CFRP wing skin damage positioning system and method based on multi-modal signal processing and bayesian optimization DSCN

Abstract

The application relates to the technical field of material nondestructive testing, in particular to a CFRP wing skin damage positioning system and method based on multi-modal signal processing and Bayesian optimization DSCN, which comprises the following steps: performing permutation entropy analysis and Higuchi fractal dimension analysis on a sensor received signal data set; performing time-frequency feature extraction on the sensor received signal data set deviating from a reference; performing damage feature screening on the signal data in an effective time period; training a DSCN structure to obtain an initial CFRP wing skin damage positioning prediction model; performing hyperparameter optimization on the initial CFRP wing skin damage positioning prediction model according to a Bayesian algorithm; and performing damage positioning prediction on a to-be-tested sample according to a final CFRP wing skin damage positioning prediction model to obtain corresponding CFRP wing skin damage positioning results. The application can accurately and efficiently extract effective damage features from complex signals, and significantly improves the precision and efficiency of CFRP wing skin damage positioning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material nondestructive testing, and particularly relates to a CFRP wing skin damage positioning system and method based on multi-modal signal processing and Bayesian optimization DSCN. BACKGROUND

[0002] The wing skin is an outer structural member of the wing of the unmanned aerial vehicle, which directly covers the wing frame to form a complete aerodynamic shape. The function of the wing skin of the unmanned aerial vehicle is mainly to maintain the aerodynamic shape of the unmanned aerial vehicle and transmit the aerodynamic load, and also to protect the internal equipment. Therefore, the integrity of the wing skin directly affects the flight performance and safety of the unmanned aerial vehicle. Carbon fiber reinforced plastic (CFRP) has become the mainstream material in the structure of the unmanned aerial vehicle due to its light weight, high strength, excellent fatigue resistance and designable anisotropy. Globally, CFRP accounts for 60%-80% of the structure of the unmanned aerial vehicle, and the wing skin is one of the main applications. The wing skin made of CFRP has a smooth surface, accurate shape and good symmetry, which can reduce air resistance, improve flight speed and efficiency, and provide sufficient lift and stability for the wing.

[0003] However, the brittle matrix and layered structure of CFRP make it sensitive to impact load and prone to internal damage. In long-term service, cyclic load can cause micro-damage accumulation in the material, initiate fatigue cracks, and if not detected in time, the cracks will expand into macro-damage, threatening flight safety. Precise positioning of internal crack damage of the wing skin is the key to preventing catastrophic failure of CFRP structures. Traditional nondestructive testing techniques have problems such as complex operation, low efficiency, detection blind area and radiation risk. In contrast, Lamb wave monitoring technology has become an ideal solution for internal damage detection of CFRP wing skin due to its millisecond response speed, long propagation distance and high sensitivity to hidden damage.

[0004] However, Lamb wave technology faces three major technical bottlenecks in the damage positioning of curved CFRP wing skin: multi-sensor signal modal aliasing noise interference, existing sensor screening methods rely on a single index, and it is difficult to distinguish between damage signals and noise; complex signal feature extraction is difficult, and traditional time-frequency analysis methods have shortcomings; high-dimensional time series signal redundancy, traditional linear dimension reduction methods face nonlinear failure and local feature loss problems. In addition, existing damage positioning models have limitations such as insufficient representation ability of shallow networks and low optimization efficiency of deep learning models, and hyperparameter optimization also faces challenges, traditional methods have problems such as dimension disaster or lack of intelligent guidance.

[0005] Chinese patent publication No. CN116642410A discloses a non-contact CFRP structure damage monitoring system and method. The non-contact CFRP structure damage monitoring system includes a vector network analyzer, a focusing antenna, and a computer. The vector network analyzer is connected to the focusing antenna through a coaxial cable, responsible for the transmission, reception, processing and analysis of microwave signals. The focusing antenna is responsible for focusing and amplifying the pulse signal generated by the vector network analyzer, irradiating to the center of the CFRP plate, and responsible for receiving the reflected signal. The capacitance of the CFRP plate is derived from the microwave reflection signal to characterize the strain in the deformation area. This scheme is difficult to accurately and efficiently extract effective damage features from complex signals in CFRP wing skin damage positioning, resulting in reduced damage positioning accuracy and efficiency. SUMMARY

[0006] To this end, the present application provides a CFRP wing skin damage positioning system and method based on multi-modal signal processing and Bayesian optimized DSCN to overcome the problem that it is difficult to accurately and efficiently extract effective damage features from complex signals in CFRP wing skin damage positioning in the prior art, resulting in reduced damage positioning accuracy and efficiency.

