A quantitative non-destructive testing method for defects in a plate structure

Abstract

This invention discloses a quantitative non-destructive testing method for defects in plate structures. The method includes: performing ultrasonic testing on the metal plate to be tested to acquire the measured signal; inputting the initial velocity model of the metal plate to be tested and the measured signal into a full waveform inversion algorithm for one iteration update to obtain the updated velocity model of the metal plate to be tested; converting the updated velocity model into a thickness model and inputting the thickness model into a trained defect reconstruction model to obtain the defect reconstruction result of the metal plate to be tested; the defect reconstruction result includes the location, size, and shape information of the defect; this invention utilizes a trained defect reconstruction model based on a full waveform inversion to accurately detect defects, thereby improving the defect detection capability in plate structures.

Description

Technical Field

[0001] This invention relates to the field of quantitative nondestructive testing technology for defects, and particularly to a method for quantitative nondestructive testing of defects in plate structures. Background Technology

[0002] Metal sheets, due to their excellent mechanical and processing properties, occupy an important position in modern industry and are widely used in aerospace structural components, automotive body panels, ship hulls, and large building structures. However, during long-term service, under the influence of alternating loads, environmental corrosion, and accidental mechanical damage, defects such as micro-cracks, corrosion pits, and localized thinning can easily develop inside the sheet metal. If these defects are not detected in time, they will lead to a decrease in the structural load-bearing capacity and even cause catastrophic accidents. Therefore, developing high-precision non-destructive testing technology for defects in metal sheet structures is of great significance for ensuring the safe operation of industrial facilities.

[0003] Ultrasonic testing technology, with its advantages of strong penetration, good directivity, and sensitivity to defects, has become one of the mainstream methods for non-destructive testing of metal plate structures. This technology can locate and quantitatively assess defects by analyzing the reflection, transmission, and scattering characteristics of ultrasonic waves propagating in a medium. In recent years, the full-waveform inversion method has been introduced into the field of ultrasonic non-destructive testing. This method constructs an objective function and iteratively updates the medium parameter model, making the forward-modeled waveform approximate the observed waveform, thereby achieving high-resolution reconstruction of the medium's acoustic parameters (such as sound velocity and density). Compared with traditional ultrasonic imaging methods, full-waveform inversion fully utilizes the entire wavefield information of ultrasonic waves, including travel time, phase, and amplitude, resulting in higher inversion accuracy and spatial resolution. Existing research shows that full-waveform inversion can effectively reconstruct the morphology of corrosion defects in plate structures, maintaining good inversion accuracy even when multiple defects coexist, with spatial resolution reaching the subwavelength level.

[0004] However, the application of full-waveform inversion technology in practical engineering inspection environments still faces many challenges. On the one hand, the inspection site is subject to complex electromagnetic interference, structural boundary reflections, and mode conversion noise, which reduces the signal-to-noise ratio of the measured signal. These noise components are amplified during the inversion process, affecting the accuracy of defect reconstruction. On the other hand, full-waveform inversion is highly dependent on the initial model. When the initial velocity model deviates significantly from the actual structure, the phase difference between the forward waveform and the measured waveform may exceed half a cycle, causing the objective function to fall into a local minimum, resulting in inversion results that deviate from the actual defect distribution. Furthermore, obtaining high-quality inversion results typically requires dozens or even hundreds of iterative calculations, which are computationally expensive and difficult to meet the needs of rapid inspection.

[0005] To address the aforementioned issues, it is necessary to develop a method for detecting defects in metal plates that balances noise resistance, computational efficiency, and inversion accuracy, in order to achieve rapid and accurate quantitative assessment of structural defects in plates under complex working conditions. Summary of the Invention

[0006] The purpose of this invention is to provide a quantitative non-destructive testing method for defects in plate structures. This method utilizes a trained defect reconstruction model to accurately detect defects based on a single full waveform inversion, thereby improving the defect detection capability in plate structures.

