A device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof

Abstract

The present invention belongs to the technical field of nondestructive testing of aluminum alloy structures, and in particular, relates to a device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof. The detection device can induce a direction-controllable uniform current field on the surface of the aluminum alloy structure to be tested; the three-dimensional characterization method realizes the three-dimensional inversion and evaluation of the complex crack morphology of the aluminum alloy structure to be tested, and accurately quantifies the crack size and angle. The present invention provides a three-dimensional characterization method, comprising the following steps: designing the structure of a generative adversarial neural network; establishing a paired database of bifurcation cracks on the surface of an aluminum alloy structure; constructing a loss function for bifurcation cracks on the surface of an aluminum alloy structure; training pix2pix to obtain a generative adversarial neural network; deploying and applying the trained generative adversarial neural network, and the trained generative adversarial neural network can be used to realize the three-dimensional characterization of bifurcation cracks on the surface of an aluminum alloy structure.

Description

Technical Field

[0001] The present invention belongs to the technical field of nondestructive testing of aluminum alloy structures, and in particular relates to a device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof. Background Art

[0002] Aluminum alloys are widely used in shipbuilding, chemical industry, aerospace, transportation and other technical fields due to their low density, good mechanical properties, excellent heat transfer performance and many other characteristics. However, after long-term operation under high load conditions, aluminum alloy structures will be affected by complex loads and harsh working environments, and their structural surfaces are prone to structural damage such as cracks, which may extend at any angle. Among them, short cracks will extend along the depth and surface direction of the structure, and then form surface cracks; surface cracks usually extend along a straight line, and then tend to branch under the action of alternating loads and extend to key parts inside the structure.

[0003] It is worth noting that for large-scale structural equipment, the existing structural overall health monitoring methods often have difficulty in timely detecting tiny local cracks and quantifying and evaluating the crack size. Especially for bifurcation cracks at different angles, the existing non-destructive testing technology (NDT) cannot guarantee high-sensitivity detection of bifurcation cracks at various angles, and it is very easy to miss detection.

[0004] In response to the above problems, technicians have tried to use an emerging non-destructive testing method, namely alternating current field testing technology (ACFM). The ACFM technology has a series of advantages such as non-contact detection, high detection efficiency, and not easily affected by coatings. It specifically obtains crack information by inducing a uniform current field and directly measuring the change in the magnetic field, so that the crack size and endpoint position can be directly obtained. Then, after further research, it was found that in the alternating current field detection technology, the unidirectional induced current used only shows a high sensitivity to cracks perpendicular to the current direction; while for the distorted magnetic field in the bifurcation area, the sensitivity of its response signal will be greatly reduced, so it is impossible to effectively detect the bifurcation area of ​​multi-angle cracks, and it is difficult to accurately obtain the extension direction of the crack. In addition, the inventors also found that the judgment of the detection signal in the prior art mostly relies on the operator's experience and knowledge, which may lead to misjudgment by the competent department, and thus cannot directly and objectively reflect the specific shape and appearance of the crack. Therefore, it is urgent for those skilled in the art to provide a technical solution that can adjust the direction of the induced current according to cracks at different angles, so as to effectively obtain crack information and achieve accurate characterization of cracks. Summary of the invention

[0005] The present invention provides a device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof. The detection device can induce a direction-controllable uniform current field on the surface of the aluminum alloy structure to be measured. With its technical support, the three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure realizes the three-dimensional inversion and evaluation of the complex crack morphology of the aluminum alloy structure to be measured, and accurately quantifies the crack size and angle.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A device for detecting bifurcation cracks on the surface of an aluminum alloy structure, comprising: a detection signal generating unit, a detection probe, and a detection signal collecting unit, wherein the detection probe is composed of an excitation coil unit and a magnetic field sensor;

[0008] The excitation coil unit is composed of a first excitation coil group placed above the surface of the aluminum alloy structure to be measured and a second excitation coil group placed above the surface of the aluminum alloy structure to be measured; wherein the first excitation coil group and the second excitation coil group are orthogonally distributed, and the first excitation coil group is composed of two first excitation coils with opposite induced current directions, and the second excitation coil group is composed of two second excitation coils with opposite induced current directions;

[0009] The magnetic field sensor is arranged at the surface of the aluminum alloy structure to be measured at the orthogonal position of the first excitation coil group and the second excitation coil group, and is used to collect the induced magnetic field signal perpendicular to the surface direction of the aluminum alloy structure to be measured.

