A method and system for testing composite panels

Abstract

This invention provides a method and system for detecting composite panels, relating to the field of panel inspection technology. The method includes ultrasonically scanning the composite panel and dividing it into multiple acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method to calculate the reflection and transmission coefficients of each layer interface. A deep learning algorithm combined with physical constraints is used to dynamically correct the acoustic parameters. The corrected parameters are then substituted into the model to reconstruct the signal and accurately locate defects. A mesh-based analysis of the stress field distribution is performed at the defect location to determine stress concentration points. By adjusting probe parameters, a fine scan of the stress concentration area is performed to obtain high-resolution data and update the defect distribution information. Finally, the system outputs detection data including defect location, stress concentration distribution, and hazard level, achieving accurate detection and assessment of defects in the composite panel.

Description

Technical Field

[0001] This invention relates to board material testing technology, and more particularly to a testing method and system for composite boards. Background Technology

[0002] Composite panels, widely used structural materials in modern industry, have always presented a key challenge in the field of nondestructive testing (NDT) due to their defect detection. While traditional ultrasonic testing methods can detect obvious defects in composite panels, for complex multi-layered structures, the complex reflection and transmission phenomena at the interfaces between different materials lead to blurred and overlapping echo signals, making it difficult to accurately distinguish defect signals from structural echoes. Furthermore, the acoustic characteristics of each layer of the composite panel are often difficult to obtain precisely, and interface characteristics also affect sound wave propagation. These factors combined result in inaccurate defect localization and insufficient sensitivity of traditional testing methods, particularly in detecting minute defects in stress concentration areas.

[0003] Existing technologies generally employ empirical parameter settings and fixed scanning strategies, which are ill-suited to the complexity of composite panel structures and materials. While some advanced methods introduce mathematical models for acoustic wave propagation analysis, the difficulty in accurately setting model parameters leads to discrepancies between theoretical predictions and actual conditions. Furthermore, existing technologies often focus on defect location information while neglecting stress field analysis around the defects, making it difficult to assess the potential hazards of defects to the structure. With the increasingly widespread application of composite panels in critical fields such as aerospace and energy, there is an urgent need to develop new detection methods capable of accurately detecting defects in composite panels and assessing their severity to ensure structural safety and reliability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for detecting composite panels, which can solve the problems in existing technologies.

[0005] A first aspect of the present invention provides a method for detecting composite panels, comprising:

[0006] The composite plate is scanned to obtain ultrasonic scanning data. The composite plate is divided into N acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method. The sound wave transfer coefficient matrix of the N acoustic characteristic layers is calculated. The reflection coefficient and transmission coefficient of each layer interface are calculated based on the sound wave transfer coefficient matrix.

[0007] The measured ultrasonic scanning data are compared with the theoretical waveforms calculated based on the reflection coefficient and transmission coefficient, and the acoustic parameters of each acoustic characteristic layer are dynamically corrected based on deep learning algorithms and physical constraints.

[0008] The corrected acoustic parameters are substituted into the sound wave propagation model, and the ultrasonic scanning data is reconstructed to obtain the real defect signal. The defect location coordinates are then determined based on the real defect signal.

[0009] The defect location coordinates are meshed, and the stress field distribution data around the defect is calculated; the acoustic characteristic change data of the defect area under the action of the stress field distribution data is obtained to determine the stress concentration location;

[0010] The ultrasonic probe's transmission power and scanning interval are adjusted for the stress concentration location to perform fine scanning and acquire high-resolution data. The defect spatial distribution information is then updated based on the high-resolution data.

[0011] The system outputs detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level.

[0012] Optional,

[0013] The composite plate is divided into N acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method. The sound wave transfer coefficient matrix of the N acoustic characteristic layers is calculated. The steps of calculating the reflection coefficient and transmission coefficient of each layer interface based on the sound wave transfer coefficient matrix include:

[0014] Empirical mode decomposition is performed on the ultrasonic scanning data to obtain intrinsic mode functions. The phase function of the intrinsic mode functions is calculated to obtain instantaneous frequency characteristics. The instantaneous frequency characteristics are combined with the square of the amplitude of the intrinsic mode functions to construct a time-frequency energy distribution matrix. The abrupt change characteristics of the time-frequency energy distribution matrix are used to determine the interlayer interface position of the composite plate, and the composite plate is divided into N acoustic characteristic layers.

[0015] The three-dimensional topographic parameters at the interlayer interface are obtained, and a stiffness calculation formula considering interface stress is constructed based on the three-dimensional topographic parameters to obtain the interface stiffness matrix characterizing the interface contact characteristics; an ideal interface transmission matrix is ​​constructed based on the wave impedance and thickness of each acoustic characteristic layer, and the interface stiffness matrix is ​​multiplied by the ideal interface transmission matrix to obtain the sound wave transmission coefficient matrix.

[0016] The sound wave transmission coefficient matrix is ​​substituted into the sound wave propagation model, which is constructed based on the material parameters of each acoustic property layer; the wave number of the N acoustic property layers is calculated according to the sound wave propagation model, and the reflection coefficient and transmission coefficient of the interlayer interface are calculated based on the wave number, the calculation of the reflection coefficient and transmission coefficient includes an interface characteristic correction term.

[0017] Optional,

[0018] The steps of comparing the measured ultrasonic scanning data with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient, and dynamically correcting the acoustic parameters of each acoustic characteristic layer based on deep learning algorithms and physical constraints, include:

[0019] The measured ultrasonic scanning data is input into the waveform encoding layer of a bidirectional long short-term memory neural network and encoded into waveform feature vectors.

[0020] The waveform feature vector is input into the multi-scale attention mechanism layer, and wavelet transform is performed on the waveform feature vector to obtain the time-frequency feature matrix. Based on the time-frequency feature matrix, a local attention vector and a global attention vector are constructed. The local attention vector and the global attention vector are adaptively fused to obtain a fused attention vector. The fused attention vector and the waveform feature vector are multiplied element-wise to obtain a weighted feature vector.

[0021] The weighted feature vector is input into the acoustic parameter prediction network, which outputs an acoustic parameter set, including wave velocity parameter, density parameter and attenuation coefficient parameter.

[0022] An optimization objective function is constructed based on multi-level physical constraint functions and waveform reconstruction errors;

[0023] The bidirectional long short-term memory neural network is self-supervised pre-trained using ultrasonic data from a defect-free region, with waveform reconstruction as the pre-training objective.

[0024] The acoustic parameter set is iteratively optimized based on the objective function. The iteration stops when the parameter change is less than a preset threshold after multiple iterations, and the corrected acoustic parameter set is obtained.

[0025] Optional,

[0026] The steps to construct the optimization objective function include:

[0027] The multi-level physical constraints include wave equation constraint terms, parameter range constraint terms, and interface continuity constraint terms; the waveform reconstruction error includes mean square error terms and phase error terms; the optimization objective function adopts a phased weight configuration strategy based on the iterative process, dynamically adjusting the weight coefficients of the waveform reconstruction error terms and physical constraint terms according to the waveform reconstruction error value, the degree of violation of physical constraints, and the difference in acoustic impedance of the material interface.

[0028] Optional,

[0029] The steps of substituting the corrected acoustic parameters into the sound wave propagation model, reconstructing the ultrasonic scanning data to obtain the true defect signal, and determining the defect location coordinates based on the true defect signal include:

[0030] Substitute the corrected acoustic parameters and the sound wave transmission coefficient matrix into the sound wave propagation model, construct an interlayer multiple reflection compensation factor based on the reflection coefficient of the adjacent interlayer interface and the interlayer sound wave propagation delay, and perform an inverse Fourier transform after multiplying the interlayer multiple reflection compensation factor with the frequency domain response of the ultrasonic scanning data to obtain the reconstructed signal.

[0031] Extract the time-domain amplitude and phase features of the reconstructed signal, perform a differential operation between the time-domain amplitude features and the theoretical reflection time calculated based on the corrected acoustic parameters, perform a differential operation between the phase features and the theoretical reflection phase, and identify the real defect signal based on the differential operation results;

[0032] Based on the corrected acoustic parameters, a sound wave propagation time equation is established, and the defect location coordinates are obtained by solving the sound wave propagation time equation according to the time characteristics of the real defect signal.

[0033] Optional,

[0034] The steps of meshing the defect location coordinates and calculating the stress field distribution data around the defect; obtaining the acoustic characteristic change data of the defect area under the action of the stress field distribution data, and determining the stress concentration location include:

[0035] An initial mesh is constructed centered on the coordinates of the defect location. The stress gradient index is obtained by calculating the square root of the sum of the squares of the partial derivatives of stress in the X and Y directions of each mesh cell. Mesh cells with stress gradient indices greater than a preset stress gradient threshold are further subdivided.

[0036] The strain value is obtained by calculating the spatial derivative of the displacement component based on the refined mesh, and the strain value is multiplied by the material elastic constant to obtain the stress field distribution data.

[0037] The change in sound velocity is calculated based on the stress field distribution data. The change in sound velocity is determined by the product of the principal stress components in the three directions and the corresponding acoustic elastic coefficients. The acoustic impedance correction value is calculated based on the change in sound velocity.

[0038] The acoustic features corresponding to the change in sound velocity and the correction value of acoustic impedance are extracted, and a feature vector containing the rate of change of sound velocity, the rate of change of acoustic impedance and the spectral energy distribution is constructed. The weighted Euclidean distance between the feature vectors is calculated, and the neighborhood radius is determined according to the average distance of the K nearest neighbors around each feature point. The feature points are then density-clustered using the neighborhood radius, and the clustering results are mapped back to spatial coordinates to determine the stress concentration location.

[0039] The stress concentration factor is obtained by calculating the ratio of the maximum stress to the nominal stress at the stress concentration location. The deviation between the measured acoustic characteristics and the theoretical acoustic characteristics is substituted into the exponential decay function to calculate the confidence index. The reliability of the stress concentration location is determined based on the confidence index.

