Subsurface defect multi-mode laser ultrasonic C scanning image reconstruction method

Abstract

The invention discloses a subsurface defect multi-mode laser ultrasonic C scanning image reconstruction method. The method comprises the following steps: carrying out two-dimensional scanning on a detected object by utilizing a laser ultrasonic detection device to obtain a multi-mode ultrasonic A scanning time domain signal data set; converting the data set into a frequency domain signal, and performing position coding according to the acquisition position of the signal; splicing the frequency domain signals with the position codes, and calculating the similarity between the signals; according to the calculated signal similarity, selecting a set number of signals to construct a graph data set; and inputting the graph data set into a trained graph neural network model to obtain a classification result of the defect signal at each position, judging whether the classification result is from a defect area or not, and performing C scanning image reconstruction according to the original acquisition position of each A scanning signal and the classification result. According to the invention, high-resolution scanning imaging of the subsurface defect can be realized, and the accuracy and reliability of subsurface defect detection are obviously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent manufacturing equipment, and more specifically, to a sub-surface defect multi-mode laser ultrasonic C-scan image reconstruction method. Background Art

[0002] Metal Additive Manufacturing (MAM) is an advanced manufacturing technology that can directly process parts with complex shapes based on three-dimensional model data. Compared with traditional subtractive manufacturing technology, MAM has the advantages of unrestricted part structure and high raw material utilization, and is widely used in aerospace and other industrial fields. The commonly used MAM technology uses high-energy sources such as lasers and electron beams to generate high temperatures to melt metal powder or wire, and realizes rapid manufacturing of parts in a layer-by-layer stacking manner. However, due to the inherent characteristics of the material, as well as the unreasonable setting of energy beam parameters (such as power, spot size, etc.) and processing parameters, defects such as cracks, holes, unfused and residual stress often occur, which will have a significant negative impact on the internal microstructure, mechanical properties and overall performance of MAM products. Therefore, there is an urgent need for an accurate non-destructive testing technology to detect defects in MAM workpieces, evaluate and further optimize the MAM process to ensure the quality and performance of the final product.

[0003] In the prior art, the methods commonly used for metal additive defect detection mainly include X-ray and ultrasonic testing. X-ray detection is based on the differences in the absorption and transmittance of X-rays by different parts of the object after the X-rays pass through it. By analyzing the changes in the intensity of the X-rays after they penetrate the object, it identifies and evaluates potential irregularities and defects inside the material. However, in materials with large thickness or density, the effect of X-ray detection is poor due to the rapid attenuation of energy, and radiation exposure poses a potential safety risk to operators. The principle of ultrasonic testing relies on the scattering, reflection and attenuation of ultrasonic waves when they propagate in the material. However, the ultrasonic testing method based on piezoelectric transducers is a contact test that requires a coupling agent or other medium to successfully couple the sound waves emitted by the transducer into the object being tested. It is not suitable for harsh environments such as high temperature and high pressure or for test pieces with complex shapes.

[0004] Laser Ultrasonic Testing (LUT) is a completely non-contact non-destructive testing technology. By using pulsed laser to irradiate the surface of an object to cause thermal expansion effect, ultrasonic waves are generated inside the material, and optical methods such as laser interference principle are used to receive ultrasonic signals. Compared with X-ray and traditional ultrasonic testing, LUT has the advantages of high sensitivity, high spatial resolution, and strong real-time detection capability. In laser ultrasonic non-destructive testing based on thermoelastic mechanism, pulsed laser will excite complex multi-mode coupled acoustic fields on the surface and inside of the material, including multiple modes of ultrasonic waves such as longitudinal waves (L), shear waves (T) and Rayleigh waves (R). These wave modes have different propagation directions, speeds and energy distributions. When the defect is located on the sub-surface, Rayleigh waves, longitudinal waves and shear waves will all be reflected on the defect surface with different time delays, causing multi-mode ultrasonic waves to produce aliased signals at the receiving point, such as Figure 1 When the amplitude of these A-scan signals is used for C-scan imaging, the C-scan images at different times are different, and the shape of the defect cannot be accurately presented, which affects the accuracy of detection, thus hindering the high-resolution imaging of sub-surface defects by the traditional C-scan imaging method. C-scan is a surface scan that indicates the projection surface condition of the object being inspected and draws the horizontal projection position of the defect on the projection surface.