[0007] To achieve the above objectives, the present invention employs the following technical solution:

[0008] In a first aspect, the present invention provides a method for quantitative nondestructive testing of defects in plate structures, comprising:

[0009] Ultrasonic testing is performed on the metal plate to be tested to obtain the measured signal;

[0010] The initial velocity model of the metal plate under test and the measured signal are input into the full waveform inversion algorithm for one iteration update to obtain the updated velocity model of the metal plate under test.

[0011] The updated velocity model is converted into a thickness model, and the thickness model is input into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

[0012] Optionally, the calculation formula for the first iteration update is expressed as follows:

[0013] ;

[0014] Specifically, This represents the velocity model after one update; Represents the initial velocity model; Indicates the optimal step size; The inverse of the Hessian matrix; This represents the gradient value in the first iteration.

[0015] Optionally, the defect reconstruction model includes: an input layer, an encoder, an intermediate layer, a decoder, and an output layer connected in sequence;

[0016] The input layer is used to input the thickness model converted from the velocity model after one update;

[0017] The encoder includes four sequentially connected downsampling layers for progressively extracting features and compressing spatial dimensions of the thickness model to obtain latent feature representations. Each downsampling layer includes a residual module and a max pooling layer connected to the residual module. The residual module includes two sequentially connected first convolutional layers, an activation function, and a batch normalization layer.

[0018] The intermediate layer includes a residual module, which is used to extract deep features from the latent feature representation to obtain a deep feature representation;

[0019] The decoder includes multiple sequentially connected upsampling layers for progressively upsampling and restoring the deep feature representation to obtain a high-resolution feature representation; each upsampling layer includes a sequentially connected deconvolution layer and a residual module; the residual module in the decoder is connected to the residual module of the corresponding layer in the encoder via skip connections.

[0020] The output layer includes a second convolutional layer, which is used to perform dimensionality reduction mapping on the high-resolution feature representation to obtain the defect reconstruction result.

[0021] Optionally, the thickness model is input into the trained defect reconstruction model to obtain the defect reconstruction results of the metal plate under test, including:

[0022] The thickness model is input into the input layer of the trained defect reconstruction model;

[0023] The thickness model in the input layer is input into the encoder, and the thickness model is progressively subjected to feature extraction and spatial dimension compression through the four downsampling layers of the encoder to obtain the latent feature representation;

[0024] The latent feature representation is input into the intermediate layer, and the residual module of the intermediate layer performs deep feature extraction on the latent feature representation to obtain the deep feature representation.

[0025] The deep feature representation is input into the decoder, and the deep feature representation is gradually upsampled and restored through multiple upsampling layers of the decoder. At the same time, the features output by the residual module of the corresponding layer in the encoder are fused by skip connections to obtain a high-resolution feature representation.

[0026] The high-resolution feature representation is dimensionality-reduced using the second convolutional layer of the output layer, and the defect reconstruction result is output.

[0027] Optionally, the training method for the defect reconstruction model includes:

[0028] Construct a training dataset, which includes multiple sets of sample pairs consisting of thickness models converted from velocity models updated by the full waveform inversion algorithm in one iteration, and the corresponding real thickness models;

[0029] The training dataset is input into the defect reconstruction model. The goal is to minimize the mean squared error loss function. The network parameters of the defect reconstruction model are iteratively updated using the backpropagation algorithm until the model converges, resulting in a well-trained defect reconstruction model.

[0030] Optionally, the mean squared error loss function is expressed as follows:

[0031] ;

[0032] in, This represents the value of the mean squared error loss function; This represents the total number of sample pairs; This represents the true thickness model in the k-th sample pair; This represents the thickness model in the k-th sample pair.

[0033] Secondly, the present invention provides a quantitative non-destructive testing system for defects in plate structures, comprising:

[0034] The detection module is used to perform ultrasonic testing on the metal plate under test and obtain the measured signal.

