A method and system for structural damage detection

Abstract

The application provides a structure damage detection method and system, comprising the following steps: first, a first longitudinal wave signal of a first preset frequency and a second longitudinal wave signal of a second preset frequency are coaxially and oppositely injected into a detection area along a first axis, and a first shear wave signal of a third preset frequency is injected into the detection area at a plurality of preset exit angles and not collinear with the first axis, so as to generate a mixed frequency sound wave signal through a target damage nonlinear effect. Then, the echo signal of each preset scanning point in the detection area is collected. Secondly, a multi-modal graph structure data is constructed based on the echo signal of the preset scanning point. Finally, a damage detection result is determined based on a multi-modal graph structure data and a civil defense graph convolutional neural network model. The method is based on a three-wave non-collinear mixing technology, and converts a multi-sound field excitation problem into a graph structure problem, accurately determines the damage detection result through a graph convolutional neural network model, and thus can effectively improve the accuracy and reliability of the material component structure damage detection.

Description

Technical Field

[0001] This invention relates to the field of ultrasonic nondestructive testing technology, specifically to a method and system for detecting structural damage. Background Technology

[0002] In aerospace, high-end equipment manufacturing, and other fields, composite metal materials are widely used in critical components due to their advantages such as high strength and lightweight. However, under extreme environments such as high temperature and high pressure, they are prone to micro-damage, which, if not detected in time, may lead to serious accidents. Therefore, accurate detection of micro-damage in composite structures is crucial for ensuring the safe operation of equipment and extending its service life.

[0003] Currently, nonlinear ultrasonic testing is commonly used as a method for micro-damage detection. However, existing nonlinear ultrasonic testing techniques face numerous technical bottlenecks and challenges when applied to manufacturing mechanical composite metal structures under extreme service environments. On one hand, the nonlinear component transformation mechanism and vibration-acoustic nonlinear modulation effect of multi-source, multi-frequency nonlinear acoustic waves under micro-damage interference remain unclear, leading to difficulties in the analysis and interpretation of the detection signals. On the other hand, existing technologies do not fully consider the influence of the external environment and boundary conditions during the testing process, resulting in a lack of sufficient flexibility and universality in practical applications, making it difficult to adapt to complex and changing engineering environments. Furthermore, existing nonlinear multivariable control methods rely on a large number of existing reference samples for learning; in practice, obtaining comprehensive and accurate reference samples is often quite difficult, limiting the widespread application of this method.

[0004] Therefore, there is an urgent need for a structural damage detection method that can solve the problems of low detection accuracy and poor reliability in existing detection technologies, so as to achieve accurate detection and spatial positioning of structural damage. Summary of the Invention

[0005] The technical problem to be solved by this invention is the inability to effectively and accurately detect and determine damage in a structure.

[0006] To address the aforementioned technical problems, this invention provides a structural damage detection method and system, specifically employing the following technical solution:

[0007] In a first aspect, the present invention provides a structural damage detection method. The method includes: First, acquiring echo signals at each preset scanning point within a detection area. The echo signal is a mixed acoustic signal generated by the nonlinear effect of a first longitudinal wave signal, a second longitudinal wave signal, and a first shear wave signal at the preset scanning point through the target damage. The frequency of the first longitudinal wave signal is a first preset frequency, the frequency of the second longitudinal wave signal is a second preset frequency, and the frequency of the first shear wave signal is a third preset frequency, the third preset frequency being the difference between the first and second preset frequencies. The first and second longitudinal wave signals are incident on the detection area at a relatively oblique angle along a first axis and resonate to generate a difference-frequency shear wave. The first shear wave signal is incident on the detection area at multiple preset exit angles and mixed with the difference-frequency shear wave; the exit point of the first shear wave signal is not located on the first axis. Then, multimodal graph structural data is constructed based on the echo signals at each preset scanning point. The multimodal graph structural data is used to characterize the nodal features and acoustic field features of the preset scanning points. Finally, the multimodal graph structural data is input into a graph convolutional neural network model, and the damage detection result is output.

[0008] This method first collects echo signals from each preset scanning point within the detection area. Then, it constructs multimodal graph structure data based on the echo signals from each preset scanning point. Finally, it inputs the multimodal graph structure data into a graph convolutional neural network model to output the damage detection result. This method is based on three-wave non-collinear mixing technology and transforms the multi-field excitation problem into a graph structure problem based on multimodal graph structure data. The graph convolutional neural network model can accurately determine the damage detection result, achieving precise reconstruction of the three-dimensional acoustic signal features of multi-source targeted excitation micro-damage characteristic signals. By integrating three-dimensional feature information with graph convolutional neural network modeling technology, it can spatially identify and precisely locate early micro-damage in ultrathin composite layers and connecting layers of composite metal materials, thereby effectively improving the accuracy and reliability of damage detection in material components.

[0009] In conjunction with the first aspect, in one alternative implementation, the above-mentioned construction of multimodal graph structure data based on the echo signal of each preset scanning point includes: First, extracting the corresponding sum-frequency amplitude, fundamental frequency amplitude, and difference-frequency amplitude from the echo signal of each preset scanning point. Then, determining the mixing nonlinear parameters based on the sum-frequency amplitude, fundamental frequency amplitude, and difference-frequency amplitude, wherein the expression for the mixing nonlinear parameters is:

[0010]

[0011] Where, N new The mixing nonlinear parameters are represented by A1, A2, and A3, respectively. Finally, multimodal graph structure data is constructed based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle.

[0012] In conjunction with the first aspect, in one alternative implementation, the construction of multimodal graph structure data based on mixing nonlinear parameters, preset scan point position information, and preset emission angle includes: First, constructing an initial three-dimensional data matrix based on the mixing nonlinear parameters, preset scan point position information, and preset emission angle. Then, normalizing the initial three-dimensional data matrix to determine the normalized three-dimensional data matrix. Next, performing PARAFAC tensor decomposition on the normalized three-dimensional data matrix to determine the spatial factor matrix. Second, extracting data based on the spatial factor matrix to determine the initial point cloud coordinate set. Further, scaling and restoring the initial point cloud coordinate set to determine the restored point cloud coordinate set. Finally, constructing the multimodal graph structure data based on the restored point cloud coordinate set and the mixing nonlinear parameters.