[0040] Optional,

[0041] The steps of adjusting the transmission power and scanning interval of the ultrasonic probe at the stress concentration location, performing fine scanning and acquiring high-resolution data, and updating the defect spatial distribution information based on the high-resolution data include:

[0042] The acoustic energy attenuation compensation coefficient is calculated based on the depth and maximum stress value at the stress concentration location. The acoustic energy attenuation compensation coefficient is multiplied by the initial transmission power to obtain the optimized transmission power. The scanning interval is determined based on the ratio of ultrasonic wavelength to probe diameter.

[0043] A spiral scanning trajectory is constructed with the stress concentration location as the center. The spiral scanning trajectory is determined by both angle and radius parameters. The angle and radius parameters are adjusted according to the scanning interval so that the spiral scanning trajectory completely covers the stress concentration location. High-resolution scanning data is obtained by performing ultrasonic scanning along the spiral scanning trajectory using the optimized transmission power.

[0044] Calculate the boundary curvature value and signal-to-noise ratio of the high-resolution scan data, and construct a weighted fusion coefficient based on the boundary curvature value and signal-to-noise ratio; use the weighted fusion coefficient to fuse the defect location coordinate information and fine scan features, and update the spatial distribution information of the defect, which includes the defect boundary contour and internal structural features.

[0045] Secondly, a testing system for composite panels is provided, comprising:

[0046] The first unit is used to scan the composite plate to obtain ultrasonic scanning data, divide the composite plate into N acoustic characteristic layers, construct a sound wave propagation model based on the recursive transfer matrix method, calculate the sound wave transfer coefficient matrix of the N acoustic characteristic layers, and calculate the reflection coefficient and transmission coefficient of each layer interface according to the sound wave transfer coefficient matrix.

[0047] The second unit is used to compare the measured ultrasonic scanning data with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient, and dynamically correct the acoustic parameters of each acoustic characteristic layer based on deep learning algorithms and physical constraints.

[0048] The third unit is used to substitute the corrected acoustic parameters into the sound wave propagation model, reconstruct the signal from the ultrasonic scanning data, obtain the real defect signal, and determine the defect location coordinates based on the real defect signal.

[0049] The fourth unit is used to perform mesh generation at the coordinates of the defect location, calculate the stress field distribution data around the defect, acquire the acoustic characteristic change data of the defect area under the action of the stress field distribution data, and determine the stress concentration location.

[0050] The fifth unit is used to adjust the transmission power and scanning interval of the ultrasonic probe for the stress concentration location, perform fine scanning and acquire high-resolution data, and update the defect spatial distribution information based on the high-resolution data;

[0051] The sixth unit is used to output detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level.

[0052] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0053] This paper constructs a sound wave propagation model using the recursive transfer matrix method, achieving an accurate description of the acoustic parameters of composite multilayer structures and overcoming the modeling difficulties of sound wave propagation at complex interfaces in traditional methods. Combining a dual optimization strategy of deep learning algorithms and physical constraints, the acoustic parameter correction process adheres to physical laws while possessing data-driven adaptability, significantly improving the accuracy and reliability of composite plate defect detection. In particular, by establishing a mapping relationship between acoustic characteristics and stress field distribution, not only can defects be accurately located, but the stress concentration around the defects can also be analyzed, providing a scientific basis for assessing the severity of defects and filling the gap in defect hazard assessment using traditional detection methods.

[0054] This paper also introduces an adaptive scanning strategy based on stress concentration regions. By dynamically adjusting the ultrasonic probe parameters and scanning trajectory, it achieves precise detection of high-risk areas, significantly improving detection resolution and sensitivity. The overall technical process of the method forms a complete system from the construction of the sound wave propagation model, parameter optimization, signal reconstruction to stress analysis and hazard assessment, providing new ideas and methods for the field of nondestructive testing of composite panels. Attached Figure Description

[0055] Figure 1 This is a schematic flowchart of a composite plate testing method according to an embodiment of the present invention;

[0056] Figure 2 A comparison chart showing how waveform reconstruction error varies with the number of iterations for different methods. Detailed Implementation

[0057] The technical solutions of the present invention will be described below with reference to the accompanying drawings. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 1 This is a schematic flowchart of a composite plate testing method according to the present invention, as shown below. Figure 1 As shown, the method includes:

[0059] The composite plate is scanned to obtain ultrasonic scanning data. The composite plate is divided into N acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method. The sound wave transfer coefficient matrix of the N acoustic characteristic layers is calculated. The reflection coefficient and transmission coefficient of each layer interface are calculated based on the sound wave transfer coefficient matrix.

[0060] The measured ultrasonic scanning data are compared with the theoretical waveforms calculated based on the reflection coefficient and transmission coefficient, and the acoustic parameters of each acoustic characteristic layer are dynamically corrected based on deep learning algorithms and physical constraints.

[0061] The corrected acoustic parameters are substituted into the sound wave propagation model, and the ultrasonic scanning data is reconstructed to obtain the real defect signal. The defect location coordinates are then determined based on the real defect signal.

[0062] The defect location coordinates are meshed, and the stress field distribution data around the defect is calculated; the acoustic characteristic change data of the defect area under the action of the stress field distribution data is obtained to determine the stress concentration location;

[0063] The ultrasonic probe's transmission power and scanning interval are adjusted for the stress concentration location to perform fine scanning and acquire high-resolution data. The defect spatial distribution information is then updated based on the high-resolution data.

[0064] The system outputs detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level. The product of the stress concentration factor and the defect volume is normalized to a quantification index of 0-10. Based on this quantification index, the defects are classified into three levels: low hazard, medium hazard, and high hazard.

[0065] Optional,

[0066] The composite plate is divided into N acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method. The sound wave transfer coefficient matrix of the N acoustic characteristic layers is calculated. The steps of calculating the reflection coefficient and transmission coefficient of each layer interface based on the sound wave transfer coefficient matrix include:

[0067] Empirical mode decomposition is performed on the ultrasonic scanning data to obtain intrinsic mode functions. The phase function of the intrinsic mode functions is calculated to obtain instantaneous frequency characteristics. The instantaneous frequency characteristics are combined with the square of the amplitude of the intrinsic mode functions to construct a time-frequency energy distribution matrix. The abrupt change characteristics of the time-frequency energy distribution matrix are used to determine the interlayer interface position of the composite plate, and the composite plate is divided into N acoustic characteristic layers.

[0068] The three-dimensional topographic parameters at the interlayer interface are obtained, and a stiffness calculation formula considering interface stress is constructed based on the three-dimensional topographic parameters to obtain the interface stiffness matrix characterizing the interface contact characteristics; an ideal interface transmission matrix is ​​constructed based on the wave impedance and thickness of each acoustic characteristic layer, and the interface stiffness matrix is ​​multiplied by the ideal interface transmission matrix to obtain the sound wave transmission coefficient matrix.

[0069] The sound wave transmission coefficient matrix is ​​substituted into the sound wave propagation model, which is constructed based on the material parameters of each acoustic property layer and is used to describe the propagation law of sound waves in the composite plate. The wave number of the N acoustic property layers is calculated according to the sound wave propagation model. The wave number characterizes the propagation characteristics of sound waves in each layer. The reflection coefficient and transmission coefficient of the interlayer interface are calculated based on the wave number. The calculation of the reflection coefficient and transmission coefficient includes an interface characteristic correction term, which is an exponential function of the normal stiffness coefficient and the arithmetic mean of the three-dimensional topography parameters.

[0070] For example, an ultrasonic probe is used to perform C-scan on the composite board with a specific step size of 0.5 mm, a scanning frequency of 5 MHz, a sampling rate of 100 MHz, and a signal length of 1024 sampling points. The acquired raw scan data presents as a set of time-domain waveforms, containing reflection information of the internal structure and potential defects of the composite board.

[0071] Empirical mode decomposition (EMD) is performed on ultrasonic scanning data. This process involves repeated filtering and iteration to extract intrinsic mode functions (EMFs) from the original signal. For example, scanning data of carbon fiber composite panels can typically be decomposed into 5-8 EMFs, each representing a specific frequency component of the signal. For a composite panel with a thickness of 10 mm, 10 filtering iterations are usually required to ensure the accuracy of the decomposition. The Hilbert transform is applied to the EMFs to obtain the analytic signal, and then the first derivative of the phase function is calculated to obtain the instantaneous frequency. Taking a composite panel scanning case as an example, the instantaneous frequency of the first calculated EMF exhibits a jump characteristic at the interface, decreasing from 4.8 MHz to 4.2 MHz.

[0072] The method for constructing the time-frequency energy distribution matrix is ​​as follows: For each time point t, the squared amplitude value of the i-th intrinsic mode function at that time point is taken as the weight, multiplied by the corresponding instantaneous frequency value, and filled into the t-th row and i-th column of the matrix. In specific implementation, the squared amplitude value of each intrinsic mode function at each time point is first calculated, and then the squared amplitude value at the same time point is multiplied by the corresponding instantaneous frequency value to form the energy distribution matrix. For the typical case of an 8-layer composite plate, after obtaining a 1024×8 matrix, it is normalized by dividing by the maximum value in the matrix.

[0073] The interlayer interface locations of the composite board are determined based on the abrupt change characteristics of the time-frequency energy distribution matrix. In practice, an energy change threshold of 20% is set; when the rate of change of energy distribution on the time axis exceeds this threshold, it is identified as a potential interface location. For an 8-layer carbon fiber reinforced epoxy resin composite board, seven distinct interface locations can typically be detected, with time positions at 23μs, 35μs, 48μs, 60μs, 72μs, 85μs, and 97μs of the original signal. Combining the sound velocity information, the time positions are converted into depth positions to determine the thickness of each layer, thus achieving N-layer division of the composite board.