[0005] In order to solve the above problems, some studies have proposed improved methods. For example, patent application CN109269986A discloses a phased array laser ultrasonic detection system, which detects defects by inducing a phased array through a laser. However, this method requires rigorous analysis of signals captured from different angles and delay intervals. Patent application CN113406010A discloses a laser ultrasonic detection method based on synthetic aperture focusing imaging. However, the synthetic aperture focusing imaging method requires an in-depth understanding of the propagation laws of ultrasonic waves in materials, the complex interactions between multimodal ultrasonic waves and defects, and the use of complex signal processing techniques.

[0006] With the rapid development of artificial intelligence technology and its wide application in various fields, its application in laser ultrasonic defect detection has also achieved certain results, significantly improving the efficiency and accuracy of defect detection. For example, patent application CN116773679A discloses a laser ultrasonic imaging method based on neural network. The method first performs empirical mode decomposition and reconstruction on each A-scan signal separately, and then inputs the decomposed signal into the neural network to extract features. The extracted feature values ​​are then converted into pixel values, and the state of each detection point is evaluated by setting a threshold, and finally the corresponding pixel image is generated. Although this method is suitable for signals with low signal-to-noise ratio, its detection accuracy depends on the selection of intrinsic mode function (IMF) in the process of empirical mode decomposition and the setting of threshold in the neural network model. In addition, the existing artificial intelligence-based detection method still needs to accurately label each signal, which often consumes a lot of manpower and time in practical applications. Moreover, when faced with multi-mode ultrasonic aliasing signals, accurate labeling becomes extremely difficult, which greatly limits the application effect of this method in sub-surface defect detection. Another challenge is that current deep learning models usually only extract features from a single signal without fully considering the relationship between signals, which makes feature extraction completely dependent on the model's autonomous learning ability. The lack of consideration of the relationship between signals may cause the model to perform poorly when processing complex signals, especially when dealing with multimodal and aliased signals. These problems show that although artificial intelligence technology has potential in laser ultrasonic defect detection, it still needs to be further optimized and developed to cope with the complexity and challenges in practical applications.

[0007] Graph Neural Network (GNN) has been widely used in classification tasks due to its high efficiency in processing graph data with non-Euclidean structures. Graph data consists of nodes and edges between nodes, which can be understood as a mesh structure rather than a picture. Among them, nodes represent entities and edges represent the relationship between them. GNN uses the relationship between nodes to perform aggregation operations, such as averaging and summing, to update the node feature vector, and extract the key features required for decision making. GNN updates the feature vector of the node through operations such as aggregation (such as averaging and summing) and convolution of the relationship between nodes, thereby extracting the key features required for decision making. Through this process, GNN is able to capture information in the graph structure and mitigate the impact of outliers, thereby enhancing the robustness of the model. However, when dealing with newly emerged nodes (i.e. nodes that have not been seen during training), general aggregation or convolution performance is limited, so further strategies are needed to improve its generalization ability. Adding an attention mechanism to the graph neural network, thereby replacing the attention mechanism of static normalized convolution, can enhance the generalization ability of unknown nodes.

[0008] In summary, the current defect detection methods mainly have the following problems:

[0009] 1) Traditional laser ultrasonic imaging methods rely on complex signal processing technology to reduce the interference caused by signal aliasing, but the final imaging effect is still far from the actual situation.

[0010] 2) Existing deep learning methods require preprocessing of A-scan signals to improve the accuracy of subsequent deep learning models in the feature extraction process, which reduces detection efficiency.

[0011] 3) Existing deep learning models only extract defect features from a single signal, failing to fully utilize the spatial structure information between signals, which increases the difficulty of feature extraction. In addition, each signal needs to be accurately labeled to support the supervised learning process, which is very time-consuming and difficult to achieve in practical applications.

[0012] 4) When the defective area occupies only a small part of the entire inspection area, the number of inspection points in the defective area is significantly less than that in the non-defective area. The non-defective signals dominate in number, resulting in a skewed dataset, which affects the accuracy and reliability of subsequent analysis. This imbalance hinders the algorithm's ability to effectively extract significant features, causing the algorithm to favor majority class signals and ignore minority class signals.

[0013] 5) When using non-focused or diffuse sound fields for imaging, accurate feature extraction becomes extremely challenging due to the complexity of the signals in the defect edge area. These A-scan signals contain both defect and non-defect features. Relying only on obviously defect-free or defective signals is not enough to accurately classify these edge signals. In addition, when relying on prior knowledge to label the signal, these edge signals may exhibit different characteristics at different times, making accurate labeling difficult.