[0013] In conjunction with the first aspect, in one alternative implementation, the aforementioned multimodal graph structure data includes: node data, node feature data, physical proximity edge data, and acoustic field feature edge data. Node data is used to characterize the point cloud in the reconstructed point cloud coordinate set within the target region, which is included within the detection region. Node feature data includes: the three-dimensional coordinate values ​​corresponding to the point cloud, mixing nonlinear parameters, and node identifiers; physical proximity edge data is used to characterize the geometric connection characteristics of the scan points within the target region; and acoustic field feature edge data is used to characterize the acoustic field spatial association characteristics of the scan points within the target region.

[0014] In conjunction with the first aspect, in one alternative implementation, the above-mentioned PARAFAC tensor decomposition of the normalized 3D data matrix to determine the spatial factor matrix includes: First, performing a rank-3 PARAFAC tensor decomposition on the normalized 3D data matrix to obtain the PARAFAC tensor decomposition result. Then, determining the spatial factor matrix based on the PARAFAC tensor decomposition result. The expression for the PARAFAC tensor decomposition result is:

[0015]

[0016] In the formula, Let a represent the normalized three-dimensional data matrix. r Denotes the spatial factor matrix, b r Represents the damage intensity factor matrix, c r Let r represent the angular response factor matrix, and r represent the rank of the decomposition. The symbol represents the outer product operation.

[0017] In conjunction with the first aspect, in one alternative implementation, the extraction of the corresponding sum-frequency amplitude, fundamental amplitude, and difference-frequency amplitude based on the echo signal at each preset scanning point includes: First, performing a Fourier transform on the echo signal at each preset scanning point to obtain a spectrum diagram corresponding to each echo signal. Then, extracting the sum-frequency amplitude, fundamental amplitude, and difference-frequency amplitude based on the spectrum diagram corresponding to the echo signal.

[0018] In conjunction with the first aspect, in one alternative implementation, the aforementioned graph convolutional neural network model includes: multiple convolutional layers and a prediction output layer; inputting multimodal graph structure data into the graph convolutional neural network model and outputting damage detection results includes: first, inputting the multimodal graph structure data into multiple convolutional layers to aggregate neighboring node features and obtain convolutional layer output results; then, inputting the convolutional layer output results into the prediction output layer for damage prediction and outputting damage detection results.

[0019] In conjunction with the first aspect, in one alternative implementation, the aforementioned damage detection result includes the probability and risk level of damage existing at the corresponding location of each node. Specifically, if the probability of damage existing at the corresponding location of a node is less than or equal to a first probability threshold, the risk level is level one; if the probability of damage existing at the corresponding location of a node is greater than the first probability threshold and less than or equal to a second probability threshold, the risk level is level two; and if the probability of damage existing at the corresponding location of a node is greater than the second probability threshold, the risk level is level three. The expression for the damage detection result is:

[0020] P i =sigmoid(Wh i (L) +b);

[0021] Among them, P i Let represent the probability that there is damage at the position corresponding to the i-th node, sigmoid() represent the activation function, W represent the training weight parameters, b represent the training bias parameters, and h represent the probability that there is damage at the i-th node. i (L) This indicates the output result of the three-body interaction.

[0022] In conjunction with the first aspect, in one alternative implementation, the method further includes: displaying the node's location with a first preset color when the node's risk level is first; displaying the node's location with a second preset color when the node's risk level is second; and displaying the node's location with a third preset color when the node's risk level is third. The first, second, and third preset colors are all different from each other.

[0023] Secondly, the present invention provides a structural damage detection system, comprising: a first transmitting probe, a second transmitting probe, a third transmitting probe, a first receiving probe, and a processing system. The first and second transmitting probes are coaxially arranged opposite each other along a first axis, while the third transmitting probe is not located on the first axis. Specifically, the first transmitting probe can be used to obliquely inject a first longitudinal wave signal of a first preset frequency into the detection area. The second transmitting probe can be used to obliquely inject a second longitudinal wave signal of a second preset frequency into the detection area, so that the second longitudinal wave signal resonates with the first longitudinal wave signal to generate a difference-frequency transverse wave. The third transmitting probe can be used to inject a first shear wave signal of a third preset frequency into the detection area at multiple preset emission angles, so that the first shear wave signal mixes with the difference-frequency transverse wave, where the third preset frequency is the difference between the first preset frequency and the second preset frequency. The first receiving probe can be used to collect echo signals at each preset scanning point within the detection area; the echo signal is a mixed acoustic signal generated by the nonlinear effect of the target damage at the preset scanning point, consisting of the first longitudinal wave signal, the second longitudinal wave signal, and the first shear wave signal. The processing system can be used to construct multimodal graph structure data based on the echo signal of each preset scanning point. The multimodal graph structure data is used to characterize the node features and acoustic field features of the preset scanning points. The multimodal graph structure data is input into a graph convolutional neural network model, and the damage detection results are output.

[0024] Thirdly, the present invention provides an electronic device, comprising: a memory and one or more processors; the memory being coupled to the processors; wherein the memory stores computer program code, the computer program code including computer instructions, which, when executed by the processor, cause the electronic device to perform the method provided by the first aspect and any of its alternative implementations.

[0025] Fourthly, the present invention provides a computer-readable storage medium including computer instructions that, when executed on an electronic device, cause the electronic device to perform the method provided by the first aspect and any alternative implementation thereof.