[0035] The first inversion module is used to input the initial velocity model of the metal plate under test and the measured signal into the full waveform inversion algorithm, perform one iteration update, and obtain the updated velocity model of the metal plate under test.

[0036] The defect reconstruction module is used to convert the updated velocity model into a thickness model and input the thickness model into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

[0037] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor;

[0038] The memory is used to store programs;

[0039] The processor is used to execute the program to implement the quantitative non-destructive testing method for plate structure defects as described in the first aspect.

[0040] Fourthly, the present invention provides a readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the quantitative non-destructive testing method for plate structure defects as described in the first aspect.

[0041] Fifthly, the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the quantitative non-destructive testing method for plate structure defects as described in the first aspect.

[0042] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0043] This invention provides a quantitative non-destructive testing method for defects in plate structures. This method combines full-waveform inversion with a deep learning model to construct an end-to-end mapping from the physical inversion results to the actual defect distribution. On one hand, full-waveform inversion fully utilizes the full-wavelength information of ultrasonic guided waves for preliminary defect localization and morphological reconstruction. On the other hand, the trained defect reconstruction model can learn the nonlinear mapping relationship between the thickness map (thickness model) converted from the velocity model output by the full-waveform inversion and the actual defect. This effectively suppresses the influence of noise components such as electromagnetic interference and boundary reflections in the measured signal on the inversion results, corrects artifacts and distortions that may occur during the full-waveform inversion process, and thus obtains more accurate and clearer defect reconstruction results, significantly improving inversion accuracy and noise resistance. Furthermore, this invention only requires one full-waveform inversion iteration to obtain a velocity model containing preliminary defect information. After converting this model into a thickness model, a second correction is performed using a data-driven defect reconstruction model. This strategy avoids the stringent requirements of traditional full-waveform inversion for a high-quality initial model, reduces the risk of getting trapped in local extrema due to a phase difference exceeding half a cycle, effectively reduces dependence on the initial model, and improves the robustness and adaptability of the method. Furthermore, this invention reduces the number of full waveform inversion iterations to one, significantly reducing the computational load of forward modeling and gradient calculation. Meanwhile, the defect reconstruction model only needs one calculation in the forward inference stage to output the reconstruction result. The combination of the two significantly shortens the detection time and significantly improves the computational efficiency while ensuring reconstruction accuracy, providing a feasible solution for rapid online detection of defects in metal plate structures.

[0044] This invention provides a quantitative non-destructive testing method for defects in plate structures. The output of this method directly includes the location, size, and shape information of the defects, realizing a leap from qualitative testing to quantitative assessment. By automatically extracting the geometric features of the defects through a deep learning model, the method outputs quantified defect parameters, providing a more intuitive and reliable basis for structural safety assessment and maintenance decisions. It constructs an end-to-end automated testing process from measured signal input to defect result output, eliminating the need for manual intervention and expert interpretation, thus lowering the operational threshold and improving testing efficiency and consistency. Attached Figure Description

[0045] Figure 1 The diagram shown is a flowchart of a quantitative non-destructive testing method for plate structure defects in one embodiment of the present invention.

[0046] Figure 2 The diagram shown is a schematic representation of a defect reconstruction model in one embodiment of the present invention.

[0047] Figure 3 The diagram shown is a schematic diagram of an ultrasonic detection system in one embodiment of the present invention; Figure 4 The diagram shown is a schematic representation of the received signals received by 36 sensors in one embodiment of the present invention. Figure 5 The image shown is an experimental signal diagram after windowing and filtering processing, according to one embodiment of the present invention. Figure 6 The image shown is a defect reconstruction result diagram of a trained defect reconstruction model (optimized U-Net) in one embodiment of the present invention; Figure 7 The image shown is a result of full waveform inversion defect reconstruction in one embodiment of the present invention; Figure 8 The figure shown is a profile depth curve at y=118mm at different frequencies for two inversion methods in one embodiment of the present invention. Figure 9 The figure shown is a comparison of the defect radius and its corresponding true value in the defect reconstruction result of the trained defect reconstruction model in one embodiment of the present invention.

