Depth learning based method for measuring thickness of composite material by laser ultrasonic with grating modulation

Abstract

The application provides a grating modulation laser ultrasonic composite material layering thickness measurement method based on deep learning, and relates to the technical field of laser ultrasonic detection. The method comprises the following steps: collecting material parameters corresponding to each component layer in a multilayer composite material, and constructing a Lamb wave dispersion physical model according to the material parameters; generating multiple sets of thickness combinations within a preset total thickness and a corresponding layer thickness variation range of each component layer, and calculating a multilayer Lamb wave dispersion curve according to the Lamb wave dispersion physical model; extracting a frequency peak value and a Lamb wave mode number from the multilayer Lamb wave dispersion curve at a target wavelength, pairing a frequency feature vector with corresponding layer thicknesses, and constructing a training sample set; training an initial thickness-frequency mapping model based on the training sample set to obtain a target thickness-frequency mapping model; collecting frequency features corresponding to the surface of a measured multilayer composite material, inputting the frequency features into the target thickness-frequency mapping model, and obtaining thicknesses corresponding to each component layer.

Description

Technical Field

[0001] This invention relates to the field of laser ultrasonic testing technology, and in particular to a deep learning-based method for measuring the thickness of composite materials through grating modulation laser ultrasonic layering. Background Technology

[0002] Functionally graded composite materials (FJCs) are novel materials whose composition, structure, morphology, or interfacial concentration varies regularly with space or time. These materials achieve integration of composition, structure, and function, and have important applications in defense engineering, high-speed loading, impact protection, aerospace, and high-tech fields. For FJCs, the thickness of the intermediate layer and the distribution of each layer directly affect the stress, strain rate, and loading waveform during service. Therefore, accurate detection of the internal layer thickness structure before impact testing or service is of great significance.

[0003] Chinese Patent CN121720390B discloses a method and system for measuring gradient material thickness based on laser ultrasonic dispersion spectrum. The method includes: modulating an excitation laser using an optical mask to form a transient grating with a periodic intensity distribution; using the transient grating to excite ultrasonic waves of a specific frequency in the material under test; and using a laser interferometer to perform line scanning to receive the Lamb wave signal generated by the structural action of the ultrasonic waves on the material under test; performing a two-dimensional Fourier transform on the Lamb wave signal to obtain a target dispersion curve; the target dispersion curve includes a mapping relationship between signal frequency and wavenumber; generating multiple sample pairs consisting of material thickness parameters and dispersion curve parameters using a transfer matrix algorithm; training a preset deep neural network based on the sample pairs until the loss function reaches a preset threshold range to obtain the target neural network; and inputting the target dispersion curve into the target neural network to obtain the thickness parameters of the material under test. However, the aforementioned methods still rely on manual point selection and empirical fitting to extract dispersive features during signal processing. This process is highly subjective and dependent on operator experience, making it prone to instability in feature extraction under complex noise environments, thus compromising the consistency and accuracy of measurement results. Therefore, it is essential to provide a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method to improve the consistency and accuracy of composite material layer thickness measurement. Summary of the Invention

[0004] In view of this, the present invention proposes a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method, which helps to improve the consistency and accuracy of composite material layer thickness measurement.

[0005] This invention provides a deep learning-based method for layered thickness measurement of composite materials using grating-modulated laser ultrasonic technology, the method comprising: Material parameters corresponding to each component layer in a multilayer composite material are collected, and a Lamb wave dispersion physical model is constructed based on the material parameters. Multiple thickness combinations are generated within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material, and the multilayer Lamb wave dispersion curve under each thickness combination is calculated according to the Lamb wave dispersion physical model. At the target wavelength, extract the frequency peak and Lamb wave mode number corresponding to each Lamb wave mode from the multilayer Lamb wave dispersion curve, pair the frequency feature vector with the corresponding layer thickness, and construct a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak and the Lamb wave mode number. The initial thickness frequency mapping model is trained based on the training sample set to obtain the target thickness frequency mapping model; The frequency characteristics corresponding to the surface of the multilayer composite material under test are collected, and the frequency characteristics are input into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

[0006] Based on the above technical solutions, preferably, the step of collecting material parameters corresponding to each component layer in the multilayer composite material and constructing a Lamb wave dispersion physical model based on the material parameters specifically includes: The material parameters corresponding to each component layer in the multilayer composite material are collected, wherein the material parameters include the density, elastic modulus, Poisson's ratio and the thickness range of each component layer. Based on the layered plate theory and the material parameters, the Lamb wave dispersion control equation is constructed and solved to obtain a multilayer Lamb wave dispersion physical model under different thickness combinations.

[0007] Based on the above technical solutions, preferably, the generation of multiple thickness combinations within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material specifically includes: The preset total thickness range and the corresponding layer thickness range of the multilayer composite material are determined according to the design requirements or prior information. The thickness range of each component layer is discretized with a preset step size. Under the condition of satisfying the thickness constraint, the thickness of each component layer is enumerated by rules or randomly sampled to generate multiple sets of thickness combinations that satisfy the thickness constraint.

[0008] More preferably, the step of pairing the frequency feature vector with the corresponding layer thickness to construct a training sample set for training the initial thickness-frequency mapping model specifically includes: For each thickness combination, extract the frequency peaks of one or more Lamb wave modes from the corresponding multilayer Lamb wave dispersion curve at the target wavelength, and obtain the Lamb wave mode number corresponding to each frequency peak. Each frequency peak and the Lamb wave mode number corresponding to the frequency peak are combined according to a preset arrangement rule to form a frequency feature vector, and the layer thickness of each component layer corresponding to the frequency feature vector is used as the output thickness vector. Training samples are constructed based on the frequency feature vector and the output thickness vector. Training samples are constructed for all thickness combinations to obtain a training sample set consisting of multiple frequency feature vectors and the output thickness vectors corresponding to the frequency feature vectors.

[0009] More preferably, the step of training the initial thickness frequency mapping model based on the training sample set to obtain the target thickness frequency mapping model specifically includes: The frequency feature vector and output thickness vector in the training sample set are paired to construct sample pairs, and the sample pairs are divided into training set and validation set; The frequency feature vector in the training set is input into the initial thickness frequency mapping model to obtain the first predicted thickness vector corresponding to the frequency feature vector; The first error value between the first predicted thickness vector and the corresponding output thickness vector is calculated according to the preset loss function, and the model parameters of the initial thickness frequency mapping model are iteratively updated based on the first error value and the preset optimization algorithm. In each training round, the frequency feature vector in the validation set is input into the updated initial thickness frequency mapping model to obtain the second predicted thickness vector corresponding to the frequency feature vector, and the second error value between the second predicted thickness vector and the corresponding output thickness vector is calculated according to the preset loss function. When the second error value meets the preset convergence condition, the training of the initial thickness frequency mapping model is stopped, and the initial thickness frequency mapping model at this time is used as the target thickness frequency mapping model.