[0014] Therefore, a new imaging method is urgently needed to effectively solve the signal aliasing problem of multi-mode ultrasonic waves in laser ultrasonic testing, so as to accurately characterize sub-surface defects and improve detection accuracy. Summary of the invention

[0015] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for reconstructing sub-surface defect multi-mode laser ultrasonic C-scan images based on graph neural network. The method comprises the following steps:

[0016] (1) Performing a two-dimensional scan on the object to be measured to obtain a multimodal ultrasonic A-scan signal data set;

[0017] (2) converting the data set into a frequency domain signal and performing position encoding according to the original acquisition position of the signal;

[0018] (3) concatenating the frequency domain signal with the position code and calculating the similarity between the signals;

[0019] (4) Treat each data collection point as an independent node, use the frequency domain signal as the node feature, and select a set number of nodes with high similarity to each other, connect them with edges, and construct a graph dataset in this way;

[0020] (5) Inputting the graph data set into the trained graph neural network model, determining whether each A-scan signal originates from a defect area, obtaining a classification result, and reconstructing a C-scan image based on the original acquisition position of each A-scan signal and the classification result.

[0021] Compared with the prior art, the advantages of the present invention are that the sub-surface defect multi-mode laser ultrasonic C-scan image reconstruction method provided by the invention determines whether the laser ultrasonic A-scan signal comes from the defect area, reconstructs the C-scan image in combination with the signal acquisition position, and uses a deep learning model to extract the defect features in the signal, thereby replacing the traditional complex signal processing process, such as full focusing technology, empirical mode decomposition, synthetic aperture focusing technology, etc. In addition, it is considered that in laser ultrasonic detection, since the defect area only occupies a small part of the overall detection area, the number of detection points in the defect area is significantly less than that in the non-defective area, resulting in the non-defective signal dominating the data set, thereby causing the data set to be unbalanced and affecting the accuracy and reliability of subsequent analysis. Moreover, when using non-focused or diffuse sound field imaging, it is difficult to accurately extract features from the signals in the defect edge area due to its complexity. These signals contain both defect and non-defect features, and when relying on prior knowledge to label the signal, the signal at the edge of the defect may show different features at different times, making it difficult to accurately assign labels. Relying only on obvious non-defective or defective signals is not enough to accurately judge these edge signals. To address these problems, the present invention proposes a semi-supervised graph neural network model that combines a graph attention mechanism and an adversarial framework. By introducing an unbalanced and negatively sampled graph attention adversarial module, the problem of misjudgment of defect area signals caused by an unbalanced data set is solved, and the accuracy of defect boundary signal classification is improved. In short, the present invention can accurately characterize sub-surface defects and significantly improve the accuracy and reliability of detection.

[0022] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0024] Figure 1It is a schematic diagram of the principle of multi-mode ultrasonic aliasing in the prior art;

[0025] Figure 2 is a flow chart of a method for reconstructing a sub-surface defect multi-mode laser ultrasonic C-scan image according to an embodiment of the present invention;

[0026] Figure 3 is a schematic diagram of the overall process of a sub-surface defect multi-mode laser ultrasonic C-scan image reconstruction method according to an embodiment of the present invention;

[0027] Figure 4 is a schematic diagram of a non-contact laser ultrasonic detection system according to an embodiment of the present invention;

[0028] Figure 5 is a flow chart of an imbalanced and negatively sampled graph attention adversarial network according to one embodiment of the present invention;

[0029] Figure 6 It is a schematic diagram comparing experimental results of different methods according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0031] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0032] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0033] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0034] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0035] The present invention is applicable to a non-contact laser ultrasonic detection system. Generally speaking, the provided method includes: exciting and receiving laser ultrasonic signals; performing fast Fourier transform on the collected multi-modal ultrasonic aliasing signals to convert them into frequency domain signals, generating corresponding position information for the collection position of each signal, and combining it with the frequency domain signal to calculate the similarity between the signals, thereby constructing a sample data set; inputting the data set into a graph neural network model to determine whether each A-scan signal originates from a defective area; and reconstructing a C-scan image based on the original collection position of each A-scan signal and the model results.

[0036] Specifically, combined Figure 2 and Figure 3 As shown, the provided sub-surface defect multi-mode laser ultrasonic C-scan image reconstruction method includes the following steps:

[0037] Step S1, building a non-contact laser ultrasonic detection system.