[0048] Figure 10 The image shown is a comparison of the horizontal center position and the corresponding true value in the defect reconstruction result of the trained defect reconstruction model in one embodiment of the present invention. Figure 11 The image shown is a comparison of the true value of the longitudinal center position in the defect reconstruction result of the trained defect reconstruction model in one embodiment of the present invention. Figure 12 The diagram shown is a schematic representation of the relative error between the actual value and the defect reconstruction result of the trained defect reconstruction model in one embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0050] Example 1

[0051] like Figure 1 As shown in the figure, this invention provides a method for quantitative non-destructive testing of defects in plate structures, comprising the following steps:

[0052] S1: Perform ultrasonic testing on the metal plate to be tested and obtain the measured signal;

[0053] S2: Input the initial velocity model of the metal plate under test and the measured signal into the full waveform inversion algorithm, perform one iteration update, and obtain the updated velocity model of the metal plate under test.

[0054] S3: Convert the updated velocity model into a thickness model, and input the thickness model into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

[0055] In this embodiment, step S1 involves ultrasonic testing of the metal plate to be tested to obtain the measured signal. Specifically, the metal plate to be tested is selected, for example, a Q235 steel plate with dimensions of 500 mm × 500 mm and a thickness of 10 mm. A circular flat-bottomed hole with a diameter of 30 mm and a depth of 2 mm is pre-set inside the plate to simulate corrosion defects. An ultrasonic guided wave detection system is used for scanning. Both the excitation and receiving sensors are piezoelectric ultrasonic probes with a center frequency of 100 kHz. A linear array is arranged on the plate surface, and the excitation point sensor position... and receiver sensor position Each area to be tested is covered separately, and full matrix data is collected to obtain the measured signal. ;

[0056] In this embodiment, step S2 involves inputting the initial velocity model of the metal plate under test and the measured signal into the full waveform inversion algorithm for one iteration update to obtain the updated velocity model of the metal plate under test. Specifically, this includes: S21: Setting a uniform, defect-free initial velocity model for the metal plate under test. S22: Initial velocity model and measured signal Input the full waveform inversion algorithm to make the analog data Compared with measured data Minimizing the residuals, the error function is defined as:

[0057] ;

[0058] In the formula: This represents the current velocity model during the first iteration, i.e. The corresponding error function value; where the simulated data The simulation data is obtained through forward modeling. S32: The initial velocity model is modeled using a quasi-Newton method (such as the L-BFGS algorithm) along the negative gradient direction. The update is performed using the formula shown below:

[0059] ;

[0060] ;

[0061] In the formula: express The gradient value of the corresponding error function; This represents the velocity model after one update; Indicates the optimal step size; The inverse of the Hessian matrix; This represents the gradient value in the first iteration. In this embodiment...

[0062] In this embodiment, the velocity model updated once is converted into a thickness model using dispersion curves.

[0063] like Figure 2 As shown, in this embodiment, in step 3, the defect reconstruction model adopts a U-Net structure, including: an input layer, an encoder, an intermediate layer, a decoder, and an output layer connected in sequence;

[0064] The input layer is used to input the thickness model converted from the velocity model after one update;

[0065] The encoder comprises four sequentially connected downsampling layers: a first downsampling layer, a second downsampling layer, a third downsampling layer, and a fourth downsampling layer. These layers are used to progressively extract features and compress the spatial dimensions of the thickness model to obtain a latent feature representation. Each downsampling layer includes a residual module and a max-pooling layer connected to the residual module. The residual module includes two sequentially connected first convolutional layers (i.e., conv 3×3 in the figure), an activation function (i.e., ReLU), and a batch normalization layer (i.e., BN).