[0026] Understandably, the beneficial effects of the structural damage detection system provided in the second aspect, the electronic equipment in the third aspect, and the computer-readable storage medium in the fourth aspect can be referenced to the beneficial effects of the first aspect and any of its possible design embodiments, and will not be repeated here. Attached Figure Description

[0027] Figure 1 A schematic flowchart of the structural damage detection method provided in the embodiments of this application;

[0028] Figure 2 A schematic diagram of the scanning path provided in the embodiments of this application;

[0029] Figure 3 This is a schematic diagram showing the distribution of probes provided in the embodiments of this application;

[0030] Figure 4 A flowchart illustrating the method for constructing multimodal graph structure data provided in an embodiment of this application;

[0031] Figure 5 This is a schematic diagram of the architecture of the structural damage detection system provided in an embodiment of this application. Detailed Implementation

[0032] The embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described below do not represent all embodiments consistent with this application. They are merely examples of systems and methods consistent with some aspects of this application as detailed in the claims.

[0033] In aerospace, high-end equipment manufacturing, and other fields, composite metal materials are widely used in critical components due to their advantages such as high strength and lightweight. However, under extreme environments such as high temperature and high pressure, they are prone to micro-damage, which, if not detected in time, may lead to serious accidents. Therefore, accurate detection of micro-damage in composite structures is crucial for ensuring the safe operation of equipment and extending its service life.

[0034] Currently, nonlinear ultrasonic testing is commonly used as a method for micro-damage detection. However, existing nonlinear ultrasonic testing techniques face numerous technical bottlenecks and challenges when applied to manufacturing mechanical composite metal structures under extreme service environments. On one hand, the nonlinear component transformation mechanism and vibration-acoustic nonlinear modulation effect of multi-source, multi-frequency nonlinear acoustic waves under micro-damage interference remain unclear, leading to difficulties in the analysis and interpretation of the detection signals. On the other hand, existing technologies do not fully consider the influence of the external environment and boundary conditions during the testing process, resulting in a lack of sufficient flexibility and universality in practical applications, making it difficult to adapt to complex and changing engineering environments. Furthermore, existing nonlinear multivariable control methods rely on a large number of existing reference samples for learning; in practice, obtaining comprehensive and accurate reference samples is often quite difficult, limiting the widespread application of this method.

[0035] Therefore, there is an urgent need for a structural damage detection method that can solve the problems of low detection accuracy and poor reliability in existing detection technologies, so as to achieve accurate detection and spatial positioning of structural damage.

[0036] To address the aforementioned issues, this application provides a structural damage detection method and system. This method can be applied to the detection and localization of damage (cracks, defects) in composite metal materials used in major equipment such as aerospace, energy, and high-speed rail, and is particularly suitable for detecting internal microcracks in complex curved surface workpieces. Specifically, this method is based on three-wave non-collinear mixing and mechanical scanning, significantly enhancing the nonlinear response in the damaged portion of the material component. Echo signals are acquired and multimodal graph structure data is constructed. Finally, based on the multimodal graph structure data and a graph convolutional neural network model, the damage detection results can be obtained. This method, based on three-wave non-collinear mixing technology and multimodal graph structure data, transforms the multi-field excitation problem into a graph structure problem. The graph convolutional neural network model accurately determines the damage detection results, achieving precise reconstruction of the three-dimensional acoustic signal features of multi-source targeted excitation micro-damage characteristic signals. By integrating three-dimensional feature information and graph convolutional neural network modeling technology, it can spatially identify and precisely locate early micro-damage in ultrathin composite layers and connecting layers of composite metal materials, thereby effectively improving the accuracy and reliability of damage detection in material components.

[0037] The solutions provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0038] Specifically, Figure 1 This is a schematic flowchart of the structural damage detection method provided in the embodiments of this application, as shown below. Figure 1 As shown, the structural damage detection method provided in this application includes the following steps S101-S103:

[0039] S101. Collect the echo signal of each preset scanning point within the detection area.

[0040] In this embodiment, the detection area is the region in the structural component of the material to be tested where the target damage exists. The target damage can be cracks, structural defects, etc. Taking a 20mm×20mm rectangular area as an example, the preset scanning points can be gridded scanning points generated with a step size of 0.5mm.

[0041] In one implementation, each preset scan point can be scanned along a zigzag path to acquire the echo signal of each preset scan point. For example, Figure 2 A schematic diagram of the scanning path provided in the embodiments of this application, as shown below. Figure 2 As shown, each preset scanning point can be scanned sequentially along the "bow"-shaped path 201 to collect echo signals.

[0042] The echo signal is a mixed acoustic signal generated by the nonlinear effect of the target damage on the first longitudinal wave signal, the second longitudinal wave signal, and the first shear wave signal at a preset scanning point. The frequency of the first longitudinal wave signal is a first preset frequency, the frequency of the second longitudinal wave signal is a second preset frequency, and the frequency of the first shear wave signal is a third preset frequency, which is the difference between the first and second preset frequencies. The first and second longitudinal wave signals are injected into the detection area at a relatively oblique angle along the first axis and resonate to generate a difference-frequency shear wave; the first shear wave signal is injected into the detection area at multiple preset exit angles and mixes with the difference-frequency shear wave. The exit point of the first shear wave signal is not located on the first axis.

[0043] For example, the first preset frequency can be 5MHz, the second preset frequency can be 4MHz, and the third preset frequency is 1MHz. The first and second preset frequencies can be preset according to the needs of actual applications, and this application does not specifically limit them.

[0044] Specifically, in the embodiments of this application, the generation of collinear difference frequency waves and the third oblique incident wave are combined to achieve three-wave non-collinear mixing, which significantly enhances the nonlinear response (nonlinear mixing effect) at the target damage site.

[0045] The nonlinear mixing effect is as follows: when the frequencies of two collinear shear waves and longitudinal waves (shear waves) meet certain conditions, their mixing can generate a new shear wave (i.e., a difference-frequency shear wave). The frequency of the difference-frequency shear wave is the difference between the fundamental frequency (i.e., the first preset frequency and the second preset frequency), its propagation direction is opposite to that of the fundamental shear wave, and its wave speed is the same as that of the fundamental shear wave. Limitations of collinear mixing include: when performing nonlinear ultrasonic mixing detection on materials, if the two incoming acoustic waves are parallel to each other or on the same axis, the technique is called collinear mixing detection. When both incoming acoustic waves are shear waves, the amplitude of the mixed wave generated by their interaction in the nonlinear region does not accumulate with propagation in the mixing region, and it can only propagate within the mixing region, unable to propagate freely outside the mixing region. However, nonlinear mixing has the following advantages: when two shear waves and longitudinal waves of specific frequencies propagate nonlinearly, their interaction in the nonlinear region can generate a third wave (such as a difference-frequency shear wave) that can propagate freely. In particular, by adjusting the incident angles of the two longitudinal waves to meet the transverse wave resonance conditions, the difference frequency energy can be efficiently converted into a transverse wave of a specific frequency.