[0010] More preferably, the step of acquiring the frequency characteristics corresponding to the surface of the multilayer composite material under test, and inputting the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test, specifically includes: A periodic excitation stripe is formed on the surface of the multilayer composite material under test by using a grating-modulated pulsed laser to excite a Lamb wave propagating in the direction perpendicular to the periodic excitation stripe. The surface of the multilayer composite material under test is scanned in one dimension in the direction perpendicular to the periodic excitation stripes using a laser interferometer or a laser Doppler vibrator. Lamb wave signals of surface displacement or surface velocity changing with time at each scanning position are collected, and the Lamb wave signals are arranged according to spatial position and time dimension to obtain the B-scan data of the multilayer composite material under test. A two-dimensional Fourier transform is performed on the B-scan data to obtain a two-dimensional spectrum. Based on the target wavelength or target wavenumber in the two-dimensional spectrum, the frequency peaks of multiple Lamb wave modes and the Lamb wave mode numbers corresponding to the frequency peaks are extracted. All the frequency peaks and Lamb wave mode numbers are arranged in a preset order to form the frequency feature vector of the multilayer composite material under test. The frequency feature vector is input into the target thickness frequency mapping model to obtain the output thickness vector corresponding to the frequency feature vector, and each thickness component in the output thickness vector is used as the thickness of each component layer in the multilayer composite material under test.

[0011] More preferably, at the target wavelength, each Lamb wave mode appearing in the multilayer Lamb wave dispersion curve is sorted from low to high according to the corresponding frequency peak value. The sorted sequence number is used as the Lamb wave mode number of each Lamb wave mode. The Lamb wave mode number is combined with the corresponding frequency peak value in the form of integer encoding or one-hot encoding to form the frequency feature vector of the initial thickness frequency mapping model.

[0012] A second aspect of this application provides a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement system, which includes a data acquisition module, a model building module, and a thickness measurement module. The data acquisition module is used to acquire the material parameters corresponding to each component layer in the multilayer composite material, and to construct a Lamb wave dispersion physical model based on the material parameters. The model building module is used to generate multiple thickness combinations within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material, and calculate the multilayer Lamb wave dispersion curve under each thickness combination according to the Lamb wave dispersion physical model. At the target wavelength, the frequency peak value and Lamb wave mode number corresponding to each Lamb wave mode are extracted from the multilayer Lamb wave dispersion curve. The frequency feature vector is paired with the corresponding layer thickness to construct a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak value and the Lamb wave mode number. The initial thickness frequency mapping model is trained based on the training sample set to obtain the target thickness frequency mapping model. The thickness measurement module is used to collect the frequency characteristics corresponding to the surface of the multilayer composite material under test, and input the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

[0013] A third aspect of this application provides an electronic device including a processor, a memory, a user interface, and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory.

[0014] A fourth aspect of this application provides a non-transitory computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method.

[0015] The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method provided by this invention has the following advantages over existing technologies: (1) By constructing multi-layer Lamb wave dispersion curves, the thickness combination of each layer is explicitly introduced in the training, so that the model can distinguish the contribution of different layers to the frequency response. Thus, the independent thickness of each component layer can be obtained in one measurement. A large number of thickness combinations and corresponding dispersion curves are generated in advance within the range of material parameters and layer thickness variations, which expands the coverage of training samples. This makes the trained model more adaptable to different material parameters and different stacked structures, and more robust. In the actual detection stage, only the frequency characteristics of the surface need to be collected and input into the trained model to quickly obtain the thickness of each layer, avoiding complex numerical inversion or repeated trial calculations, and significantly shortening the measurement time. The Lamb wave frequency characteristics are obtained by using grating-modulated laser ultrasound. No coupling agent or contact sensor is required for the test piece, and the structural integrity is not damaged. Key information is learned from the frequency feature vector composed of frequency peak and mode number through deep learning, reducing the subjectivity and uncertainty brought about by manual feature design, mode selection and inversion parameter tuning, thereby improving the consistency and accuracy of composite material layer thickness measurement.

[0016] (2) By using a grating-modulated pulsed laser to form periodic stripes on the surface, only Lamb wave modes that propagate along the vertical direction of the stripes and match the stripe period are excited. This can obtain a multi-mode Lamb wave signal with concentrated energy at the target wavelength / wavenumber, improve the effective signal energy and mode selectivity, and reduce the interference of irrelevant modes and noise. A laser interferometer or laser Doppler vibrator is used to perform a one-dimensional spatial scan in the direction perpendicular to the excitation stripes to obtain a two-dimensional Lamb wave signal. This maintains the advantages of non-contact, non-destructive, and high bandwidth, and also obtains complete information about the wave propagation process in space. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method provided by this invention; Figure 2 A schematic diagram of the training process for the thickness frequency mapping model provided by this invention; Figure 3 This is a schematic diagram of the structure of the multilayer composite material layer thickness measurement experimental device provided by the present invention; Figure 4 A schematic diagram comparing the actual and predicted thickness values ​​of each layer in the five-layer sample provided by this invention; Figure 5 This is a schematic diagram of the grating-modulated laser ultrasonic composite material layer thickness measurement system provided by the present invention; Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention.

[0019] Figure reference numerals: 1. Grating-modulated laser ultrasonic composite material layer thickness measurement system; 11. Data acquisition module; 12. Model building module; 13. Thickness measurement module; 2. Electronic equipment; 21. Processor; 22. Communication bus; 23. User interface; 24. Network interface; 25. Memory. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention discloses a deep learning-based method for layered thickness measurement of composite materials using grating-modulated laser ultrasonic technology, with reference to... Figure 1 and Figure 2 The method includes steps S1 to S5.