[0066] The intermediate layer includes a residual module, which is used to extract deep features from the latent feature representation to obtain a deep feature representation;

[0067] The decoder includes multiple sequentially connected upsampling layers, namely a first upsampling layer, a second upsampling layer, a third upsampling layer, and a fourth upsampling layer, which are used to progressively upsample and restore the deep feature representation to obtain a high-resolution feature representation. The upsampling layer includes sequentially connected deconvolutional layers (i.e., up-conv 2×2 in the figure) and residual modules. The residual modules in the decoder are connected to the residual modules of the corresponding layers in the encoder through skip connections.

[0068] The output layer includes a second convolutional layer (i.e., conv 1×1 in the figure), which is used to perform dimensionality reduction mapping on the high-resolution feature representation to obtain the defect reconstruction result.

[0069] Based on the above defect reconstruction model, in this embodiment, step S3 inputs the thickness model into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test, including:

[0070] S31: Input the thickness model into the input layer of the trained defect reconstruction model;

[0071] S32: Input the thickness model in the input layer into the encoder, and perform feature extraction and spatial dimension compression on the thickness model step by step through the four downsampling layers of the encoder to obtain the latent feature representation;

[0072] S33: Input the latent feature representation into the intermediate layer, and perform deep feature extraction on the latent feature representation through the residual module of the intermediate layer to obtain the deep feature representation;

[0073] S34: Input the depth feature representation into the decoder, and perform upsampling recovery operation on the depth feature representation step by step through multiple upsampling layers of the decoder. At the same time, the features output by the residual module of the corresponding layer in the encoder are fused by skip connection to obtain the high-resolution feature representation.

[0074] S35: Use the second convolutional layer of the output layer to perform dimensionality reduction mapping on the high-resolution feature representation and output the defect reconstruction result.

[0075] Specifically, in step S34: the depth feature representation is input into the first upsampling layer, and firstly, it is upsampled through a deconvolution layer to double the size of the feature map and halve the number of channels to obtain the first upsampled feature; then, the first upsampled feature is combined with the fourth residual feature output by the residual module in the fourth downsampling layer of the encoder; the channel dimensions are concatenated and fused through skip connections to obtain the first fused feature; then, the first fused feature is input into the residual module of the first upsampling layer for nonlinear mapping to output the first decoded feature.

[0076] The first decoding feature is input into the second upsampling layer. It is first upsampled by the deconvolution layer, which doubles the size and halves the number of channels to obtain the second upsampled feature. Then, the second upsampled feature is concatenated and fused with the third residual feature output by the residual module in the third downsampling layer of the encoder to obtain the second fused feature. It is then processed by the residual module of the second upsampling layer to output the second decoding feature.

[0077] The second decoding feature is input into the third upsampling layer. First, it is upsampled by the deconvolution layer, doubling the size and halving the number of channels to obtain the third upsampled feature. Then, the third upsampled feature is concatenated and fused with the second residual feature output by the residual module in the second downsampling layer of the encoder to obtain the third fused feature. Finally, it is processed by the residual module of the third upsampling layer to output the third decoding feature.

[0078] The third decoding feature is input into the fourth upsampling layer. First, it is upsampled by the deconvolution layer, doubling the size and halving the number of channels to obtain the fourth upsampled feature. Then, the fourth upsampled feature is concatenated and fused with the first residual feature output by the residual module in the first downsampling layer of the encoder to obtain the fourth fused feature. Finally, it is processed by the residual module of the fourth upsampling layer to output the fourth decoding feature, which is the high-resolution feature representation.

[0079] In this embodiment, the training method for the defect reconstruction model includes:

[0080] Construct a training dataset, which includes multiple sets of sample pairs consisting of thickness models converted from velocity models updated by the full waveform inversion algorithm in one iteration, and the corresponding real thickness models;

[0081] The training dataset is input into the defect reconstruction model. The goal is to minimize the mean squared error loss function. The network parameters of the defect reconstruction model are iteratively updated using the backpropagation algorithm until the model converges, resulting in a well-trained defect reconstruction model.