[0010] Preferably, two sinusoidal signals with the same frequency, phase and amplitude ratio of K are respectively input into the first excitation coil group and the second excitation coil group, then the X-direction induced current Jx and the Y-direction induced current Jy induced on the surface of the aluminum alloy structure under test respectively satisfy:

[0011] Formula (1);

[0012] Formula (2);

[0013] The superimposed current Js obtained by vector addition of the induced current Jx in the X direction and the induced current Jy in the Y direction satisfies:

[0014] Formula (3);

[0015] Among them, the angle between the superimposed current Js and the horizontal direction ,satisfy:

[0016] Formula (4).

[0017] Preferably, the superposition current Js is normalized;

[0018] Among them, the normalized superposition current Js satisfies:

[0019] Formula (5).

[0020] On the other hand, a three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure comprises the following steps:

[0021] Step 1: Design the structure of the generative adversarial neural network;

[0022] Step 2: Establish a paired database of bifurcation cracks on the surface of aluminum alloy structures;

[0023] Step 3: Construct the loss function of bifurcation crack on the surface of aluminum alloy structure;

[0024] Step 4: Train pix2pix to obtain a generative adversarial neural network;

[0025] Step 5: Deploy and apply the trained generative adversarial neural network. The trained generative adversarial neural network can be used to realize the three-dimensional characterization of bifurcation cracks on the surface of aluminum alloy structures.

[0026] Preferably, the step 1 is specifically described as:

[0027] Design the layer structure, connection method, number of convolution kernels, convolution kernel size, and step size of the generator network in the generative adversarial neural network;

[0028] Design the layer structure and connection method of the discriminator network in the generative adversarial neural network.

[0029] Preferably, the step 2 is specifically described as:

[0030] The real image of the bifurcation crack on the surface of the aluminum alloy structure is collected, and after standardization and grayscale processing, the sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is obtained;

[0031] Extract the magnetic field intensity signal perpendicular to the surface of the aluminum alloy structure to obtain the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure, and after standardization, normalization and grayscale processing, obtain the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure;

[0032] The sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is feature-paired with the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure to establish a paired database of the bifurcation crack on the surface of the aluminum alloy structure.

[0033] Preferably, the step three is specifically described as:

[0034] The loss function of the bifurcation crack on the surface of the aluminum alloy structure is constructed to meet the following requirements:

[0035] g_loss=g_bce_loss+g_l1_loss;

[0036] Among them, g_loss is the total loss value of the generative adversarial neural network; g_bce_loss is the binary cross entropy loss function, and g_l1_loss is the L1 loss function;

[0037] The binary cross entropy loss function satisfies:

[0038] g_bce_loss=bce_loss(g_fake_predict,paddle.ones_like(g_fake_predict));

[0039] L1 loss function satisfies:

[0040] ;

[0041] Among them, g_fake_predict is the discriminant probability output by the discriminator network, fake_B and real_B are the target image and the target real image generated by the generator network respectively, and a is a constant.

[0042] The present invention provides a device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof. The device for detecting bifurcation cracks on the surface of an aluminum alloy structure includes a detection signal generating unit, a detection probe, and a detection signal acquisition unit. The detection probe is composed of an excitation coil unit and a magnetic field sensor; the excitation coil unit is composed of a first excitation coil group placed above the surface of the aluminum alloy structure to be measured and a second excitation coil group placed above the surface of the aluminum alloy structure to be measured, and the magnetic field sensor is arranged on the surface of the aluminum alloy structure to be measured at an orthogonal position between the first excitation coil group and the second excitation coil group. The characterization method is based on and includes the following steps: designing the structure of a generative adversarial neural network; establishing a paired database of bifurcation cracks on the surface of an aluminum alloy structure; constructing a loss function for bifurcation cracks on the surface of an aluminum alloy structure; training pix2pix to obtain a generative adversarial neural network; and deploying and applying the trained generative adversarial neural network.