[0007] To achieve the above-mentioned purpose, in one aspect, the present application provides a CFRP wing skin damage positioning method based on multi-modal signal processing and Bayesian optimized DSCN, comprising the following steps:

[0008] S1. In the numerical simulation experiment of the CFRP wing skin, a wing skin undamaged model is established, and the damage position of the CFRP wing skin is set on multiple spatial coordinates of the wing skin undamaged model. The corresponding Lamb wave time domain signal data is obtained according to the damage position of the CFRP wing skin, and the corresponding sensor received signal data set is obtained according to the Lamb wave time domain signal data;

[0009] S2. Based on the preset double-index sensor signal screening strategy, the sensor received signal data set is subjected to permutation entropy analysis and Higuchi fractal dimension analysis to obtain the sensor received signal data set deviating from the reference;

[0010] S3. The sensor received signal data set deviating from the reference is subjected to time-frequency feature extraction to obtain effective time period signal data;

[0011] S4. Based on the functional principal component analysis strategy, the effective time period signal data is subjected to damage feature screening to obtain a damage feature screening result;

[0012] S5. The DSCN structure is trained according to the damage feature screening result to obtain an initial CFRP wing skin damage positioning prediction model;

[0013] S6. Perform hyperparameter optimization on the initial CFRP wing skin damage positioning prediction model according to the Bayesian algorithm to obtain a final CFRP wing skin damage positioning prediction model;

[0014] S7. Perform damage positioning prediction on the to-be-tested sample according to the final CFRP wing skin damage positioning prediction model to obtain the corresponding CFRP wing skin damage positioning result.

[0015] Compared with the prior art, the beneficial effects of the present application are:

[0016] 1. By using permutation entropy analysis and Higuchi fractal dimension analysis to screen the sensor received signal data set, the permutation entropy can quantify the randomness of the signal and reflect the dynamic complexity of the signal, and the Higuchi fractal dimension can describe the self-similarity and non-linear characteristics of the signal. Through the double-index strategy of permutation entropy and Higuchi fractal dimension, the sensor received signal deviating from the reference can be identified from multiple dimensions, effectively filtering out irrelevant signals caused by environmental noise, sensor errors, etc. Compared with the existing technology which only indirectly represents strain through capacitance change, the present application can screen the original signal received by the sensor, more accurately retain the effective information related to damage, and avoid the interference of invalid features in the complex signal.

[0017] 2. By performing time-frequency feature extraction on the sensor received signal data set deviating from the reference, effective time period signal data can be obtained. Time-frequency analysis can expand the signal in both time and frequency dimensions, clearly presenting the frequency component changes of the signal at different times. When the CFRP wing skin damage occurs, the frequency characteristics of the relevant signals of the CFRP wing skin will change in a certain time period. Through time-frequency feature extraction, the frequency change information closely related to the damage can be captured. Compared with the existing technology which only analyzes the signal from a single dimension, the present application can more comprehensively and accurately obtain the damage signal characteristics, improving the efficiency of extracting effective damage features from the sensor received signal, thereby improving the overall efficiency and accuracy of damage positioning.

[0018] 3. By using the Bayesian algorithm to perform hyperparameter optimization on the initial CFRP wing skin damage positioning prediction model, the optimal parameter combination is automatically searched, avoiding the blindness of manual parameter adjustment. Compared with the existing technology which only relies on microwave reflection signal to derive a single feature representation of capacitance, the present application can extract key damage features from multi-dimensional time-frequency characteristics through functional principal component analysis strategy, ensure the optimal configuration of model parameters through Bayesian algorithm, and through the synergistic effect of functional principal component analysis strategy and Bayesian algorithm, the final CFRP wing skin damage positioning prediction model can more accurately capture the mapping relationship between damage features and position, improving the accuracy of damage positioning.

[0019] Further, the S1 comprises the following steps:

[0020] S11, a finite element simulation software is used to establish a wing skin undamaged model, and a plurality of internal crack damage positions are determined in the wing skin undamaged model based on a preset crack damage rule, thereby forming a CFRP wing skin simulation structure containing undamaged and different damage states, wherein the preset crack damage rule comprises presetting internal crack damage positions in a high stress area of the CFRP wing skin, and setting according to a typical damage mode of the CFRP wing skin;

[0021] S12, a process of excitation and propagation of Lamb waves in the CFRP wing skin simulation structure is simulated, thereby obtaining an original time domain signal data set;

[0022] S13, each preset sensor receiving point is located from the original time domain signal data set, and fluctuation response characteristics of each sensor receiving point in a time domain dimension are analyzed, thereby obtaining a sensor receiving signal data set, wherein the sensor receiving signal data set comprises an undamaged state data set and a damage state data set.

