Micro-damage nonlinear Lamb wave nondestructive measurement detection and imaging method

Abstract

The invention provides a micro-damage non-linear Lamb wave nondestructive measurement detection and imaging method, which comprises the following steps of: exciting a non-linear Lamb wave to generate non-linear interaction with micro-damage in a tested sample by adopting a composite signal containing two different frequency components, and generating a non-linear characteristic response; synchronously acquiring a time domain signal and a frequency domain signal through a hybrid sensing array; respectively carrying out feature extraction, and then carrying out weighted fusion to obtain an enhanced micro-damage nonlinear feature vector; constructing a quantitative model fusing physical information, inputting the micro-damage nonlinear feature vector and the material physical parameters of the tested sample into the model, and outputting a quantitative measurement result of the micro-damage geometric parameters; executing a dynamic adaptive imaging algorithm based on the quantitative measurement result and the multi-modal signal, and generating a micro-damage two-dimensional imaging result; and feeding back a two-dimensional imaging result to a control unit of the hybrid sensing array, and adjusting spatial layout parameters of the sensing array so as to optimize the micro-damage detection precision of a subsequent detection period.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of nondestructive testing, and particularly relates to a micro-damage nonlinear Lamb wave nondestructive metrology detection and imaging method. BACKGROUND

[0002] In the industrial fields of aerospace, high-end equipment manufacturing, energy and chemical industry, metal and composite structural components are prone to produce micro defects such as closed cracks, micro pores and early fatigue damage with a size of microns to sub-millimeters under the action of complex alternating loads and harsh environments. Such micro damages have a high concealment as the macro mechanical properties do not obviously decay in the early stage, but they have a significant damage accumulation and rapid expansion characteristics under external excitation, which is a key inducement to cause sudden failure of the structure and seriously threatens the service safety and reliability of the equipment. Therefore, it has become a bottleneck problem to be urgently broken through in the field of structural health management to develop a nondestructive testing and monitoring technology capable of early detection, accurate quantitative evaluation and even visualization positioning of micro damages.

[0003] At present, the widely used nondestructive testing technologies in the industry, such as traditional ultrasonic testing and ray detection based on body waves, have a detection resolution and signal-to-noise ratio limited by physical principles, and are not sensitive to micro damages much smaller than the wavelength, have a high recognition threshold, and are difficult to effectively capture the weak nonlinear acoustic signal characteristics such as micro-damage and material nonlinear interaction generated high-order harmonics and mixing. Guided wave detection technology represented by Lamb wave has attracted attention due to its unique advantage of rapid scanning of large-scale plate and tubular structures, however, most of the existing technical systems are established within the framework of linear wave theory. The linear Lamb wave method mainly relies on linear signal parameters such as amplitude attenuation, time difference, phase change of the wave, which are not sensitive to the weak nonlinear perturbation caused by micro damages, resulting in an inherent "linear detection threshold" that cannot effectively identify and quantitatively characterize micro damages below the threshold, and most applications still remain at the qualitative judgment level of whether there is damage.

[0004] In the aspect of sensing, a single sensing mechanism is difficult to comprehensively capture the complex nonlinear wave response caused by micro damages. The piezoelectric ceramic (PZT) as a representative of active sensing elements has a fast response speed and strong driving ability, and can well obtain time-domain dynamic waveforms, but its signals are easily affected by electromagnetic interference in the field, and has limited sensitivity to signal components outside a specific frequency band. The optical sensor represented by the fiber Bragg grating (FBG) has high stability and anti-interference ability in the frequency domain, but its ability to capture transient impact and wideband dynamic strain is relatively weak. The fragmentation of time-domain and frequency-domain signal sensing ability leads to incomplete signal dimension obtained from a single sensor, and it is difficult to construct an "information map" reflecting the complete nonlinear characteristics of micro damages.

[0005] Furthermore, at the level of signal processing and damage assessment models, current mainstream methods tend to adopt a purely data-driven modeling approach. These models heavily rely on training with a large number of labeled damage samples, while real-world samples of micro-damages are extremely expensive to obtain and scarce, leading to a high risk of overfitting and poor generalization performance. More importantly, these "black box" models rely entirely on the statistical correlation between input and output, ignoring the physical laws that Lamb wave propagation must follow (such as the elastic wave equation). This can cause the model's predictions to violate basic physical principles, resulting in "physical inconsistencies," and casting doubt on the physical credibility and reliability of the quantitative results (such as damage size and depth).

[0006] Furthermore, from a system perspective, traditional detection systems typically employ a pre-defined sensor network layout, lacking the ability to adaptively adjust based on the specific morphology and damage state of the object being measured. Corresponding imaging algorithms often have resolutions limited by the Rayleigh criterion, making it difficult to exceed wavelength limits. The entire detection process is mostly open-loop, meaning it terminates after a single excitation, acquisition, and imaging operation, failing to feed the imaging or measurement results back to the sensing system for dynamic optimization of subsequent detection strategies (such as adjusting sensor layout to focus on suspected areas). Consequently, it is difficult to consistently maintain high-precision detection capabilities and robustness under complex and changing operating conditions.

[0007] In summary, existing technologies still face a series of interconnected technical bottlenecks in the early and accurate perception of multi-scale micro-damage, physically reliable quantitative assessment, and adaptive and efficient detection imaging, which limit their application effectiveness in industrial-grade high-reliability scenarios. Summary of the Invention

[0008] To address the shortcomings and deficiencies of existing technologies, this invention provides a non-linear Lamb wave non-destructive metrological detection and imaging method and system for micro-damage. It aims to solve the problems of traditional linear detection technology, such as high lower limit of micro-damage identification, low quantitative accuracy, weak cross-material generalization ability, single sensor signal dimension, and lack of closed-loop optimization mechanism.