[0043] The aluminum alloy structure surface bifurcation crack detection device and the three-dimensional characterization method thereof having the above-mentioned structural features and step features have at least the following technical advantages compared with the prior art:

[0044] 1. A uniform current field with controllable direction can be induced on the surface of the aluminum alloy structure under test; and the change and adjustment of the induced current angle can be achieved through normalization processing, which is helpful to obtain the maximum distortion of the crack (at any angle);

[0045] 2. It solves the drawbacks of existing detection technologies (such as rotating AC electromagnetic field detection methods) that the energy is not focused and the energy is wasted;

[0046] 3. It has a high detection sensitivity for bifurcated cracks in any direction. With a small number of training sets, it can realize three-dimensional inversion and evaluation of complex crack morphology, and achieve accurate quantification of crack size and angle. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0048] Figure 1 A schematic diagram of the structure of the device for detecting bifurcation cracks on the surface of an aluminum alloy structure provided by the present invention;

[0049] Figure 2 It is a schematic diagram of the combination of the first excitation coil group and the second excitation coil group in the excitation coil unit;

[0050] Figure 3a The amplitude ratio K is Schematic diagram of the direction of superimposed current when ;

[0051] Figure 3b It is a schematic diagram of the direction of the superimposed current when the amplitude ratio K is 1;

[0052] Figure 3c The amplitude ratio K is Schematic diagram of the direction of superimposed current when ;

[0053] Figure 3d It is a schematic diagram of the superimposed current trend when the amplitude ratio K is 0;

[0054] Figure 4a is a current schematic diagram of the superimposed current Js without normalization processing;

[0055] Figure 4b is a current schematic diagram of the superimposed current Js after normalization;

[0056] Figure 5a It is a schematic diagram of the magnetic field image when the amplitude ratio K is 0;

[0057] Figure 5b The amplitude ratio K is Schematic diagram of magnetic field image when ;

[0058] Figure 5c Schematic diagram of magnetic field image when amplitude ratio K is 1;

[0059] Figure 5d When the amplitude ratio K is Schematic diagram of magnetic field image;

[0060] Figure 5e Schematic diagram of magnetic field image when amplitude ratio K is 5;

[0061] Figure 6 For Figure 5a-5e Schematic diagram of superimposed magnetic field image in the Bz direction of bifurcated crack on the surface of aluminum alloy structure after contrast superposition;

[0062] Figure 7 Schematic diagram of the flow of the three-dimensional characterization method for bifurcated cracks on the surface of aluminum alloy structure provided by the present invention;

[0063] Figure 8 Schematic diagram of the flow framework for standardizing and grayscaling the real image of bifurcated cracks on the surface of aluminum alloy structure, and standardizing, normalizing and grayscaling the magnetic field image of bifurcated cracks on the surface of aluminum alloy structure;

[0064] Fig. 9 Schematic diagram of the three-dimensional characterization result of bifurcated cracks obtained based on the three-dimensional characterization method for bifurcated cracks on the surface of aluminum alloy structure;

[0065] Fig.10a Schematic diagram of the crack angle error result obtained by randomly selecting 30 samples from the test set;

[0066] Fig.10b Schematic diagram of the crack length error result obtained by randomly selecting 30 samples from the test set;

[0067] Fig.10c Schematic diagram of the crack length error result obtained by randomly selecting 20 superimposed images from the test set. Detailed implementation manner

[0068] The present invention provides a detection device for bifurcated cracks on the surface of aluminum alloy structure and its three-dimensional characterization method. The detection device can induce a uniformly controllable current field on the surface of the aluminum alloy structure to be measured. With its technical support, the three-dimensional characterization method for bifurcated cracks on the surface of aluminum alloy structure realizes the three-dimensional inversion and evaluation of the complex crack morphology of the aluminum alloy structure to be measured, and accurately quantifies the crack size and angle.