[0022] Step S1: Collect the material parameters corresponding to each component layer in the multilayer composite material, and construct a Lamb wave dispersion physical model based on the material parameters; Step S2: Generate multiple sets of thickness combinations within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material, and calculate the multilayer Lamb wave dispersion curve under each thickness combination according to the Lamb wave dispersion physical model. Step S3: Extract the frequency peak and Lamb mode number corresponding to each Lamb mode from the multilayer Lamb dispersion curve at the target wavelength, pair the frequency feature vector with the corresponding layer thickness, and construct a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak and the Lamb mode number. Step S4: Train the initial thickness frequency mapping model based on the training sample set to obtain the target thickness frequency mapping model; Step S5: Collect the frequency characteristics corresponding to the surface of the multilayer composite material under test, and input the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

[0023] In this embodiment, by constructing multilayer Lamb wave dispersion curves and explicitly introducing layer thickness combinations during training, the model can distinguish the contribution of different layers to the frequency response. Thus, the independent thickness of each component layer can be obtained in a single measurement. A large number of thickness combinations and corresponding dispersion curves are generated in advance within the range of material parameters and layer thickness variations, expanding the coverage of training samples. This makes the trained model more adaptable to different material parameters and different stacked structures, and more robust. In the actual detection stage, only the frequency characteristics of the surface need to be collected and input into the trained model to quickly obtain the thickness of each layer, avoiding complex numerical inversion or repeated trial calculations, significantly shortening the measurement time. Lamb wave frequency characteristics are obtained by using grating-modulated laser ultrasound, which does not require coupling agent or contact sensors on the tested part and does not damage the structural integrity. Key information is learned from the frequency feature vector composed of frequency peaks and mode numbers through deep learning, reducing the subjectivity and uncertainty brought about by manual feature design, mode selection, and inversion parameter tuning, thereby improving the consistency and accuracy of composite material layer thickness measurement.

[0024] Based on the above embodiments, as an optional embodiment, step S1, which involves collecting material parameters corresponding to each component layer in the multilayer composite material and constructing a Lamb wave dispersion physical model based on the material parameters, may further include the following: Step S11: Collect the material parameters corresponding to each component layer in the multilayer composite material. The material parameters include the density, elastic modulus, Poisson's ratio, and the thickness range of each component layer.

[0025] In this step, the material parameters include the density of each constituent layer. Elastic modulus Poisson's ratio and the range of thickness values ​​for each component layer. ,in, Indicates the layer number, Indicates the total number of layers in a multilayer composite material. Indicates the first Density of layer material, Indicates the first The elastic modulus of the layer material, Indicates the first Poisson's ratio of the layer material, and the range of layer thickness values. Used to limit the The thickness variation range of the layer when constructing the dispersion physical model and generating thickness combinations. Indicates the first Maximum thickness of the layer material Indicates the first Minimum thickness of the layer material, Indicates the first The thickness of the layer material.

[0026] Step S12: Based on the layered plate theory and material parameters, construct the Lamb wave dispersion control equation and solve the Lamb wave dispersion control equation to obtain a multilayer Lamb wave dispersion physical model under different thickness combinations.

[0027] In this step, the density of each component layer is utilized. Elastic modulus Poisson's ratio and the range of thickness values ​​for each component layer. According to the layered plate theory, the governing equations reflecting the relationship between stress, strain and displacement in multilayer plates are established, and boundary conditions of displacement continuity and stress continuity are applied at the interfaces of each layer, while boundary conditions of zero stress are applied on the free surface of the plate.

[0028] In frequency With wavenumber Given the conditions, the corresponding longitudinal wave number and transverse wave number are calculated based on the material parameters of each component layer. The wave field state vector and stress-displacement transfer matrix of each layer are constructed. Then, the wave field state of the lower surface is associated with the wave field state of the upper surface through interlayer transfer, forming the overall Lamb wave dispersion control equation of the multilayer plate.

[0029] The Lamb wave dispersion control equation is written in terms of frequency. With wavenumber Lamb wave dispersion governing equation ,in, This represents the set of parameters comprised of the material parameters and thicknesses of each constituent layer. And numerical solution methods were used to solve the problem under different thickness combinations. The solution is performed to obtain the multilayer Lamb wave dispersion curves under the corresponding thickness combinations, thereby constructing a multilayer Lamb wave dispersion physical model under different thickness combinations.

[0030] Based on the above embodiments, as an optional embodiment, step S2, which involves generating multiple thickness combinations within the preset total thickness and the thickness variation range of each component layer of the multilayer composite material, may further include the following: Step S21: Determine the preset total thickness range of the multilayer composite material and the corresponding thickness range of each component layer based on design requirements or prior information.

[0031] In this step, the minimum and maximum values ​​of the overall thickness of the multilayer composite material can be determined based on structural design drawings, material manuals, or historical test data. The interval between these minimum and maximum values ​​is set as the preset total thickness range. At the same time, based on the design tolerances, manufacturing process deviations, and prior experience of each component layer, the minimum and maximum layer thicknesses of each component layer are determined, and the interval between these minimum and maximum layer thicknesses is set as the corresponding layer thickness range.

[0032] Step S22: Discretize the thickness range of each component layer with a preset step size. Under the condition of satisfying the thickness constraint, perform regular enumeration or random sampling on the thickness of each component layer to generate multiple sets of thickness combinations that satisfy the thickness constraint.

[0033] In this step, the thickness range of each component layer is discretized according to a preset step size to obtain several candidate layer thickness values ​​for each component layer. The preset step size can be set according to the thickness measurement resolution requirements and computational complexity requirements, so that the discretized candidate layer thickness values ​​can cover the thickness variation range of each component layer relatively fully while ensuring computational efficiency.

[0034] After obtaining the candidate layer thickness values ​​of each component layer, under the premise of satisfying the thickness constraints, the candidate layer thickness values ​​of each component layer are enumerated by rules or randomly sampled to generate multiple sets of thickness combinations that satisfy the thickness constraints. The thickness constraints include that the total thickness formed by the sum of the layer thicknesses of each component layer is within the preset total thickness value range, and the layer thickness of each component layer falls within its corresponding layer thickness value range.

[0035] In one example, during the generation of the multiple sets of thickness combinations, additional constraints can be applied to the relative relationship between the thicknesses of each component layer according to the structural design requirements. For example, the thickness ratio or thickness difference of certain adjacent layers can be constrained to be within a preset range to further improve the rationality and representativeness of the generated thickness combinations.

[0036] Based on the above embodiments, as an optional embodiment, step S3, which involves pairing the frequency feature vector with the corresponding layer thicknesses to construct a training sample set for training the initial thickness-frequency mapping model, may further include the following: Step S31: For each thickness combination, extract the frequency peaks of one or more Lamb wave modes from the corresponding multilayer Lamb wave dispersion curve at the target wavelength, and obtain the Lamb wave mode number corresponding to each frequency peak.