[0009] The scheme first uses a composite signal containing two different frequency components to drive a sparsely arranged piezoelectric ceramic array to excite a nonlinear Lamb wave in the test sample. This wave interacts nonlinearly with the micro-damage, generating a nonlinear characteristic response containing sum, difference, and harmonic frequencies, providing a precise feature carrier for micro-damage identification. Subsequently, a hybrid sensing array composed of piezoelectric sensors and fiber optic grating sensors is used to simultaneously acquire the time-domain displacement signal and frequency-domain strain signal corresponding to the above characteristic response. During the acquisition process, a reference clock adapted to the synchronization requirements of multiple channels is generated by the control unit. Preprocessing techniques such as wavelet denoising, zero drift elimination, and temperature drift compensation are combined to ensure signal quality. At the same time, the sensing array is dynamically deployed based on the digital twin model of the test sample to maximize the efficiency of micro-damage signal acquisition.

[0010] In the feature extraction stage, the scheme inputs the time-domain and frequency-domain signals into parallel time-domain processing branches and frequency-domain processing branches, respectively. After extracting features in each domain through one-dimensional convolution, the two types of features are fused by element-wise weighted multiplication using attention weights calculated based on the matching degree between time-domain feature energy and frequency-domain frequency. At the same time, comparative learning optimization is introduced with query samples, positive samples with damage, and negative samples without damage. Finally, a fixed-dimensional enhanced micro-damage nonlinear feature vector is output.

[0011] In terms of quantitative metrology, the scheme constructs a deep transfer coupled quantitative model that integrates physical information. The aforementioned micro-damage feature vectors and the material physical parameters of the test sample (including third-order nonlinear elastic constants, second-order elastic constants, and density) are input into the model. The multi-source information is weighted and fused through an attention mechanism, and the nonlinear Lamb wave equation is embedded as a physical constraint. The model prediction conforms to the physical propagation law by calculating the wave displacement response residual. At the same time, a deep transfer module based on the maximum mean difference is introduced to achieve feature alignment between the digital twin virtual sample (source domain) and the real sample (target domain). Even with a small number of samples, it can still maintain cross-material detection accuracy. Finally, it outputs quantitative results of geometric parameters such as micro-damage size and depth, with a metrological error of no more than 3%.

[0012] In the imaging and optimization phase, the scheme, based on the aforementioned quantitative results and multimodal signals, constructs a forward model using compressed sensing theory. Through inversion calculations, it generates two-dimensional imaging results of micro-damage, achieving an imaging resolution of less than 1 / 10 of the wavelength. Simultaneously, time-frequency attention weights are introduced to enhance signal components in the damage feature region, improving the imaging signal-to-noise ratio. Furthermore, the imaging results are fed back to the sensor array control unit, dynamically adjusting the array spatial layout parameters to form a closed-loop system of "detection-measurement-imaging-optimization." After 1-2 iterations, the system's anti-interference capability against complex working conditions such as vibration and temperature fluctuations is significantly improved, and the detection accuracy across metals and composite materials remains at no less than 90%. This meets the long-term needs of aerospace, high-end equipment manufacturing, and other fields for high-precision and robust micro-damage detection.

[0013] The specific technical solution adopted by this invention to solve its technical problem is as follows:

[0014] A nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves includes:

[0015] A composite signal containing two different frequency components is used to drive a sparsely arranged piezoelectric ceramic array to excite a nonlinear Lamb wave in the test sample. The nonlinear Lamb wave interacts nonlinearly with the micro-damage in the test sample, generating a nonlinear characteristic response containing sum and difference frequencies or harmonics.

[0016] The time-domain and frequency-domain signals corresponding to the nonlinear characteristic response are simultaneously acquired by a hybrid sensing array composed of piezoelectric sensors and fiber Bragg grating sensors.

[0017] The time-domain signal and the frequency-domain signal are respectively input into the parallel time-domain processing branch and the frequency-domain processing branch for feature extraction. Then, the extracted time-domain features and frequency-domain features are weighted and fused using an attention mechanism to obtain an enhanced micro-damage nonlinear feature vector.

[0018] A quantitative model integrating physical information is constructed. The nonlinear feature vector of the micro-damage and the material physical parameters of the test sample are input into the model. The nonlinear Lamb wave equation is introduced as a physical constraint during the training and prediction process of the model, and the quantitative measurement results of the geometric parameters of the micro-damage are output.

[0019] Based on the quantitative measurement results and multimodal signals, a dynamic adaptive imaging algorithm is executed to generate two-dimensional imaging results of micro-damage; the two-dimensional imaging results are fed back to the control unit of the hybrid sensor array to adjust the spatial layout parameters of the sensor array in order to optimize the micro-damage detection accuracy in subsequent detection cycles.

[0020] Furthermore, the composite signal containing two different frequency components is formed by superimposing two sinusoidal signals of adjacent frequencies, and drives the sparse piezoelectric ceramic array after the gain is adjusted by an adaptive power amplifier according to the characteristics of the test sample.

[0021] Furthermore, in the hybrid sensing array, the piezoelectric sensor acquires the time-domain displacement response signal, and the fiber Bragg grating sensor acquires the frequency-domain strain signal based on Bragg wavelength modulation. Before acquisition, the multi-mode signal is preprocessed, including using wavelet transform combined with an improved soft-hard compromise threshold function for noise reduction, eliminating zero drift of the piezoelectric sensor signal through moving average, and introducing a compensation coefficient based on the digital twin temperature field of the test sample to eliminate temperature drift of the fiber Bragg grating sensor signal. The synchronous acquisition is achieved through a reference clock generated by the control unit. The reference clock frequency is adapted to the multi-channel synchronization requirements. Based on the reference clock, a trigger signal is sent to discretely sample the preprocessed signal. The sampling period and the number of sampling points are set to cover the complete Lamb wave propagation period.