[0069] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0070] The present invention provides a device for detecting bifurcation cracks on the surface of an aluminum alloy structure. Figure 1 As shown, it includes: a detection signal generating unit, a detection probe, and a detection signal collecting unit. Further, the detection probe is composed of an excitation coil unit and a magnetic field sensor. Figure 2 As shown, the excitation coil unit is composed of a first excitation coil group placed above the surface of the aluminum alloy structure to be measured and a second excitation coil group placed above the surface of the aluminum alloy structure to be measured. It is worth noting that the first excitation coil group and the second excitation coil group are orthogonally distributed; the first excitation coil group and the second excitation coil group are orthogonally distributed, and the first excitation coil group is composed of two first excitation coils with opposite induced current directions, and the second excitation coil group is composed of two second excitation coils with opposite induced current directions.

[0071] As a preferred embodiment of the present invention, two sinusoidal (excitation) signals with the same frequency, phase and amplitude ratio of K are respectively input into the first excitation coil group and the second excitation coil group, and the X-direction induced current Jx and the Y-direction induced current Jy induced on the surface of the aluminum alloy structure under test respectively satisfy:

[0072] Formula (1);

[0073] Formula (2);

[0074] The superimposed current Js obtained by vector addition of the induced current Jx in the X direction and the induced current Jy in the Y direction satisfies:

[0075] Formula (3);

[0076] Among them, the angle between the superimposed current Js and the horizontal direction ,satisfy:

[0077] Formula (4).

[0078] The purpose of this arrangement is to induce induced currents in the X direction and the Y direction on the surface of the aluminum alloy structure under test by loading sinusoidal (excitation) signals to the first excitation coil group and the second excitation coil group respectively. And further, by adding the induced current vectors, a superimposed current Js with a direction varying with the amplitude ratio K is induced on the surface of the aluminum alloy structure under test. For example, the amplitude ratio K is selected to be , 0, we can get the corresponding different superimposed current directions (the incident angle of the superimposed current satisfies: ), specific reference is as follows Figure 3a-3d As shown. It is worth noting that according to the different directions of the bifurcation cracks on the surface of the aluminum alloy structure, the amplitude ratio K of the two sinusoidal (excitation) signals can be adjusted to ensure that the superimposed current Js is as perpendicular to the direction of the bifurcation crack as possible, thereby obtaining the maximum detection sensitivity. Compared with the existing rotating AC electromagnetic field detection technology (the excitation current synthesized by the existing rotating AC electromagnetic field detection technology rotates continuously, and the induced current in the specified direction cannot be obtained; and the response signal is a synthetic amount of continuous rotation, and the current distortion in a certain direction cannot be obtained, so it is impossible to accurately realize high-sensitivity detection of defects in a certain direction, and it is easy to miss the detection of some features of complex cracks); the above-mentioned technical means can ensure the controllable adjustment of the current in any direction, and the superimposed current at each moment is in a single direction, so that the morphology of complex structural defects can be effectively reconstructed.

[0079] In addition, as a preferred embodiment of the present invention, the superimposed current Js may be further normalized. The superimposed current Js after normalization satisfies:

[0080] Formula (5).

[0081] Specific comparison Figure 4a , Figure 4b ( Figure 4a is a current schematic diagram of the superimposed current Js without normalization. Figure 4b (Figure 3) is the current schematic diagram of the superimposed current Js after normalization processing. It can be found that through normalization processing, superimposed currents of the same magnitude can be synthesized, thereby highlighting the influence of current direction on the magnetic field distortion at the crack endpoint and eliminating the adverse effects of uneven current amplitude on the detection results.