[0037] In this step, record the first... The multilayer Lamb wave dispersion physical model corresponding to the thickness combination is the dispersion relation function under that thickness combination. Based on this dispersion relation function, the target wavenumber corresponding to the preset target wavelength is determined. At the specified location, the multilayer Lamb wave dispersion curve or dispersion amplitude distribution corresponding to this thickness combination on the frequency-wavenumber plane is calculated. At the target wavenumber... Peak search is performed on the dispersion curve or dispersion amplitude distribution along the frequency direction to extract the frequency values ​​corresponding to one or more local maxima, which are used as the frequency peaks of each Lamb wave mode at the target wavelength under the thickness combination.

[0038] No. The group thickness combination can be expressed as:

[0039] This thickness combination is at the target wavenumber. amplitude function at point It can be represented as:

[0040] Along frequency For amplitude function Perform peak search to obtain If there are multiple frequency peaks, then the frequency peaks can be represented as:

[0041] For each frequency peak The Lamb wave mode number is determined according to the preset mode numbering rules:

[0042] in, Indicates the first The thickness vector of the group thickness combination. Indicates the first In the group thickness combination, the first Layer thickness of the material Indicates the first Group thickness combination at target wavenumber The magnitude function at that point, Indicates frequency, Indicates frequency Target wavenumber and the The set of parameters corresponding to the group thickness combination The amplitude function of the multilayer Lamb wave response under the given conditions; Indicates the first The group thickness combination detected at the target wavelength One frequency peak, This indicates the number of Lamb wave modes selected for each thickness combination. This represents the local peak extraction function. Indicates the first Under the group thickness combination, the first The Lamb wave mode number corresponding to each frequency peak. This represents the modal numbering function.

[0043] Step S32: Combine each frequency peak and the Lamb wave mode number corresponding to the frequency peak according to a preset arrangement rule to form a frequency feature vector, and use the layer thickness of each component layer corresponding to the frequency feature vector as the output thickness vector, and construct training samples based on the frequency feature vector and the output thickness vector.

[0044] In this step, for the first The combination of thicknesses will be obtained at the target wavelength. The frequency peaks are arranged sequentially according to a preset arrangement rule (e.g., sorted by frequency value from smallest to largest, or sorted by mode number from smallest to largest) to form the frequency component sequence of the thickness combination. Based on this, the Lamb wave mode numbers corresponding to each frequency peak are encoded in the same order and spliced ​​or combined with the frequency component sequence to obtain a frequency feature vector with fixed dimension, which is used to fully characterize the multimodal frequency response characteristics of the thickness combination at the target wavelength.

[0045] The first The thicknesses of each component layer in the thickness combination are arranged in order of layer number to obtain an output thickness vector that corresponds one-to-one with the frequency feature vector. This output thickness vector is used to characterize the true thickness distribution of each component layer in the multilayer composite material under this thickness combination. Using the frequency feature vector as the input vector and the corresponding output thickness vector as the output vector, an input-output training sample is formed and stored in the training sample set.

[0046] In one example, the first The thickness combination is obtained in step S31 Frequency peaks and its modal number Frequency feature vectors are formed by combining elements according to a preset arrangement rule. :

[0047] The first In the group thickness combination, the layer thicknesses of each component layer are arranged in order of layer number to form the output thickness vector. :

[0048] in, Indicates the first The frequency feature vector corresponding to the group thickness combination Represents the feature construction function, Indicates the first The output thickness vector corresponding to the combination of thickness groups. This represents the transpose of a vector.

[0049] Step S33: Construct training samples for all thickness combinations to obtain a training sample set consisting of multiple frequency feature vectors and the output thickness vectors corresponding to the frequency feature vectors.

[0050] In this step, the total number of thickness combinations generated is denoted as . The final training sample set is represented as:

[0051] in, Represents the training sample set, Indicates the first Training samples corresponding to the group thickness combination.

[0052] In one example, at the target wavelength, the Lamb wave modes appearing in the multilayer Lamb wave dispersion curve are sorted from low to high according to the corresponding frequency peaks. The sorted numbers are used as the Lamb wave mode numbers of each Lamb wave mode. The Lamb wave mode numbers are combined with the corresponding frequency peaks in the form of integer encoding or one-hot encoding to form the frequency feature vector of the initial thickness frequency mapping model.

[0053] Based on the above embodiments, as an optional embodiment, step S4, which involves training the initial thickness frequency mapping model based on the training sample set to obtain the target thickness frequency mapping model, may further include the following: Step S41: Pair the frequency feature vector and the output thickness vector in the training sample set to construct sample pairs, and divide the sample pairs into training set and validation set.

[0054] In this step, the frequency feature vector of each sample in the training sample set is matched one-to-one with its corresponding output thickness vector to form several sample pairs. Then, according to a preset data partitioning ratio (e.g., 8:2 or 7:3) or a random partitioning strategy, the sample pairs are divided into a training set for model parameter learning and a validation set for performance evaluation, thereby ensuring that the model training and validation processes are independent of each other.

[0055] Step S42: Input the frequency feature vectors in the training set into the initial thickness frequency mapping model to obtain the first predicted thickness vector corresponding to the frequency feature vectors.

[0056] In this step, for each sample pair in the training set, the frequency feature vector is used as the model input and fed into the initial thickness frequency mapping model. Through the forward calculation process of the model, the first predicted thickness vector corresponding to the frequency feature vector is obtained, which is used to approximately characterize the predicted thickness of each component layer of the multilayer composite material.

[0057] Step S43: Calculate the first error value between the first predicted thickness vector and the corresponding output thickness vector according to the preset loss function, and iteratively update the model parameters of the initial thickness frequency mapping model based on the first error value and the preset optimization algorithm.

[0058] In this step, for each sample pair in the training set, the first predicted thickness vector output by the initial thickness frequency mapping model and the output thickness vector corresponding to the sample are substituted into a preset loss function to calculate the first error value between the current model prediction and the actual thickness. Based on this, according to the first error value, a preset optimization algorithm (such as a gradient descent-based optimization algorithm, Adam optimization algorithm, etc.) is used to iteratively update the trainable parameters in the initial thickness frequency mapping model one or more times to gradually reduce the prediction error on the training set.

[0059] Step S44: In each training round, the frequency feature vector in the validation set is input into the updated initial thickness frequency mapping model to obtain the second predicted thickness vector corresponding to the frequency feature vector, and the second error value between the second predicted thickness vector and the corresponding output thickness vector is calculated according to the preset loss function.