[0023] In the scheme, the finite element simulation software refers to an engineering simulation finite element software containing an element library and a material model library, which can process complex boundary and load conditions, the wing skin undamaged model refers to a model simulating a structure form and mechanical characteristics of the wing skin without internal cracks and other damages based on the finite element simulation software, and the CFRP wing skin simulation structure refers to a structure containing undamaged and different damage states formed by determining a plurality of internal crack damage positions according to the preset crack damage rule based on the wing skin undamaged model.

[0024] The CFRP wing skin simulation structure containing undamaged and different damage states is constructed by the finite element simulation software, the excitation and propagation of Lamb waves in the structure are simulated, the sensor receiving points are located from the original time domain signal data set, and the fluctuation response characteristics are analyzed, thereby obtaining the sensor receiving signal data set containing undamaged and damage state data, which is helpful for accurately analyzing damage characteristics.

[0025] Further, the S2 comprises the following steps:

[0026] S21, permutation entropy analysis is performed on the undamaged state data set and the damage state data set according to a permutation entropy analysis algorithm in a preset double-index sensor signal screening strategy, thereby obtaining permutation entropy values of each sensor under the undamaged state of the CFRP wing skin and the damage state of the CFRP wing skin , wherein a mathematical expression of the permutation entropy value is:

[0027] , denotes a probability of occurrence of the i-th permutation pattern, denotes an embedding dimension, denotes an embedding dimension factorial of n;

[0028] S22, according to the preset double-index sensor signal screening strategy Higuchi fractal dimension algorithm is respectively carried out Higuchi fractal dimension analysis to the undamaged state data set and the damaged state data set, and the Higuchi fractal dimension value of each sensor under the undamaged state of the CFRP wing skin and the damaged state of the CFRP wing skin is obtained;

[0029] S23, the permutation entropy value and the Higuchi fractal dimension value are integrated to obtain a double-index feature matrix;

[0030] S24, according to the undamaged state data set, the permutation entropy change reference and the Higuchi fractal dimension change reference of each sensor are set, and according to the permutation entropy value and the Higuchi fractal dimension value of each sensor under the damaged state of the CFRP wing skin, the corresponding permutation entropy change rate and Higuchi fractal dimension change rate are determined;

[0031] S25, according to the permutation entropy change reference and the Higuchi fractal dimension change reference, the double-index feature matrix is jointly analyzed to obtain a candidate sensitive sensor list meeting a preset sensitive sensor condition, and the preset sensitive sensor condition includes that the permutation entropy change rate exceeds the permutation entropy change reference, or the Higuchi fractal dimension change rate exceeds the Higuchi fractal dimension change reference;

[0032] S26, the sensor receiving signal corresponding to each candidate sensitive sensor in the candidate sensitive sensor list is added to the sensor receiving signal data set deviating from the reference.

[0033] In the scheme, the preset double-index sensor signal screening strategy refers to a method of comprehensively utilizing permutation entropy analysis and Higuchi fractal dimension analysis to screen sensor signals. The two analyses are performed on the undamaged and damaged state data sets respectively, and the sensitive sensors are screened out after integrating the results and combining with the change criteria. The permutation entropy change criterion refers to the change reference standard of the permutation entropy values of each sensor set based on the undamaged state data set, which is used to measure whether the change of the permutation entropy values of the sensor under the damaged state exceeds the normal range. The Higuchi fractal dimension change criterion refers to the change reference standard of the Higuchi fractal dimension values of each sensor set according to the undamaged state data set, which is used to judge whether the change of the Higuchi fractal dimension values of the sensor under the damaged state is abnormal. The permutation entropy change rate refers to the change degree of the permutation entropy values of each sensor under the damaged state of the CFRP wing skin relative to the undamaged state. The Higuchi fractal dimension change rate refers to the change degree of the Higuchi fractal dimension values of each sensor under the damaged state of the CFRP wing skin relative to the undamaged state. The candidate sensitive sensor list refers to the sensor list screened out according to the comparison of the permutation entropy and Higuchi fractal dimension change rate with the corresponding change criterion. The sensor receiving signal data set deviating from the criterion refers to the data set formed by adding the sensor receiving signal corresponding to the candidate sensitive sensor list.