[0082] To further facilitate the understanding of the present invention by those skilled in the art, the following specific implementation process is provided as an explanation. First, the detection signal generating unit is selected to generate two sinusoidal signals with a frequency of 1kHz, an initial phase of 0°, and amplitudes of 5V and 5×KV respectively; it is loaded into the first excitation coil group and the second excitation coil group of the excitation coil unit; by changing the value of K, the direction of the (inductive) superimposed current is changed. In the process of detecting bifurcated cracks on the surface of the aluminum alloy structure, the superimposed current direction is made as perpendicular to the crack as possible; the superimposed current will be deflected when passing through the crack endpoint, thereby generating a secondary magnetic field disturbance phenomenon. Then, the magnetic field sensor picks up the magnetic field signal perpendicular to the surface direction of the aluminum alloy structure being measured. Further, the filter amplifier is used to amplify and filter out the interference signal; finally, the detection signal acquisition unit converts the electrical signal into a digital signal for processing by a computer, and finally draws a magnetic field image.

[0083] It should be noted that the specific dimensions of the bifurcation cracks on the surface of the aluminum alloy structure can be referred to as follows: the transverse crack length is 20mm, the oblique crack length is 10mm, the width is 0.5mm, the depth is 4mm, and the bifurcation angle is 45°. The magnetic field image in the Bz direction can be drawn as follows: Figure 5a-5e ,in, Figure 5a It is a schematic diagram of the magnetic field image when the amplitude ratio K is 0; Figure 5b The amplitude ratio K is Schematic diagram of magnetic field image when ; Figure 5c It is a schematic diagram of the magnetic field image when the amplitude ratio K is 1; Figure 5d The amplitude ratio K is Schematic diagram of magnetic field image when ; Figure 5e is a schematic diagram of the magnetic field image when the amplitude ratio K is 5. Figure 5a-5e After comparing and superimposing the magnetic field images shown in FIG. 1 , the superimposed magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure in the Bz direction is obtained. Figure 6 shown.

[0084] On the other hand, the present invention also provides a three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure, such as Figure 7 As shown, the following steps are included:

[0085] Step 1: Design the structure of the generative adversarial neural network.

[0086] Specifically, the step 1 is described as follows:

[0087] Design the layer structure, connection method, number of convolution kernels, convolution kernel size, and step size of the generator network in the generative adversarial neural network;

[0088] Design the layer structure and connection method of the discriminator network in the generative adversarial neural network.

[0089] It is worth noting that, as an optional implementation, the generator network can be composed of 8 downsampling layers and 8 upsampling layers. In order to extract more levels of features in the magnetic field image in the Bz direction, the number of convolution kernels in the convolution layer is set to 64, the convolution kernel size is set to 4×4, and the step size is 1, so that the features of the local area can be effectively captured, ensuring that the convolution operation does not excessively reduce the size of the Bz image and retaining more spatial information. Jump connections are used between the downsampling layer and the upsampling layer in order to reintroduce the spatial information lost in the downsampling process into the upsampling process, so as to more accurately characterize the three-dimensional contour of the crack. During the upsampling process, U-Net will splice the output of the upsampling layer with the output feature map of the corresponding downsampling layer, so the number of input channels of the upsampling part is usually set to twice the number of output channels of the downsampling part.

[0090] The discriminator network uses the PatchGAN structure, which consists of five layers, including two convolutional layers and three custom convolutional blocks. Through a series of convolution and activation operations, it extracts the multi-level features of the image and finally outputs a probability value to indicate the authenticity of the input image.

[0091] Step 2: Establish a paired database of bifurcation cracks on the surface of aluminum alloy structures.

[0092] After completing step 1, further implement step 2. Specifically, step 2 can be described as:

[0093] The real image of the bifurcation crack on the surface of the aluminum alloy structure is collected, and after standardization and grayscale processing, the sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is obtained;

[0094] Extract the magnetic field intensity signal perpendicular to the surface of the aluminum alloy structure to obtain the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure, and after standardization, normalization and grayscale processing, obtain the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure;

[0095] The sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is feature-paired with the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure to establish a paired database of the bifurcation crack on the surface of the aluminum alloy structure.