[0022] Furthermore, the initial deployment and adjustment of the hybrid sensor array are based on the digital twin model of the test sample: the sensor layout optimization algorithm is run by the computing workstation, with the goal of maximizing the signal-to-noise ratio of the observed signal, and the optimal layout parameters are obtained and sent to the control unit, which drives the hardware link to complete the sensor array layout; when adjusting the spatial layout of the sensor array, the goal is to balance the signal-to-noise ratio and hardware resource consumption, and the final layout parameters are determined by a preset trade-off coefficient. The adjustment is used to adapt to the micro-damage signal acquisition requirements of subsequent detection cycles.

[0023] Furthermore, both the parallel temporal and frequency domain processing branches employ one-dimensional convolution to extract corresponding domain features. The convolution process includes preset convolution weights and bias terms. The attention mechanism calculates the feature weights of each domain by matching the temporal feature energy with the frequency feature frequency, and then performs element-wise weighted fusion of the extracted temporal and frequency features. Contrastive learning is also introduced during feature extraction to construct a triple containing query samples, positive samples with micro-damage, and negative samples without damage. The feature distance between samples is optimized through a loss function, and finally, a fixed-dimensional micro-damage nonlinear feature vector is output.

[0024] Furthermore, the material physical parameters of the test sample include the third-order nonlinear elastic constant, the second-order elastic constant, and the material density. The input dimension of the quantization model is determined by the dimension of the micro-damage nonlinear feature vector and the number of material physical parameters. The nonlinear Lamb wave equation serves as a physical constraint. A quantization model is constructed through a physical information neural network. The residual of the wave equation is calculated on the predicted Lamb wave displacement response value output by the model. The magnitude of the residual is used to measure the degree of fit between the model prediction result and the physical propagation law of Lamb waves. The quantization model also includes a deep transfer module, which aligns the features of the source domain of the corresponding digital twin virtual sample with the target domain of the corresponding real sample of the test sample based on the maximum mean difference, ensuring the model's cross-material generalization ability.

[0025] Furthermore, the quantization model is trained using a multi-objective weighted loss function. The total loss includes data loss, physical loss, and transfer loss: the data loss is calculated based on the difference between the predicted values ​​of the micro-damage geometric parameters and the true labels; the physical loss is calculated based on the residuals of the wave equation; and the transfer loss is calculated based on the maximum mean difference between the source domain and the target domain. The training process uses an adaptive optimizer, and the learning rate is adaptively adjusted according to a preset decay coefficient until the loss converges.

[0026] Furthermore, the dynamic adaptive imaging algorithm combines compressed sensing theory, treats micro-damage as a sparsely distributed scattering source, constructs a forward model based on the Lamb wave equation, and obtains the two-dimensional imaging matrix of micro-damage through inversion calculation; during the imaging process, time-frequency attention weights are introduced to enhance the signal components of the micro-damage feature region, and the time-frequency attention weights are consistent with the time-domain and frequency-domain feature weights obtained in the feature extraction process.

[0027] Furthermore, a micro-damage nonlinear Lamb wave nondestructive metrological testing and imaging system includes:

[0028] Excitation module: It consists of a composite signal generator and a sparsely arranged piezoelectric ceramic array. The composite signal generator can generate a composite signal containing two different frequency components, which drives the piezoelectric ceramic array to excite a nonlinear Lamb wave in the test sample. The nonlinear Lamb wave interacts nonlinearly with the micro-damage in the test sample to generate a nonlinear characteristic response containing sum and difference frequencies or harmonics.

[0029] Acquisition module: a hybrid sensing array consisting of piezoelectric sensors and fiber optic grating sensors, and a control unit for synchronous control. The hybrid sensing array can synchronously acquire time-domain and frequency-domain signals corresponding to nonlinear characteristic responses, and the control unit can generate a reference clock to achieve synchronous acquisition of multi-channel signals.

[0030] Feature extraction module: It is used to extract features by inputting time-domain signals and frequency-domain signals into parallel time-domain processing branches and frequency-domain processing branches respectively, and to weight and fuse time-domain features and frequency-domain features through attention mechanism to output enhanced micro-damage nonlinear feature vector;

[0031] Quantitative modeling module: Used to construct a quantitative model that integrates physical information. The nonlinear feature vector of micro-damage and the material physical parameters of the test sample are input into the model. During the model training and prediction process, the nonlinear Lamb wave equation is introduced as a physical constraint, and the quantitative measurement results of the geometric parameters of micro-damage are output.

[0032] Imaging optimization module: It is used to execute dynamic adaptive imaging algorithm based on quantitative measurement results and multimodal signals to generate two-dimensional imaging results of micro-damage, and feed the two-dimensional imaging results back to the control unit to adjust the spatial layout parameters of the hybrid sensor array in order to optimize the micro-damage detection accuracy in subsequent detection cycles.

[0033] And a computer device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the method described above.

[0034] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0035] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0036] This technology effectively breaks through the lower limit of micro-damage identification in traditional linear detection techniques. By capturing weak nonlinear characteristics through the coupling effect of nonlinear Lamb waves with micro-damage, it achieves early and accurate identification of micro-damage. At the same time, it establishes a measurement benchmark for the geometric parameters of micro-damage based on a quantitative model that integrates physical information, thus solving the core pain point of "easy to qualitatively identify but difficult to quantify" in existing technologies and providing a unified and reliable quantitative basis for micro-damage assessment.