[0060] In this step, after completing one round of parameter updates for the training set samples, the frequency feature vectors in the validation set are sequentially input into the currently updated initial thickness frequency mapping model. The second predicted thickness vector corresponding to each frequency feature vector is obtained through forward computation. Subsequently, the second predicted thickness vector and the output thickness vector corresponding to the validation set are substituted into the same preset loss function to calculate the second error value on the validation set, which is used to measure the generalization performance of the current model on samples that have not participated in the training.

[0061] Step S45: When the second error value meets the preset convergence condition, stop training the initial thickness frequency mapping model and use the initial thickness frequency mapping model at this time as the target thickness frequency mapping model.

[0062] In this step, the preset convergence conditions include: the second error value on the validation set is less than or equal to a preset error threshold, and / or the second error value on the validation set no longer decreases significantly within a certain number of consecutive training rounds. Based on the above convergence criteria, when the preset convergence conditions are met, the training process of the initial thickness-frequency mapping model is considered complete.

[0063] In one example, the initial thickness-frequency mapping model can be represented by the following functional form: For the first in the training set a sample pair The mapping relationship of the initial thickness-frequency mapping model can be expressed as:

[0064] In a typical case where mean squared error is used as the preset loss function, the loss function on the training set can be expressed as:

[0065] In each training epoch, a gradient descent-based optimization algorithm can be used to adjust the parameters. Perform iterative updates:

[0066] In the validation set The corresponding second error value can be expressed as:

[0067] in, Indicates the first The group thickness combination corresponds to the first predicted thickness vector of the training sample. This represents the initial thickness-frequency mapping model. Indicates the current model parameters The training loss function is as follows: Represents the training set, Indicates by the first The next round of model parameter vector obtained from the parameter update in the first round of training. Indicates the first The model parameter vector obtained after one round of training. Indicates the learning rate. This represents the loss function with respect to the model parameters. The gradient of the first The values ​​taken during training rounds, Indicates the first The second error value on the validation set after one round of training. Represents the validation set. Represents the 2-norm operator.

[0068] Based on the above embodiments, as an optional embodiment, step S5, which involves acquiring the frequency characteristics corresponding to the surface of the multilayer composite material under test and inputting the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test, may further include the following: Step S51: Periodic excitation stripes are formed on the surface of the multilayer composite material under test using a grating-modulated pulsed laser to excite Lamb waves propagating in the direction perpendicular to the periodic excitation stripes.

[0069] In this step, by arranging a phase grating, interference optical element, or spatial light modulator in the laser optical path, the pulsed laser beam is modulated to form several parallel periodic excitation fringes with a preset grating period spacing on the surface of the multilayer composite material being tested. The periodic excitation fringes form a one-dimensional spatial periodic thermoelastic or ablation excitation source array within the plate surface. The fringe spacing (i.e., grating period) of the periodic excitation fringes is pre-designed according to the target wavelength or target wavenumber so that the Lamb wave mode propagating in the perpendicular direction along the periodic excitation fringes is preferentially excited at the spatial frequency corresponding to the grating period. Furthermore, by adjusting the energy, pulse width, repetition frequency, and spot size of the pulsed laser, a Lamb wave response signal with a sufficient signal-to-noise ratio can be obtained without damaging the material.

[0070] In one example, the pulsed laser can be replaced with other pulsed laser sources capable of producing sufficiently short pulse widths and sufficient energy densities; the grating modulation unit can be a fixed grating mask, a programmable SLM, a DMD, or other spatial light modulation device capable of forming periodic stripe spots.

[0071] Step S52: A one-dimensional spatial scan of the surface of the multilayer composite material under test is performed in the direction perpendicular to the periodic excitation stripes using a laser interferometer or a laser Doppler vibrator. Lamb wave signals of surface displacement or surface velocity changing with time are collected at each scan position. The Lamb wave signals are arranged according to spatial position and time dimension to obtain the B-scan data of the multilayer composite material under test.

[0072] In this step, a spatial scanning measurement line is pre-set in the vertical direction of the periodic excitation stripes. The laser interferometer measurement spot or the laser Doppler vibrator measurement point are arranged or moved along the measurement line according to the preset spatial step size. At each scanning position, the Lamb wave time-domain signal of the surface displacement or surface velocity of the multilayer composite material under test is recorded as a function of time. The Lamb wave time-domain signals collected at each scanning position are sorted and stored according to the two-dimensional coordinates of "spatial position-time" to construct a two-dimensional data array, namely B-scan data, that reflects the propagation characteristics of Lamb waves in space and time, thereby fully characterizing the propagation behavior of Lamb waves in the tested multilayer composite material under periodic excitation conditions.

[0073] In one example, in addition to a laser interferometer, other non-contact or contact high-frequency ultrasonic receiving devices may be used depending on the detection scenario, but a laser interferometer is preferred to maintain the advantages of non-contact detection.

[0074] Step S53: Perform a two-dimensional Fourier transform on the B-scan data to obtain a two-dimensional spectrum. Based on the target wavelength or target wavenumber in the two-dimensional spectrum, extract the frequency peaks of multiple Lamb wave modes and the Lamb wave mode numbers corresponding to the frequency peaks. Arrange all the frequency peaks and Lamb wave mode numbers in a preset order to form the frequency feature vector of the multilayer composite material under test.

[0075] In this step, the B-scan data obtained in step S52 is subjected to a two-dimensional Fourier transform in both spatial and temporal dimensions to obtain a two-dimensional spectrum with frequency-wavenumber or frequency-wavelength as the coordinate axis. The energy distribution of different Lamb wave modes under different combinations of frequency and wavenumber (or wavelength) in the two-dimensional spectrum is represented in the form of spectral lines or spectral peaks. Based on the pre-set target wavelength or target wavenumber, the peak value of the spectral amplitude distribution is searched along the frequency direction on the corresponding section of the two-dimensional spectrum, and the frequency values ​​corresponding to several local maxima are extracted as the frequency peak values ​​of multiple Lamb wave modes of the multilayer composite material under test at the target wavelength or target wavenumber.

[0076] Furthermore, based on the preset Lamb wave mode numbering rules, such as sorting by frequency from low to high at the target wavelength, or combining the theoretical dispersion curve for mode identification, a unique Lamb wave mode number is assigned to each frequency peak, resulting in several pairs of modal feature data; all frequency peaks and their corresponding Lamb wave mode numbers are arranged and spliced ​​in a preset order to form the frequency feature vector corresponding to the tested multilayer composite material.