[0096] It should be noted that in the process of obtaining sample data of the magnetic field image of bifurcation cracks on the surface of the aluminum alloy structure, the amplitude ratio K can be changed multiple times, and the magnetic field intensity signal perpendicular to the Bz direction of the surface of the aluminum alloy structure under test can be extracted to realize the detection simulation of bifurcation cracks of different sizes and angles, thereby obtaining the (multi-perspective) magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure in the Bz direction.

[0097] The process of standardizing and graying the real image of the bifurcation crack on the surface of the aluminum alloy structure and the process of standardizing, normalizing and graying the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure can be specifically referred to as Figure 8 The standardized processing process can be referred to as follows: the real image of the bifurcation crack on the surface of the aluminum alloy structure and the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure are formatted into an image with a length of 30 mm, a width of 30 mm, and a pixel size of 256*256.

[0098] The normalization process can be referred to as follows: In order to eliminate the influence of the uneven current amplitude caused by the change of K value, the following formula is used to normalize the original data, that is:

[0099] ;

[0100] In the formula, is the data before normalization, is the normalized data.

[0101] The grayscale processing process can be referred to as follows: in order to save the computational complexity of the neural network, the real image of the bifurcation crack on the surface of the aluminum alloy structure and the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure are uniformly converted into grayscale images for processing; and merged into paired images in the X direction, and the synthesized pixels are 512*256 grayscale images.

[0102] Through the above processing, 960 sets of paired data samples were obtained, thus constructing a paired database of bifurcation cracks on the surface of aluminum alloy structures; and further divided it into a training set and a test set for use in subsequent steps.

[0103] Step 3: Construct the aluminum alloy structure table.

[0104] After completing step 2, further implement step 3. As a preferred embodiment of the present invention, step 3 is specifically described as:

[0105] The loss function of the bifurcation crack on the surface of the aluminum alloy structure is constructed to meet the following requirements:

[0106] g_loss=g_bce_loss+g_l1_loss;

[0107] Among them, g_loss is the total loss value of the generative adversarial neural network; g_bce_loss is the binary cross entropy loss function, and g_l1_loss is the L1 loss function;

[0108] The binary cross entropy loss function satisfies:

[0109] g_bce_loss=bce_loss(g_fake_predict,paddle.ones_like(g_fake_predict));

[0110] L1 loss function satisfies:

[0111] ;

[0112] Among them, g_fake_predict is the discriminant probability output by the discriminator network, fake_B and real_B are the target image and the target real image generated by the generator network respectively, and a is a constant.

[0113] Step 4: Train pix2pix to obtain a generative adversarial neural network.

[0114] After completing step three, further implement step four. It is worth noting that in the process of training pix2pix to obtain a generative adversarial neural network, it is preferred to alternately train the generator network and the discriminator network. Among them, the network training method and hyperparameter setting are important factors in improving the overall network performance. The Adam optimizer is selected for network training, which requires less memory and can calculate different learning rates for different parameters. The learning rate is set to 1e-4, the batch size is set to 4, and the training rounds are set to 150. After experimental verification, the above operations can obtain the optimal solution and save the parameters of the generator.

[0115] Step 5: Deploy and apply the trained generative adversarial neural network. The trained generative adversarial neural network can be used to realize the three-dimensional characterization of bifurcation cracks on the surface of aluminum alloy structures.

[0116] After completing step 4, further implement step 5. Specifically, deploy and apply the generative adversarial neural network obtained through training. The output result of the generative adversarial neural network is the three-dimensional characterization result of the bifurcation crack. For details, please refer to Fig. 9 shown.