[0074] A surface roughness measuring instrument was used to sample and measure the interlayer interfaces of the composite plate, recording parameters such as peak height, valley depth, and average roughness. Taking the third interface of a certain composite plate as an example, the measured average roughness was 2.3 μm, the peak-to-valley difference was 8.5 μm, and the standard deviation of the sampling points was 1.2 μm. A stiffness calculation formula considering interfacial stress was constructed based on the three-dimensional morphology parameters. In practical applications, interfacial stiffness is positively correlated with the interfacial contact area, the material's elastic modulus, and the interfacial stress. For an interface with an average roughness of 2.3 μm, when the interfacial pressure is 5 MPa, the calculated normal stiffness coefficient is approximately 2.8 × 10⁻⁶. 14 N / m 3 The tangential stiffness coefficient is approximately 9.3 × 10⁻⁶. 13 N / m 3 .

[0075] The construction of the interface stiffness matrix needs to consider the stiffness characteristics in both the normal and tangential directions. For a typical composite plate interface, the matrix dimension is 2×2, with the diagonal elements representing the normal and tangential stiffness coefficients, respectively, and the off-diagonal elements ideally being zero. In actual calculations, the off-diagonal elements can be appropriately adjusted according to the anisotropic characteristics of the interface, typically taking values ​​no more than 5% of the diagonal elements.

[0076] An ideal interface transfer matrix is ​​constructed based on the wave impedance and thickness of each acoustic layer. For carbon fiber composite panels, the typical acoustic impedance range is 6 × 10⁻⁶. 6 kg / (m 2 ·s) to 9×10 6 kg / (m2 ·s). Taking an 8-layer composite board as an example, the acoustic impedances of each layer from top to bottom are 7.2, 8.5, 8.3, 7.8, 8.1, 8.6, 7.9 and 8.2 (×10) respectively. 6 kg / (m 2 The thicknesses of each layer are 1.2, 1.3, 1.25, 1.3, 1.2, 1.35, 1.3 and 1.1 mm, respectively.

[0077] The acoustic wave transmission coefficient matrix is ​​obtained by multiplying the interface stiffness matrix by the ideal interface transmission matrix. For each interface, the calculated transmission coefficient matrix has a dimension of 4×4, which includes the reflection, transmission, and mode conversion characteristics of acoustic waves at the interface. In actual calculations, the typical values ​​of the main diagonal elements of the transmission coefficient matrix of the third interface are 0.92 and 0.88, indicating high transmission efficiency, while the typical values ​​of the off-diagonal elements are less than 0.15, indicating a weaker mode conversion effect.

[0078] The sound wave transmission coefficient matrix is ​​substituted into the sound wave propagation model, which is constructed based on the material parameters of each acoustic property layer. The model input parameters include: longitudinal wave velocity (typically 5900-6500 m / s), transverse wave velocity (typically 3100-3400 m / s), and density (typically 1550-1650 kg / m³) for each layer. 3 The model calculates the propagation characteristics of the sound wave in each layer and the reflection / transmission behavior at the interface for a specific frequency (e.g., 5 MHz).

[0079] The wavenumber of each acoustic characteristic layer is calculated based on the sound wave propagation model. The wavenumber characterizes the propagation characteristics of sound waves in each layer. Typical longitudinal wavenumbers have a real part of approximately 5000-5500 rad / m and an imaginary part of approximately 10-20 rad / m; transverse wavenumbers have a real part of approximately 9200-9800 rad / m and an imaginary part of approximately 25-40 rad / m. The real part of the wavenumber determines the wavelength, while the imaginary part determines the attenuation characteristics. The reflection and transmission coefficients are calculated using a method that combines the difference in wave impedance between the materials on both sides of the interface and the interface stiffness. For the interface between layer m and layer m+1, the reflection coefficient is calculated as the ratio of the difference in wave impedance between the two materials to the sum of their wave impedances, multiplied by an interface characteristic correction term; the transmission coefficient is calculated as twice the product of wave impedances divided by the sum of their wave impedances, multiplied by the interface characteristic correction term. The interface characteristic correction term is determined by an exponential function consisting of the normal stiffness coefficient and the arithmetic mean of the three-dimensional topographic parameters. This correction considers the influence of the interface contact state on sound wave propagation. For the third interface, the calculated longitudinal wave reflection coefficient is approximately 0.08 and the transmission coefficient is approximately 0.92; the transverse wave reflection coefficient is approximately 0.12 and the transmission coefficient is approximately 0.88.

[0080] This method employs empirical mode decomposition and time-frequency analysis to accurately identify interlayer interfaces in complex composite plates, overcoming the limitations of traditional methods in handling multi-layered signals. The acoustic wave propagation model based on the recursive transfer matrix method can efficiently handle wave propagation problems in multi-layered structures, exhibiting both high computational efficiency and good accuracy.

[0081] Optional,

[0082] The steps of comparing the measured ultrasonic scanning data with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient, and dynamically correcting the acoustic parameters of each acoustic characteristic layer based on deep learning algorithms and physical constraints, include:

[0083] The measured ultrasonic scanning data is input into the waveform encoding layer of a bidirectional long short-term memory neural network, whereby the waveform encoding layer encodes the measured ultrasonic scanning data into a waveform feature vector.

[0084] The waveform feature vector is input to the multi-scale attention mechanism layer, which performs wavelet transform on the waveform feature vector to obtain a time-frequency feature matrix. Based on the time-frequency feature matrix, a local attention vector and a global attention vector are constructed. The local attention vector and the global attention vector are adaptively fused to obtain a fused attention vector. The fused attention vector and the waveform feature vector are multiplied element-wise to obtain a weighted feature vector.

[0085] The weighted feature vector is input into the acoustic parameter prediction network, which outputs an acoustic parameter set, including wave velocity parameter, density parameter and attenuation coefficient parameter.

[0086] An optimization objective function is constructed based on multi-level physical constraint functions and waveform reconstruction errors;

[0087] The bidirectional long short-term memory neural network is self-supervised pre-trained using ultrasonic data from a defect-free region, with waveform reconstruction as the pre-training objective.

[0088] The acoustic parameter set is iteratively optimized based on the objective function. The iteration stops when the parameter change is less than a preset threshold after multiple iterations, and the corrected acoustic parameter set is obtained.

[0089] For example, measured ultrasonic scanning data is input into the waveform encoding layer of a bidirectional long short-term memory neural network. This network employs a four-layer structure, with each layer containing 128 hidden units and using the tanh activation function. The input data is a 1024-byte ultrasonic A-scan signal, which is preprocessed by time-domain normalization before being fed into the network. The waveform encoding layer captures the temporal features and contextual relationships of the waveform through forward and backward information transmission, ultimately encoding the measured waveform into a 256-dimensional feature vector. For instance, after encoding, the high-frequency noise components in the ultrasonic signal of a carbon fiber composite board are effectively suppressed, and key features are enhanced. The first 128 dimensions of the feature vector mainly represent low-frequency structural information, while the last 128 dimensions mainly represent high-frequency detail information.

[0090] The waveform feature vector is input into a multi-scale attention mechanism layer. This layer first performs a continuous wavelet transform on the waveform feature vector, using the Mexican cap wavelet as the mother wavelet. The scale parameter is set to 1 to 32, with a step size of 1, for a total of 32 scales, resulting in a time-frequency feature matrix of size 256×32. In practical applications, for ultrasonic signals with a center frequency of 5MHz, the energy of the wavelet-transformed time-frequency feature matrix is ​​most concentrated in the 4-6MHz frequency band and the 20-60μs time interval, showing obvious interlayer interface characteristics.

[0091] Local and global attention vectors are constructed based on the time-frequency feature matrix. The local attention calculation process involves dividing the time-frequency feature matrix into 4×4 sub-blocks, each of size 64×8. The average energy of each sub-block is calculated as its representative value. These 16 representative values ​​are then converted into local attention scores using a softmax function, forming a 16-dimensional local attention vector. The global attention calculation process involves calculating the average values ​​in the row and column directions of the entire time-frequency feature matrix, resulting in a 256-dimensional row vector and a 32-dimensional column vector. These two vectors are concatenated and reduced to 64 dimensions using a fully connected layer, then passed through a softmax function to obtain the global attention vector. The local and global attention vectors are adaptively fused to obtain a fused attention vector. The fusion process employs a gating mechanism, using a two-layer perceptron network to calculate the fusion weights. The input is the concatenation of the local and global attention vectors, and the output is a fusion coefficient between 0 and 1. When the signal noise is low, the global attention weight is high (typically 0.7-0.8); while when the signal contains defect information with obvious local features, the local attention weight will increase accordingly (typically 0.6-0.7). The dimension of the fused attention vector is 256, consistent with the original waveform feature vector.

[0092] The weighted feature vector is obtained by performing element-wise multiplication of the fused attention vector and the waveform feature vector. This step selectively enhances the original features, enabling the network to focus on key information in the waveform.

[0093] The weighted feature vector is input into an acoustic parameter prediction network, which employs a three-layer fully connected structure with hidden layer sizes of 128, 64, and 32, and the activation function is ReLU. The network outputs a set of acoustic parameters, including wave velocity, density, and attenuation coefficient parameters for each layer. For an 8-layer composite plate, the output dimension is 8 × 3 = 24, corresponding to the three acoustic parameters for each of the eight layers. In practical applications, the wave velocity parameters output by the prediction network typically range from 5900 to 6500 m / s, and the density parameters range from 1550 to 1650 kg / m³. 3 The attenuation coefficient parameter ranges from 20 to 35 Np / m.

[0094] An optimization objective function is constructed based on multi-level physical constraint functions and waveform reconstruction errors. The physical constraints include three levels: first, wave equation constraints, ensuring acoustic parameters satisfy wave propagation laws, with a weight of 0.3; second, parameter range constraints, guaranteeing acoustic parameters remain within a physically reasonable range, with a weight of 0.2; and finally, interface continuity constraints, ensuring displacement and stress continuity conditions are met at interlayer interfaces, with a weight of 0.2. Waveform reconstruction errors include the mean square error of the time-domain waveform amplitude (typically with a weight of 0.2) and the phase error based on the Hilbert transform (typically with a weight of 0.1).