[0037] Significantly improving the reliability and integrity of signal acquisition, a hybrid sensing array composed of piezoelectric sensors and fiber optic grating sensors is adopted to complementarily cover the dynamic response in the time domain and the characteristic distribution in the frequency domain, making up for the shortcomings of single sensors, such as limited signal dimension, weak anti-interference ability, or insufficient dynamic response. At the same time, combined with the dynamic deployment of the sensing array based on the digital twin model, the signal acquisition efficiency in the target area is further optimized, ensuring the effective acquisition of micro-damage features.

[0038] To enhance the accuracy of micro-damage feature mining, a parallel time-domain and frequency-domain processing branch combined with an attention-weighted fusion mechanism is used to selectively extract nonlinear features related to micro-damage. This is further enhanced by contrastive learning optimization to reduce the influence of noise interference and irrelevant features, providing high-quality feature input for subsequent quantitative measurement and imaging.

[0039] To improve the cross-scenario adaptability and physical rationality of the detection model, the Lamb wave equation is embedded in the quantification model as a physical constraint to avoid the "physical inconsistency" problem that is prone to occur in pure data-driven models. At the same time, a deep transfer module is introduced, which can still achieve stable detection across different material types such as metals and composite materials when real samples of micro-damage are scarce, reducing the dependence on a large number of real samples.

[0040] To optimize micro-damage imaging and system robustness, a dynamic adaptive imaging algorithm combining compressed sensing theory and time-frequency attention mechanism is used to improve imaging resolution and damage area identification, clearly presenting the spatial morphology of micro-damage. Furthermore, through a closed-loop mechanism of "detection-measurement-imaging-optimization," the sensor array layout is adjusted using imaging results feedback, continuously improving the system's anti-interference capability against complex working conditions such as vibration and temperature fluctuations, meeting the long-term high-reliability requirements for structural health monitoring in aerospace, high-end equipment manufacturing, energy and chemical industries. Attached Figure Description

[0041] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0042] Figure 1 This is a roadmap of the measurement method according to an embodiment of the present invention;

[0043] Figure 2 These are physical diagrams of embodiments of the present invention;

[0044] Figure 3 This is a chart showing the resolution and quantitative accuracy of micro-damage detection in an embodiment of the present invention.

[0045] Figure 4 This is a diagram illustrating the efficiency of imaging signal-to-noise ratio improvement and closed-loop optimization in an embodiment of the present invention. Detailed Implementation

[0046] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in detail:

[0047] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0048] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0049] This invention discloses a non-destructive metrological detection and imaging technology for micro-damage using nonlinear Lamb waves, aiming to solve the technical problems of high lower limit of micro-damage identification, low quantitative accuracy, and weak cross-material generalization in traditional linear detection. First, an integrated link is constructed, consisting of an excitation unit composed of a composite signal generator and a sparse PZT array, and a PZT-fiber Bragg grating hybrid sensor array, enabling intelligent dynamic deployment of the sensors based on a digital twin model. Second, a nonlinear Lamb wave is excited using a dual-frequency composite signal, utilizing the nonlinear coupling between the wave and the micro-damage to generate harmonic and difference frequency characteristic carriers. Subsequently, multi-mode signals are synchronously acquired via multi-level signal conditioning and an FPGA, and nonlinear features are accurately mined through a dual-branch contrastive learning-time-frequency attention network. Furthermore, a deep transfer-coupled quantization model integrating physical information is constructed, embedding physical constraints of the Lamb wave equation and a cross-domain transfer module to achieve quantitative measurement of micro-damage geometric parameters. Finally, a high-precision two-dimensional image is generated by combining compressed sensing and dynamic adaptive imaging algorithms, and the imaging results are used to optimize the sensor layout. This invention breaks through the traditional linear detection lower limit, achieves subwavelength imaging resolution, has strong cross-material generalization ability, and significantly improves detection robustness and quantitative accuracy, making it suitable for various non-destructive testing scenarios for micro-damage in structures.

[0050] This invention overcomes the limitations of linear detection, enabling accurate early identification of micro-damage; it constructs a quantitative model with physical constraints, establishing a measurement benchmark for damage geometric parameters (size, depth); it integrates multimodal sensing and intelligent algorithms to improve the comprehensiveness and cross-domain adaptability of feature representation; and it forms a closed loop of "detection-measurement-imaging-optimization" to dynamically improve detection robustness and accuracy, meeting the high-reliability monitoring requirements for micro-damage of key components.

[0051] The following detailed embodiments, in conjunction with the accompanying drawings, further illustrate and describe the specific implementation process of the present invention:

[0052] like Figure 1 , Figure 2 As shown, this embodiment provides a specific implementation process for a micro-damage nonlinear Lamb wave nondestructive metrology and imaging technology, which can be referred to in the following steps:

[0053] Step (1): Construct an excitation unit consisting of a composite signal generator, an adaptive power amplifier, and a sparse PZT piezoelectric ceramic array; construct an integrated "acquisition-processing-control" link with a PZT-fiber grating hybrid sensor array, a multi-level signal conditioning unit, an FPGA timing controller, and an edge computing workstation; after fixing the test sample, send adaptive layout instructions to the FPGA through the edge computing workstation to realize the intelligent dynamic deployment of the sensor array and maximize the acquisition efficiency of micro-damage signals.

[0054] Step (2): Use a composite signal generator to generate a nonlinear excitation signal that is adapted to the characteristics of the sample; after being amplified by an adaptive power amplifier, drive the sparse PZT piezoelectric ceramic array to excite a nonlinear Lamb wave in the sample. The wave propagates precisely along the surface of the sample and is nonlinearly coupled with the micro-damage, providing a feature carrier for micro-damage identification.

