Three-dimensional Reconstruction Method for Surface Corrosion Defects of a Rotating Electromagnetic Field Structure

Abstract

The present invention provides a three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure. First, a magnetic field image sample library of surface corrosion defects of the rotating electromagnetic field structure is established through simulation and experiments. Then, a three-dimensional reconstruction network for surface corrosion defects of the structure is constructed. The network structure uses Pix2Pix, and the loss function combines a data loss function and a physical loss function. Furthermore, the established reconstruction network is trained to obtain the optimal solution. Finally, the trained network is deployed on a rotating electromagnetic field detection system. The present invention first provides an effective method for the three-dimensional reconstruction of surface corrosion defects of the structure. Moreover, since it is difficult to obtain a large number of samples of real surface corrosion defects of the structure and the production is expensive, the present invention realizes the accurate reconstruction of the three-dimensional contour of surface corrosion defects of the structure under small samples by means of a dual-drive method of physics and data, providing reliable data support for the safety assessment and maintenance decision-making of the structure.

Description

Technical Field

[0001] The present invention relates to the field of non-destructive testing signal processing, and in particular to a three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure Background Art

[0002] The rotating electromagnetic field is a new type of electromagnetic non-destructive testing technology developed on the basis of detection technologies such as alternating current electromagnetic fields and eddy currents. While having the advantages of non-contact detection and high quantization accuracy, it makes up for the shortcoming that the sensitivity of conventional electromagnetic non-destructive testing technologies has directionality, and can achieve high-sensitivity detection of defects in any direction of the structure

[0003] However, the three-dimensional profile of the surface corrosion defect of the structure and the spatial magnetic field response signal of the rotating electromagnetic field have a multiple non-linear mapping relationship. The conventional inverse problem solving equation is an ill-conditioned equation and there is no unique solution. The rapid development of artificial intelligence technology provides a new idea for the three-dimensional reconstruction of the surface corrosion defect of the structure. However, it relies on a large number of sample trainings. There are few industrial field samples of structural corrosion defects, and the cost of manual production is relatively high. In view of the above problems, the present invention proposes a new three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure. For the first time, the Pix2Pix network is applied to the three-dimensional reconstruction of structural corrosion defects of a rotating electromagnetic field. Then, according to the characteristics of the three-dimensional morphology of the corrosion defect, physical and data loss functions are constructed, and finally the three-dimensional reconstruction and accurate evaluation of the corrosion defect under small samples are realized Summary of the Invention

[0004] The problem to be solved by the present invention is to propose a three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure, to realize the accurate reconstruction of the three-dimensional profile of the structural corrosion defect under small samples, to provide effective data support for the safety assessment of the structure, and to effectively improve the intrinsic safety level of the equipment

[0005] The present invention provides a three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure, and its steps include

[0006] The first step: Establish a magnetic field image sample library for surface corrosion defects of a rotating electromagnetic field structure, and the sample library includes magnetic field images and real defect images

[0007] The second step: Establish a three-dimensional reconstruction network for surface corrosion defects of the structure; Since Pix2Pix can automatically capture the data distribution of the real sample set and obtain the mapping relationship between the input image and the output image, it has currently been widely applied to image restoration, handwritten digit generation, face generation, image inversion, etc. In this method, both magnetic field images and defect images can be obtained. Therefore, the main structure of the network adopts Pix2Pix; The input during network training is the magnetic field image, and the output is the generated defect image

[0008] The original loss function of the Pix2Pix network is not entirely suitable for reconstructing the three-dimensional contour of corrosion defects. Due to the limited number of defect samples, using the original loss function will result in a large error. To obtain a high-precision three-dimensional contour reconstruction result of structural corrosion defects under small samples, according to the characteristics of corrosion three-dimensional contour reconstruction, physical and data loss functions are established to constrain the optimization direction of the network. The total loss function of the network is:

[0009]

[0010] where λ1, λ2, and λ3 are weight coefficients. and are data loss functions. and are physical loss functions. The proportion of the data loss function or the physical loss function in the total loss function can be adjusted through the weight coefficients.

[0011] The main role is to establish the adversarial relationship of the network, and the cross-entropy loss function is adopted. The main role is to obtain the difference degree between the generated image and the real image, and the L1 loss function is adopted.