[0038] Figure 4 It is a non-contact laser ultrasonic testing system. The system includes a data acquisition card, a computer, a pulsed laser, a dual-wave hybrid interferometer, a control unit, and a two-dimensional mobile scanning platform. First, the pulsed laser generates a pulsed laser, which is reflected by a reflector and focused into a line laser by a cylindrical lens, and then irradiated on the surface of the object to stimulate multi-mode ultrasonic waves. The ultrasonic signal is received by the dual-wave hybrid interferometer, then digitized and recorded by the data acquisition card, thereby obtaining a multi-mode ultrasonic A-scan time domain signal carrying the internal characteristics of the sample under test.

[0039] Step S2, using a non-contact laser ultrasonic detection system to collect multi-mode ultrasonic A-scan time domain signals and construct a sample image data set.

[0040] Using the non-contact laser ultrasonic testing system mentioned above, the metal additive sample to be tested is scanned in two dimensions to obtain a multi-mode aliasing data set A(x, y, t). Then, the time domain signal is converted into a frequency domain signal using the fast Fourier transform formula of formula (1), and then normalized according to formula (2) to reduce the impact of high-frequency noise and highlight key frequency features. These normalized signals are used as the input of the deep learning model.

[0041]

[0042] Where F[k] represents the kth frequency component of the signal s[n] in the frequency domain, s[n] represents the nth sampling point in the time domain signal; N is the total number of sampling points of the signal; k is the frequency index, and j is the imaginary unit, that is, is a complex exponential function.

[0043]

[0044] Among them, S norm represents the normalized result, s n is each frequency domain component of the frequency domain signal, s min is the minimum value of the frequency domain component, s max is the maximum value.

[0045] In order to ensure that the graph dataset can accurately capture the spatial proximity relationship between nodes, the absolute position encoding of the X and Y coordinates is introduced for each signal during the construction phase. The specific encoding formulas are shown in Equations (3) and (4). By encoding the position according to the acquisition location of the signal, it is effectively prevented that ultrasonic signals with different internal structural features due to their distant locations are mistakenly identified as similar, thereby avoiding incorrect or irrelevant node connections.

[0046]

[0047] Among them, i represents the acquisition position of the signal (X and Y coordinates), j represents the position in the position code, for example, when j is an even number, the generated code value is placed at the 2jth position of the position code, and when j is an odd number, the code value is placed at the 2j+1th position. d represents the total dimension size of the position code, that is, the length of the code.

[0048] Finally, the frequency domain signal is concatenated with the position code, and the similarity between the signals is calculated using formula (5), and a certain number of signals with the highest similarity are taken as the nodes connected to them.

[0049]

[0050] Among them, X k and Y k Respectively represent the signal v i and v j The kth dimension eigenvalue, N represents the dimension of the A-scan signal frequency component and position encoding. In one embodiment, the dimension of the position encoding is set to half of the frequency domain signal dimension to prevent the position encoding from excessively affecting the selection of neighboring nodes. Due to the variability of the relative positions of different defects in the scan area, the position encoding may interfere with the algorithm's ability to make accurate judgments during the training process, so the position encoding is only used during the dataset construction stage.

[0051] Step S3: Use the sample graph dataset to train the graph neural network model until the set loss target is met.

[0052] There are many types of graph neural network models, such as graph convolutional network GCN, graph autoencoder GAE, etc.

[0053] In one embodiment, the proposed graph neural network model is an inductive semi-supervised model for multimodal ultrasound signal classification, or an unbalanced and negatively sampled graph attention adversarial network. The process of C-scan image reconstruction using the graph neural network model is as follows: Figure 5 As shown in Figure 2. The model generally consists of an unbalanced graph attention adversarial network (or unbalanced graph attention adversarial module) and a negative sampling graph attention adversarial network (or negative sampling graph attention adversarial module). The frameworks of these two modules are composed of a generator and a discriminator, and both use the graph attention mechanism as a feature extractor. The key role of the graph attention mechanism is to extract features by calculating the attention coefficients of the target node and its neighboring nodes, see equations (6) and (7). At the same time, the generator generates synthetic data to challenge the discriminator, which must accurately distinguish between real data and false data. In the unbalanced graph attention adversarial network, the generator first generates false nodes to balance the number of nodes of different categories and form a uniformly distributed graph structure. The discriminator's graph attention mechanism extracts features from the uniformly distributed graph structure. In the negative sampling graph attention adversarial network, the generator combines the graph attention adversarial network with the probabilistic negative sampling strategy to generate high-quality negative samples. These negative samples help the discriminator's graph attention mechanism to effectively extract significant defect features from unlabeled signals. The label information is used to optimize the algorithm parameters through back propagation, thereby improving the classification accuracy. After the model converges, the C-scan image can be reconstructed based on the classification results and the original position of the signal.