[0077] Step S54: Input the frequency feature vector into the target thickness frequency mapping model to obtain the output thickness vector corresponding to the frequency feature vector, and use each thickness component in the output thickness vector as the thickness of each component layer in the multilayer composite material being tested.

[0078] In this step, the frequency feature vector of the multilayer composite material under test is used as input and fed into the target thickness frequency mapping model obtained through offline training. Through the forward calculation process of this model, an output thickness vector corresponding one-to-one with the frequency feature vector is output. Each component in the output thickness vector corresponds to the predicted thickness value of each constituent layer in the multilayer composite material under test, according to the preset layer numbering order. By directly using each thickness component in the output thickness vector as the measured thickness of the corresponding layer, the layer-by-layer thickness measurement result of the multilayer composite material under test in the thickness direction can be obtained, realizing non-destructive layer-by-layer thickness inversion of the multilayer composite material under test.

[0079] In this embodiment, periodic stripes are formed on the surface by grating-modulated pulsed laser, exciting only Lamb wave modes that propagate perpendicular to the stripes and match the stripe period. This allows for the acquisition of energy-concentrated multimode Lamb wave signals at the target wavelength / wavenumber, improving effective signal energy and mode selectivity while reducing interference from irrelevant modes and noise. A laser interferometer or laser Doppler vibrator is used to perform a one-dimensional spatial scan perpendicular to the excitation stripes, obtaining a two-dimensional Lamb wave signal. This maintains the advantages of being non-contact, non-destructive, and having high bandwidth, while also acquiring complete information about the wave's propagation in space, providing a sufficient data foundation for subsequent accurate analysis in the wavenumber-frequency domain. The frequency peaks of multiple modes and their corresponding mode numbers are arranged in a preset order to form a frequency feature vector, ensuring consistency in the dimension and order of input features among different tested components. This reduces uncertainties caused by missed mode detection or disordered sorting, making the thickness-frequency mapping model more stable and reliable during the inference phase.

[0080] like Figure 3 As shown, Figure 3 This is a schematic diagram of the experimental apparatus for measuring the thickness of multilayer composite materials. The apparatus mainly includes a grating-modulated pulsed laser excitation device for generating periodic excitation stripes on the surface of the multilayer composite material under test, and a laser measurement device for detecting the vibration response of the surface of the multilayer composite material under test. The test object is a multilayer composite plate specimen.

[0081] The multilayer composite material plate under test is laid horizontally and may include multiple material layers and possible delamination defects or interface debonding areas. A laser measuring device is arranged on one side above the plate surface to perform non-contact detection of the normal displacement or normal velocity of the plate surface at a predetermined measurement position. The laser measuring device is preferably a laser interferometer or a laser Doppler vibrator, and its output end is electrically connected to the data acquisition and processing unit to record the Lamb wave response signal in the time domain.

[0082] On the upper surface of the multilayer composite material plate under test, several parallel periodic excitation stripes are formed by a grating-modulated pulsed laser excitation device. Figure 3The image is represented by a striped pattern. A grating-modulated pulsed laser is directed vertically onto the surface of the specimen. After modulation by the grating or interference optical elements, a spatially periodic energy distribution with a preset stripe spacing is formed on the specimen surface, thereby exciting Lamb waves that propagate in the plate perpendicular to the excitation stripes. The figure illustrates the propagation of the Lamb waves within the plate using waveform curves. The Lamb waves are reflected and transmitted at multilayer structures and at possible delamination defects.

[0083] With the above structural arrangement, the grating-modulated pulsed laser excitation device is responsible for exciting a Lamb wave signal with a specific wavelength and propagation direction on the surface of the multilayer composite material plate under test. The laser measurement device is responsible for performing one-dimensional spatial scanning measurement of the surface response of the Lamb wave above the plate. With the subsequent data processing steps, the thickness of each component layer and delamination defects inside the multilayer composite material under test can be inverted and evaluated.

[0084] In one example, taking a five-layer gradient composite material, the Lamb wave dispersion curves for different layer thickness combinations are first calculated based on the Poisson's ratio, density, and Young's modulus of each layer. Then, a specific wavelength is selected. The frequency values ​​corresponding to this wavelength were extracted and used as inputs to the target thickness frequency mapping model. The frequency values ​​extracted from the five layers of material at specific wavelengths are shown in Table 1.

[0085] Table 1 Modal number Frequency value (kHz) 1 1495.73 2 1678.88 3 2228.33 4 2533.58 5 3021.98 After inputting the aforementioned frequency values ​​into the trained target thickness frequency mapping model, the predicted thickness of each of the five material layers can be obtained. In the example results, the average error between the predicted and actual values ​​is approximately 7.4%, indicating that this method can effectively invert the thickness of multilayer materials using the Lamb wave frequency characteristics at a specific wavelength.

[0086] like Figure 4 As shown, Figure 4 This is a diagram showing the comparison between the actual and predicted thickness values ​​of each layer in the five-layer sample. The horizontal axis represents the layer number, from layer 1 to layer 5; the vertical axis represents the thickness in μm. Figure 4 The text uses two different filling methods for the bar chart to represent the actual thickness of each layer and the predicted thickness obtained through the target thickness frequency mapping model. The diagonally filled bars represent the actual thickness, and the dotted filled bars represent the predicted thickness. The value marked at the top of the bars is the corresponding thickness value.

[0087] from Figure 4It can be seen that for layers 1, 3, and 4, the predicted thickness is basically consistent with the actual thickness, with only minor deviations. For layers 2 and 5, the predicted thickness is somewhat underestimated compared to the actual thickness, but the overall error is still within an acceptable range. This comparative result demonstrates that the proposed layered thickness measurement method based on the target thickness frequency mapping model can accurately invert the thickness of each component layer of the multilayer composite material, verifying the effectiveness and reliability of the method in practical applications.

[0088] Based on the above method, this application discloses a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement system, referencing... Figure 5 The grating-modulated laser ultrasonic composite material layer thickness measurement system 1 includes a data acquisition module 11, a model construction module 12, and a thickness measurement module 13, wherein... The data acquisition module 11 is used to acquire the material parameters corresponding to each component layer in the multilayer composite material, and to construct a Lamb wave dispersion physical model based on the material parameters; The model building module 12 is used to generate multiple sets of thickness combinations within the range of the preset total thickness and the thickness variation of each component layer of the multilayer composite material. It calculates the multilayer Lamb wave dispersion curve under each thickness combination according to the Lamb wave dispersion physical model. At the target wavelength, it extracts the frequency peak value and Lamb wave mode number corresponding to each Lamb wave mode from the multilayer Lamb wave dispersion curve. It pairs the frequency feature vector with the corresponding layer thickness to build a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak value and the Lamb wave mode number. The initial thickness frequency mapping model is trained based on the training sample set to obtain the target thickness frequency mapping model. The thickness measurement module 13 is used to collect the frequency characteristics corresponding to the surface of the multilayer composite material under test, and input the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

[0089] In one example, the data acquisition module 11 is used to acquire the material parameters corresponding to each component layer in the multilayer composite material. The material parameters include the density, elastic modulus, Poisson's ratio, and the thickness range of each component layer. Based on the layered plate theory and the material parameters, the Lamb wave dispersion control equation is constructed and solved to obtain the multilayer Lamb wave dispersion physical model under different thickness combinations.

