Regularization-based acfm defect inversion intelligent judgment method

Abstract

The application discloses a kind of ACFM defect inversion intelligent judging methods based on regularization, it is related to nondestructive testing defect imaging technical field, including: with the help of high-resolution detection probe one-time scanning method obtains the magnetic field signal Bz amplitude matrix on aluminum material defect, based on Biot-Savart law obtains the magnetic field distribution characteristics of square current-carrying line frame at any point P in space to get transfer magnetic field bz amplitude, obtain each matrix in disturbance current matrix formula, add regularization parameter to realize ill-conditioned linear equation solution, further solve out disturbance current matrix, carry out two-dimensional fast inverse fourier transform to disturbance current matrix, take out matrix real part, inverse zero frequency translation, obtain Y direction current distribution and take gray scale diagram to obtain defect inversion disturbance current distribution diagram, intuitively show defect surface profile, realize defect intelligent judging by training RCNN model, it is favorable to realize the visual evaluation and remaining life accurate prediction of aluminum material crack.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing defect imaging technology, and in particular to a regularized ACFM defect inversion intelligent judgment method. Background Technology

[0002] During the operation of my country's aerospace vehicles, the complex and harsh environment of rain, snow, and sandstorms can easily cause damage to load-bearing components, vulnerable parts, and equipment surfaces. This leads to a decrease in the fracture toughness of materials, further accelerating the formation and propagation of cracks, seriously threatening the safe operation of aerospace vehicles. Alternating Current Field Measurement (ACFM) technology is an emerging electromagnetic non-destructive testing technology with features such as non-contact testing, no need to clean adhering substances, and quantitative assessment. It is widely used in aerospace vehicle inspection. It utilizes the uniform current induced by the detection probe on the surface of a conductive specimen. The current disturbs the space magnetic field around defects, causing distortion. Defects are detected and assessed by measuring the distorted magnetic field. When no defects are present, the current on the surface of the conductive specimen is uniform, and the space magnetic field is undisturbed.

[0003] Existing AC electromagnetic field detection technology relies on characteristic signals Bx and Bz, or a butterfly diagram composed of them, for judgment. Bx and Bz signals are magnetic field signals parallel to the specimen surface (parallel to the probe scanning direction) and perpendicular to the specimen surface, respectively. These characteristic signals can only detect the existence of cracks perpendicular to the induced current, and cannot achieve imaging display of defect morphology. At the same time, due to the complex working conditions at the detection site and the different experience of operators, it is easy to lead to misjudgment or missed detection of defects, making it difficult to achieve intelligent machine judgment.

[0004] Therefore, it is necessary to propose an intuitive method that can realize intelligent judgment of surface defects in aluminum materials. This method can not only present the surface contour of defects but also realize intelligent judgment of defects, providing accurate data support for defect assessment and life prediction of aluminum materials. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a regularized ACFM-based intelligent defect inversion method, which visually presents the surface contour of defects in aluminum materials and enables intelligent machine learning-based judgment, providing accurate and visual data support for defect assessment and life prediction of aluminum materials.

[0006] This invention provides a regularized ACFM defect inversion intelligent judgment method, including:

[0007] Step 1: A high-resolution AC electromagnetic field array detection probe with m sensors (m ≥ 7) and a sensor spacing of 0.5 mm - 3 mm is used. The magnetic field signal amplitude Bz at different locations on the same plane above the aluminum material defect is acquired through a single scan. The probe scanning direction is defined as the X direction. n location points are extracted along the X-direction scanning path. m sensor points are extracted in the direction perpendicular to the probe scanning direction. The magnetic field signal amplitudes Bz at these location points form an m x n magnetic field signal matrix. Performing a Fourier transform on matrix A yields the complex matrix of the magnetic field signal.

[0008] Step 2: Set the coordinates of any point P in space as (x, y, z), the vacuum permeability as μ0, the current as I, the number of observation grids in the X direction NGx as m, the number of observation grids in the Y direction NGy as n, the lift-off height of the observation point as z, and the side length of the square current-carrying wireframe as 2L. Based on the Biot-Savart law, determine the magnetic field distribution characteristics at any point P in space of the square current-carrying wireframe as the transfer magnetic field bz, and derive the formula:

[0009]

[0010] The amplitude of the transferred magnetic field bz forms an m-row n-column transfer matrix.

[0011] Step 3: Perform a Fourier transform on the transition matrix B to obtain an m-row, n-column complex transition matrix. Taking the conjugate of the transition complex matrix F1 yields an m-row, n-column conjugate transition complex matrix.