[0034] Through the preset double-index sensor signal screening strategy, permutation entropy and Higuchi fractal dimension analysis are performed on the undamaged data set and the damaged state data set respectively, a double-index feature matrix is obtained, the change rate of each sensor is determined by combining the change criterion set in the undamaged state, and then the candidate sensitive sensor list is screened out and the sensor receiving signal data set deviating from the criterion is formed. This can accurately identify sensors sensitive to damage, remove irrelevant signal interference, and highlight damage features.

[0035] Further, the S3 comprises the following steps:

[0036] S31, performing time-frequency localization analysis on the sensor receiving signal data set deviating from the criterion by continuous wavelet transform to obtain a time-frequency feature data set, wherein the mathematical expression of the energy distribution of the time-frequency plane in the time-frequency localization analysis process is:

[0037] wherein, a scale parameter, a translation parameter, a wavelet basis function, a time domain signal, a time variable;

[0038] ​S32, input the time-frequency feature dataset into a Markov transition field for state transition analysis, and construct a state transition probability matrix, wherein the state transition probability The mathematical expression of the state transition probability matrix is as follows:

[0039] represents the number of transitions from state i to state j, represents the number of transitions from state i to state k, represents a smoothing factor, and k represents the number of states;

[0040] S33, obtain effective time period signal data by signal evolution law of the state transition probability matrix and filtering characteristic time period.

[0041] In the scheme, the time-frequency feature dataset refers to a data set obtained by performing time-frequency localization analysis on a sensor signal data set deviating from a reference using continuous wavelet transform, the state transition probability matrix refers to a matrix constructed by inputting the time-frequency feature dataset into a Markov transition field for state transition analysis, the signal evolution law of the state transition probability matrix refers to analyzing the transition trend and mode of the signal between different states based on the state transition probability matrix, so as to reveal the change law of the signal over time, and the effective time period signal data refers to signal data in a specific time period containing key damage information filtered by analyzing the signal evolution law of the state transition probability matrix.

[0042] The time-frequency feature dataset is obtained by performing time-frequency analysis on the signal deviating from the reference using continuous wavelet transform, the time-frequency characteristics of the signal are fully presented, the time-frequency feature dataset is input into the Markov transition field to construct the state transition probability matrix, the signal evolution law is analyzed, the trend of the signal state change can be accurately grasped, the effective time period signal data is filtered based on this, irrelevant and redundant information is removed, and the key features related to damage are highlighted.

[0043] Further, the S4 comprises the following steps:

[0044] S41, standardize the effective time period signal data based on a functional principal component analysis strategy, obtain standardized effective time period signal data, and construct a mean function according to the standardized effective time period signal data;

[0045] S42, center the standardized effective time period signal data according to the mean function, and obtain centered effective time period signal data;

[0046] S43, expand the centered effective time period signal data according to a B-spline basis function, obtain a basis function matrix, and construct a projection matrix according to the basis function matrix; ​

[0047] S44, projecting the centering processed effective time period signal data according to the projection matrix to obtain projected signal data;

[0048] S45, performing covariance calculation on the projected signal data to obtain a covariance matrix, and processing the covariance matrix according to an L2 regularization method to obtain a regularized covariance matrix;

[0049] S46, performing Cholesky decomposition on the regularized covariance matrix to obtain an intermediate result containing basis coefficients, and performing damage feature screening on the projected signal data according to the intermediate result containing the basis coefficients to obtain a damage feature screening result.

[0050] In the scheme, the functional principal component analysis strategy refers to a method for extracting main features from data, reducing data dimension, and highlighting key information in the data, the B-spline basis function refers to a basis function that can flexibly fit various complex function shapes for expanding functional data, the projection matrix refers to a matrix constructed by the basis function matrix, used for projecting the centering processed effective time period signal data into a new space, the L2 regularization method refers to adding a small regularization parameter on the diagonal line of the covariance matrix in the covariance matrix processing, and the Cholesky decomposition refers to a method of decomposing a symmetric positive definite matrix into a lower triangular matrix and its transpose product.