[0055] Step (3): The PZT-fiber grating hybrid sensor array acquires multi-modal signals. After preprocessing by a multi-level signal conditioning unit, the FPGA timing controller completes the synchronous acquisition of multi-channel signals. In the edge computing workstation, a dual-branch contrastive learning-time-frequency attention network is used to achieve accurate mining of micro-damage features.

[0056] Step (4): In the edge computing workstation, based on the extracted nonlinear feature parameters, a deep migration micro-damage-nonlinear Lamb wave feature coupling quantification model that integrates physical information is constructed; this model breaks through the lower limit of micro-damage identification of traditional linear detection and realizes the quantitative measurement benchmark of micro-damage geometric parameters.

[0057] Step (5): Based on the coupled model and multimodal feature data, a dynamic adaptive imaging algorithm is executed in the edge computing workstation to generate a two-dimensional high-precision imaging result of micro-damage; at the same time, the imaging result is fed back to the FPGA timing controller to trigger the secondary optimization deployment of the sensor array, forming a closed-loop system of "detection-measurement-imaging-optimization", which greatly improves the robustness and accuracy of micro-damage detection.

[0058] As a preferred embodiment, step (1) is as follows:

[0059] Once the test sample is fixed, the edge computing workstation runs a sensor layout optimization algorithm based on the sample's digital twin model. The objective function of this algorithm is... Defined in specific layout parameters Below, observed signal The amount of information contained, such as the expected signal-to-noise ratio.

[0060] Optimal layout parameters The calculation formula is:

[0061]

[0062] The formula for calculating the signal-to-noise ratio is:

[0063]

[0064] In the formula, and Let Variance be the variance of the signal and noise. The optimization process aims to maximize information indicators such as SNR in the target region. .

[0065] Calculated optimal layout instructions The instruction is sent to the FPGA timing controller. The FPGA interprets this instruction into specific hardware control logic, driving its internal multi-channel analog switch matrix to physically connect the specified excitation and reception channels, thereby realizing the intelligent dynamic deployment of the sensor array.

[0066] As a preferred embodiment, step (2) is as follows:

[0067] This invention employs composite signals to optimize nonlinear interactions, assuming the center frequency of the desired excited Lamb wave mode is... The baseband signal generated by the load signal generator Represented as:

[0068]

[0069] In the formula, , For two adjacent input frequencies; , The amplitude. This weak electrical signal Amplification is required to drive the PZT array elements. The adaptive amplifier provides the following driving voltage:

[0070]

[0071] In the formula, G represents the gain.

[0072] Amplified signal The sparse PZT array elements selected in driving step (1) are used to excite Lamb waves in the sample via the inverse piezoelectric effect. When the propagating Lamb waves... When encountering micro-damage such as closed cracks, nonlinear phenomena beyond reflection and transmission occur. For contact-type micro-damage, the stress-strain relationship is no longer linear, but exhibits hysteresis, bilinear, or more complex nonlinear constitutive relationships. Let the displacement field be u, then the wave equation is:

[0073]

[0074] In the formula, It is a second-order elastic constant; The third elastic constant represents the classical nonlinearity of the material. The ellipsis in the formula represents a higher-order correction term related to the nonlinear response of micro-damage.

[0075] incident wave field As an excitation, when applied to a nonlinear system, the system output contains new frequency components not present in the input frequency. This is for dual-frequency excitation. In the scattered field, the following will appear:

[0076] High-frequency harmonics: such as ;

[0077] Sum frequency and difference frequency: ;

[0078] These nonlinear spectral sidebands constitute the feature carriers for micro-damage identification.

[0079] As a preferred embodiment, step (3) is as follows:

[0080] This invention proposes using a hybrid sensing signal of "PZT time domain + FBG frequency domain", with a PZT subarray acquiring the Lamb wave time domain displacement response. The original signal is:

[0081]

[0082] In the formula, It is an effective signal containing damage characteristics; It is Gaussian noise; This represents zero drift error.

[0083] The FBG subarray acquires the frequency domain signal of the strain field based on Bragg wavelength modulation, which is then converted to:

[0084]

[0085] In the formula, Angular frequency; Effective signal frequency domain distribution; This is scattering noise; This is due to temperature drift error.

[0086] As a preferred approach, the corresponding demodulation scheme employs a tunable matched grating demodulation method: the wavelength modulation signal output from the FBG subarray is first boosted by a narrowband optical amplifier (to prevent weak signals from being drowned out by noise), and then input into the tunable matched grating demodulation module; this module dynamically adjusts the temperature or stress parameters of the matched grating through control commands issued by the edge computing workstation, so that the Bragg wavelength of the matched grating matches the sensing wavelength of the FBG subarray in real time; when the two wavelengths match, the optical power output by the demodulation module reaches its peak value, and the wavelength offset of the FBG subarray is obtained by recording the matched grating parameters corresponding to the peak value; finally, combined with the wavelength-strain sensitivity coefficient of the FBG sensor (which can be obtained in advance through static calibration), the wavelength offset is converted into a strain field frequency domain signal to complete the demodulation.

[0087] At the same time, the FPGA timing controller sends a synchronous trigger signal to ensure that the dual-mode signals are time-aligned.

[0088] This invention employs an improved soft-hard compromise threshold function for signal noise reduction:

[0089]

[0090] In the formula, This is a db8 ​​wavelet transform. The threshold for Stein's unbiased risk estimation.

[0091] PZT signals are eliminated by using a moving average to remove zero drift:

[0092]

[0093] In the formula, The moving average window length is set to 50. The sampling period.

[0094] FBG signal introduces a compensation coefficient based on digital twin temperature field. :

[0095]

[0096] in, This is the difference between the actual temperature of the test area of ​​the sample and the system's reference operating temperature.