[0012] Step four, set a regularization parameter α that is greater than 0, according to Find the disturbance current matrix J; perform a two-dimensional fast inverse Fourier transform on the disturbance current matrix J to obtain a two-dimensional inverse Fourier transform matrix J1; extract the real part of the two-dimensional inverse Fourier transform matrix J1 to obtain the real part matrix J2; perform an inverse zero-frequency shift on the real part matrix J2 to obtain the inverse zero-frequency shift matrix J3.

[0013] Step 5, set i = 1:m, j = 1:n, according to The Y-direction current distribution matrix Jy is obtained by calculating the inverse zero-frequency translation matrix J3, and the grayscale image is obtained to obtain the defect inversion disturbance current distribution map M.

[0014] Step 6: Establish an image library of defect inversion disturbance current distribution maps M for different cracks. Mark the white areas as regions of interest by ROI and define the defect category as "liewen". Apply transfer learning to build an RCNN model and use the image library as the training database to train the model, thereby realizing RCNN intelligent defect determination.

[0015] This invention provides a regularized ACFM defect inversion intelligent judgment method. It uses a high-resolution detection probe to obtain the magnetic field signal amplitude matrix Bz above aluminum material defects via a single scan. A Fourier transform is performed on this matrix, and the magnetic field distribution characteristics at any point P in space using the Biot-Savart law are used to obtain the transferred magnetic field amplitude bz. Another Fourier transform is performed on this matrix, and its conjugate complex number is obtained to acquire the various matrices in the perturbation current matrix formula. Regularization parameters are added to solve the ill-conditioned linear equation, further resolving the perturbation current matrix. A two-dimensional fast Fourier inverse transform is performed on the perturbation current matrix, the real part of the matrix is ​​extracted, an inverse zero-frequency shift is performed, the current distribution in the Y direction is obtained, and a grayscale image is extracted to obtain the defect inversion perturbation current distribution map, which visually displays the defect surface contour. Intelligent defect judgment is achieved by training an RCNN model. Attached Figure Description

[0016] Figure 1 Flowchart of the ACFM defect inversion intelligent judgment method based on regularization provided by the present invention;

[0017] Figure 2 Photographs of surface cracks on aluminum test blocks provided in embodiments of the present invention;

[0018] Figure 3 This is a schematic diagram of the scanning path of the high-resolution AC electromagnetic field array detection probe provided in an embodiment of the present invention;

[0019] Figure 4 This is a color image of the magnetic field signal matrix A formed by scanning different locations using the magnetic field signal Bz amplitude at a single location using AC electromagnetic field detection technology, as provided in an embodiment of the present invention.

[0020] Figure 5 A schematic diagram illustrating the calculation of the magnetic field distribution characteristics at any point P in space using a square current-carrying wireframe provided in an embodiment of the present invention;

[0021] Figure 6 A color image formed by the transfer matrix B of the transfer magnetic field bz amplitude provided in the embodiments of the present invention;

[0022] Figure 7 This is a color image formed by processing the disturbance current matrix to obtain the inverse zero-frequency translation matrix J3, as provided in an embodiment of the present invention.

[0023] Figure 8 The defect surface contour visualization image presented by the defect inversion perturbation current distribution map M provided in the embodiments of the present invention;

[0024] Figure 9 The image formed by marking the white area of ​​the ROI as the region of interest and defining the defect category as "liewen" in the embodiment of the present invention;

[0025] Figure 10 The image formed based on the principle of RCNN model for transfer learning provided in this embodiment of the invention;

[0026] Figure 11 The image formed by training the RCNN network structure provided in the embodiments of the present invention;

[0027] Figure 12 This invention provides an embodiment of an image formed by intelligently determining defects using a trained RCNN model. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are all within the scope of protection of this invention.

[0029] In this embodiment of the invention, the method is applied to a regularized ACFM defect inversion intelligent judgment method. First, the amplitude matrix of the magnetic field signal Bz above the aluminum material defect is obtained by a single scan using a high-resolution array detection probe. A Fourier transform is performed on this matrix, and the magnetic field distribution characteristics at any point P in space of the square current-carrying wireframe are obtained based on the Biot-Savart law to obtain the amplitude of the transferred magnetic field bz. A Fourier transform is then performed on this matrix, and its conjugate complex number is obtained to acquire the matrices in the perturbation current matrix formula. Regularization parameters are added to solve the ill-conditioned linear equation, further resolving the perturbation current matrix. A two-dimensional fast Fourier inverse transform is performed on the perturbation current matrix, the real part of the matrix is ​​extracted, an inverse zero-frequency shift is performed, the current distribution in the Y direction is obtained, and a grayscale image is obtained to obtain the defect inversion perturbation current distribution map, which visually displays the defect surface contour. Intelligent defect judgment is achieved by training an RCNN model, which is beneficial for realizing the visual assessment of aluminum material cracks and accurate prediction of remaining life.