[0012] The expression of

[0013]

[0014] The expression of

[0015]

[0016] The physical loss function is constructed based on the evaluation indexes of defects. For corrosion-type defects, the main evaluation indexes are the volume and maximum depth of corrosion. Since the collected magnetic field images contain noise, it is easy to have pixel points with large deviations in the generated defect images. To reduce the maximum depth error, the average value of the largest 5 pixel points in the image is taken as the maximum depth of the defect, and the deviation between the generated defect depth and the actual defect depth is calculated and corrected by adding weight coefficients to construct the loss function. For the volume of the defect, only the deviation between the real corrosion and the generated corrosion is calculated in the basic loss function. Since the values in the non-defect areas are all 0, the pixel values with large deviations are generally located in the corrosion area. To obtain a more accurate defect volume, the weights of the 15 pixel points with the largest deviation are increased to form the loss function. The finally constructed loss function for corrosion-type defects is:

[0017] The construction process of

[0018] Both the real defect image (M*M) and the generated defect image (M*M) are converted into arrays with 1 column and M*M rows, and the two arrays are sorted in descending order respectively;

[0019] Take the first 5 rows of the two arrays after rearrangement respectively to form new arrays A(1*5) and B(1*5), The calculation process of is as follows:

[0020]

[0021] In the formula, A i represents the i-th element of array A, B i represents the i-th element of array B, and |·| represents taking the absolute value;

[0022] The construction process of is as follows:

[0023] The deviation matrix V between the real defect image and the generated defect image can be expressed as:

[0024] V = y - G(x, z)

[0025] In the formula, y ∈ R M×M is the real defect image, G(x, z) ∈ R M×M is the generated defect image, x is the magnetic field image, and z is the noise signal; since the deviation matrix V is not all positive, its absolute value needs to be taken to obtain the absolute deviation matrix V':

[0026] V' = abs(V i,j ), i = 1, 2,...M, j = 1, 2, M

[0027] Convert the absolute deviation matrix into an array with 1 column and M*M rows, and sort it in descending order, and select the first 15 rows to form a new array C ∈ R 1×15 , The calculation process is as follows:

[0028]

[0029] In the formula, C i represents the i-th row of array C;

[0030] Step 3: Use the established sample library of magnetic field images of surface corrosion defects of the rotating electromagnetic field structure to train the established three-dimensional reconstruction network of surface corrosion defects of the structure, and obtain the optimal solution;

[0031] Step 4: Deploy the trained network to the rotating alternating current electromagnetic field detection system.

[0032] Furthermore, the magnetic field image is constructed by the magnetic flux density perpendicular to the surface of the structure to be measured. Since there are few non-ferromagnetic corrosion samples in the industrial field and it is expensive to manufacture artificially, the magnetic flux density is measured by a rotating electromagnetic field simulation model or an experimental system.

[0033] Furthermore, the real defect image is constructed by the three-dimensional features of the real defect, and the length and width of the real defect image are the same.

[0034] Furthermore, M is the number of pixels in the length or width of the image.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] (1) The present invention first applies the Pix2Pix network to the three-dimensional reconstruction of structural corrosion defects, providing an effective intelligent solution for defect inversion and making up for the deficiency that the traditional physical equation solving method has no unique solution.

[0037] (2) Aiming at the problem that there are few samples of structural surface corrosion defects in the industrial field and the production is expensive, resulting in low accuracy of reconstructing the three-dimensional contour of corrosion by neural network, a physical loss function is added on the basis of the conventional data loss function, improving the accuracy of reconstructing the three-dimensional contour of corrosion.

[0038] (3) The physical loss function is constructed based on the evaluation indexes of defects. For corrosion-type defects, the main evaluation indexes are the volume and maximum depth of corrosion; therefore, the average value of the largest 5 pixel points in the image is taken as the maximum depth of the defect, and the deviation between the generated defect depth and the actual defect depth is calculated to form the loss function; for the volume of the defect, only the deviation between the real corrosion and the generated corrosion is calculated in the basic loss function. Since the values in the non-defect area are all 0, the pixel values with larger deviations are generally located in the corrosion area; in order to obtain a more accurate defect volume, the weights of the 15 pixel points with the largest deviations are increased to form the loss function; through the construction of the loss function, the inversion of the maximum depth of the corrosion defect is improved, and through the construction of the loss function, the inversion of the corrosion volume is improved, and both further improve the accuracy of the three-dimensional reconstruction of the corrosion defect. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is the flowchart of the three-dimensional reconstruction of the structural corrosion defect in the embodiment of the present application

[0040] Figure 2 It is the magnetic field image in the embodiment of the present application

[0041] Figure 3 It is the defect image in the embodiment of the present application

[0042] Figure 4 For the three - dimensional reconstruction network of the structural surface corrosion defect in the embodiment of the present application Detailed implementation manners

[0043] To make the objectives, technical solutions and advantages of the present invention clearer, the present 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 the present invention, rather than all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of the present invention.