[0046] In this embodiment, specifically, two transmitting probes (i.e., the first transmitting probe and the second transmitting probe) emit two collinearly propagating high-frequency ultrasonic waves (i.e., the first longitudinal wave signal and the second longitudinal wave signal). These generate a low-frequency difference-frequency signal within the material structure (target damage) through contact acoustic nonlinearity, forming a highly directional difference-frequency wave (i.e., a difference-frequency shear wave). Then, a third oblique-incident probe (i.e., the third transmitting probe) emits a signal matched to the difference-frequency shear wave. An angle-adjustable design achieves nonlinear mixing of the three waves. This nonlinear configuration at the target damage location (e.g., the plastic zone at the crack tip or the crack itself) excites strong nonlinear interactions, generating new frequency components (especially sum-frequency signals). A receiving probe (i.e., the first receiving probe) captures the modulated sideband signal (i.e., the echo signal). By adjusting the incident angle of the first shear wave signal, spatial control of the mixing region can be achieved, thereby enabling selective detection of damage in specific directions.

[0047] For example, Figure 3 This is a schematic diagram of the probe setup provided in the embodiments of this application, such as... Figure 3 As shown in (a), (b), and (c), the first transmitting probe Tx1 and the second transmitting probe Tx2 are coaxially arranged opposite each other on the material structure component 300 to be tested along the first axis 301. The third transmitting probe Tx3 is not located on the first axis 301; instead, it can be positioned on the center line of the material structure component 300. The first receiving probe Rx can be positioned at the center of the probe group (i.e., the first transmitting probe Tx1, the second transmitting probe Tx2, and the third transmitting probe Tx3).

[0048] The preset emission angle of the third transmitting probe Tx3 can be set from 30° to 70°, with 9 angles (i.e., 30°, 35°, ..., 70°) in 5° increments.

[0049] In order for the first longitudinal wave signal emitted by the first transmitting probe Tx1 and the second transmitting probe Tx2 to resonate and generate a difference-frequency transverse wave, the wave vector angle φ between the first longitudinal wave signal and the second longitudinal wave signal inside the material structure component to be tested needs to satisfy:

[0050]

[0051] in,

[0052]

[0053] In the formula, c represents the transverse and longitudinal wave velocity ratio, c p c represents the longitudinal wave velocity. s c represents the transverse wave velocity. p and c sThe wave velocity of the material structure component to be tested can be obtained by measuring the wave velocity of the acrylic wedge alone. 'a' represents the frequency ratio, 'f1' represents the first preset frequency, and 'f2' represents the second preset frequency.

[0054] The refraction angles of the first and second longitudinal wave signals within the structure of the material to be tested are determined by geometric relationships. Using Snel's theorem, the refraction angle γ can be determined to satisfy:

[0055]

[0056] S102. Construct multimodal graph structure data based on the echo signal of each preset scanning point.

[0057] Furthermore, the echo signals acquired by S101 are analyzed and processed to convert them into graph data, constructing multimodal graph structure data. This transforms the multi-field coupling problem into a graph structure problem, facilitating damage localization using a graph convolutional neural network model. The multimodal graph structure data can be used to characterize the nodal features and acoustic field features of preset scanning points.

[0058] In some embodiments, S102 may specifically include the following steps S1021-S1023:

[0059] S1021. Extract the corresponding sum frequency amplitude, fundamental frequency amplitude and difference frequency amplitude based on the echo signal of each preset scanning point.

[0060] Specifically, the amplitude of the echo signal can first be extracted to determine the mixing nonlinear parameters.

[0061] In some embodiments, S1021 may specifically include: first, performing a Fourier transform on the echo signal at each preset scanning point to obtain a spectrum diagram corresponding to each echo signal. Then, extracting the sum frequency amplitude, fundamental frequency amplitude, and difference frequency amplitude based on the spectrum diagram corresponding to the echo signal.

[0062] S1022. Determine the mixing nonlinear parameters based on the sum-frequency amplitude, fundamental frequency amplitude, and difference-frequency amplitude. The expression for the mixing nonlinear parameters is:

[0063]

[0064] Where, N new The parameters represent the mixing nonlinearity, with A1 representing the sum frequency amplitude, A2 representing the fundamental frequency amplitude, and A3 representing the difference frequency amplitude.

[0065] Specifically, the sum-frequency amplitude A1 is the core nonlinear response signal, and its intensity directly reflects the strength of the nonlinear mixing effect caused by structural damage. The fundamental amplitude A2 can be used to characterize one of the original high-frequency excitation energies incident on the structural damage region. The difference-frequency amplitude A3 can be used to characterize the low-frequency excitation energy incident on the structural damage region.

[0066] Mixing nonlinear parameter N new It is a normalized nonlinear parameter. The numerator and sum-frequency amplitude A1 can be used to characterize the result (the sum-frequency signal intensity generated by the nonlinear effect). The denominator, the basis wave amplitude A2 and the difference-frequency amplitude A3 can be used to characterize the cause.

[0067] The more severe the structural damage (e.g., larger crack opening size, length, or plastic zone), the stronger its nonlinear behavior (e.g., "breathing effect" or plastic deformation), and the larger the sum-frequency amplitude A1 produced under the same incident energy. Therefore, the mixing nonlinear parameter N... new The higher the value, the better. Mixing nonlinear parameter N new It can quantify the degree of nonlinear damage in local areas.