[0054]

[0055] where v i represents the source node, v j Represents the source node v i Neighbor nodes connected by edges, e ij Indicates v i With v j The attention coefficient between Yes i The set of first-order neighbors, including v i itself, the purpose is to add self-loops in the calculation without losing its main features.

[0056] In one embodiment, by analyzing the similarity of frequency domain signals, a graph dataset with a non-Euclidean topology is constructed, which is labeled G=(V, E, A), where v∈V represents a node set, and each node corresponds to an ultrasonic signal received at a different location. E represents the connection relationship between these signals, and the adjacency matrix A is a two-dimensional array used to represent the connectivity between nodes, where a value of 1 indicates that there is a connection, and a value of 0 indicates that there is no connection, as shown in formula (8):

[0057]

[0058] (1) Unbalanced Graph Attention Adversarial Network

[0059] In the unbalanced subgraph In the example, V contains a smaller number of nodes v less and a larger number of nodes v more , the relationship between the two is |v less |<<|v more |. First, through a fully connected layer Generate fake nodes where |v im | indicates the number of false nodes generated by the generator |v im |=|v less |-|v more |,d o Represents the dimension of the data. Then, the adjacency matrix A between the false nodes and the minority nodes is calculated by formula (9): im .

[0060]

[0061] By calculating v im With v less The similarity between the generated nodes and the real nodes is used to determine the link weight between them, and normalization is performed to ensure that the sum of each row is 1. Finally, GAT (Graph Attention Network) is used as the discriminator. The input of GAT is a new graph with uniform distribution. Where V′ is the sum of v im The new node set v′∈V′ formed by the original node v. E′ is A′ is the new adjacency matrix obtained based on E′. For example, the loss function of the generator is expressed as:

[0062]

[0063] in, and The negative log likelihood loss is calculated based on the discriminator's evaluation of the false nodes, in order to make the false nodes more similar to the real nodes. is the predicted probability. It is to calculate the Euclidean distance between the node features after the false nodes and the minority real nodes are embedded, with the aim of making the generated node features close to the real minority node features. is the L2 regularizer, θ is the training weight set of the generator, and the regularization coefficient is α. The loss function of the discriminator is, for example, expressed as:

[0064]

[0065] in, and Both are binary cross-entropy. Used to determine whether the node is a false node or a real node. It is to distinguish between minority nodes and majority nodes. The purpose is to increase the embedding distance of real nodes of different categories. is the L2 regularizer, Ω is the training weight set. The regularization coefficient is β. In one embodiment, the training objective function of the unbalanced graph attention adversarial network is shown in formula (12):

[0066]

[0067] in, represents the adversarial loss function between the discriminator D and the generator G, is the expected symbol, used to calculate the average loss, v~P data(v) Indicates that from all data P data(v) Sampling node v, P data(v) Including real nodes and fake nodes generated by the generator, v im ~P im(vim) represents a fake node P generated from the generator im(vim) Sampling node v im ; is the loss function of the discriminator; is the loss function of the generator.

[0068] (2) Negative Sampling Graph Attention Adversarial Network

[0069] The present invention proposes a probability-based negative sampling graph adversarial module, which generates a certain number of high-quality negative samples (nodes of different categories) for the model by distinguishing the differences between labeled signals and unlabeled signals. This strategy aims to encourage embedding similarity between nodes of the same category, while ensuring that negative samples are far away from target nodes in the embedding space. While solving the gradient vanishing problem that may occur in the uniform negative sampling method, it improves the model's ability to identify complex signals in defect edge areas. First, the generator calculates the similarity between labeled signals and unlabeled signals, and based on this similarity, further calculates the probability of each unlabeled signal being a negative sample, and selects nodes with higher probability and appropriate number as negative samples. For example, the function of the generator is expressed as:

[0070]

[0071] Among them, v iis the current node, v′ j is a candidate node. NSG Represents the union of all node embeddings in the generator. Using the Softmax function to calculate the sum of all nodes will face the problem of extremely low efficiency, so the mini-batch method is used to optimize the calculation of Softmax. For the discriminator, its function is expressed as:

