, α3 and β3 are respectively d Fitting, get coefficient -w d Fitting equation α3=σ1*w d +λ1 and β3=σ2*w d +λ2, where the four coefficients are determined by the motional eddy current detection probe; σ1 and λ1 represent α3 and w d The coefficients of the linear relationship between σ2 and λ2 represent β3 and w d The coefficients of the linear relationship between them, because α3, β3 must be determined by w d To determine, different d Lower α3 and β3 are different; S62, collect the same diameter D of each experimental group in S52 pi Defect data, with ΔB as the horizontal axis and depth d p The depth-ΔB equation d is obtained by linear fitting of the vertical axis p =α4*ΔB+β4, where α4 and β4 are constant coefficients, which are determined by the motional eddy current detection probe and width. When the probe remains unchanged, α4 and β4 are respectively p Fitting, get coefficient -D p Fitting equation α4=σ3*D p +λ3 and β4=σ4*D p +λ4, where the four coefficients are determined by the motional eddy current probe; S63, extract the characteristic signal ΔB of the unknown defect, first calculate the known crack width w d Or the diameter of the corrosion hole is substituted into the coefficient equation to obtain the depth-ΔB equation, and then substituted ΔB into the depth-ΔB equation to obtain the depth d of the defect d or d p .

2. The method for classifying and quantitatively identifying multiple defects on the surface of a metal component based on motional eddy currents according to claim 1, characterized in that: In step 4, x is selected as 10 mm.

Images