[0095] The objective function is optimized using a phased weighting strategy. In the initial iterations (usually the first 50 iterations), the waveform reconstruction error weight is relatively high (0.8), while the physical constraint weight is relatively low (0.2) to quickly approximate the measured waveform. When the waveform reconstruction error drops below a threshold (usually 20% of the initial error), the physical constraint weight is increased to 0.5, and the waveform reconstruction error weight is decreased to 0.5 to ensure the physical rationality of the solution. When a region with an acoustic impedance difference greater than 30% is detected at the material interface, the waveform reconstruction error weight in that region is increased by 50% to strengthen the fitting of abnormal regions. When the degree of violation of any physical constraint exceeds a preset threshold (usually 10% of the initial violation), the weight of that constraint term is doubled until the degree of constraint violation drops below the threshold.

[0096] Before applying deep learning algorithms, the network needs to be pre-trained. A bidirectional long short-term memory neural network was self-supervised pre-trained using ultrasound data from a defect-free region, with waveform reconstruction as the pre-training objective. Specifically, the ultrasound A-scan signal from the defect-free region was input into the network, and the original signal was reconstructed through an encoder-decoder structure. Training was performed by minimizing the reconstruction error. Pre-training used a batch size of 64, a learning rate of 0.001, 100 training epochs, and the Adam optimizer. The pre-trained network effectively captured the essential features of the ultrasound signal, laying the foundation for subsequent parameter prediction tasks.

[0097] The acoustic parameter set is iteratively optimized based on the objective function using gradient descent with an initial learning rate of 0.01, dynamically adjusted using cosine annealing. Iteration stops when the parameter changes over five consecutive iterations are all less than a preset threshold (typically 0.1% of the initial parameter value), yielding the corrected acoustic parameter set. In typical applications, the optimization process usually converges after 200-300 iterations, and the error between the final acoustic parameters and the actual physical parameters is typically less than 5%.

[0098] This method combines the advantages of deep learning and physical constraints to achieve precise correction of the acoustic parameters of composite boards. Compared with traditional methods, it exhibits significant improvements in adaptability and accuracy. A bidirectional long short-term memory network captures the temporal characteristics of the waveform, a multi-scale attention mechanism enhances the representation of key information, and physical constraints ensure that the parameter correction results conform to the laws of sound wave propagation. An adaptive fusion strategy dynamically adjusts the weights of local and global information based on signal characteristics, enabling the algorithm to maintain stable performance under different noise environments and defect types.

[0099] Optional,

[0100] The steps to construct the optimization objective function include:

[0101] The multi-level physical constraints include: wave equation constraint terms constructed based on sound wave propagation theory, used to ensure that the acoustic parameter set satisfies the wave equation; parameter range constraint terms constructed based on material acoustic properties, used to ensure that the acoustic parameter set is within the physically feasible range; and interface continuity constraint terms constructed based on interlayer interface properties, used to ensure that sound waves satisfy displacement and stress continuity conditions when propagating at interlayer interfaces.

[0102] The waveform reconstruction error includes: a mean square error term based on the amplitude of the time-domain waveform, used to characterize the amplitude difference between the measured waveform and the reconstructed waveform; and a phase error term based on the Hilbert transform, used to characterize the phase difference between the measured waveform and the reconstructed waveform.

[0103] The optimization objective function adopts a phased weight configuration strategy based on the iterative process. The weight coefficients of the waveform reconstruction error term and the physical constraint term are dynamically adjusted according to the waveform reconstruction error value, the degree of violation of physical constraints and the difference in acoustic impedance of the material interface.

[0104] For example, the wave equation constraint terms are implemented by checking whether the acoustic parameters satisfy the propagation equation of sound waves in anisotropic media. For each acoustic characteristic layer, the longitudinal wave velocity, transverse wave velocity, and density parameters of that layer are taken and substituted into the sound wave propagation equation to check the residual values. The residual values ​​are calculated using the finite difference method. For ultrasound at a frequency of 5 MHz, the spatial step is set to 0.05 mm and the time step is set to 0.01 μs. When the residual value exceeds a preset threshold (e.g., 1% of the longitudinal wave velocity), the corresponding wave equation constraint term is increased. In practice, for typical carbon fiber composite panels, when the longitudinal wave velocity varies in the range of 5900-6500 m / s, the weighting coefficient of the wave equation constraint term is set to 0.3; when the velocity value exceeds this range, the weighting coefficient is increased to 0.5 to force the parameters to return to a reasonable range.

[0105] The parameter range constraints, constructed based on the acoustic properties of the materials, constitute the second layer of physical constraints. These constraints ensure that each acoustic parameter remains within a physically feasible range. For carbon fiber composites, the reasonable range for longitudinal wave velocity is 5500-7000 m / s, for transverse wave velocity it is 3000-3800 m / s, for density it is 1450-1750 kg / m³, and for attenuation coefficient it is 15-50 Np / m. The parameter range constraints are implemented using a soft constraint approach. When the parameter value approaches the boundary, the constraint value increases gradually; when the parameter value exceeds the boundary, the constraint value increases sharply. Specifically, when the parameter deviates from the center value by less than 10%, the constraint growth coefficient is 0.05; when it deviates by 10%-20%, the coefficient is 0.2; when it deviates by 20%-30%, the coefficient is 0.5; and when it deviates by more than 30%, the coefficient is 1.0. This segmented approach allows for some optimization space while ensuring parameter rationality.

[0106] The interface continuity constraint term, constructed based on the interlayer interface characteristics, is the third layer of physical constraints. This constraint ensures that the displacement and stress continuity conditions are met when sound waves propagate at the interlayer interface. At the interface, the normal displacement, tangential displacement, normal stress, and tangential stress of the upper and lower layers must satisfy the continuity condition. In implementation, for each interface, the displacement and stress difference between the two sides of the interface are calculated separately. When the difference is greater than a preset threshold, the corresponding interface continuity constraint term is increased. For the third interface, the displacement continuity threshold is set to 1 μm, and the stress continuity threshold is set to 0.1 MPa. In practical applications, when the material properties on both sides of the interface differ significantly (e.g., the fiber direction change exceeds 45 degrees), the weight of the continuity constraint needs to be adjusted accordingly, typically ranging from 0.15 to 0.25.

[0107] Waveform reconstruction error comprises two parts: a mean square error (MSE) term based on the amplitude of the time-domain waveform and a phase error term based on the Hilbert transform. The MSE term characterizes the amplitude difference between the measured and reconstructed waveforms. It is calculated by summing the squares of the amplitude differences at corresponding sampling points of the two waveforms and then dividing by the number of sampling points. For a 1024-point A-scan signal, the MSE typically decreases rapidly in the first 20 iterations, dropping from an initial value (e.g., 0.5) to a lower level (e.g., 0.05). The phase error term based on the Hilbert transform characterizes the phase difference between the measured and reconstructed waveforms. First, the measured and reconstructed waveforms are subjected to Hilbert transforms respectively to obtain the corresponding analytic signals. Then, the absolute value of the phase difference between the two analytic signals is calculated. Finally, the phase differences at all sampling points are summed and normalized. In spectral analysis, the phase difference in the 3-7MHz range is the primary focus, as the phase error in this range has the greatest impact on reconstruction quality. In practice, the weighting coefficient of the phase error term is usually set to half that of the mean square error term, such as 0.1, because phase information has an important influence on the shape of the reconstructed waveform, but its contribution to the overall fit is less than that of the amplitude information.

[0108] The objective function is optimized using a phased weight allocation strategy based on an iterative process, dynamically adjusting the weight coefficients of each item. Specifically, this is implemented in three phases:

[0109] In the first stage (less than 50 iterations), the waveform reconstruction error term has a higher weight, while the physical constraint term has a lower weight. A typical configuration is: mean square error term weight 0.5, phase error term weight 0.2, wave equation constraint term weight 0.1, parameter range constraint term weight 0.1, and interface continuity constraint term weight 0.1. The goal of this stage is to rapidly reduce the waveform reconstruction error, making the theoretical waveform closer to the measured waveform.

[0110] When the mean squared error drops below 20% of its initial value (e.g., from 0.5 to below 0.1), the second stage begins. At this point, the weight of the waveform reconstruction error term is appropriately reduced, while the weight of the physical constraint term is increased. A typical configuration is: mean squared error term weight 0.3, phase error term weight 0.15, wave equation constraint term weight 0.2, parameter range constraint term weight 0.15, and interface continuity constraint term weight 0.2. The second stage emphasizes the physical rationality of the parameters to prevent overfitting.

[0111] When a region with an acoustic impedance difference greater than 30% is detected at the material interface, the process enters the third stage, where a processing strategy is implemented for that region. At this stage, the weight of the mean square error term for that region is increased by 50%, while the weight of the interface continuity constraint term is decreased by 30%. For example, if the acoustic impedance ratio of a certain interface is 1.35 (exceeding the threshold of 1.3), the weight of the mean square error term for the region near that interface increases from 0.3 to 0.45, while the weight of the interface continuity constraint term decreases from 0.2 to 0.14. This adjustment takes into account that at interfaces with large acoustic impedance differences, reflection and transmission phenomena are more complex, waveform fitting is more difficult, and the interface continuity condition may not be fully satisfied.

[0112] An adaptive adjustment mechanism was also implemented throughout the optimization process. When the degree of violation of any physical constraint exceeds a preset threshold, the weight of that constraint term is doubled until the degree of constraint violation decreases below the threshold. For example, when the P-wave velocity of a certain layer exceeds the reasonable range (7000 m / s) by more than 10%, the weight of the parameter range constraint term is increased from 0.15 to 0.3. This mechanism ensures that physically unreasonable results are not obtained during parameter optimization.