[0072]

[0073] Among them, v j Indicates the negative sampling node with higher probability, H NSD represents the union of all node embeddings in the discriminator. The loss function of the generator is expressed as:

[0074]

[0075] in, Indicates the similarity of positive samples (i.e., actually connected node pairs), ν p Represents the node v i Neighbor nodes connected by edges. Calculate the expected loss of negative sample pairs, H NSG and H NSD Represent the node embedding vectors from the generator and discriminator respectively, aiming to minimize the similarity of negative sample node pairs while maximizing the similarity of actually connected node pairs. It is to calculate the binary cross entropy according to its true label, aiming to learn the intrinsic characteristics of the data from the labeled data, where N is the number of labeled nodes. In short, the negative sampling generator samples high-quality negative nodes using a probability distribution. However, the embedding output of the generator is a discrete index. Therefore, the stochastic gradient descent method (SGD) cannot be used directly for optimization. The generator loss can be optimized using a policy gradient-based reinforcement learning method, expressed as:

[0076]

[0077] Among them, W NSG represents the parameter matrix in the generator, The gradient is By the discriminator D NS (·) Weighted calculation. In the field of reinforcement learning, D in the above formula can be NS (·) is considered as a reward function. If the discriminator judges that the data of the generator is real data, a "reward" can be provided to the generator. The generator is trained to generate data as realistic as possible to obtain higher rewards. For each negative sample pair (v i ,v j), the strategy adopted by the generator will penalize some low-quality negative sample pairs by reducing their corresponding probabilities and encourage the discriminator to assign high-quality negative vertices. The loss function of the discriminator is expressed as:

[0078]

[0079] in, represents a batch in the training process, v p Indicates that v i The directly connected nodes, v n is the negative node sampled using the generator, is the regularization term, ||·|| F is the Frobenius norm of the node feature, and λ is the regularized harmonization factor (for example, set to 1e-5). NS It can be optimized using gradient descent technology. The training objective function of the negative sampling graph attention adversarial network is shown in formula (18):

[0080]

[0081] in, represents the adversarial objective function between the discriminator D and the generator G, is the expected symbol, used to calculate the average loss; v j ~D(·|v i ;H NSD ) indicates that node vj is a negative sample of target node vi, and its node embedding vector in the discriminator D is H NSD ; Similarly, v j ~G(·|v i ;H NSG ) represents node v j The node v is generated by the generator G i Negative samples, whose node embedding vector in generator G is H NSG ; is the loss function of the discriminator; is the loss function of the generator.

[0082] By continuously iterating the basic process described above until the model converges, a trained graph neural network model can be obtained.

[0083] Step S4: For the multi-mode ultrasound aliasing signal actually collected, the classification result is obtained using the trained graph neural network model, and the C-scan image is reconstructed in combination with the original position of the signal.

[0084] After the graph neural network model is trained, it can be used for actual C-scan image reconstruction. For example, the specific application steps of the model include: performing a two-dimensional scan on the object to be measured to obtain a multi-mode ultrasonic A-scan signal data set; converting the data set into a frequency domain signal and performing position encoding according to the original acquisition position of the signal; splicing the frequency domain signal with the position encoding and calculating the similarity between the signals; based on the calculated similarity between the signals, selecting a set number of signals to construct a graph data set; inputting the graph data set into the trained graph neural network model to determine whether each A-scan signal originates from the defect area, obtain the classification result, and reconstruct the C-scan image based on the original acquisition position of each A-scan signal and the classification result.

[0085] In order to further verify the effect of the present invention, an experimental verification was carried out. In general, the experimental process includes: performing two-dimensional scanning and acquisition of A-scan signals on metal additive samples, converting the A-scan signals from the time domain to the frequency domain through fast Fourier transform, and establishing a graph data set based on the similarity between frequency domain signals. A graph neural network model is used to determine whether each A-scan signal originates from a defective area; and accurate reconstruction of the C-scan image of sub-surface defects is achieved based on the classification results and the original position of the signal.

[0086] For example, multiple AlSi10Mg and TC4 metal additive samples made by powder bed fusion technology and containing artificial subsurface defects of different shapes and sizes are scanned in two dimensions to obtain a multi-mode ultrasonic aliasing data set A(x,y,t), and the data set information is shown in Table 1.