[0117] It is worth noting that the actual size of the bifurcation crack used in the experiment is: the transverse crack is 20mm long, the oblique crack is 10mm long, the depth is 4mm, and the bifurcation angle is 150°; while the transverse crack length in the inversion result of the generative adversarial neural network is 19.264mm, the absolute error is 0.736mm; the oblique crack length is 9.549mm, the absolute error is 0.451mm; the depth is 3.935mm, the absolute error of the depth is 0.065mm; the angle is 134.272°, and the absolute error of the angle is 0.728°. It can be found that the above inversion quantification results are relatively accurate.

[0118] The error between the neural network output and the true value was further analyzed and evaluated. First, 30 samples in the test set were randomly selected and input into the neural network for 3D reconstruction. The indicators for evaluating the size of the bifurcation crack morphology were set as the length of the crack and the bifurcation angle. The crack angle error results under the conditions of 30 samples are shown in the figure below. Fig.10a As shown in the figure, the crack length error results under 30 sample conditions are as follows Fig.10b It is worth noting that these 30 random images include superimposed images and single current incident images, with an average length error of 1.54 mm. Analysis of the results shows that sample images with errors exceeding 3 mm are all from single current incident images.

[0119] For further optimization, 20 superimposed images were selected from the test set for further length error quantification. The crack length error results are as follows: Fig.10c As shown. It can be found that the length error quantification result is significantly reduced at this time, and the average length error is 0.66mm. In other words, compared with random images, the use of superimposed images can reduce the length error quantification by about 57%. In terms of angle quantification, among the 30 random images in the test set, only 26 samples can invert the crack bifurcation angle, and those that fail to invert the oblique cracks are all magnetic field images generated under a single current incidence. In other words, the samples generated by the superimposed image method can quantify the angle, and the average error is 0.66°.

[0120] The above experimental results indicate that the three-dimensional characterization method for bifurcation cracks on the surface of aluminum alloy structures provided by the present invention has obvious technical advantages in terms of the quantification accuracy of bifurcation cracks on the surface of aluminum alloy structures, and the superimposed current provided by the bifurcation crack detection device for the surface of aluminum alloy structures provided by the present invention has obvious technical advantages in drawing magnetic field images.

[0121] The present invention provides a device for detecting bifurcation cracks on the surface of an aluminum alloy structure and a three-dimensional characterization method thereof. The device for detecting bifurcation cracks on the surface of an aluminum alloy structure includes a detection signal generating unit, a detection probe, and a detection signal acquisition unit. The detection probe is composed of an excitation coil unit and a magnetic field sensor; the excitation coil unit is composed of a first excitation coil group placed above the surface of the aluminum alloy structure to be measured and a second excitation coil group placed above the surface of the aluminum alloy structure to be measured, and the magnetic field sensor is arranged on the surface of the aluminum alloy structure to be measured at an orthogonal position between the first excitation coil group and the second excitation coil group. The characterization method is based on and includes the following steps: designing the structure of a generative adversarial neural network; establishing a paired database of bifurcation cracks on the surface of an aluminum alloy structure; constructing a loss function for bifurcation cracks on the surface of an aluminum alloy structure; training pix2pix to obtain a generative adversarial neural network; and deploying and applying the trained generative adversarial neural network.

[0122] The aluminum alloy structure surface bifurcation crack detection device and the three-dimensional characterization method thereof having the above-mentioned structural features and step features have at least the following technical advantages compared with the prior art:

[0123] 1. A uniform current field with controllable direction can be induced on the surface of the aluminum alloy structure under test; and the change and adjustment of the induced current angle can be achieved through normalization processing, which is helpful to obtain the maximum distortion of the crack (at any angle);

[0124] 2. It solves the drawbacks of existing detection technologies (such as rotating AC electromagnetic field detection methods) that the energy is not focused and the energy is wasted;

[0125] 3. It has a high detection sensitivity for bifurcated cracks in any direction. With a small number of training sets, it can realize three-dimensional inversion and evaluation of complex crack morphology, and achieve accurate quantification of crack size and angle.