[0113] Figure 2 The graph compares the waveform reconstruction error of different methods with the number of iterations. The horizontal axis represents the number of iterations (0-250), and the vertical axis represents the waveform reconstruction error (0-0.5). The error reduction rate of the genetic algorithm (dashed line) and the simplex method (dotted line) gradually slows down, while the error of the present invention (solid line) decreases to 0.23 in the first 50 iterations. After entering the second stage, due to the adoption of a phased weight configuration strategy based on the iteration process, the weight coefficients of each error term and constraint term are dynamically adjusted, and the error reduction rate remains fast, eventually reaching 0.01. This fully demonstrates that the multi-level physical constraint optimization method of the present invention has higher parameter correction accuracy.

[0114] This method combines multi-level physical constraints with waveform reconstruction errors, effectively solving the problem of acoustic parameter correction for composite boards. By comprehensively considering time-domain amplitude and phase errors, the accuracy of waveform reconstruction is improved; and through a staged weighting strategy, a balance between reconstruction accuracy and physical rationality is achieved.

[0115] Optional,

[0116] The steps of substituting the corrected acoustic parameters into the sound wave propagation model, reconstructing the ultrasonic scanning data to obtain the true defect signal, and determining the defect location coordinates based on the true defect signal include:

[0117] Substitute the corrected acoustic parameters and the sound wave transmission coefficient matrix into the sound wave propagation model, construct an interlayer multiple reflection compensation factor based on the reflection coefficient of the adjacent interlayer interface and the interlayer sound wave propagation delay, and perform an inverse Fourier transform after multiplying the interlayer multiple reflection compensation factor with the frequency domain response of the ultrasonic scanning data to obtain the reconstructed signal.

[0118] Extract the time-domain amplitude and phase features of the reconstructed signal, perform a differential operation between the time-domain amplitude features and the theoretical reflection time calculated based on the corrected acoustic parameters, perform a differential operation between the phase features and the theoretical reflection phase, and identify the real defect signal based on the differential operation results;

[0119] Based on the corrected acoustic parameters, a sound wave propagation time equation is established, and the defect location coordinates are obtained by solving the sound wave propagation time equation according to the time characteristics of the real defect signal.

[0120] For example, the corrected acoustic parameters and sound wave transfer coefficient matrix are substituted into the sound wave propagation model. The acoustic parameters include the longitudinal wave velocity, transverse wave velocity, density, and attenuation coefficient of each layer, while the sound wave transfer coefficient matrix contains the characteristics of sound wave propagation at the interface. Taking an 8-layer carbon fiber composite board as an example, the acoustic parameters of the third layer after the aforementioned correction steps are: longitudinal wave velocity 6320 m / s, transverse wave velocity 3280 m / s, and density 1620 kg / m³. 3 The attenuation coefficient is 28 Np / m. The main diagonal elements of the acoustic transmission coefficient matrix at the interface between the third and fourth layers are 0.92 and 0.88, respectively, indicating high transmission efficiency.

[0121] For each interface between adjacent layers, the amplitude attenuation of multiple reflections is calculated based on the interface reflection coefficient, and the propagation delay is calculated based on the thickness of each layer and the sound velocity. For example, the third layer has a thickness of 1.25 mm, a longitudinal wave velocity of 6320 m / s, and a propagation delay of 0.40 μs; the reflection coefficient at the interface between the third and fourth layers is 0.08, corresponding to a 92% attenuation for the first reflection. Based on these parameters, an interlayer multiple reflection compensation factor is constructed by superimposing the responses of multiple reflections. Multiple reflections of up to three orders are typically considered, as reflections of higher orders have attenuated to less than 1% of the original signal and can be ignored.

[0122] The original ultrasonic A-scan signal is subjected to a Fourier transform to obtain the frequency domain response. Then, the frequency domain response is multiplied by a multiple reflection compensation factor to achieve frequency domain compensation. Finally, an inverse Fourier transform is performed to obtain the reconstructed time domain signal. Frequency domain compensation is crucial for effectively separating interlayer multiple reflections from the actual defect signal. In practical applications, for ultrasonic waves with a center frequency of 5MHz, compensation is mainly performed within the 2-8MHz range, as frequency components outside this range have a relatively small impact on the compensation effect.

[0123] Compared to the original signal, the reconstructed signal effectively suppresses interlayer multiple reflections. For example, in a typical 8-layer composite board test, the amplitude of the multiple reflection signal at the interface between the third and fourth layers in the original signal is about 15% of the main reflection signal, while after reconstruction, the amplitude of this multiple reflection signal is reduced to below 3%, and the signal-to-noise ratio is improved by about 5 times. The peak values ​​in the reconstructed signal are more prominent, and the distinction between defect signals and structural signals is significantly improved.

[0124] The temporal amplitude and phase features of the reconstructed signal are extracted. Temporal amplitude features include local maxima and their occurrence times. A sliding window method is used, with the window size set to 1.5 times the ultrasonic period (typically 0.3 μs), and the window sliding step size equal to the sampling interval (typically 0.01 μs). Amplitude maxima are detected within each window, and their time position and amplitude are recorded. For a typical 8-layer composite board, seven main peaks corresponding to the interlayer interfaces and several secondary peaks can be detected. Phase features are obtained by calculating the instantaneous phase using Hilbert transform, and the time position and amplitude of phase transition points are recorded. Phase transitions are an important feature for identifying defect signals because defects often cause abnormal changes in the acoustic wave phase.

[0125] The temporal amplitude characteristics are differentially analyzed with the theoretical reflection time calculated based on the corrected acoustic parameters. The theoretical reflection time is calculated based on the thickness of each layer and the sound velocity. For example, for the aforementioned 8-layer composite board, the theoretical reflection times at the interfaces of the first three layers are 24.2 μs, 36.1 μs, and 48.3 μs, respectively. The actual detected peak time is then differentially analyzed with the theoretical time to obtain the time deviation. The time deviation of normal interlayer reflections is typically within ±0.2 μs, while defect signals often exhibit significant time advance or delay, with deviations ranging from ±0.5 μs to ±2.0 μs.

[0126] The phase characteristics are compared with the theoretical reflection phase using a differential calculation. The theoretical reflection phase is calculated based on the acoustic impedance ratio of the materials on both sides of the interface. When the acoustic impedance ratio is greater than 1, the phase change is π; when the acoustic impedance ratio is less than 1, the phase change is 0. For example, the acoustic impedance ratio of the third interface is 1.06, and the theoretical reflection phase change is π. The phase deviation is obtained by differentiating the actual detected phase jump from the theoretical phase change. The phase deviation of normal interlayer reflection is usually within ±π / 6, while the phase deviation of defect signals is often greater than ±π / 3, and some are even close to π.

[0127] The true defect signal is identified based on the differential calculation results. Defect signal identification employs dual criteria: a time deviation exceeding ±0.3 μs and a phase deviation exceeding ±π / 4. A signal is considered defective when both conditions are met simultaneously. Different threshold combinations are set for different types of defects, such as delamination and inclusion. For example, the time deviation threshold for delamination defects is set to ±0.3 μs, and the phase deviation threshold to ±π / 4; while the time deviation threshold for inclusion defects is set to ±0.5 μs, and the phase deviation threshold to ±π / 3. This differentiated setting improves the identification accuracy of different defect types.

[0128] A sound wave propagation time equation is established based on the modified acoustic parameters. This equation describes the relationship between the sound wave propagation time from the transducer emission point to the defect and back to the receiving point and the defect's location coordinates. The equation includes the thickness and sound velocity information of each layer, as well as the transducer's location coordinates. For the common case of perpendicular incidence, the equation simplifies to the sum of the sound wave propagation times in each layer. For example, if the defect is located in the fourth layer, the sound wave must pass through the first three layers in sequence, reflect, and then return; the total propagation time is the sum of the individual propagation times.

[0129] The defect location coordinates are obtained by solving the acoustic wave propagation time equation based on the temporal characteristics of the actual defect signal. The solution process employs Newton's iteration method, with the initial value set as the estimated depth corresponding to the time interval. The convergence condition for the iteration is that the coordinate change between two consecutive iterations is less than 0.01 mm or the number of iterations reaches 20. For a defect signal with a time interval of 58.6 μs, the calculated defect depth is 3.85 mm, located in the middle of the fourth layer, with its lateral coordinate consistent with the probe position. When using an oblique incidence or focusing probe, the equation solution needs to consider the changes in the acoustic wave propagation path, making the calculation process more complex, but the principle remains the same.

[0130] To verify the accuracy of the positioning results, cross-validation can be performed using multiple scans from different locations. For example, scanning a defect in a carbon fiber composite panel at five different locations yielded calculated defect depths of 3.83 mm, 3.86 mm, 3.85 mm, 3.84 mm, and 3.87 mm, with an average of 3.85 mm and a standard deviation of 0.014 mm, indicating good consistency and reliability of the positioning results. Compared with subsequent slice verification results, the positioning error is less than 0.1 mm, meeting the requirements for engineering applications.

[0131] This method has three significant advantages: by constructing an interlayer multiple reflection compensation factor, it effectively eliminates multiple reflection interference in complex multilayer structures and improves the signal-to-noise ratio of defect signals; by combining the dual criteria of time-domain amplitude and phase characteristics, it significantly improves the accuracy of defect identification and reduces the false positive rate; and the sound wave propagation time equation established based on precisely corrected acoustic parameters makes the calculation of defect location coordinates more accurate.

[0132] Optional,

[0133] The steps of meshing the defect location coordinates and calculating the stress field distribution data around the defect; obtaining the acoustic characteristic change data of the defect area under the action of the stress field distribution data, and determining the stress concentration location include:

[0134] An initial mesh is constructed centered on the coordinates of the defect location. The stress gradient index is obtained by calculating the square root of the sum of the squares of the partial derivatives of stress in the X and Y directions of each mesh cell. Mesh cells with stress gradient indices greater than a preset stress gradient threshold are further subdivided.