[0044] The present invention provides a three - dimensional reconstruction method for the structural surface corrosion defect of a rotating electromagnetic field, and its steps include:

[0045] In the first step, a magnetic field image sample library of the structural surface corrosion defect of the rotating electromagnetic field is established. The sample library includes magnetic field images and real defect images, such as Figure 2 and 3 as shown;

[0046] In the second step, a three - dimensional reconstruction network for the structural surface corrosion defect is established. The main structure of the network adopts Pix2Pix. When the network is trained, the input is the magnetic field image, and the output is the generated defect image, such as Figure 4 as shown, and the number of pixels in length and width is 100 respectively;

[0047] The loss function is:

[0048]

[0049] where λ1, λ2, and λ3 are weight coefficients, which are set to 1, 0.5, and 1.5 respectively;

[0050] The expression of

[0051]

[0052] The expression of

[0053]

[0054] The construction process of

[0055] Both the real defect image (100 * 100) and the generated defect image (100 * 100) are converted into an array with 1 column and 10000 rows, and the two arrays are respectively arranged in descending order;

[0056] Take the first 5 rows of the two rearranged arrays respectively to form new arrays A(1*5) and B(1*5). The calculation process of

[0057]

[0058] In the formula, A i represents the i-th element of array A, and B i represents the i-th element of array B, and |·| represents taking the absolute value;

[0059] The construction process of

[0060] The deviation matrix V between the real defect image and the generated defect image can be expressed as:

[0061] V = y - G(x, z)

[0062] In the formula, y ∈ R M×M is the real defect image, G(x, z) ∈ R M×M is the generated defect image, x is the magnetic field image, and z is the noise signal; since the deviation matrix V is not all positive, its absolute value needs to be taken to obtain the absolute deviation matrix V':

[0063] V' = abs(V i,j ), i = 1, 2,...M, j = 1, 2,...M

[0064] Convert the absolute deviation matrix into an array with 1 column and 100*100 rows, and arrange it in descending order, and select the first 15 rows to form a new array C ∈ R 1×15 , The calculation process is:

[0065]

[0066] In the formula, C i represents the i-th row of array C;

[0067] Step 3: Use the three-dimensional reconstruction network of the surface corrosion defect magnetic field image of the established rotating electromagnetic field structure to train and obtain the optimal solution;

[0068] Step 4: Deploy the trained network to the rotating alternating current electromagnetic field detection system.

[0069] The above specific embodiments have further elaborated in detail the purpose, technical solution and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure, characterized in that: S1: Establish a magnetic field image sample library for surface corrosion defects of the rotating electromagnetic field structure, where the sample library includes magnetic field images and real defect images; S2: Establish a three-dimensional reconstruction network for surface corrosion defects of the structure. The main structure of the network uses Pix2Pix. When the network is trained, the input is the magnetic field image in S1, and the output is the generated defect image; The loss function is: where λ1, λ2, and λ3 are weight coefficients; The expression is: The expression is: The construction process is as follows: Both the real defect image and the generated defect image are converted into an array with 1 column and M*M rows, where the sizes of the real defect image and the generated defect image are both M*M, and the two arrays are sorted in descending order respectively; Take the first 5 rows of the two rearranged arrays respectively to form new arrays A(1*5) and B(1*5). The calculation process of Where, A i represents the i-th element of array A, B i represents the i-th element of array B, and |·| represents taking the absolute value; The construction process is as follows: The deviation matrix V between the real defect image and the generated defect image can be expressed as: V = y - G(x, z) where \(y\in R\) M×M is the true defect image, \(G(x,z)\in R\) M×M is the generated defect image, \(x\) is the magnetic field image, and \(z\) is the noise signal; since the deviation matrix \(V\) is not all positive, its absolute value needs to be taken to obtain the absolute deviation matrix \(V'\): V′ = abs(V i,j ), i = 1, 2,... M, j = 1, 2,... M Convert the absolute deviation matrix into an array with 1 column and M * M rows, and arrange it in descending order. Select the first 15 rows to form a new array C ∈ R 1×15 , The calculation process is as follows: where C i represents the i-th row of the array C; S3: Use the magnetic field image sample library for surface corrosion defects of the rotating electromagnetic field structure established in S1 to train the three-dimensional reconstruction network for surface corrosion defects of the structure established in S2, and obtain the optimal solution; S4: Deploy the trained network to the rotating alternating magnetic field detection system.

2. The three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure according to claim 1, characterized in that, The magnetic field image is constructed by the magnetic flux density perpendicular to the surface of the structure to be measured, and the magnetic flux density is measured by a rotating electromagnetic field simulation model or an experimental system.

3. A three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure according to claim 1, characterized in that, The real defect image is constructed by the real three-dimensional characteristics of the defect, and the length and width of the real defect image are the same.

4. A three-dimensional reconstruction method for surface corrosion defects of a rotating electromagnetic field structure according to claim 1, characterized in that, The M is the number of pixels in the length or width of the image.

Images