[0030] Example 1

[0031] Figure 1 The flowchart of the ACFM defect inversion intelligent judgment method based on regularization provided in the embodiments of the present invention includes:

[0032] S1. Prepare an aluminum test block. The surface of the test block has a rectangular crack with a length of 5.0 mm, a depth of 4.0 mm, and a width of 0.2 mm. Figure 2 As shown in the diagram. A schematic of the high-resolution detection probe using 64 sensors to scan the path simultaneously is shown in the diagram. Figure 3As shown, the probe scanning direction is defined as the X direction, and 64 sensors with a spacing of 1 mm are defined in the Y direction. A one-time scanning method is used to detect cracks on the test block, and the amplitude of the magnetic field signal Bz is obtained. The magnetic field signal matrix A is formed with m rows and n columns of 64. Some elements of the magnetic field signal matrix A are as follows:

[0033]

[0034] To visually represent the magnitude of the elements in the magnetic field preference matrix A, a planar color plot is drawn using the magnitudes of the magnetic field signals Bz at corresponding points on the horizontal (length) and vertical (width) axes, as shown below. Figure 4 As shown, it can be seen that the amplitude of the magnetic field signal Bz in the magnetic field signal matrix A is disturbed around the crack, showing positive and negative peaks, and the surface contour of the crack cannot be directly displayed.

[0035] According to the formula [Bz]=[conv([bz],[J])], the surface profile of the defect can be visualized by inverting the perturbation current distribution through the distorted magnetic field. The magnetic field signal matrix A formed by the amplitude of the magnetic field signal Bz is Fourier transformed to obtain the complex magnetic field signal matrix F with 64 rows and 64 columns.

[0036] To further calculate the disturbance current matrix J, after processing the magnetic field signal matrix A formed by the amplitude of the magnetic field signal Bz, it is necessary to calculate and process the transferred magnetic field bz, and proceed to step S2.

[0037] S2: Based on the Biot-Savart law, the magnetic field distribution characteristics of a square current-carrying wire frame at any point P(x, y, z) in space are derived, such as... Figure 5 As shown, based on the idea of ​​segmented calculation and superposition summation, the four segments of the square current-carrying wire frame are calculated in segments and then superimposed and summed to obtain the amplitude of the transfer magnetic field bz, and form a transfer matrix B with 64 rows m and 64 columns n.

[0038] To visually represent the magnitude of the elements in the transfer matrix B, a planar color plot is drawn using the amplitude of the transfer magnetic field bz at corresponding points on the horizontal (length) and vertical (width) axes, as shown below. Figure 6 As shown, the transferred magnetic field bz has a peak at the middle position, and the magnetic field strength is 0 at other positions.

[0039] To further calculate the disturbance current matrix J, and obtain the transfer matrix B composed of the transfer magnetic field bz, it is necessary to process it accordingly before proceeding to step S3.

[0040] S3: Performing a Fourier transform on the transition matrix B yields a complex transition matrix F1 with 64 rows and 64 columns (m and n). Taking the conjugate of the complex matrix F1 yields the conjugate complex transition matrix F2 with 64 rows and 64 columns (m and n).

[0041] After obtaining the matrix result, proceed to step S4 to further calculate the disturbance current matrix J.

[0042] S4. Based on the obtained complex matrix F of the magnetic field signal, the complex transition matrix F1, and the conjugate complex transition matrix F2, the formula can be derived. Because the denominator of this linear equation can contain a value of 0, resulting in no solution, it is an ill-conditioned linear equation. The method of solving ill-conditioned problems by adding conditions is called regularization. By setting a regularization parameter α that is greater than 0, the formula is derived. Find the disturbance current matrix J. Apply a two-dimensional inverse fast Fourier transform to the disturbance current matrix J to obtain a two-dimensional inverse Fourier transform matrix J1. Extract the real part of the two-dimensional inverse Fourier transform matrix J1 to obtain the real part matrix J2. Then perform a zero-frequency shift Fourier transform on the real part matrix J2 to rearrange it back into the inverse zero-frequency shift matrix J3. Plot the planar color graph O, as shown. Figure 7 As shown. Figure 7 The obtained two-dimensional diagram of the disturbance current shows a peak and a trough at both ends of the 0-degree crack, which is consistent with the distribution law of the current disturbance.