[0135] The strain value is obtained by calculating the spatial derivative of the displacement component based on the refined mesh, and the strain value is multiplied by the material elastic constant to obtain the stress field distribution data.

[0136] The change in sound velocity is calculated based on the stress field distribution data. The change in sound velocity is determined by the product of the principal stress components in the three directions and the corresponding acoustic elastic coefficients. The acoustic impedance correction value is calculated based on the change in sound velocity.

[0137] The acoustic features corresponding to the change in sound velocity and the correction value of acoustic impedance are extracted, and a feature vector containing the rate of change of sound velocity, the rate of change of acoustic impedance and the spectral energy distribution is constructed. The weighted Euclidean distance between the feature vectors is calculated, and the neighborhood radius is determined according to the average distance of the K nearest neighbors around each feature point. The feature points are then density-clustered using the neighborhood radius, and the clustering results are mapped back to spatial coordinates to determine the stress concentration location.

[0138] The stress concentration factor is obtained by calculating the ratio of the maximum stress to the nominal stress at the stress concentration location. The deviation between the measured acoustic characteristics and the theoretical acoustic characteristics is substituted into the exponential decay function to calculate the confidence index. The reliability of the stress concentration location is determined based on the confidence index.

[0139] For example, an initial mesh is constructed centered on the defect location coordinates, using a rectangular mesh structure, extending outwards to an area five times the defect size. For a circular defect with a diameter of 2 mm, the initial mesh coverage is 10 mm × 10 mm, with an initial mesh size of 0.5 mm × 0.5 mm, containing a total of 400 mesh elements. To capture regions with significant stress gradient changes, the square root of the sum of the squares of the partial derivatives of stress in the X and Y directions for each mesh element is calculated as a stress gradient index. The partial derivatives of stress are calculated using the central difference method, dividing the difference in stress values ​​between adjacent mesh points by the mesh spacing. For example, mesh elements located at the defect edge typically have a stress gradient index in the range of 25-35 MPa / mm, while in areas far from the defect, this index is typically less than 5 MPa / mm.

[0140] Mesh elements with stress gradient indices exceeding a preset stress gradient threshold are refined. In practical applications, the stress gradient threshold is set to 20 MPa / mm, and mesh elements exceeding this threshold are divided into four equal sub-mesh units. The refinement process can be iterative; for regions with particularly high stress gradients (such as defect tips), up to three refinements can be performed to achieve a minimum mesh size of 0.0625 mm × 0.0625 mm. In a case involving a circular defect, the initial 400 mesh elements increased to 1260 after refinement, with the highest mesh density within a 0.5 mm radius around the defect, covering approximately 680 mesh elements.

[0141] The strain value is obtained by calculating the spatial derivative of the displacement components based on the refined mesh. The displacement field is calculated using the finite element method, with the boundary condition set as the working stress of the composite plate, typically a tensile stress of 50 MPa. A quadratic interpolation function is used to solve the displacement field to improve accuracy. For anisotropic carbon fiber composites, the displacement field calculation needs to consider the angle between the principal axes of the material and the loading direction. For example, when the fiber direction is at a 45-degree angle to the loading direction, the displacement field exhibits significant asymmetry, and the maximum displacement around the defect can reach 1.8 times the far-field displacement.

[0142] The stress field distribution data are obtained by multiplying the strain value by the material's elastic constants. For orthotropic carbon fiber composites, the elastic constant matrix contains nine independent components. A typical carbon fiber / epoxy composite has a longitudinal elastic modulus of 140 GPa, a transverse elastic modulus of 10 GPa, an in-plane shear modulus of 5.5 GPa, and a principal Poisson's ratio of 0.3. The stress calculation takes into account the anisotropic characteristics of the material, obtaining the stress tensor at each point. At the edge of a circular defect, the maximum principal stress can reach three times the far-field stress, resulting in significant stress concentration. At the defect tip, the stress concentration is even more severe, with the maximum principal stress reaching 5-7 times the far-field stress.

[0143] The change in sound velocity is calculated based on stress field distribution data. The change in sound velocity is determined by the product of the principal stress components in the three directions and their corresponding acoustoelastic coefficients. For carbon fiber composites, the typical value for the longitudinal wave acoustoelastic coefficient is -10 × 10⁻⁶. -11 Pa -1 The typical value of the transverse wave acoustic elastic coefficient is -15 × 10⁻⁶. -11 Pa -1This means that when the material is subjected to a tensile stress of 50 MPa, the longitudinal wave velocity changes by approximately -0.5%, and the transverse wave velocity changes by approximately -0.75%. In stress concentration regions, this change is more significant; for example, at the defect tip, the longitudinal wave velocity can decrease by up to -3.5%, and the transverse wave velocity by up to -5.2%. The acoustic impedance correction is determined by both the change in sound velocity and the change in density. Because the material volume changes slightly under stress, the density also changes accordingly, but this change is usually small; for a stress level of 50 MPa, the density change is approximately -0.15%. Therefore, the change in acoustic impedance is primarily determined by the change in sound velocity. In the high-stress region at the defect tip, the acoustic impedance can decrease by up to -3.7%, a change sufficient to produce a measurable difference in reflected signals during ultrasonic testing.

[0144] Acoustic features corresponding to changes in sound velocity and acoustic impedance correction values ​​are extracted to construct a feature vector containing the rate of change of sound velocity, the rate of change of acoustic impedance, and the spectral energy distribution. The rate of change of sound velocity is directly calculated as described above; the rate of change of acoustic impedance is determined by combining the rate of change of sound velocity and the rate of change of density; the spectral energy distribution is obtained by comparing the ultrasonic spectrum before and after stress. For ultrasonic waves with a center frequency of 5 MHz, the spectral changes caused by stress are mainly manifested as a small drift in the center frequency (usually within ±0.2 MHz) and a change in bandwidth (usually within ±5%). The feature vector of each grid point contains five components: the rate of change of longitudinal wave sound velocity, the rate of change of transverse wave sound velocity, the rate of change of acoustic impedance, the spectral center frequency shift, and the rate of change of bandwidth.

[0145] The weighted Euclidean distance between eigenvectors is calculated, and the neighborhood radius is determined based on the average distance of the K nearest neighbors around each feature point. In the weighted Euclidean distance calculation, the weights of each feature component are set according to its sensitivity; typically, the weight for the rate of change of P-wave velocity is 0.3, the weight for the rate of change of S-wave velocity is 0.3, the weight for the rate of change of acoustic impedance is 0.2, the weight for the frequency shift of the spectral center is 0.1, and the weight for the rate of change of bandwidth is 0.1. The number of nearest neighbors, K, is set to 5% of the total number of points; for example, for 1260 grid points, K is set to 63. The neighborhood radius is the average distance of the K nearest neighbors. The neighborhood radius varies for different regions; the neighborhood radius in stress concentration regions is usually smaller, approximately 0.7 times the average distance in the feature space.

[0146] Density clustering of feature points is performed using neighborhood radius, and the clustering results are mapped back to spatial coordinates to determine the location of stress concentration. The density clustering algorithm is based on the density reachability of points, dividing high-density areas into clusters. Clustering parameters include a minimum number of points, set to K / 3 (21 points); and a distance threshold set to the calculated neighborhood radius. The clustering results typically yield 3-5 clusters, with the densest cluster corresponding to the most significant stress concentration region. Mapping the clustering results back to spatial coordinates provides the precise location and shape of the stress concentration region. In a circular defect case, the most significant stress concentration region is located at ±45 degrees from the defect edge, forming two local regions, each approximately 0.8 mm × 0.5 mm in size.

[0147] The stress concentration factor is obtained by calculating the ratio of the maximum stress to the nominal stress at the stress concentration location. For the circular defect example above, the maximum stress at the stress concentration location is 152 MPa, and the nominal stress in the far field is 50 MPa; therefore, the stress concentration factor is 3.04. Different shapes of defects have different stress concentration factors. For example, when the major axis of an elliptical defect is perpendicular to the loading direction, the stress concentration factor can be as high as 4.2; while when the major axis is parallel to the loading direction, the stress concentration factor drops to around 2.1.

[0148] The confidence index is calculated by substituting the deviation between measured and theoretical acoustic characteristics into an exponential decay function. Measured acoustic characteristics are derived from the analysis of reflected waveforms in ultrasonic scanning data, including amplitude, phase, and spectral variations of the reflected waves; theoretical acoustic characteristics are derived from the stress field calculations described above. The deviation is expressed as normalized root mean square error, ranging from 0.05 to 0.2. The confidence index is calculated using an exponential decay function: a deviation of 0.05 corresponds to a confidence index of 0.95; a deviation of 0.1 corresponds to a confidence index of 0.90; and a deviation of 0.2 reduces the confidence index to 0.82. This calculation method ensures that the confidence index is inversely proportional to the consistency between measured data and theoretical predictions. Regions with a confidence index greater than 0.9 are considered high-reliability stress concentration locations; regions with a confidence index between 0.8 and 0.9 are considered medium-reliability stress concentration locations; and regions with a confidence index below 0.8 require additional verification methods. For example, for critical structural components, the confidence index of stress concentration locations is usually required to be no less than 0.85 to ensure the reliability of the assessment results.

[0149] The adaptive mesh refinement strategy of this method improves computational efficiency and accuracy by using a finer mesh in regions with large stress gradients and a coarser mesh in regions with gentle stress changes, thus rationally allocating computational resources. A quantitative relationship between the stress field and acoustic properties is established through the acoustic elastic effect, realizing the organic combination of ultrasonic detection results and stress analysis. The density clustering algorithm based on multidimensional features improves the robustness of stress concentration location identification and reduces the impact of noise and measurement errors.

[0150] Optional,

[0151] The steps of adjusting the transmission power and scanning interval of the ultrasonic probe at the stress concentration location, performing fine scanning and acquiring high-resolution data, and updating the defect spatial distribution information based on the high-resolution data include:

[0152] The acoustic energy attenuation compensation coefficient is calculated based on the depth and maximum stress value at the stress concentration location. The acoustic energy attenuation compensation coefficient is multiplied by the initial transmission power to obtain the optimized transmission power. The scanning interval is determined based on the ratio of ultrasonic wavelength to probe diameter.