[0068] As damage (taking a crack as an example) propagates, the stress concentration area at the crack tip increases, leading to an enhanced acoustic nonlinear response and an increase in the mixing nonlinear parameter N. new The value of will increase significantly. When the crack partially closes under certain conditions (such as compressive loading), the acoustic nonlinear response will weaken, and the mixing nonlinear parameter N will increase. new The value of will decrease. Therefore, through the mixing nonlinear parameter N new The mixing nonlinear parameter N can characterize the degree of structural damage. new It can be effectively used as a detection indicator for structural damage.

[0069] S1023. Construct multimodal graph structure data based on mixing nonlinear parameters, preset scanning point position information, and preset emission angle.

[0070] Furthermore, based on the mixing nonlinear parameters determined in S1022, as well as the position information of the preset scanning points and the preset emission angle, multimodal graph structure data can be constructed.

[0071] In some embodiments, Figure 4 This is a flowchart illustrating the method for constructing multimodal graph structured data provided in an embodiment of this application, as shown below. Figure 4 As shown, S1023 can be specifically implemented through the following steps S201-S206:

[0072] S201. Construct an initial three-dimensional data matrix based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle.

[0073] Specifically, based on the mixing nonlinear parameter N new And the location information of the preset scanning points and the preset emission angle (x, y, θ) i By combining the data, an initial three-dimensional data matrix can be constructed.

[0074] For example, taking a scenario with 9 preset emission angles, each corresponding to 40×40 preset scanning points, the initial 3D data matrix X can be represented as:

[0075] X∈R 40×40×9 .

[0076] S202. Normalize the initial three-dimensional data matrix to determine the normalized three-dimensional data matrix.

[0077] Then, the initial three-dimensional data matrix can be normalized to eliminate the dimensional differences of the original data, mapping data of different dimensions or magnitudes to a unified scale, and obtaining a normalized three-dimensional data matrix.

[0078] Specifically, the global mean and standard deviation can first be determined based on the initial 3D data matrix X, and then normalization can be performed based on the global mean and standard deviation. The normalized 3D data matrix... The expression is:

[0079]

[0080] in, This represents the normalized three-dimensional data matrix. This represents the global mean. It represents the standard deviation.

[0081] S203. Perform PARAFAC tensor decomposition on the normalized three-dimensional data matrix to determine the spatial factor matrix.

[0082] In some embodiments, the normalized three-dimensional data matrix can be decomposed into a rank-3 PARAFAC tensor to obtain the PARAFAC tensor decomposition result. Then, the spatial factor matrix is ​​determined based on the PARAFAC tensor decomposition result.

[0083] The expression for the PARAFAC tensor decomposition result is as follows:

[0084]

[0085] In the formula, Let a represent the normalized three-dimensional data matrix. r Denotes the spatial factor matrix, b r Represents the damage intensity factor matrix, c rLet r represent the angular response factor matrix, and r represent the rank of the decomposition. The symbol represents the outer product operation.

[0086] Specifically, using the rank-3 PARAFAC tensor decomposition balances decomposition accuracy and processing efficiency. The space factor matrix a r The column vector is essentially a three-dimensional point cloud (XYZ coordinate point cloud) containing the damage. Damage intensity factor matrix b r The damage intensity of each point in the point cloud can be quantified, and its value is related to the mixing nonlinear parameter N. new Directly related.

[0087] S204. Extract data based on the spatial factor matrix to determine the initial point cloud coordinate set.

[0088] Specifically, the spatial factor matrix a can be extracted. r The first three columns of the first spatial factor matrix a1 are used as the initial point cloud coordinate set, and the initial point cloud coordinate set P raw for:

[0089] P raw =[a1(:;1),a1(:;2),a1(:;3)];

[0090] Where a1(:;1) represents the first column of the first spatial factor matrix a1, serving as the X coordinate. a1(:;2) represents the second column of the first spatial factor matrix a1, serving as the Y coordinate. a1(:;3) represents the third column of the first spatial factor matrix a1, serving as the Z coordinate.

[0091] S205. Perform coordinate scaling and restoration on the initial point cloud coordinate set to determine the restored point cloud coordinate set.

[0092] Specifically, reconstruct the point cloud coordinate set P mm The expression is:

[0093]

[0094] Where S1 represents the size of the detection region, and max(a1) represents the maximum value in the first spatial factor matrix a1, i.e., the normalization coefficient.

[0095] S206. Construct multimodal graph structure data based on the restored point cloud coordinate set and mixing nonlinear parameters.

[0096] Finally, based on the set of restored point cloud coordinates determined in S205 and the mixing nonlinear parameters determined in S1022, multimodal graph structure data can be constructed.

[0097] In some embodiments, the multimodal graph structure data may specifically include: node data, node feature data, physical proximity edge data, and acoustic field feature edge data.

[0098] Node data is used to represent the point cloud in the reconstructed point cloud coordinate set within the target area, which is included in the detection area. For example, taking a 20mm × 20mm rectangular area as the detection area, a 1mm × 1mm rectangular area can be selected as the target area based on the initially determined location of the target damage, thereby reducing the amount of data processed and improving detection efficiency.

[0099] Node feature data includes: the 3D coordinates of the point cloud, mixing nonlinear parameters, and node identifiers. Specifically, node feature data can include node feature vectors.

[0100]

[0101] Among them, h i Let x represent the node feature vector of the i-th node. i ,y i ,z i () represents the three-dimensional coordinate value corresponding to the i-th node. This represents the mixing nonlinear parameters of the i-th node, where i represents the node identifier.

[0102] Physical proximity edge data is used to characterize the geometric connectivity features of scanned points within a target region. Physical proximity edge data includes the physical proximity edge features between two nodes.

[0103] Specifically, physically adjacent edges can connect nodes with a distance less than a first distance threshold (e.g., 0.5 mm) to represent the geometric continuity of the material. This ensures that adjacent nodes (distance < 0.5 mm) share material continuity constraints, conforming to the laws of solid mechanics. The characteristics of the physically adjacent edges between two nodes satisfy the following expression:

[0104]

[0105] in, p represents the physical proximity edge characteristics between nodes i and j. i Let p be the three-dimensional coordinates of node i. j Let p be the three-dimensional coordinates of node j. i and p j Belongs to the set of reconstructed point cloud coordinates P mm L1 represents the first distance threshold. For example, for metal alloys, since the grain size ranges from 0.1 to 0.8 mm, the first distance threshold L1 can be 0.5 mm. ||||2 represents the Euclidean distance.