[0090] In one example, the model building module 12 is used to determine the preset total thickness range of the multilayer composite material and the corresponding thickness range of each component layer according to the design requirements or prior information; the thickness range of each component layer is discretized with a preset step size; and under the condition of satisfying the thickness constraint, the thickness of each component layer is enumerated by rules or randomly sampled to generate multiple sets of thickness combinations that satisfy the thickness constraint.

[0091] In one example, the model building module 12 is used to extract the frequency peaks of one or more Lamb wave modes from the corresponding multilayer Lamb wave dispersion curve at the target wavelength for each thickness combination, and obtain the Lamb wave mode number corresponding to each frequency peak; combine each frequency peak and the Lamb wave mode number corresponding to the frequency peak according to a preset arrangement rule to form a frequency feature vector, and use the layer thickness of each component layer corresponding to the frequency feature vector as the output thickness vector, and construct training samples based on the frequency feature vector and the output thickness vector; construct training samples for all thickness combinations to obtain a training sample set composed of multiple frequency feature vectors and the output thickness vectors corresponding to the frequency feature vectors.

[0092] In one example, the model building module 12 is used to pair frequency feature vectors and output thickness vectors in the training sample set to construct sample pairs, and divide the sample pairs into training set and validation set; input the frequency feature vectors in the training set into the initial thickness frequency mapping model to obtain the first predicted thickness vector corresponding to the frequency feature vector; calculate the first error value between the first predicted thickness vector and the corresponding output thickness vector according to the preset loss function, and iteratively update the model parameters of the initial thickness frequency mapping model based on the first error value and the preset optimization algorithm; in each training round, input the frequency feature vectors in the validation set into the updated initial thickness frequency mapping model to obtain the second predicted thickness vector corresponding to the frequency feature vector, and calculate the second error value between the second predicted thickness vector and the corresponding output thickness vector according to the preset loss function; when the second error value meets the preset convergence condition, stop training the initial thickness frequency mapping model, and use the initial thickness frequency mapping model at this time as the target thickness frequency mapping model.

[0093] In one example, the thickness measurement module 13 is used to form periodic excitation stripes on the surface of the multilayer composite material under test using a grating-modulated pulsed laser to excite Lamb waves propagating in the direction perpendicular to the periodic excitation stripes; a laser interferometer or laser Doppler vibrator is used to perform a one-dimensional spatial scan of the surface of the multilayer composite material under test in the direction perpendicular to the periodic excitation stripes, and Lamb wave signals of surface displacement or surface velocity changing with time at each scan position are collected. The Lamb wave signals are arranged according to spatial position and time dimension to obtain the B-scan data of the multilayer composite material under test; a two-dimensional Fourier transform is performed on the B-scan data to obtain a two-dimensional spectrum, and according to the target wavelength or target wavenumber in the two-dimensional spectrum, the frequency peaks of multiple Lamb wave modes and the Lamb wave mode numbers corresponding to the frequency peaks are extracted. All the frequency peaks and Lamb wave mode numbers are arranged in a preset order to form the frequency feature vector of the multilayer composite material under test; the frequency feature vector is input into the target thickness frequency mapping model to obtain the output thickness vector corresponding to the frequency feature vector, and each thickness component in the output thickness vector is used as the thickness of each component layer in the multilayer composite material under test.

[0094] In one example, at the target wavelength, the Lamb wave modes appearing in the multilayer Lamb wave dispersion curve are sorted from low to high according to the corresponding frequency peaks. The sorted numbers are used as the Lamb wave mode numbers of each Lamb wave mode. The Lamb wave mode numbers are combined with the corresponding frequency peaks in the form of integer encoding or one-hot encoding to form the frequency feature vector of the initial thickness frequency mapping model.

[0095] Please see Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 2 may include: at least one processor 21, at least one network interface 24, user interface 23, memory 25, and at least one communication bus 22.

[0096] The communication bus 22 is used to enable communication between these components.

[0097] The user interface 23 may include a display screen and a camera. Optionally, the user interface 23 may also include a standard wired interface and a wireless interface.

[0098] The network interface 24 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0099] The processor 21 may include one or more processing cores. The processor 21 connects to various parts of the server using various interfaces and lines, and performs various server functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 25, and by calling data stored in the memory 25. Optionally, the processor 21 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 21 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 21.

[0100] The memory 25 may include random access memory (RAM) or read-only memory. Optionally, the memory 25 may include non-transitory computer-readable storage medium. The memory 25 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 25 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 25 may also be at least one storage device located remotely from the aforementioned processor 21. Figure 6 As shown, the memory 25, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an application program for a deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method.

[0101] exist Figure 6In the electronic device 2 shown, the user interface 23 is mainly used to provide an input interface for the user and to obtain the user input data; while the processor 21 can be used to call the application program stored in the memory 25, which is a deep learning-based grating modulation laser ultrasonic composite material layer thickness measurement method. When executed by one or more processors, the electronic device performs one or more methods as described in the above embodiments.

[0102] A non-transitory computer-readable storage medium stores instructions that, when executed by one or more processors, cause a computer to perform one or more methods as described in the above embodiments.

[0103] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0104] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0105] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the shown or discussed mutual couplings or direct couplings or communication connections may be through some service interfaces; indirect couplings or communication connections between apparatuses or units may be electrical or other forms.