[0153] A spiral scanning trajectory is constructed with the stress concentration location as the center. The spiral scanning trajectory is determined by both angle and radius parameters. The angle and radius parameters are adjusted according to the scanning interval so that the spiral scanning trajectory completely covers the stress concentration location. High-resolution scanning data is obtained by performing ultrasonic scanning along the spiral scanning trajectory using the optimized transmission power.

[0154] Calculate the boundary curvature value and signal-to-noise ratio of the high-resolution scan data, and construct a weighted fusion coefficient based on the boundary curvature value and signal-to-noise ratio; use the weighted fusion coefficient to fuse the defect location coordinate information and fine scan features, and update the spatial distribution information of the defect, which includes the defect boundary contour and internal structural features.

[0155] For example, sound waves experience energy attenuation when propagating in composite materials. This attenuation is related to material properties, sound wave frequency, and propagation distance. The calculation of the sound wave energy attenuation compensation coefficient considers two main factors: first, geometric diffusion and material absorption attenuation caused by depth; and second, acoustic impedance changes caused by stress concentration. For the depth factor, the attenuation compensation coefficient increases with increasing depth, described by an exponential relationship. For the stress concentration factor, the attenuation compensation coefficient is proportional to the maximum stress value. For example, for a stress concentration region located at a depth of 3.5 mm with a maximum stress of 150 MPa, the calculated attenuation compensation coefficient is 1.42, which means that the transmitted power needs to be increased by 42% to obtain sufficient echo signal strength.

[0156] The initial transmit power of an ultrasonic probe is typically set to an excitation voltage of 50-100V, corresponding to an acoustic power of 0.1-0.2W. For example, with an initial transmit power of 80V, multiplying by an attenuation compensation factor of 1.42 results in an optimized transmit power of 113.6V. The upper limit is set to twice the initial power. When the calculated result exceeds this upper limit, the upper limit is used, and the signal averaging frequency is increased accordingly to improve the signal-to-noise ratio.

[0157] The scanning spacing is determined based on the ratio of ultrasonic wavelength to probe diameter. For carbon fiber composites, the longitudinal wave velocity is approximately 6000 m / s, and the wavelength corresponding to a 5 MHz frequency is 1.2 mm. Probe diameters are typically 6-12 mm; for example, when using an 8 mm diameter probe, the wavelength-to-probe diameter ratio is 0.15. According to acoustic field theory and the sampling theorem, the scanning spacing should not exceed half the wavelength, i.e., 0.6 mm. However, considering the need for higher resolution in stress concentration areas, the actual scanning spacing is further reduced to one-quarter of the wavelength, i.e., 0.3 mm. This setting ensures a good balance between high-frequency detail and spatial resolution.

[0158] A helical scanning trajectory is constructed centered on the stress concentration location. Compared to traditional grid scanning, helical scanning offers higher efficiency and more uniform spatial coverage. The helical trajectory is determined by angle and radius parameters. The angle parameter represents the rotation angle around the center point, and the radius parameter represents the distance to the center point. The angle parameter starts at 0 degrees and increases in fixed steps until the required number of revolutions is completed; the radius parameter increases linearly with the angle. For a specific helical trajectory, the distance between adjacent trajectory points can be controlled by adjusting the angle step size and the radius growth rate, making it equal to or slightly less than the set scan interval of 0.3 mm. The angle and radius parameters are adjusted according to the scan interval to ensure that the helical scanning trajectory completely covers the stress concentration location. For example, for an elliptical stress concentration area of ​​0.8 mm × 0.5 mm, considering edge effects, the scanning range needs to be expanded to 1.2 mm × 0.9 mm. Starting from the center of the ellipse, with an angle step size of 15 degrees and a radius growth rate of 0.02 mm / 15 degrees, the distance between the trajectory points in one spiral (360 degrees) is approximately 0.29 mm, which is less than the set scanning distance of 0.3 mm, ensuring sufficient spatial coverage. A total of 6 spirals are required to completely cover the target area, corresponding to a maximum radius of 0.72 mm, generating a total of 144 scanning points.

[0159] High-resolution scan data is acquired by ultrasonic scanning along a helical scanning trajectory using optimized transmission power. During the scan, the probe stops at each trajectory point and emits ultrasonic waves, receiving the echo signals. To improve the signal-to-noise ratio (SNR), multiple repeated measurements are performed at each point, and the average value is taken, typically 16-32 times. For the optimized transmission power of 113.6V, averaging 24 signals can improve the SNR by approximately 4.9 times. The raw data obtained from the scan is the A-scan signal corresponding to each trajectory point, typically containing 1024-2048 sampling points at a sampling rate of 100MHz, corresponding to a time window of 10.24-20.48μs.

[0160] For high-resolution scan data, the boundary curvature value reflects the shape characteristics of the defect boundary and is an important indicator for distinguishing different types of defects. Curvature calculation is based on the echo time difference between adjacent scan points. The specific process is as follows: First, the time position of the defect echo in each A-scan signal is extracted to construct a defect depth distribution map; then, the depth distribution map is smoothed and edge detected to obtain the defect boundary; finally, the curvature value at each point on the boundary is calculated. For circular defects, the boundary curvature value is relatively uniform, with a standard deviation of less than 0.2; while for crack-like defects, the curvature value increases significantly at the tip, with the maximum curvature value reaching 3-5 times the average. The signal-to-noise ratio (SNR) is calculated as the ratio of the defect echo signal amplitude to the standard deviation of the background noise. High-quality scan data typically has an SNR greater than 20 dB. Weighted fusion coefficients are used to control the weights of initial defect location information and fine scan features in the fusion process. The construction method is as follows: regions with high boundary curvature values ​​(such as defect tips) are given higher weights for fine scanning features, while regions with low curvature values ​​(such as straight defect edges) are given higher weights for initial defect location information. Simultaneously, regions with high signal-to-noise ratios (SNR) have increased weights for fine scanning features, while regions with low SNR have increased weights for initial information. In practical applications, when the curvature value exceeds 1.5 and the SNR exceeds 25 dB, the weight of fine scanning features can reach 0.85; when the curvature value is less than 0.5 and the SNR is less than 15 dB, the weight of initial defect location information increases to 0.7.

[0161] A weighted fusion coefficient is used to fuse defect location coordinates and fine scan features to update the spatial distribution information of the defect. The fusion process employs a weighted average method; for each point on the defect boundary, its final location coordinates are a weighted average of the initial location and the fine scan location. For example, if the initial depth of a point is 3.85 mm, the fine scan depth is 3.92 mm, and the weighted fusion coefficient is 0.75 (fine scan weight), then the fused depth is 3.9025 mm. For the internal structural features of the defect, extraction is primarily based on fine scan data, including internal echo intensity distribution, phase changes, and spectral characteristics. These features can be used to determine the nature of the defect, such as whether it contains inclusions or secondary cracks.

[0162] The updated defect spatial distribution information includes defect boundary contours and internal structural features. The boundary contours are represented as a 3D point cloud, describing the defect's shape and size. For example, for an actual elliptical defect, the updated major axis length is 2.24 mm, the minor axis length is 1.38 mm, and the area is 2.43 mm², representing an improvement in dimensional accuracy of approximately 40% compared to the initial estimate. Internal structural features include acoustic impedance distribution maps, phase jump maps, and spectral energy distribution maps, which collectively reveal the defect's internal composition and properties. For instance, low values ​​in the central region of the acoustic impedance distribution map (more than 30% lower than the surrounding material) indicate a possible cavity within the defect; while high values ​​in certain regions (more than 15% higher than the surrounding material) suggest the possible presence of high-density impurities.

[0163] For specific defect types, characteristic parameters can be extracted for quantitative characterization. For example, for delamination defects, their area, perimeter, thickness, and average acoustic impedance can be extracted; for crack-like defects, their length, opening width, tip radius, and propagation direction can be extracted. These parameters, combined with the material's mechanical properties, can be used to assess the degree of impact of defects on structural integrity. In a case study of a carbon fiber / epoxy resin composite board, a crack defect with a length of 2.24 mm and a tip radius of 0.08 mm was found. Its propagation direction made an angle of 15 degrees with the principal stress direction. The assessment results indicated that this defect had a risk of continued propagation under working loads, requiring repair or enhanced monitoring.

[0164] This method improves the detection sensitivity of deep and high-stress areas by using acoustic energy attenuation compensation technology at stress concentration locations. The spiral scanning trajectory design not only improves scanning efficiency but also achieves uniform coverage of the defect area. Based on the weighted fusion strategy of boundary curvature and signal-to-noise ratio, it makes full use of the complementary advantages of initial scanning and fine scanning, and significantly enhances the accuracy of boundary shape characterization.

[0165] Secondly, a testing system for composite panels is provided, comprising:

[0166] The first unit is used to scan the composite plate to obtain ultrasonic scanning data, divide the composite plate into N acoustic characteristic layers, construct a sound wave propagation model based on the recursive transfer matrix method, calculate the sound wave transfer coefficient matrix of the N acoustic characteristic layers, and calculate the reflection coefficient and transmission coefficient of each layer interface according to the sound wave transfer coefficient matrix.

[0167] The second unit is used to compare the measured ultrasonic scanning data with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient, and dynamically correct the acoustic parameters of each acoustic characteristic layer based on deep learning algorithms and physical constraints.

[0168] The third unit is used to substitute the corrected acoustic parameters into the sound wave propagation model, reconstruct the signal from the ultrasonic scanning data, obtain the real defect signal, and determine the defect location coordinates based on the real defect signal.

[0169] The fourth unit is used to perform mesh generation at the coordinates of the defect location, calculate the stress field distribution data around the defect, acquire the acoustic characteristic change data of the defect area under the action of the stress field distribution data, and determine the stress concentration location.