[0106] Sound field feature edge data is used to characterize the spatial correlation features of the sound field at scan points within the target region. Sound field feature edge data includes the sound field feature edge features between two nodes.

[0107] Specifically, the sound field feature edges can be connected to those with similar mixing nonlinear parameters N. new The nodes are used to represent the spatial correlation of the sound field. If the mixing nonlinearity parameter N of the two nodes... new If the nodes are close together, it indicates that their local nonlinear behaviors are similar, and they may belong to the same damage region. By aggregating information from such nodes, damage boundaries can be identified more accurately. The acoustic field feature edges of the two nodes satisfy the following expression:

[0108]

[0109] in, The acoustic field features and edge features of nodes i and j are represented. This represents the mixing nonlinearity parameter of the i-th node. Let S1 represent the mixing nonlinearity parameter of the j-th node, and S1 represent the similarity threshold, for example, the similarity threshold S1 can be 0.7. σ1 represents the hyperparameter controlling the similarity sensitivity, for example, it can be the bandwidth parameter, and the expression for the bandwidth parameter σ1 is:

[0110]

[0111] S103. Input the multimodal graph structure data into the graph convolutional neural network model and output the damage detection results.

[0112] In some embodiments, the graph convolutional neural network model includes: multiple convolutional layers, a three-body interaction layer, and a prediction output layer. Then, in step S103, the multimodal graph structure data is input into the graph convolutional neural network model, and the damage detection result is output, including:

[0113] First, the multimodal graph structure data is input into multiple convolutional layers to aggregate the features of neighboring nodes and obtain the output results of the convolutional layers.

[0114] Then, the output of the convolutional layer is input into the prediction output layer to perform damage prediction, and the damage detection result is output.

[0115] Specifically, the expression for the output of the (l+1)th convolutional layer in the graph convolutional neural network model is:

[0116]

[0117] in, Let N(i) represent the output of the (l+1)th convolutional layer of node i, and let N(i) represent the set of neighboring nodes of node i. This represents the characteristics of node i's neighbors. i This represents the degree of node i, which is equal to the number of its neighbors plus 1. j W represents the degree of node j. (l) Let σ represent the weight matrix, and σ² represent the Sigmoid activation function.

[0118] The loss function L for training the graph convolutional neural network model is:

[0119]

[0120] Among them, y i This represents the true damage value of the i-th damage, for example, it can be the measured crack propagation amount. This represents the predicted damage value for the i-th damage, such as the predicted crack propagation.

[0121] In some embodiments, the damage detection results include the probability and risk level of damage at each node's corresponding location. Specifically, if the probability of damage at the node's corresponding location is less than or equal to a first probability threshold (e.g., the first probability threshold can be 0.4), the risk level is Level 1, meaning the risk of damage is low. If the probability of damage at the node's corresponding location is greater than the first probability threshold and less than or equal to a second probability threshold (e.g., the first probability threshold can be 0.7), the risk level is Level 2, meaning the risk of damage is moderate and has some impact on the structure. If the probability of damage at the node's corresponding location is greater than the second probability threshold, the risk level is Level 3, meaning the risk of damage is high and has a significant impact on the structure.

[0122] The expression for the damage detection result is as follows:

[0123] P i =sigmoid(Wh i (conv) +b);

[0124] Among them, P i Let represent the probability that there is damage at the position corresponding to the i-th node, sigmoid() represent the activation function, W represent the training weight parameters, b represent the training bias parameters, and h represent the probability that there is damage at the i-th node. i (conv) This is the output of the last convolutional layer.

[0125] In some embodiments, the method can also visually display the location of a node using different colors based on different risk levels. Specifically, the method further includes:

[0126] When the risk level of a node is at the first level, the location of the node is displayed in the first preset color.

[0127] When the risk level of a node is level two, the location of the node is displayed in the second preset color.

[0128] When the risk level of a node is level three, the location of the node is displayed in the third preset color; the first preset color, the second preset color, and the third preset color are different from each other.

[0129] For example, the first preset color can be green, the second preset color can be yellow, and the third preset color can be red.

[0130] The structural damage detection method provided in the above embodiments of this application firstly acquires echo signals from each preset scanning point within the detection area. Then, multimodal graph structural data is constructed based on the echo signals from each preset scanning point. Finally, the multimodal graph structural data is input into a graph convolutional neural network model, and the damage detection result is output. This method is based on three-wave non-collinear mixing technology and transforms the multi-field excitation problem into a graph structural problem based on multimodal graph structural data. The graph convolutional neural network model can accurately determine the damage detection result, achieving precise reconstruction of the three-dimensional acoustic signal features of multi-source targeted excitation micro-damage characteristic signals. By integrating three-dimensional feature information with graph convolutional neural network modeling technology, it can spatially identify and precisely locate early micro-damage in ultrathin composite layers and connecting layers of composite metal materials, thereby effectively improving the accuracy and reliability of damage detection in material components.

[0131] This application also provides a structural damage detection system. Figure 5 This is a schematic diagram of the architecture of the structural damage detection system provided in the embodiments of this application, such as... Figure 5 As shown, the structural damage detection system 500 includes: a first transmitting probe 501, a second transmitting probe 502, a third transmitting probe 503, a first receiving probe 504, and a processing system 505. The first transmitting probe 501 and the second transmitting probe 502 are coaxially arranged opposite each other along a first axis, while the third transmitting probe 503 is not located on the first axis.

[0132] The first transmitting probe 501 can be used to tilt and transmit a first longitudinal wave signal of a first preset frequency into the detection area.