[0043] To clearly display the surface outline of the defect, proceed to step S5.

[0044] S5, set i = 1:64, j = 1:64, according to The Y-direction current distribution matrix Jy is obtained by calculating the current distribution in the inverse zero-frequency translation matrix J3. A grayscale image is then taken to obtain the defect inversion disturbance current distribution map M, as shown below. Figure 8 As shown. Figure 8 The white area represents a region without current, clearly showing the surface outline of the defect as a crack shape. The black areas at both ends of the crack represent regions of current accumulation, indicating high-density current areas. Figure 4 In comparison, this image clearly and intuitively shows the surface contour of the crack, improving the visualization of defect detection, increasing the accuracy of defect feature recognition, and reducing the false negative rate.

[0045] S6. Establish an image library of defect inversion perturbation current distribution maps M for different cracks, mark the white areas as regions of interest (ROIs) and define the defect category as "liewen". Figure 9 As shown. Figure 9 The area highlighted in yellow is the crack area, defined as "liewen".

[0046] An RCNN model is built using transfer learning, and transfer learning is performed based on the CIFAR10Net model, such as... Figure 10 As shown. Figure 10 This clearly demonstrates that the CIFAR10Net model is the source model, and the RCNN model is the target model to be created. It replicates the design and parameters of all models except the output model from the source model, training the output layer from scratch and fine-tuning the parameters of the remaining layers. The resulting RCNN network structure is shown below. Figure 11 As shown. Figure 11 The network structure is clearly shown to include one input layer, three convolutional layers, three pooling layers, two fully connected layers, and one output-like layer, categorized as "Background" and "liewen".

[0047] An image database is used as the training database to train a model, thereby enabling RCNN intelligent defect detection. The defect inversion disturbance current distribution map M of cracks in the image database is identified, such as... Figure 12 As shown. Figure 12 It can clearly present the surface contour of cracks and determine the defect type as cracks with a confidence level of 89.5%, realizing intelligent defect judgment. This makes it easier for operators to perform intelligent defect judgment in complex working conditions, reducing the workload of operators and ensuring the accuracy of defect judgment.

[0048] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A regularized ACFM defect inversion intelligent judgment method, characterized in that, Including step one: A high-resolution AC electromagnetic field array probe with m sensors (m ≥ 7) and a sensor spacing of 0.5 mm–3 mm is used to acquire the magnetic field signal amplitude Bz at different locations on the same plane above an aluminum material defect through a single scan. The probe scanning direction is defined as the X direction. n location points are extracted along the scanning path in the X direction, and m sensor points are extracted in the direction perpendicular to the probe scanning direction. The magnetic field signal amplitudes Bz at these location points form an m-row, n-column magnetic field signal matrix. Performing a Fourier transform on matrix A yields the complex matrix of the magnetic field signal. Step Two: Given an arbitrary point P in space with coordinates (x, y, z), a vacuum permeability of μ0, a current of I, and observation grid numbers of m in the X direction (NGx) and n in the Y direction (NGy), and an observation point lift-off height of z, and a square current-carrying wireframe with a side length of 2L, the magnetic field distribution characteristics at any point P in space are determined based on the Biot-Savart law, denoted as the transfer magnetic field bz. The formula is: The amplitude of the transferred magnetic field bz forms an m-row n-column transfer matrix. Step 3: Performing a Fourier transform on the transition matrix B yields an m-row, n-column complex transition matrix. Taking the conjugate of the transition complex matrix F1 yields an m-row, n-column conjugate transition complex matrix. Step Four: Set a regularization parameter α that is greater than 0, based on Find the disturbance current matrix J; perform a two-dimensional inverse fast Fourier transform on the disturbance current matrix J to obtain a two-dimensional inverse Fourier transform matrix J1; extract the real part of the two-dimensional inverse Fourier transform matrix J1 to obtain the real part matrix J2; perform an inverse zero-frequency shift on the real part matrix J2 to obtain the inverse zero-frequency shift matrix J3. Step 5: Let i = 1:m, j = 1:n, according to The Y-direction current distribution matrix Jy is obtained by calculating the current distribution in the Y direction from the inverse zero-frequency translation matrix J3. The grayscale image is then used to obtain the defect inversion disturbance current distribution map M. Step Six: An image library of defect inversion disturbance current distribution maps M for different cracks is established. White areas are marked as regions of interest by ROI and the defect category is defined as "liewen". Transfer learning is applied to build an RCNN model. The image library is used as the training database to train the model, thereby realizing RCNN intelligent defect determination.

Images