[0170] The fifth unit is used to adjust the transmission power and scanning interval of the ultrasonic probe for the stress concentration location, perform fine scanning and acquire high-resolution data, and update the defect spatial distribution information based on the high-resolution data;

[0171] The sixth unit is used to output detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level.

[0172] Thirdly, a computer-readable storage medium is provided, having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

Claims

1. A method for testing composite panels, characterized in that, include: Ultrasonic scanning data is obtained by scanning the composite panel. The composite panel is divided into N acoustic characteristic layers. A sound wave propagation model is constructed based on the recursive transfer matrix method. The sound wave transfer coefficient matrix of the N acoustic characteristic layers is calculated. The reflection coefficient and transmission coefficient of each layer interface are calculated based on the sound wave transfer coefficient matrix. Specifically, this includes: performing empirical mode decomposition on the ultrasonic scanning data to obtain intrinsic mode functions (EMFs); calculating the phase function of the EMFs to obtain instantaneous frequency characteristics; combining the instantaneous frequency characteristics with the squared amplitude of the EMFs to construct a time-frequency energy distribution matrix; and determining the interlayer interface positions of the composite panel based on the abrupt change characteristics of the time-frequency energy distribution matrix. The composite panel is divided into N acoustic property layers. Three-dimensional topographic parameters at the interlayer interfaces are obtained. Based on these parameters, a stiffness calculation formula considering interfacial stress is constructed to obtain an interface stiffness matrix characterizing the interface contact characteristics. An ideal interface transfer matrix is ​​constructed based on the wave impedance and thickness of each acoustic property layer. The interface stiffness matrix is ​​multiplied by the ideal interface transfer matrix to obtain a sound wave transfer coefficient matrix. The sound wave transfer coefficient matrix is ​​substituted into a sound wave propagation model, which is constructed based on the material parameters of each acoustic property layer. The wave number of each of the N acoustic property layers is calculated according to the sound wave propagation model. Based on the wave number, the reflection coefficient and transmission coefficient of the interlayer interfaces are calculated. The measured ultrasonic scanning data is compared with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient. The acoustic parameters of each acoustic characteristic layer are dynamically corrected based on deep learning algorithms and physical constraints. Specifically, this involves: inputting the measured ultrasonic scanning data into the waveform encoding layer of a bidirectional long short-term memory neural network to encode it into a waveform feature vector; inputting the waveform feature vector into a multi-scale attention mechanism layer, performing wavelet transform on the waveform feature vector to obtain a time-frequency feature matrix, constructing local attention vectors and global attention vectors based on the time-frequency feature matrix, adaptively fusing the local attention vector and the global attention vector to obtain a fused attention vector, and then combining the fused attention vector with the waveform feature vector... The eigenvectors are multiplied element-wise to obtain a weighted eigenvector; the weighted eigenvectors are input into an acoustic parameter prediction network, which outputs an acoustic parameter set including wave velocity, density, and attenuation coefficient parameters; an optimization objective function is constructed based on a multi-level physical constraint function and waveform reconstruction error; the bidirectional long short-term memory neural network is self-supervised pre-trained using ultrasonic data from a defect-free region, with waveform reconstruction as the pre-training objective; the acoustic parameter set is iteratively optimized based on the optimization objective function, and iteration stops when the parameter change in multiple consecutive iterations is less than a preset change threshold, resulting in a corrected acoustic parameter set; The corrected acoustic parameters are substituted into the sound wave propagation model, and the ultrasonic scanning data is reconstructed to obtain the real defect signal. The defect location coordinates are then determined based on the real defect signal. The process involves: dividing the defect location into a mesh and calculating the stress field distribution data around the defect; acquiring acoustic characteristic change data of the defect area under the influence of the stress field distribution data to determine the stress concentration location; specifically, constructing an initial mesh centered on the defect location coordinates; calculating the square root of the sum of squares of the partial derivatives of stress in the X and Y directions for each mesh element to obtain a stress gradient index; refining mesh elements with stress gradient indices greater than a preset stress gradient threshold; calculating the spatial derivative of the displacement components based on the refined mesh to obtain strain values; multiplying the strain values ​​by the material elastic constants to obtain stress field distribution data; and calculating the change in sound velocity based on the stress field distribution data, where the change in sound velocity is calculated by the principal stress components in the three directions and the corresponding... The product of the acoustic elastic coefficients is determined, and the acoustic impedance correction value is calculated based on the change in sound velocity. Acoustic features corresponding to the change in sound velocity and the acoustic impedance correction value are extracted to construct a feature vector containing the rate of change in sound velocity, the rate of change in acoustic impedance, and the spectral energy distribution. The weighted Euclidean distance between the feature vectors is calculated, and the neighborhood radius is determined based on the average distance of the K nearest neighbors around each feature point. Density clustering of the feature points is performed using the neighborhood radius, and the clustering results are mapped back to spatial coordinates to determine the stress concentration location. The ratio of the maximum stress to the nominal stress within the stress concentration location is calculated to obtain the stress concentration coefficient. The deviation between the measured acoustic features and the theoretical acoustic features is substituted into the exponential decay function to calculate the confidence index, and the reliability of the stress concentration location is determined based on the confidence index. The ultrasonic probe's transmission power and scanning interval are adjusted for the stress concentration location to perform fine scanning and acquire high-resolution data. The defect spatial distribution information is then updated based on the high-resolution data. The system outputs detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level.

2. The method according to claim 1, characterized in that, The steps to construct the optimization objective function include: The multi-level physical constraints include: wave equation constraint terms constructed based on sound wave propagation theory; parameter range constraint terms constructed based on material acoustic properties; and interface continuity constraint terms constructed based on interlayer interface properties. The waveform reconstruction error includes: a mean square error term based on the amplitude of the time-domain waveform; and a phase error term based on the Hilbert transform. The optimization objective function adopts a phased weight configuration strategy based on the iterative process. The weight coefficients of the waveform reconstruction error term and the physical constraint term are dynamically adjusted according to the waveform reconstruction error value, the degree of violation of physical constraints and the difference in acoustic impedance of the material interface.

3. The method according to claim 1, characterized in that, The steps of substituting the corrected acoustic parameters into the sound wave propagation model, reconstructing the ultrasonic scanning data to obtain the true defect signal, and determining the defect location coordinates based on the true defect signal include: Substitute the corrected acoustic parameters and the sound wave transmission coefficient matrix into the sound wave propagation model, construct an interlayer multiple reflection compensation factor based on the reflection coefficient of the adjacent interlayer interface and the interlayer sound wave propagation delay, and perform an inverse Fourier transform after multiplying the interlayer multiple reflection compensation factor with the frequency domain response of the ultrasonic scanning data to obtain the reconstructed signal. Extract the time-domain amplitude and phase features of the reconstructed signal, perform a differential operation between the time-domain amplitude features and the theoretical reflection time calculated based on the corrected acoustic parameters, perform a differential operation between the phase features and the theoretical reflection phase, and identify the real defect signal based on the differential operation results; Based on the corrected acoustic parameters, a sound wave propagation time equation is established, and the defect location coordinates are obtained by solving the sound wave propagation time equation according to the time characteristics of the real defect signal.

4. The method according to claim 1, characterized in that, The steps of adjusting the transmission power and scanning interval of the ultrasonic probe at the stress concentration location, performing fine scanning and acquiring high-resolution data, and updating the defect spatial distribution information based on the high-resolution data include: The acoustic energy attenuation compensation coefficient is calculated based on the depth and maximum stress value at the stress concentration location. The acoustic energy attenuation compensation coefficient is multiplied by the initial transmission power to obtain the optimized transmission power. The scanning interval is determined based on the ratio of ultrasonic wavelength to probe diameter. A spiral scanning trajectory is constructed with the stress concentration location as the center. The spiral scanning trajectory is determined by both angle and radius parameters. The angle and radius parameters are adjusted according to the scanning interval so that the spiral scanning trajectory completely covers the stress concentration location. High-resolution scanning data is obtained by performing ultrasonic scanning along the spiral scanning trajectory using the optimized transmission power. Calculate the boundary curvature value and signal-to-noise ratio of the high-resolution scan data, and construct a weighted fusion coefficient based on the boundary curvature value and signal-to-noise ratio; use the weighted fusion coefficient to fuse the defect location coordinate information and fine scan features, and update the spatial distribution information of the defect, which includes the defect boundary contour and internal structural features.

5. A detection system for composite panels, used to implement the method according to any one of claims 1-4, characterized in that, include: The first unit is used to scan the composite plate to obtain ultrasonic scanning data, divide the composite plate into N acoustic characteristic layers, construct a sound wave propagation model based on the recursive transfer matrix method, calculate the sound wave transfer coefficient matrix of the N acoustic characteristic layers, and calculate the reflection coefficient and transmission coefficient of each layer interface according to the sound wave transfer coefficient matrix. The second unit is used to compare the measured ultrasonic scanning data with the theoretical waveform calculated based on the reflection coefficient and transmission coefficient, and dynamically correct the acoustic parameters of each acoustic characteristic layer based on deep learning algorithms and physical constraints. The third unit is used to substitute the corrected acoustic parameters into the sound wave propagation model, reconstruct the signal from the ultrasonic scanning data, obtain the real defect signal, and determine the defect location coordinates based on the real defect signal. The fourth unit is used to perform mesh generation at the coordinates of the defect location, calculate the stress field distribution data around the defect, acquire the acoustic characteristic change data of the defect area under the action of the stress field distribution data, and determine the stress concentration location. The fifth unit is used to adjust the transmission power and scanning interval of the ultrasonic probe for the stress concentration location, perform fine scanning and acquire high-resolution data, and update the defect spatial distribution information based on the high-resolution data; The sixth unit is used to output detection data, which includes defect location coordinates, stress concentration distribution data, and defect hazard level.

6. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 4.

Images