[0097] As a preferred embodiment, the implementation of the digital twin temperature field can refer to the following implementation process:

[0098] Basic model construction: In the digital twin platform of the edge computing workstation, a three-dimensional finite element heat conduction model is constructed based on the geometric parameters of the test sample (such as size, shape, and material properties); physical parameters such as thermal conductivity, specific heat capacity, and thermal expansion coefficient of the sample material are embedded in the model, and boundary conditions of the test environment (such as air convection heat transfer coefficient and ambient temperature fluctuation range) are defined.

[0099] Temperature field simulation calculation: Solve the heat conduction equation using finite element software (such as ANSYS, ABAQUS) to obtain the temperature distribution cloud map of the entire sample area and determine the temperature field data at the deployment location of the FBG sensor.

[0100] Experimental calibration and correction: High-precision thermocouples are attached to key locations on the sample surface (including FBG sensor deployment points), and actual temperature data is collected under testing conditions; the simulated temperature is compared with the actual temperature, and the boundary conditions of the finite element model are corrected by the least squares method (such as adjusting the convective heat transfer coefficient) to ensure that the error between the simulated temperature and the actual temperature is ≤0.5℃, thus ensuring the accuracy of the digital twin temperature field.

[0101] Compensation coefficient correlation: Based on the corrected temperature field data, a mapping relationship between the temperature drift compensation coefficient and temperature change is established and called in real time to the FBG signal temperature drift compensation link.

[0102] Using an 8th-order elliptic filter, power frequency and high-frequency noise are filtered out, while nonlinear characteristic frequencies are preserved, and the transfer function is... exist Take 1 at time. To protect broadband.

[0103] Multi-channel sampling time difference can lead to phase shift and misjudgment of damage location. FPGA achieves synchronization through the following steps:

[0104] Internal PLL generates a reference clock:

[0105]

[0106] In the formula, =100MHz.

[0107] Based on the clock-triggered signal, the conditioned signal is discretely sampled:

[0108]

[0109] In the formula, The sampling period is k = 0, 1, 2, ..., N-1, N = 4096, covering the complete wave propagation period.

[0110] This invention innovatively employs dual-branch contrastive learning-time-frequency attention network (DBCL-TFAN) feature mining to address the problem of insufficient sensitivity to weak features:

[0111] Input layer:

[0112] PZT time domain data With FBG frequency domain data Input the two branches separately;

[0113] Feature extraction and attention enhancement:

[0114] Branch 1 (PZT) is obtained through 1D convolution. , These are the temporal features extracted after PZT branch convolution; The convolution weights for the PZT branch; This is the bias term for the PZT branch. Branch 2 (FBG) yields... , The frequency domain features are extracted after FBG branch convolution; The convolution weights for the BG branch; This is the bias term for the FBG branch. Then, it is weighted in the time domain. , for Energy, frequency domain weights Weighted, For frequency matching degree, we get , , It is the element-wise product.

[0115] Comparative learning and output:

[0116] Construct triplet samples and query Positive sample p contains damage, while negative sample n does not. The loss function is... optimization, The square of the L2 norm is used to finally fuse features. A 16-dimensional micro-damage feature vector is output through a fully connected layer. .

[0117] As a preferred embodiment, step (4) is as follows:

[0118] The model's input is data-driven feature information fused with physical information, defined as:

[0119]

[0120]

[0121] In the formula, These are the third-order nonlinear elastic constant, the second-order elastic constant, and the density of the material, respectively. =20; =6.

[0122] Multi-source inputs are weighted and fused using an attention mechanism to highlight feature components strongly correlated with damage. The fusion process is as follows:

[0123]

[0124] In the formula, A vector of physical parameters; , This is a learnable weight matrix. For attention weights, , For attention parameters.

[0125] Based on the Physical Information Neural Network (PINN), the nonlinear Lamb wave equation from step 2 is used as a soft-constrained embedded model. The residuals of the physical equations are calculated using automatic differentiation techniques, achieving collaborative modeling driven by both data and mechanisms. (This is followed by a seemingly unrelated sentence about fusion features.) Decoding is performed to obtain the predicted values ​​of the Lamb wave displacement response. The residual terms of the wave equation are calculated using automatic differentiation:

[0126]

[0127] The smaller the residual, the more the model's prediction results conform to the laws of physical propagation, effectively avoiding the "black box" defect of purely data-driven models.

[0128] This invention employs a deep migration module based on Maximum Mean Difference (MMD) to address the problems of scarce real-world micro-damage samples and insufficient generalization ability across materials. It defines source domain features. Features of the target domain The MMD in the regenerating nucleus Hilbert space (RKHS) is:

[0129]

[0130] In the formula, , The number of samples in the source domain and the number of samples in the target domain are respectively. The RKHS mapping function uses a Gaussian kernel. Calculate the kernel matrix. Minimize the MMD to achieve cross-domain feature distribution alignment, ensuring that the model can maintain high prediction accuracy even with a small number of real samples.

[0131] The model employs a multi-objective weighted loss function to simultaneously optimize data fitting accuracy, physical consistency, and cross-domain generalization ability.

[0132]

[0133]

[0134]

[0135]

[0136] In the formula, For data loss; These are the predicted damage size and depth values ​​output by the model. This is a real label. The physical loss is M, where M is the number of physical constraint sampling points. This is migration loss.

[0137] The Adam optimizer is used to minimize the total loss, and the adaptive learning rate adjustment strategy is as follows: , Let be the learning rate in the gen-th round. =0.95 is the decay coefficient, and the iteration continues until the loss converges.