[0051] By using the functional principal component analysis strategy, the effective time period signal data is first standardized and centered, then the B-spline basis function expansion is used to construct the projection matrix for projection, then the L2 regularization method is used to process the covariance matrix, and finally the Cholesky decomposition is performed, which can effectively extract the main features of the signal data, reduce the data dimension, improve the efficiency and accuracy of data processing, and enhance the stability of the covariance matrix through regularization processing, which can more accurately reflect the damage characteristics of the CFRP wing skin.

[0052] Further, the S5 comprises the following steps:

[0053] S51, performing network initialization on the DSCN structure according to a progressive architecture construction method, wherein the network initialization includes setting the DSCN structure to a single hidden layer and setting the hidden layer to a single node, and all hidden layer nodes are fully connected to the output layer of the DSCN structure;

[0054] S52, constructing training samples according to the damage feature screening result, training the DSCN structure according to the training samples, and dynamically adjusting node parameters through a supervised learning mechanism to obtain optimized node parameters and corresponding model performance indicators;

[0055] S53, gradually increase the number of hidden layers and nodes in the depth direction and the width direction based on the optimized node parameters and the corresponding model performance indicators, to obtain an expanded DSCN structure;

[0056] S54, perform output weight analysis on the expanded DSCN structure, calculate the connection weights between all hidden layer nodes and the output layer by the least square method, and establish an initial CFRP wing skin damage positioning prediction model.

[0057] In this scheme, the DSCN (Deep Stochastic Configuration Networks) structure refers to a deep neural network model based on random configuration, combining the hierarchical feature extraction capability of deep learning and the random configuration method. The progressive architecture construction method refers to a method of gradually constructing and optimizing the neural network structure. The supervised learning mechanism refers to adjusting the network node parameters by the known damage feature screening results when training the DSCN structure. The network calculates the error between the predicted output and the actual label based on the input training sample and the corresponding label, and then dynamically adjusts the node parameters by the back propagation algorithm to minimize the error.

[0058] By using the progressive architecture construction method, the DSCN structure is initialized, the number of hidden layers and nodes is gradually increased, the network complexity can be flexibly adjusted according to the training sample and the model performance indicator, and the node parameters are dynamically adjusted by using the damage feature screening results as the training sample through the supervised learning mechanism, so that the DSCN structure can better learn the features and rules in the data. Finally, the output weight is determined by the least square method, the initial CFRP wing skin damage positioning prediction model is established, and the accuracy and adaptability of the model are improved.

[0059] Further, the S6 comprises the following steps:

[0060] S61, determining the input feature dimension and the hyperparameter set of the initial CFRP wing skin damage positioning prediction model according to the Bayesian algorithm;

[0061] S62, determining the test error of the predicted coordinates and the real coordinates of the test set samples based on the input feature dimension and the hyperparameter set, and constructing a target optimization function with the minimum test error as the target;

[0062] S63, performing Bayesian probability modeling analysis on the target optimization function by the prior distribution and the mixed kernel function, and constructing a Gaussian process model;

[0063] S64, constructing a time-weighted expected improvement acquisition function based on the prediction mean, standard deviation and current optimal observation value of the Gaussian process model, and selecting the hyperparameter set of the next evaluation point through the time-weighted expected improvement acquisition function;

[0064] S65, updating the Gaussian process model according to the hyperparameter set of the next evaluation point in combination with the model prediction value and the actual test error until a preset convergence condition is met, and outputting the optimal hyperparameter combination;

[0065] S66, inputting the optimal hyperparameter combination into the initial CFRP wing skin damage positioning prediction model for hyperparameter optimization to obtain a final CFRP wing skin damage positioning prediction model.

[0066] In the scheme, the input feature dimension refers to the number of features used to input the initial CFRP wing skin damage positioning prediction model, the hyperparameter set refers to the parameters that need to be manually set before model training, such as learning rate, regularization coefficient, etc., different hyperparameter combinations will affect the training effect and prediction performance of the model, the target optimization function refers to a function constructed to minimize the test error of the predicted coordinates and the true coordinates of the test set samples, used to evaluate the performance of the model under different input feature dimensions and hyperparameter combinations, the Gaussian process model refers to a model providing probability prediction of the target function obtained by Bayesian probability modeling analysis of the target optimization function, the time-weighted expected improvement acquisition function is constructed based on the prediction mean, standard deviation and current optimal observation value of the Gaussian process model, used to select the hyperparameter set of the next evaluation point, and the preset convergence condition refers to a standard for judging whether the model optimization process is ended, such as reaching the maximum number of iterations, the target function value changing less than a threshold, etc., and the optimal hyperparameter combination is output when the condition is met.