[0126] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A device for detecting bifurcation cracks on the surface of an aluminum alloy structure, comprising: a detection signal generating unit, a detection probe, and a detection signal collecting unit, characterized in that: The detection probe is composed of an excitation coil unit and a magnetic field sensor; The excitation coil unit is composed of a first excitation coil group placed above the surface of the aluminum alloy structure to be measured and a second excitation coil group placed above the surface of the aluminum alloy structure to be measured; wherein the first excitation coil group and the second excitation coil group are orthogonally distributed, and the first excitation coil group is composed of two first excitation coils with opposite induced current directions, and the second excitation coil group is composed of two second excitation coils with opposite induced current directions; The magnetic field sensor is arranged at the surface of the aluminum alloy structure to be measured at the orthogonal position of the first excitation coil group and the second excitation coil group, and is used to collect the induced magnetic field signal perpendicular to the surface direction of the aluminum alloy structure to be measured; Two sinusoidal signals with the same frequency, phase and amplitude ratio of K are respectively input into the first excitation coil group and the second excitation coil group. Then, the X-direction induced current Jx and the Y-direction induced current Jy induced on the surface of the aluminum alloy structure under test respectively satisfy: Formula (1); Formula (2); The superimposed current Js obtained by vector addition of the induced current Jx in the X direction and the induced current Jy in the Y direction satisfies: Formula (3); Among them, the angle between the superimposed current Js and the horizontal direction ,satisfy: Formula (4).

2. The device for detecting bifurcation cracks on the surface of an aluminum alloy structure according to claim 1, characterized in that: Normalize the superimposed current Js; Among them, the normalized superposition current Js satisfies: Formula (5).

3. A three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure, the characterization method is based on a device for detecting bifurcation cracks on the surface of an aluminum alloy structure as claimed in any one of claims 1 to 2, characterized in that: The steps include: Step 1: Design the structure of the generative adversarial neural network; Step 2: Establish a paired database of bifurcation cracks on the surface of aluminum alloy structures; Step 3: Construct the loss function of bifurcation crack on the surface of aluminum alloy structure; The step three is specifically described as: The loss function of the bifurcation crack on the surface of the aluminum alloy structure is constructed to meet the following requirements: g_loss=g_bce_loss+g_l1_loss; Among them, g_loss is the total loss value of the generative adversarial neural network; g_bce_loss is the binary cross entropy loss function, and g_l1_loss is the L1 loss function; The binary cross entropy loss function satisfies: g_bce_loss=bce_loss(g_fake_predict,paddle.ones_like(g_fake_predict)); L1 loss function satisfies: ； Among them, g_fake_predict is the discriminant probability output by the discriminator network, fake_B and real_B are the target image and the target real image generated by the generator network respectively, and a is a constant; Step 4: Train pix2pix to obtain a generative adversarial neural network; Step 5: Deploy and apply the trained generative adversarial neural network. The trained generative adversarial neural network can be used to realize the three-dimensional characterization of bifurcation cracks on the surface of aluminum alloy structures.

4. A three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure according to claim 3, characterized in that: The step 1 is specifically described as: Design the layer structure, connection method, number of convolution kernels, convolution kernel size, and step size of the generator network in the generative adversarial neural network; Design the layer structure and connection method of the discriminator network in the generative adversarial neural network.

5. The three-dimensional characterization method for bifurcation cracks on the surface of an aluminum alloy structure according to claim 3, characterized in that: The step 2 is specifically described as: The real image of the bifurcation crack on the surface of the aluminum alloy structure is collected, and after standardization and grayscale processing, the sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is obtained; Extract the magnetic field intensity signal perpendicular to the surface of the aluminum alloy structure to obtain the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure, and after standardization, normalization and grayscale processing, obtain the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure; The sample data of the real image of the bifurcation crack on the surface of the aluminum alloy structure is feature-paired with the sample data of the magnetic field image of the bifurcation crack on the surface of the aluminum alloy structure to establish a paired database of the bifurcation crack on the surface of the aluminum alloy structure.

Images