[0133] The second transmitting probe 502 can be used to obliquely inject a second longitudinal wave signal of a second preset frequency into the detection area so that the second longitudinal wave signal resonates with the first longitudinal wave signal to generate a difference frequency transverse wave.

[0134] The third transmitting probe 503 can be used to inject a first shear wave signal of a third preset frequency into the detection area at multiple preset emission angles, so that the first shear wave signal is mixed with the difference frequency transverse wave, and the third preset frequency is the difference between the first preset frequency and the second preset frequency.

[0135] The first receiving probe 504 can be used to collect echo signals at each preset scanning point within the detection area; the echo signal is a mixed acoustic signal generated by the nonlinear effect of the target damage at the preset scanning point, consisting of the first longitudinal wave signal, the second longitudinal wave signal, and the first shear wave signal.

[0136] The processing system 505 can be used to construct multimodal graph structure data based on the echo signal of each preset scanning point. The multimodal graph structure data is used to characterize the node features and acoustic field features of the preset scanning points. The multimodal graph structure data is input into a graph convolutional neural network model, and the damage detection result is output.

[0137] In some embodiments, such as Figure 5 As shown, the structural damage detection system 500 further includes a display module 506. The display module 506 is used to: display the location of the node in a first preset color when the node's risk level is first; display the location of the node in a second preset color when the node's risk level is second; and display the location of the node in a third preset color when the node's risk level is third. The first, second, and third preset colors are all different from each other.

[0138] The structural damage detection system provided in the above embodiments of this application firstly transmits a first longitudinal wave signal through a first transmitting probe, a second longitudinal wave signal through a second transmitting probe, and a first shear wave signal through a third transmitting probe. Then, the echo signal of each preset scanning point within the detection area can be collected through a first receiving probe. Next, a multimodal graph structure data can be constructed based on the echo signal of each preset scanning point through a processing system. Finally, the multimodal graph structure data is input into a graph convolutional neural network model through the processing system, and the damage detection result is output. This system is based on three-wave non-collinear mixing technology and transforms the multi-field excitation problem into a graph structure problem based on multimodal graph structure data. The graph convolutional neural network model can accurately determine the damage detection result, achieving precise reconstruction of the three-dimensional acoustic signal features of multi-source targeted excitation micro-damage characteristic signals. By integrating three-dimensional feature information and graph convolutional neural network modeling technology, it can spatially identify and precisely locate early micro-damage in ultrathin composite layers and connecting layers of composite metal materials, thereby effectively improving the accuracy and reliability of damage detection in material components.

[0139] This invention also provides an electronic device, which may include a display screen, a memory, and one or more processors. The display screen, memory, and processors are coupled. The memory stores computer program code, which includes computer instructions. When the processor executes the computer instructions, the electronic device can perform the various methods or steps executed in the above-described embodiments of the structural damage detection method. Of course, this electronic device includes, but is not limited to, the display screen, memory, and one or more processors described above.

[0140] This invention also provides a computer-readable storage medium for storing computer instructions for running the above-described structural damage detection method.

[0141] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0142] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0143] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0144] Similar parts between the embodiments provided in this application can be referred to mutually. The specific implementation methods provided above are only a few examples under the overall concept of this application and do not constitute a limitation on the scope of protection of this application. For those skilled in the art, any other implementation methods extended from the solution of this application without creative effort shall fall within the scope of protection of this application.

Claims

1. A method for detecting structural damage, characterized in that, The method includes: The system collects echo signals from each preset scanning point within the detection area. The echo signals are mixed acoustic signals generated by the nonlinear effect of a first longitudinal wave signal, a second longitudinal wave signal, and a first shear wave signal at the preset scanning point. The frequency of the first longitudinal wave signal is a first preset frequency, the frequency of the second longitudinal wave signal is a second preset frequency, and the frequency of the first shear wave signal is a third preset frequency, which is the difference between the first and second preset frequencies. The first and second longitudinal wave signals are incident on the detection area at a relatively oblique angle along a first axis and resonate to generate a difference-frequency shear wave. The first shear wave signal is incident on the detection area at multiple preset exit angles and mixes with the difference-frequency shear wave. The exit point of the first shear wave signal is not located on the first axis. Multimodal graph structure data is constructed based on the echo signal of each preset scanning point. The multimodal graph structure data is used to characterize the node features and acoustic field features of the preset scanning points. The multimodal graph structure data is input into a graph convolutional neural network model, and the damage detection results are output. The construction of multimodal graph structure data based on the echo signals of each preset scanning point includes: Based on the echo signal of each preset scanning point, the corresponding sum frequency amplitude value, fundamental frequency amplitude value, and difference frequency amplitude value are extracted respectively. The mixing nonlinear parameters are determined based on the sum frequency amplitude, the fundamental frequency amplitude, and the difference frequency amplitude. The expression for the mixing nonlinear parameters is as follows: ； in, This represents the mixing nonlinearity parameter. This represents the sum-frequency amplitude value. This represents the fundamental frequency amplitude. This represents the difference frequency amplitude; The multimodal graph structure data is constructed based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle. The construction of the multimodal graph structure data based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle includes: An initial three-dimensional data matrix is ​​constructed based on the position information of the preset scanning points and the preset emission angle. The initial three-dimensional data matrix is ​​normalized to determine the normalized three-dimensional data matrix; Perform PARAFAC tensor decomposition on the normalized three-dimensional data matrix to determine the spatial factor matrix; Data is extracted based on the spatial factor matrix to determine the initial point cloud coordinate set; The initial point cloud coordinate set is scaled and restored to determine the restored point cloud coordinate set; The multimodal graph structure data is constructed based on the restored point cloud coordinate set and the mixing nonlinear parameters; The multimodal graph structure data includes: node data, node feature data, physical proximity edge data, and acoustic field feature edge data; wherein... The node data is used to characterize the point cloud in the set of reconstructed point cloud coordinates within the target area, which is included within the detection area; The node feature data includes: the three-dimensional coordinate values, mixing nonlinear parameters, and node identifiers corresponding to the point cloud; the physical proximity edge data is used to characterize the geometric connection features of the scanning points within the target area; and the sound field feature edge data is used to characterize the sound field spatial correlation features of the scanning points within the target area.