[0082] The mean squared error loss function is expressed as follows:

[0083] ;

[0084] in, This represents the value of the mean squared error loss function; This represents the total number of sample pairs; This represents the true thickness model in the k-th sample pair; This represents the thickness model in the k-th sample pair.

[0085] This embodiment proposes a quantitative non-destructive testing method for plate structure defects, which can be divided into a training phase and a testing phase. In the training phase, a defect-free metal plate is used as the initial model. This model, along with the measured signal obtained through an ultrasonic guided wave detection system, is input into a full waveform inversion algorithm. A single full waveform inversion is used to update the thickness model. A defect reconstruction model is then established, using the updated thickness model as input and the actual thickness model as the output target, to train the mapping relationship between the defect reconstruction model and the thickness model. In the testing phase, the measured guided wave signal of the metal plate under test is obtained using an ultrasonic detection system and input into the full waveform inversion process for a single thickness model update. This model is then input into the trained defect reconstruction model to reconstruct the defect morphology.

[0086] Example 2

[0087] Based on the quantitative non-destructive testing method for plate structure defects provided in Example 1, this example further illustrates the present invention in conjunction with specific embodiments.

[0088] An ultrasonic testing system was set up in the laboratory. The test piece was an aluminum plate measuring 500mm × 500mm × 3mm, with a flat-bottomed hole defect in the center measuring 6mm in diameter and 1.5mm in depth. Sensors were arranged on a circle with a radius of 50mm centered on the plate's center, in 10° increments, for a total of 36 sensors. A schematic diagram of the ultrasonic testing system is shown below. Figure 3 As shown in the figure, x represents the horizontal coordinate, y represents the vertical coordinate, and r represents the defect radius.

[0089] Based on the quantitative non-destructive testing method for plate structure defects provided in Example 1, the specific steps of this example are as follows:

[0090] Step 01: Using a defect-free uniform metal plate as the initial velocity model, the measured received signal is acquired using an ultrasonic guided wave detection system;

[0091] Step 02: Input the initial velocity model and the measured signal into the full waveform inversion algorithm, perform one iterative update, and obtain the initially updated velocity model;

[0092] Step 03: Convert the updated velocity model into a thickness model using dispersion curves, thereby constructing the training dataset;

[0093] Step 04: Construct the U-Net neural network model, i.e. the defect reconstruction model. Take the thickness model output from a full waveform inversion as the input and the real thickness model as the output target, and train the network to learn the nonlinear mapping relationship between the two.

[0094] Step 05: The defect reconstruction model adopts a 4-layer U-Net structure, as shown in Table 1. A residual module is introduced at each sampling stage to prevent network degradation and improve the ability to extract small defect features.

[0095] Table 1 Defect Reconstruction Model Parameters

[0096] Step 06: Optimize the hyperparameters, including learning rate, batch size and optimization algorithm, to improve network convergence speed and reconstruction accuracy. The resulting hyperparameters are shown in Table 2.

[0097] Table 2 Optimal Solution of Network Parameters

[0098] Step 07: First, select one sensor for excitation, and the remaining 35 for reception. The excitation signal is a sinusoidal signal modulated with a Hanning window at a center frequency of 100kHz. The sensors are sequentially excited, ultimately obtaining 35×36 groups of received signals. The received signal duration is 25ms, and the sampling rate is 1.25MHz. The received signals from the one excitation sensor and the other 35 sensors are as follows: Figure 4 As shown. The direct wave in the received signal is captured by windowing, such as... Figure 5 As shown, the start and end points of the time window are determined by the group velocity of the A0 mode and the transmitted waveform of the signal generator.

[0099] Step 08: Input the initial velocity model and the measured signal into the full waveform inversion algorithm, perform one iteration update, and obtain the preliminary velocity model of the target to be measured;

[0100] Step 09: Convert the initial velocity model into a thickness model and input it into the trained defect reconstruction model;

[0101] Step 10: The model output is the reconstruction result of the defect, including the location, size and shape information of the defect.