[0106] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0107] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0108] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0109] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A deep learning-based method for layered thickness measurement of composite materials using grating-modulated laser ultrasonic technology, characterized in that... The method includes: Material parameters corresponding to each component layer in a multilayer composite material are collected, and a Lamb wave dispersion physical model is constructed based on the material parameters. Multiple thickness combinations are generated within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material, and the multilayer Lamb wave dispersion curve under each thickness combination is calculated according to the Lamb wave dispersion physical model. At the target wavelength, extract the frequency peak and Lamb wave mode number corresponding to each Lamb wave mode from the multilayer Lamb wave dispersion curve, pair the frequency feature vector with the corresponding layer thickness, and construct a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak and the Lamb wave mode number. The initial thickness frequency mapping model is trained based on the training sample set to obtain the target thickness frequency mapping model; The frequency characteristics corresponding to the surface of the multilayer composite material under test are collected, and the frequency characteristics are input into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

2. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 1, characterized in that, The process of collecting material parameters corresponding to each constituent layer in the multilayer composite material and constructing a Lamb wave dispersion physical model based on these material parameters specifically includes: The material parameters corresponding to each component layer in the multilayer composite material are collected, wherein the material parameters include the density, elastic modulus, Poisson's ratio and the thickness range of each component layer. Based on the layered plate theory and the material parameters, the Lamb wave dispersion control equation is constructed and solved to obtain a multilayer Lamb wave dispersion physical model under different thickness combinations.

3. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 1, characterized in that, The process of generating multiple thickness combinations within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material specifically includes: The preset total thickness range and the corresponding layer thickness range of the multilayer composite material are determined according to the design requirements or prior information. The thickness range of each component layer is discretized with a preset step size. Under the condition of satisfying the thickness constraint, the thickness of each component layer is enumerated by rules or randomly sampled to generate multiple sets of thickness combinations that satisfy the thickness constraint.

4. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 1, characterized in that, The step of pairing frequency feature vectors with corresponding layer thicknesses to construct a training sample set for training the initial thickness-frequency mapping model specifically includes: For each thickness combination, extract the frequency peaks of one or more Lamb wave modes from the corresponding multilayer Lamb wave dispersion curve at the target wavelength, and obtain the Lamb wave mode number corresponding to each frequency peak. Each frequency peak and the Lamb wave mode number corresponding to the frequency peak are combined according to a preset arrangement rule to form a frequency feature vector, and the layer thickness of each component layer corresponding to the frequency feature vector is used as the output thickness vector. Training samples are constructed based on the frequency feature vector and the output thickness vector. Training samples are constructed for all thickness combinations to obtain a training sample set consisting of multiple frequency feature vectors and the output thickness vectors corresponding to the frequency feature vectors.

5. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 4, characterized in that, The step of training the initial thickness-frequency mapping model based on the training sample set to obtain the target thickness-frequency mapping model specifically includes: The frequency feature vector and output thickness vector in the training sample set are paired to construct sample pairs, and the sample pairs are divided into training set and validation set; The frequency feature vector in the training set is input into the initial thickness frequency mapping model to obtain the first predicted thickness vector corresponding to the frequency feature vector; The first error value between the first predicted thickness vector and the corresponding output thickness vector is calculated according to the preset loss function, and the model parameters of the initial thickness frequency mapping model are iteratively updated based on the first error value and the preset optimization algorithm. In each training round, the frequency feature vector in the validation set is input into the updated initial thickness frequency mapping model to obtain the second predicted thickness vector corresponding to the frequency feature vector, and the second error value between the second predicted thickness vector and the corresponding output thickness vector is calculated according to the preset loss function. When the second error value meets the preset convergence condition, the training of the initial thickness frequency mapping model is stopped, and the initial thickness frequency mapping model at this time is used as the target thickness frequency mapping model.

6. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 1, characterized in that, The process of acquiring frequency characteristics corresponding to the surface of the multilayer composite material under test and inputting these frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test specifically includes: A periodic excitation stripe is formed on the surface of the multilayer composite material under test by using a grating-modulated pulsed laser to excite a Lamb wave propagating in the direction perpendicular to the periodic excitation stripe. The surface of the multilayer composite material under test is scanned in one dimension in the direction perpendicular to the periodic excitation stripes using a laser interferometer or a laser Doppler vibrator. Lamb wave signals of surface displacement or surface velocity changing with time at each scanning position are collected, and the Lamb wave signals are arranged according to spatial position and time dimension to obtain the B-scan data of the multilayer composite material under test. A two-dimensional Fourier transform is performed on the B-scan data to obtain a two-dimensional spectrum. Based on the target wavelength or target wavenumber in the two-dimensional spectrum, the frequency peaks of multiple Lamb wave modes and the Lamb wave mode numbers corresponding to the frequency peaks are extracted. All the frequency peaks and Lamb wave mode numbers are arranged in a preset order to form the frequency feature vector of the multilayer composite material under test. The frequency feature vector is input into the target thickness frequency mapping model to obtain the output thickness vector corresponding to the frequency feature vector, and each thickness component in the output thickness vector is used as the thickness of each component layer in the multilayer composite material under test.

7. The deep learning-based grating-modulated laser ultrasonic composite material layer thickness measurement method as described in claim 1, characterized in that, At the target wavelength, the Lamb wave modes appearing in the multilayer Lamb wave dispersion curve are sorted from low to high according to their corresponding frequency peaks. The sorted sequence numbers are used as the Lamb wave mode numbers of each Lamb wave mode. The Lamb wave mode numbers are combined with the corresponding frequency peaks in the form of integer encoding or one-hot encoding to form the frequency feature vector of the initial thickness frequency mapping model.

8. A grating-modulated laser ultrasonic composite material layer thickness measurement system based on deep learning, characterized in that, The grating-modulated laser ultrasonic composite material layer thickness measurement system includes a data acquisition module, a model construction module, and a thickness measurement module. The data acquisition module is used to acquire the material parameters corresponding to each component layer in the multilayer composite material, and to construct a Lamb wave dispersion physical model based on the material parameters. The model building module is used to generate multiple thickness combinations within the range of the preset total thickness and the thickness variation of each component layer corresponding to the multilayer composite material, and calculate the multilayer Lamb wave dispersion curve under each thickness combination according to the Lamb wave dispersion physical model. At the target wavelength, the frequency peak value and Lamb wave mode number corresponding to each Lamb wave mode are extracted from the multilayer Lamb wave dispersion curve. The frequency feature vector is paired with the corresponding layer thickness to construct a training sample set for training the initial thickness frequency mapping model. The frequency feature vector includes the frequency peak value and the Lamb wave mode number. The initial thickness frequency mapping model is trained based on the training sample set to obtain the target thickness frequency mapping model. The thickness measurement module is used to collect the frequency characteristics corresponding to the surface of the multilayer composite material under test, and input the frequency characteristics into the target thickness frequency mapping model to obtain the thickness of each component layer in the multilayer composite material under test.

9. An electronic device, characterized in that, The device includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images