2. The method according to claim 1, characterized in that, The step of performing PARAFAC tensor decomposition on the normalized three-dimensional data matrix to determine the spatial factor matrix includes: The normalized three-dimensional data matrix is ​​subjected to rank 3 PARAFAC tensor decomposition to obtain the PARAFAC tensor decomposition result. The spatial factor matrix is ​​determined based on the results of the PARAFAC tensor decomposition. The expression for the PARAFAC tensor decomposition result is as follows: ； In the formula, This represents the normalized three-dimensional data matrix. Represents the spatial factor matrix, Represents the damage intensity factor matrix. Represents the angle response factor matrix. The rank of the factorization is represented by ∘, and the ∘ symbol represents the outer product operation.

3. The method according to claim 1, characterized in that, The step of extracting the corresponding sum-frequency amplitude, fundamental frequency amplitude, and difference-frequency amplitude based on the echo signal of each preset scanning point includes: Perform Fourier transform on the echo signal of each preset scanning point to obtain the spectrum diagram corresponding to each echo signal; The sum frequency amplitude, the fundamental frequency amplitude, and the difference frequency amplitude are extracted from the spectrum diagram corresponding to the echo signal.

4. The method according to any one of claims 1-3, characterized in that, The graph convolutional neural network model includes multiple convolutional layers and a prediction output layer; the step of inputting the multimodal graph structure data into the graph convolutional neural network model and outputting the damage detection result includes: The multimodal graph structure data is input into the multiple convolutional layers to aggregate neighboring node features and obtain the convolutional layer output results; The output of the convolutional layer is input into the prediction output layer for damage prediction, and the damage detection result is output.

5. The method according to claim 4, characterized in that, The damage detection results include the probability and risk level of damage at the corresponding location of each node; If the probability of damage at the location corresponding to the node is less than or equal to the first probability threshold, the risk level is the first level. If the probability of damage at the location corresponding to the node is greater than a first probability threshold and less than or equal to a second probability threshold, the risk level is level two; if the probability of damage at the location corresponding to the node is greater than the second probability threshold, the risk level is level three. The expression for the damage detection result is: ； in, Indicates the first The probability that there is damage at the corresponding location of each node. This represents the activation function. Indicates the training weight parameters, This represents the training bias parameters. This indicates the output result of the three-body interaction.

6. The method according to claim 5, characterized in that, The method further includes: When the risk level of the node is level one, the location corresponding to the node is displayed in a first preset color; When the risk level of the node is level two, the location corresponding to the node is displayed in a second preset color; When the risk level of the node is level three, the location corresponding to the node is displayed in a third preset color; the first preset color, the second preset color, and the third preset color are different from each other.

7. A structural damage detection system, characterized in that, The system includes: a first transmitting probe, a second transmitting probe, a third transmitting probe, a first receiving probe, and a processing system; wherein the first transmitting probe and the second transmitting probe are coaxially opposite each other along a first axis, and the third transmitting probe is not located on the first axis; The first transmitting probe is used to obliquely transmit a first longitudinal wave signal of a first preset frequency into the detection area; The second transmitting probe is used to obliquely transmit a second longitudinal wave signal of a second preset frequency into the detection area, so that the second longitudinal wave signal resonates with the first longitudinal wave signal to generate a difference frequency transverse wave; The third transmitting probe is used to inject a first shear wave signal of a third preset frequency into the detection area at multiple preset emission angles, so that the first shear wave signal is mixed with the difference frequency transverse wave, and the third preset frequency is the difference between the first preset frequency and the second preset frequency. The first receiving probe is used to collect the echo signal of each preset scanning point in the detection area; the echo signal is a mixed acoustic signal generated by the nonlinear effect of the target damage of the first longitudinal wave signal, the second longitudinal wave signal and the first shear wave signal at the preset scanning point. The processing system is used to construct multimodal graph structure data based on the echo signal of each preset scanning point. The multimodal graph structure data is used to characterize the node features and acoustic field features of the preset scanning points. The multimodal graph structure data is input into a graph convolutional neural network model, and damage detection results are output. The construction of multimodal graph structure data based on the echo signals of each preset scanning point includes: Based on the echo signal of each preset scanning point, the corresponding sum frequency amplitude value, fundamental frequency amplitude value, and difference frequency amplitude value are extracted respectively. The mixing nonlinear parameters are determined based on the sum frequency amplitude, the fundamental frequency amplitude, and the difference frequency amplitude. The expression for the mixing nonlinear parameters is as follows: ； in, This represents the mixing nonlinearity parameter. This represents the sum-frequency amplitude value. This represents the fundamental frequency amplitude. This represents the difference frequency amplitude; The multimodal graph structure data is constructed based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle. The construction of the multimodal graph structure data based on the mixing nonlinear parameters, the position information of the preset scanning points, and the preset emission angle includes: An initial three-dimensional data matrix is ​​constructed based on the position information of the preset scanning points and the preset emission angle. The initial three-dimensional data matrix is ​​normalized to determine the normalized three-dimensional data matrix; Perform PARAFAC tensor decomposition on the normalized three-dimensional data matrix to determine the spatial factor matrix; Data is extracted based on the spatial factor matrix to determine the initial point cloud coordinate set; The initial point cloud coordinate set is scaled and restored to determine the restored point cloud coordinate set; The multimodal graph structure data is constructed based on the restored point cloud coordinate set and the mixing nonlinear parameters; The multimodal graph structure data includes: node data, node feature data, physical proximity edge data, and acoustic field feature edge data; wherein... The node data is used to characterize the point cloud in the set of reconstructed point cloud coordinates within the target area, which is included within the detection area; The node feature data includes: the three-dimensional coordinate values, mixing nonlinear parameters, and node identifiers corresponding to the point cloud; the physical proximity edge data is used to characterize the geometric connection features of the scanning points within the target area; and the sound field feature edge data is used to characterize the sound field spatial correlation features of the scanning points within the target area.

Images