[0087] Table 1 Specific information of metal additive manufacturing samples of different materials

[0088]

[0089] After the image reconstruction experiment was carried out using the present invention, the final result was compared with that of X-ray and scanning acoustic microscope (SAM). Figure 6 The experimental results show that compared with scanning acoustic microscopy imaging, the graph neural network-based method can more clearly outline the defect boundary. The relative error (RE) can be calculated by formula (19) to quantify the size deviation between the reconstructed image of the algorithm and the standard X-ray imaging. The results are shown in Table 2.

[0090]

[0091] Among them, L T Indicates the actual size of the defect. L PRepresents the defect size calculated by the model, which is obtained by calculating the sum of the areas of all points marked as defects.

[0092] Table 2 Dimensional deviation between reconstructed image and X-ray image

[0093] Dataset C1 C2 C3 C4 C5 RE 5.08％±0.15％ 7.21％±0.41％ 6.22％±0.14％ 7.17％±0.43％ 9.53％±0.35％

[0094] In summary, compared with the prior art, the present invention has the following advantages:

[0095] 1) The present invention uses laser ultrasonic non-contact non-destructive testing technology. Under the thermoelastic mechanism, pulsed laser is used to excite multi-mode ultrasonic waves, and dual-wave hybrid laser interferometer is used to receive ultrasonic signals to obtain A-scan signals. Then, the defect features in the A-scan signal are extracted through deep learning, replacing the traditional complex signal processing technology, judging whether the A-scan signal comes from the defect area, and reconstructing the C-scan image according to the original position of the signal, thereby achieving high-resolution and accurate imaging of sub-surface defects.

[0096] 2) Taking into account the technologies such as full focusing technology and synthetic aperture focusing technology that require complex signal processing methods, the present invention uses a non-full focusing non-contact laser ultrasound system combined with a deep learning model, which has higher application flexibility and can scan quickly, thereby providing results faster without waiting for the full focusing system to complete the complex focusing processing process.

[0097] 3) Usually, defect features are only extracted from a single signal. The present invention adopts a graph neural network, which replaces the traditional preprocessing process through its special feature extraction method. By capturing the spatial relationship and topological structure information between signals, the defect features are comprehensively extracted, thereby improving the accuracy of feature extraction.

[0098] 4) Considering that it is impossible to accurately label each signal in practical applications, the present invention adopts a semi-supervised deep learning model, which reduces the dependence on precise labels and enhances the generalization ability of the model when processing unlabeled data. In addition, considering that current deep learning models usually only extract features from a single signal, over-rely on the model's autonomous learning ability, and require accurate labels, the present invention adopts a semi-supervised graph neural network, which can capture the relationship between signals, improve the accuracy of feature extraction, and reduce the dependence on precise labels.

[0099] 5) In order to solve the data imbalance problem, the present invention constructs an unbalanced graph adversarial module to balance the potential defect features of the signal extracted from the training dataset by generating false nodes.

[0100] 6) The present invention uses probabilistic negative sampling to extract defect features from unlabeled signals by relying on labeled signals, effectively addressing the problem of mixed defect and non-defect features in edge signals and improving the accuracy of edge area signal classification.

[0101] 7) In terms of data processing and analysis, the present invention innovatively uses the graph attention mechanism as a feature extractor, effectively solving the problem of insufficient feature extraction capabilities of traditional methods.

[0102] It should be noted that, without violating the spirit and scope of the present invention, those skilled in the art may make appropriate changes or modifications to the above embodiments. For example, the present invention has universality and flexibility, and its detection object scope significantly exceeds the limitations of sub-surface defects, and can be effectively applied to defect identification in deep and complex structures. The present invention is not limited to the detection of metal additive manufacturing samples, but shows wide compatibility with various manufacturing methods and material types. For example, for traditional subtractive manufacturing, casting, emerging 3D printing, composite material preparation, etc., the present invention can flexibly adapt to achieve efficient and accurate non-destructive testing. Although the core of the present invention focuses on laser ultrasonic non-destructive testing technology, its design concept and implementation path reserve sufficient space for the integration of diversified technical means, such as piezoelectric ultrasound.

[0103] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0104] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0105] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0106] The computer program instructions for performing the operation of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, Python, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present invention.

[0107] Various aspects of the present invention are described herein with reference to the flow charts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each box of the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer-readable program instructions.

[0108] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0109] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0110] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that it is equivalent to implement it by hardware, implement it by software, and implement it by combining software and hardware.