[0067] By using the Bayesian algorithm to determine the input feature dimension and the hyperparameter set, a basis is provided for model optimization, the model performance is evaluated by constructing the target optimization function, the Gaussian process model is obtained by Bayesian probability modeling, which can accurately predict the target function, the time-weighted expected improvement acquisition function is used to effectively select the hyperparameter set of the next evaluation point, accelerate the optimization process, and update the Gaussian process model according to the next evaluation point until the optimal hyperparameter combination is output when the preset convergence condition is met, which is input into the initial CFRP wing skin damage positioning prediction model to obtain the final CFRP wing skin damage positioning prediction model, which can automatically optimize the model parameters, improve the accuracy and stability of the CFRP wing skin damage positioning prediction model, reduce the positioning error, and improve the damage positioning effect.

[0068] Further, the S7 comprises the following steps:

[0069] S71, multi-modal signal acquisition is performed on the to-be-tested CFRP wing skin to obtain a to-be-tested sample, and the to-be-tested sample is preprocessed to obtain a preprocessed to-be-tested sample;

[0070] S72, the preprocessed to-be-tested sample is input into the final CFRP wing skin damage positioning and prediction model for damage positioning and prediction, a spatial comparison graph containing real three-dimensional coordinates and predicted three-dimensional coordinates of the damage is obtained, and spatial constraint analysis is performed on the spatial comparison graph to obtain a corresponding CFRP wing skin damage positioning result.

[0071] In this scheme, the state information of the to-be-tested CFRP wing skin is comprehensively and accurately obtained by acquiring the to-be-tested sample through multi-modal signal acquisition and preprocessing, the preprocessed to-be-tested sample is input into the final CFRP wing skin damage positioning and prediction model for damage positioning and prediction, a spatial comparison graph containing real three-dimensional coordinates and predicted three-dimensional coordinates of the damage is obtained, the damage distribution can be intuitively presented, and the damage positioning result is obtained by performing spatial constraint analysis on the spatial comparison graph, thereby improving the accuracy of damage positioning.

[0072] On the other hand, the application also provides a CFRP wing skin damage positioning system based on multi-modal signal processing and Bayesian optimization DSCN, comprising:

[0073] A sensor received signal data set acquisition module is configured to, in a numerical simulation experiment of a CFRP wing skin, establish a wing skin undamaged model, set a CFRP wing skin damage position on a plurality of spatial coordinates of the wing skin undamaged model, acquire corresponding Lamb wave time domain signal data according to the CFRP wing skin damage position, and acquire corresponding sensor received signal data set according to the Lamb wave time domain signal data.

[0074] A sensor signal screening module is configured to perform permutation entropy analysis and Higuchi fractal dimension analysis on the sensor received signal data set according to a preset double-index sensor signal screening strategy, and obtain a sensor received signal data set deviating from a benchmark.

[0075] A time-frequency feature extraction module is configured to perform time-frequency feature extraction on the sensor received signal data set deviating from the benchmark to obtain effective time period signal data.

[0076] A damage feature screening module is configured to perform damage feature screening on the effective time period signal data according to a functional principal component analysis strategy to obtain a damage feature screening result.

[0077] A model training module is configured to train a DSCN structure according to the damage feature screening result to obtain an initial CFRP wing skin damage positioning and prediction model.

[0078] a model optimization module, configured to perform hyperparameter optimization on the initial CFRP wing skin damage positioning prediction model according to a Bayesian algorithm, to obtain a final CFRP wing skin damage positioning prediction model;

[0079] a damage positioning prediction module, configured to perform damage positioning prediction on the to-be-tested sample according to the final CFRP wing skin damage positioning prediction model, to obtain a corresponding CFRP wing skin damage positioning result. BRIEF DESCRIPTION OF DRAWINGS

[0080] Figure 1 a flowchart of a CFRP wing skin damage positioning method based on multi-modal signal processing and Bayesian optimization DSCN in an embodiment of the present application;

[0081] Figure 2 a structural diagram of a CFRP wing skin damage positioning system based on multi-modal signal processing and Bayesian optimization DSCN in an embodiment of the present application;

[0082] Figure 3 a performance difference diagram of different prediction models in an embodiment of the present application;

[0083] Figure 4 a diagram of error distribution of different prediction models on different test samples in an embodiment of the present application. DETAILED DESCRIPTION

[0084] The following will be further described in detail through specific embodiments:

[0085] Referring to Figure 1 shown, which is a flowchart of a CFRP wing skin damage positioning method based on multi-modal signal processing and Bayesian optimization DSCN in an embodiment of the present application, including the following steps:

[0086] S1, in a numerical simulation experiment of a CFRP wing skin, a wing skin undamaged model is established, a CFRP wing skin damage position is set on a plurality of spatial coordinates of the wing skin undamaged model, a corresponding Lamb wave time domain signal data is obtained according to the CFRP wing skin damage position, and a corresponding sensor received signal data set is obtained according to the Lamb wave time domain signal data;

[0087] S2, based on a pre-set double-index sensor signal screening strategy, permutation entropy analysis and Higuchi fractal dimension analysis are performed on the sensor received signal data set, to obtain a sensor received signal data set deviating from a benchmark;

[0088] S3, time-frequency feature extraction is performed on the sensor received signal data set deviating from the benchmark, to obtain effective time period signal data;

[0089] S4, damage feature screening is performed on the effective time period signal data based on a functional principal component analysis strategy, and a damage feature screening result is obtained;

[0090] S5, the DSCN structure is trained according to the damage feature screening result, and an initial CFRP wing skin damage positioning prediction model is obtained;

[0091] S6, the initial CFRP wing skin damage positioning prediction model is super parameter optimized according to a Bayesian algorithm, and a final CFRP wing skin damage positioning prediction model is obtained;

[0092] S7, damage positioning prediction is performed on a to-be-tested sample according to the final CFRP wing skin damage positioning prediction model, and a corresponding CFRP wing skin damage positioning result is obtained.

[0093] Specifically, S1 includes the following steps:

[0094] S11, a wing skin undamaged model is established based on a finite element simulation software, and a plurality of internal crack damage positions are determined in the wing skin undamaged model based on a preset crack damage rule, forming a CFRP wing skin simulation structure containing an undamaged state and different damage states, the preset crack damage rule including presetting internal crack damage positions in a high stress area of the CFRP wing skin, and setting according to a typical damage mode of the CFRP wing skin;

[0095] S12, the excitation and propagation process of Lamb waves in the CFRP wing skin simulation structure is simulated, and an original time domain signal data set is obtained;

[0096] S13, each sensor receiving point is positioned from the original time domain signal data set, and the fluctuation response characteristics of each sensor receiving point in the time domain dimension are analyzed, obtaining a sensor receiving signal data set, the sensor receiving signal data set including an undamaged state data set and a damage state data set.

[0097] In this embodiment, the excitation and propagation process of Lamb waves in the CFRP wing skin simulation structure in the finite element simulation software is simulated. First, a complete wing skin geometric model is constructed, including the undamaged state and a plurality of preset internal crack damage positions. The crack can be simulated by reducing the local material stiffness or introducing cohesive element to ensure that its geometric characteristics (such as length, angle) meet the actual damage scene. Subsequently, the CFRP laminate is given the orthotropic material properties, including the elastic modulus, Poisson's ratio and shear modulus, and the fiber direction of each layer is defined. Select an appropriate excitation signal (such as a sine wave modulated by a Hanning window), apply displacement or force load at the selected position on the surface of the wing skin, simulate the excitation effect of the piezoelectric sensor, and the excitation direction needs to match the target Lamb wave mode. Solve by explicit dynamic analysis step, set reasonable grid size (satisfy the wavelength resolution) and time step (ensure stability), record the displacement or stress field output during wave propagation, after the simulation is completed, extract the time domain displacement / acceleration data of each preset sensor receiving point, form the original time domain signal data set.

[0098] When analyzing the time domain fluctuation response characteristics of the sensor receiving point, first locate the sensor position from the original data set to ensure that the crack area is avoided to capture the directly propagating wave signal. Extract the time domain signal through the History Output function of the finite element simulation software, and perform time alignment and filtering processing to remove noise and retain the effective frequency band. Subsequently, analyze the amplitude variation, wave packet arrival time (ToF) and waveform distortion of the time domain signal, calculate the signal energy (such as RMS value) or perform cross-correlation analysis, and compare the differences between the damaged state and the undamaged baseline signal. Finally, store the analyzed characteristics according to the sensor position and damage working condition, and construct a sensor receiving signal data set containing the undamaged state and different damage states. The whole process needs to pay attention to the dispersion effect of CFRP and the crack modeling accuracy, and optimize the calculation efficiency to ensure the feasibility of simulation.