[0102] Taking an eccentric defect with a test concentration radius of 6 mm, a lateral center position of 122 mm, and a longitudinal center position of 118 mm as an example... Figure 6 The image shows the defect reconstruction results of the trained defect reconstruction model (optimized U-Net). Figure 7 The results of full-waveform inversion defect reconstruction are presented. It can be seen that, compared with the full-waveform inversion method, the inversion method based on the trained defect reconstruction model can better restore the defect thickness and contour structure information, and does not produce artifacts. Figure 8 The figures show the profile depth curves at y=118mm at different frequencies for the two inversion methods. It can be seen that, compared with the full waveform inversion method, the inversion method based on the optimized U-Net network can better restore the defect thickness and contour structure information, and does not produce artifacts.

[0103] Table 3 shows the relative error of the reconstruction results from the two methods and the computation time:

[0104] It can be seen that full waveform inversion produces virtually no positioning error, but the radius error is relatively large. Inversion based on a trained defect reconstruction model has a smaller radius error, but it does produce positioning errors. The computation time shows that inversion based on the trained defect reconstruction model takes significantly less time than full waveform inversion, achieving short-time defect prediction.

[0105] further, Figure 9 , Figure 10 and Figure 11 The defect radius (inversion radius), lateral center position, and longitudinal center position in the reconstruction results are compared with their corresponding true values. It can be seen that the defect reconstruction results of the defect reconstruction model of the present invention are very close to the true values. Figure 12 The relative errors between the reconstruction results and the true values ​​are shown. The results indicate that the relative error of the defect radius reconstructed using the defect reconstruction model does not exceed 2.7%, the relative error of the lateral center position does not exceed 3.83%, and the relative error of the longitudinal center position does not exceed 2.51%.

[0106] In summary, the imaging results demonstrate that this method can separate wave packets of defect echoes of different modes with limited input data, and can be used for ultrasonic wave packet separation of pipelines to achieve accurate imaging.

[0107] The above are typical applications of the present invention, but the applications of the present invention are not limited thereto.

[0108] Example 3

[0109] This invention describes a system for implementing the quantitative nondestructive testing method for plate structure defects provided in Embodiment 1, comprising:

[0110] The detection module is used to perform ultrasonic testing on the metal plate under test and obtain the measured signal.

[0111] The first inversion module is used to input the initial velocity model of the metal plate under test and the measured signal into the full waveform inversion algorithm, perform one iteration update, and obtain the updated velocity model of the metal plate under test.

[0112] The defect reconstruction module is used to convert the updated velocity model into a thickness model and input the thickness model into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

[0113] The specific functions of each module described above are explained in the relevant content of the method in Embodiment 1, and will not be repeated here.

[0114] Example 4 This example describes an electronic device, including: a memory and a processor;

[0115] The memory is used to store programs;

[0116] The processor is used to execute the program to implement the quantitative non-destructive testing method for plate structure defects in Embodiment 1.

[0117] This embodiment also introduces a readable storage medium storing a computer program, characterized in that, when the computer program is executed by a processor, it implements the quantitative non-destructive testing method for plate structure defects in Embodiment 1.

[0118] This embodiment also introduces a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the quantitative non-destructive testing method for plate structure defects in Embodiment 1.

[0119] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0120] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0121] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0122] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A quantitative non-destructive testing method for defects in plate structures, characterized in that, include: Ultrasonic testing is performed on the metal plate to be tested to obtain the measured signal; The initial velocity model of the metal plate under test and the measured signal are input into the full waveform inversion algorithm for one iteration update to obtain the updated velocity model of the metal plate under test. The updated velocity model is converted into a thickness model, and the thickness model is input into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

2. The quantitative non-destructive testing method for plate structure defects according to claim 1, characterized in that, The calculation formula for a single iteration update is expressed as follows: ； Specifically, This represents the velocity model after one update; Represents the initial velocity model; Indicates the optimal step size; The inverse of the Hessian matrix; This represents the gradient value in the first iteration.