[0111] Embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the marketplace, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A sub-surface defect multi-mode laser ultrasonic C-scan image reconstruction method, comprising the following steps: Performing a two-dimensional scan on the object to be measured to obtain a multi-mode ultrasonic A-scan time domain signal data set; Converting the data set into a frequency domain signal and performing position encoding according to the acquisition position of the signal; splicing the frequency domain signal with the position code, and calculating the similarity between the signals; Construct a graph dataset. The graph dataset is constructed by treating each data collection point as an independent node, using the frequency domain signal as the node feature, and selecting a set number of nodes whose similarities exceed a set threshold and connecting them with each other through edges. The graph data set is input into the trained graph neural network model to determine whether each A-scan signal originates from a defect area, obtain a classification result, and reconstruct a C-scan image based on the original acquisition position of each A-scan signal and the classification result.

2. The method according to claim 1, characterized in that The graph neural network model includes a first graph attention adversarial network and a second graph attention adversarial network. The first graph attention adversarial network includes a first generator and a first discriminator, and the second graph attention adversarial network includes a second generator and a second discriminator. The first generator is used to generate false sample nodes, and the first discriminator is used to extract features through the first graph neural network and judge the category and authenticity of the nodes. The second generator is used to generate a negative sample set, and the second discriminator extracts features through the second graph neural network and judges the categories of different nodes.

3. The method according to claim 1, characterized in that: In the process of training the graph neural network model, the loss function of the first graph attention adversarial network is set to: in: in, represents the adversarial loss function between the first discriminator D and the first generator G, is the expectation symbol, is the loss function of the first generator, is the loss function of the first discriminator, and The negative log-likelihood loss is calculated based on the discriminator's evaluation of the false nodes. is the predicted probability, It is the Euclidean distance between the false nodes and the embedded node features of a few real nodes. is the L2 regularizer, θ is the training weight set of the generator, the regularization coefficient is α, and are both binary cross entropy, Used to determine whether the node is a false node or a real node. Used to distinguish different types of nodes. is the L2 regularizer, Ω is the training weight set, the regularization coefficient is β, and v less Indicates a small number of nodes, v more Indicates a large number of nodes, |v im |=|v less |-|v more | represents the number of false nodes generated by the generator, d o Indicates the dimension of the data, v~P data(v) Indicates that from all data P data(v) Sampling node v, P data(v) Including real nodes and fake nodes generated by the first generator, v im ~P im(vim) represents the false node P generated from the first generator im(vim) Sampling node v im .

4. The method according to claim 1, characterized in that In the process of training the graph neural network model, the loss function of the second graph attention adversarial network is set to: Among them, G NS is the function of the second generator, expressed as: Among them, v i is the node, v′ j is a candidate node, H NSG represents the union of all node embeddings in the second generator; D NS is the function of the second discriminator, expressed as: Among them, v j Represents the current node v i Negative sample nodes, H NSD represents the union of all node embeddings in the second discriminator, is the loss function of the second generator, is the loss function of the second discriminator, represents the adversarial objective function between the second discriminator D and the second generator G, is the expectation symbol; v j ~D(·|v i ;H NSD ) indicates that node vj is a negative sample of node vi, and its node embedding vector in the second discriminator D is H NSD ;v j ~G(·|v i ;H NSG ) indicates that node vj is generated by the second generator G. i The negative sample, whose node embedding vector in the second generator G is H NSG .

5. The method according to claim 4, characterized in that The loss function of the second generator Set to:

6. The method according to claim 4, characterized in that The loss function of the second discriminator Set to: in, Indicates the similarity of calculating positive samples. Indicates the expected value of the loss of the negative sample pair, H NSG represents the node embedding vector from the second generator, H NSD represents the node embedding vector from the second discriminator, It is to calculate the binary cross entropy based on the true label, where N is the number of nodes with labels; Represents a batch of data in the training process, v p Indicates that v i The directly connected nodes, v n are negative vertices sampled using the generator, is the regularization term, ||·|| F is the Frobenius norm of the node feature, and λ is the regularization harmonic factor.

7. The method according to claim 1, characterized in that The acquired signal is position-encoded according to the following formula: Among them, i represents the acquisition position of the signal, j represents the position in the position encoding, and d represents the total dimension of the position encoding.

8. The method according to claim 1, characterized in that The similarity between signals is calculated according to the following formula: Among them, X k and Y k Respectively represent the signal v i and v j The eigenvalue of the kth dimension, N represents the dimension of the frequency component and position encoding of the A-scan signal.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images