[0138] As a preferred embodiment, step (5) is as follows:

[0139] Treating micro-damage as sparsely distributed "scattering sources," a forward model is constructed based on the Lamb wave equation. Combined with compressed sensing theory, high-resolution imaging is achieved. The inversion formula is:

[0140]

[0141] In the formula, A two-dimensional imaging matrix for micro-damage; The forward modeling matrix for Lamb wave propagation; This is the fusion vector of the PZT time-domain and FBG frequency-domain signals; For sparse regularization parameters.

[0142] Introducing time-frequency attention weights in step 3 Dynamically enhance signal components containing damage features, and extend the imaging model as follows:

[0143]

[0144] This design enables the imaging system to focus on the nonlinear feature regions of micro-damage in real time, improving the imaging signal-to-noise ratio by 30%.

[0145] Imaging results The feedback is sent to the FPGA timing controller in step 1, which optimizes the sensor layout using the following objective function:

[0146]

[0147] In the formula, These are the spatial layout parameters for the sensor array; For layout Signal-to-noise ratio at the following levels; Hardware resource consumption for layout; This is a weighting factor.

[0148] like Figure 3 , Figure 4 As shown, compared with the prior art, the advantages of the above solutions provided by the embodiments of the present invention are as follows:

[0149] (1) Breaking through the lower limit of micro-damage identification and achieving high-precision quantitative measurement: Through the nonlinear Lamb wave coupling effect and the deep migration model that integrates physical information, micro-damage with size ≤ 1 / 20 of wavelength can be captured, breaking through the traditional linear detection threshold; the measurement error of damage geometric parameters (size, depth) is ≤ 3%, and a unified benchmark for quantitative assessment of micro-damage has been established, solving the pain point of traditional technology being "easy to qualitative but difficult to quantitative".

[0150] (2) Multimodal sensing collaboration improves signal acquisition reliability: The “PZT time domain + FBG frequency domain” hybrid sensor array is adopted to complementarily cover the time domain dynamic response and frequency domain feature distribution, and the anti-electromagnetic interference capability is improved by 40% compared with a single PZT sensor; combined with the intelligent dynamic layout based on digital twin, the signal-to-noise ratio of the target area is maximized, and the micro-damage feature acquisition efficiency is improved by more than 50%.

[0151] (3) Strong cross-material generalization ability and reduced sample dependence: The deep transfer learning module aligns the features of the digital twin virtual sample (source domain) generated based on the digital twin model of the test sample with the real sample (target domain). Under the condition of few samples (real samples ≥ 20 sets), the detection accuracy of cross metal / composite materials is maintained at ≥ 90%, which solves the problem of poor model generalization caused by the scarcity of real samples of micro-damage and is suitable for multiple industrial scenarios.

[0152] (4) Subwavelength imaging resolution and excellent visualization effect: The dynamic adaptive imaging algorithm integrates compressed sensing and physical prior, and the imaging resolution is ≤1 / 10 of the wavelength, which is 2-3 orders of magnitude higher than traditional linear imaging; the introduction of time-frequency attention weights strengthens the damage feature area, and the imaging signal-to-noise ratio is improved by 30%, which can clearly present the spatial location and morphological details of micro-damage.

[0153] (5) Closed-loop iterative optimization, continuous improvement of detection robustness: Construct a "detection-measurement-imaging-optimization" closed-loop system, and the imaging results are fed back in real time to optimize the sensor layout. After 1-2 iterations, the system's anti-interference ability for complex working conditions (such as vibration and temperature fluctuations) is improved by 45%, meeting the long-term high reliability requirements for structural health monitoring in aerospace, high-end equipment and other fields.

[0154] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0155] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0156] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0157] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

[0158] This invention is not limited to the preferred embodiment described above. Anyone inspired by this invention can derive various other forms of non-destructive metrological testing and imaging methods for micro-damage nonlinear Lamb waves. All equivalent variations and modifications made within the scope of the claims of this invention shall fall within the scope of this invention.

Claims

1. A nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves, characterized in that, include: A composite signal containing two different frequency components is used to drive a sparsely arranged piezoelectric ceramic array to excite a nonlinear Lamb wave in the test sample. The nonlinear Lamb wave interacts nonlinearly with the micro-damage in the test sample, generating a nonlinear characteristic response containing sum and difference frequencies or harmonics. The time-domain and frequency-domain signals corresponding to the nonlinear characteristic response are simultaneously acquired by a hybrid sensing array composed of piezoelectric sensors and fiber Bragg grating sensors. The time-domain signal and the frequency-domain signal are respectively input into the parallel time-domain processing branch and the frequency-domain processing branch for feature extraction. Then, the extracted time-domain features and frequency-domain features are weighted and fused using an attention mechanism to obtain an enhanced micro-damage nonlinear feature vector. A quantitative model integrating physical information is constructed. The nonlinear feature vector of the micro-damage and the material physical parameters of the test sample are input into the model. The nonlinear Lamb wave equation is introduced as a physical constraint during the training and prediction process of the model, and the quantitative measurement results of the geometric parameters of the micro-damage are output. Based on the quantitative measurement results and multimodal signals, a dynamic adaptive imaging algorithm is executed to generate two-dimensional imaging results of micro-damage; the two-dimensional imaging results are fed back to the control unit of the hybrid sensor array to adjust the spatial layout parameters of the sensor array in order to optimize the micro-damage detection accuracy in subsequent detection cycles.

2. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: The composite signal containing two different frequency components is formed by superimposing two sinusoidal signals of adjacent frequencies. After the gain of the adaptive power amplifier is adjusted according to the characteristics of the test sample, it drives the sparse piezoelectric ceramic array.

3. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: In the hybrid sensing array, the piezoelectric sensor acquires the time-domain displacement response signal, and the fiber Bragg grating sensor acquires the frequency-domain strain signal based on Bragg wavelength modulation. Before acquisition, the multi-mode signal is preprocessed, including using wavelet transform combined with an improved soft-hard compromise threshold function for noise reduction, eliminating zero drift of the piezoelectric sensor signal by moving average, and introducing a compensation coefficient based on the digital twin temperature field of the test sample to eliminate temperature drift of the fiber Bragg grating sensor signal. The synchronous acquisition is achieved by a reference clock generated by the control unit. The reference clock frequency is adapted to the multi-channel synchronization requirements. Based on the reference clock, a trigger signal is sent to discretely sample the preprocessed signal. The sampling period and the number of sampling points are set to cover the complete Lamb wave propagation period.

4. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: The initial deployment and adjustment of the hybrid sensor array are based on the digital twin model of the test sample: the sensor layout optimization algorithm is run by the computing workstation, with the maximum signal-to-noise ratio of the observed signal as the objective function, and the optimal layout parameters are obtained and sent to the control unit, which drives the hardware link to complete the sensor array layout. When adjusting the spatial layout of the sensor array, the goal is to balance the signal-to-noise ratio with the consumption of hardware resources. The final layout parameters are determined by a preset trade-off coefficient. The adjustment is used to adapt to the micro-damage signal acquisition requirements of subsequent detection cycles.

5. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: Both the parallel temporal and frequency domain processing branches employ one-dimensional convolution to extract corresponding domain features. The convolution process includes preset convolution weights and bias terms. The attention mechanism calculates the feature weights of each domain by matching the temporal feature energy with the frequency feature frequency, and then performs element-wise weighted fusion of the extracted temporal and frequency features. Contrastive learning is also introduced during feature extraction to construct triplets containing query samples, positive samples with micro-damage, and negative samples without damage. The feature distance between samples is optimized through a loss function, and finally, a fixed-dimensional micro-damage nonlinear feature vector is output.

6. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: The material physical parameters of the test sample include the third-order nonlinear elastic constant, the second-order elastic constant, and the material density. The input dimension of the quantization model is determined by the dimension of the micro-damage nonlinear feature vector and the number of material physical parameters. The nonlinear Lamb wave equation serves as a physical constraint. A quantization model is constructed through a physical information neural network. The residual of the wave equation is calculated on the predicted Lamb wave displacement response output by the model. The magnitude of the residual is used to measure the degree of fit between the model prediction result and the physical propagation law of Lamb waves. The quantization model also includes a deep transfer module, which aligns the source domain of the corresponding digital twin virtual sample with the target domain of the corresponding real sample based on the maximum mean difference, ensuring the model's cross-material generalization ability.

7. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 1, characterized in that: The quantization model is trained using a multi-objective weighted loss function. The total loss includes data loss, physical loss, and transfer loss: data loss is calculated based on the difference between the predicted values ​​of micro-damage geometric parameters and the true labels; physical loss is calculated based on the residuals of the wave equation; and transfer loss is calculated based on the maximum mean difference between the source domain and the target domain. The training process uses an adaptive optimizer, with the learning rate adaptively adjusted according to a preset decay coefficient, iterating until the loss converges.

8. The nondestructive metrological detection and imaging method for micro-damage nonlinear Lamb waves according to claim 5, characterized in that: The dynamic adaptive imaging algorithm combines compressed sensing theory, treats micro-damage as a sparsely distributed scattering source, constructs a forward model based on the Lamb wave equation, and obtains the two-dimensional imaging matrix of micro-damage through inversion calculation; during the imaging process, time-frequency attention weights are introduced to enhance the signal components of the micro-damage feature region, and the time-frequency attention weights are consistent with the time-domain and frequency-domain feature weights obtained in the feature extraction process.

9. A micro-damage nonlinear Lamb wave nondestructive metrological testing and imaging system, characterized in that, include: Excitation module: It consists of a composite signal generator and a sparsely arranged piezoelectric ceramic array. The composite signal generator can generate a composite signal containing two different frequency components, which drives the piezoelectric ceramic array to excite a nonlinear Lamb wave in the test sample. The nonlinear Lamb wave interacts nonlinearly with the micro-damage in the test sample to generate a nonlinear characteristic response containing sum and difference frequencies or harmonics. Acquisition module: a hybrid sensing array consisting of piezoelectric sensors and fiber optic grating sensors, and a control unit for synchronous control. The hybrid sensing array can synchronously acquire time-domain and frequency-domain signals corresponding to nonlinear characteristic responses, and the control unit can generate a reference clock to achieve synchronous acquisition of multi-channel signals. Feature extraction module: It is used to extract features by inputting time-domain signals and frequency-domain signals into parallel time-domain processing branches and frequency-domain processing branches respectively, and to weight and fuse time-domain features and frequency-domain features through attention mechanism to output enhanced micro-damage nonlinear feature vector; Quantitative modeling module: Used to construct a quantitative model that integrates physical information. The nonlinear feature vector of micro-damage and the material physical parameters of the test sample are input into the model. During the model training and prediction process, the nonlinear Lamb wave equation is introduced as a physical constraint, and the quantitative measurement results of the geometric parameters of micro-damage are output. Imaging optimization module: It is used to execute dynamic adaptive imaging algorithm based on quantitative measurement results and multimodal signals to generate two-dimensional imaging results of micro-damage, and feed the two-dimensional imaging results back to the control unit to adjust the spatial layout parameters of the hybrid sensor array in order to optimize the micro-damage detection accuracy in subsequent detection cycles.

10. A computer device, characterized in that, It includes a processor and a memory, the memory storing a computer program, and when the processor executes the computer program, it implements the micro-damage nonlinear Lamb wave nondestructive metrological detection and imaging method according to any one of claims 1-8.

Images