3. The quantitative non-destructive testing method for plate structure defects according to claim 2, characterized in that, The defect reconstruction model includes: an input layer, an encoder, an intermediate layer, a decoder, and an output layer connected in sequence; The input layer is used to input the thickness model converted from the velocity model after one update; The encoder includes four sequentially connected downsampling layers for progressively extracting features and compressing spatial dimensions of the thickness model to obtain latent feature representations. Each downsampling layer includes a residual module and a max pooling layer connected to the residual module. The residual module includes two sequentially connected first convolutional layers, an activation function, and a batch normalization layer. The intermediate layer includes a residual module, which is used to extract deep features from the latent feature representation to obtain a deep feature representation; The decoder includes multiple sequentially connected upsampling layers for progressively upsampling and restoring the deep feature representation to obtain a high-resolution feature representation; each upsampling layer includes a sequentially connected deconvolution layer and a residual module; the residual module in the decoder is connected to the residual module of the corresponding layer in the encoder via skip connections. The output layer includes a second convolutional layer, which is used to perform dimensionality reduction mapping on the high-resolution feature representation to obtain the defect reconstruction result.

4. The quantitative non-destructive testing method for plate structure defects according to claim 3, characterized in that, The thickness model is input into the trained defect reconstruction model to obtain the defect reconstruction results of the metal plate under test, including: The thickness model is input into the input layer of the trained defect reconstruction model; The thickness model in the input layer is input into the encoder, and the thickness model is progressively subjected to feature extraction and spatial dimension compression through the four downsampling layers of the encoder to obtain the latent feature representation; The latent feature representation is input into the intermediate layer, and the residual module of the intermediate layer performs deep feature extraction on the latent feature representation to obtain the deep feature representation. The deep feature representation is input into the decoder, and the deep feature representation is gradually upsampled and restored through multiple upsampling layers of the decoder. At the same time, the features output by the residual module of the corresponding layer in the encoder are fused by skip connections to obtain a high-resolution feature representation. The high-resolution feature representation is dimensionality-reduced using the second convolutional layer of the output layer, and the defect reconstruction result is output.

5. The quantitative non-destructive testing method for plate structure defects according to claim 4, characterized in that, The training method for the defect reconstruction model includes: Construct a training dataset, which includes multiple sets of sample pairs consisting of thickness models converted from velocity models updated by the full waveform inversion algorithm in one iteration, and the corresponding real thickness models; The training dataset is input into the defect reconstruction model. The goal is to minimize the mean squared error loss function. The network parameters of the defect reconstruction model are iteratively updated using the backpropagation algorithm until the model converges, resulting in a well-trained defect reconstruction model.

6. The quantitative non-destructive testing method for plate structure defects according to claim 5, characterized in that, The mean squared error loss function is expressed as follows: ； in, This represents the value of the mean squared error loss function; This represents the total number of sample pairs; This represents the true thickness model in the k-th sample pair; This represents the thickness model in the k-th sample pair.

7. A quantitative non-destructive testing system for defects in plate structures, characterized in that, include: The detection module is used to perform ultrasonic testing on the metal plate under test and obtain the measured signal. The first inversion module is used to input the initial velocity model of the metal plate under test and the measured signal into the full waveform inversion algorithm, perform one iteration update, and obtain the updated velocity model of the metal plate under test. The defect reconstruction module is used to convert the updated velocity model into a thickness model and input the thickness model into the trained defect reconstruction model to obtain the defect reconstruction result of the metal plate under test; the defect reconstruction result includes the location, size and shape information of the defect.

8. An electronic device, characterized in that, include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement the quantitative non-destructive testing method for plate structure defects as described in any one of claims 1-6.

9. A readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the quantitative non-destructive testing method for plate structure defects as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the quantitative non-destructive testing method for plate structure defects as described in any one of claims 1-6.

Images