[0099] Specifically, S2 includes the following steps:

[0100] S21, according to the permutation entropy analysis algorithm in the preset double-index sensor signal screening strategy, permutation entropy analysis is performed on the undamaged state data set and the damage state data set respectively, and the permutation entropy values of each sensor in the undamaged state of the CFRP wing skin and the damaged state of the CFRP wing skin are obtained , wherein the mathematical expression of the permutation entropy value is:

[0101] , represents the probability of the i-th permutation mode appearing, represents the embedding dimension, represents the embedding dimension factorial of the embedding dimension;

[0102] S22, performing Higuchi fractal dimension analysis on the undamaged state data set and the damaged state data set respectively according to the Higuchi fractal dimension algorithm in the preset double-index sensor signal screening strategy, to obtain Higuchi fractal dimension values of each sensor under the undamaged state of the CFRP wing skin and the damaged state of the CFRP wing skin;

[0103] S23, integrating the permutation entropy values and the Higuchi fractal dimension values to obtain a double-index feature matrix;

[0104] S24, setting permutation entropy change criteria and Higuchi fractal dimension change criteria for each sensor according to the undamaged state data set, and determining corresponding permutation entropy change rates and Higuchi fractal dimension change rates according to the permutation entropy values and the Higuchi fractal dimension values of each sensor under the damaged state of the CFRP wing skin;

[0105] S25, performing joint analysis on the double-index feature matrix according to the permutation entropy change criteria and the Higuchi fractal dimension change criteria, to obtain a candidate sensitive sensor list meeting preset sensitive sensor conditions, the preset sensitive sensor conditions including that the permutation entropy change rate exceeds the permutation entropy change criteria, or the Higuchi fractal dimension change rate exceeds the Higuchi fractal dimension change criteria;

[0106] S26, adding the sensor received signal corresponding to each candidate sensitive sensor in the candidate sensitive sensor list to the sensor received signal data set deviating from the criteria.

[0107] In this embodiment, the calculation steps of the Higuchi fractal dimension value are as follows:

[0108] For the time series corresponding to the undamaged state data set and the damaged state data set perform segmentation processing, and set the maximum number of segments The time series is divided into segments.

[0109] Calculate the curve length of each segment , the formula is:

[0110] ; wherein, is the number of segments, and N is the number of the time series corresponding to the undamaged state data set and the damaged state data set. Linear fitting is performed on and , and the slope is the Higuchi fractal dimension value.

[0111] Specifically, S3 includes the following steps:

[0112] S31, performing time-frequency localization analysis on the sensor-received signal data set deviating from the reference through continuous wavelet transform to obtain a time-frequency feature data set, wherein the mathematical expression of energy distribution on the time-frequency plane in the time-frequency localization analysis process is:

[0113] wherein, denotes a scale parameter, denotes a translation parameter, denotes a wavelet base function, denotes a time-domain signal, denotes a time variable;

[0114] S32, inputting the time-frequency feature data set into a Markov transition field to perform state transition analysis, and constructing a state transition probability matrix, wherein the mathematical expression of the state transition probability

[0115] denotes the number of transitions from state i to state j, denotes the number of transitions from state i to state k, denotes a smoothing factor, and k denotes the number of states;

[0116] S33, obtaining effective time period signal data by signal evolution rule of the state transition probability matrix and filtering characteristic time period.

[0117] Specifically, S4 includes the following steps:

[0118] S41, performing standardization processing on the effective time period signal data based on a functional principal component analysis strategy to obtain standardized effective time period signal data, and constructing a mean function according to the standardized effective time period signal data;

[0119] S42, performing centering processing on the standardized effective time period signal data according to the mean function to obtain centered effective time period signal data;

[0120] S43, performing base function expansion on the centered effective time period signal data according to a B-spline base function to obtain a base function matrix, and constructing a projection matrix according to the base function matrix;

[0121] S44, projecting the centered effective time period signal data according to the projection matrix to obtain projected signal data;

[0122] S45, performing covariance calculation on the projected signal data to obtain a covariance matrix, and processing the covariance matrix according to an L2 regularization method to obtain a regularized covariance matrix;​​​

[0123] S46. Perform Cholesky decomposition on the regularized covariance matrix to obtain intermediate results containing basis coefficients. Based on the intermediate results containing basis coefficients, perform damage feature screening on the projected signal data to obtain damage feature screening results.

[0124] In this embodiment, the mathematical basis of the functional principal component analysis strategy is the spectral decomposition of the covariance operator. The effective time period signal data is defined as a function. It is a square-integrable stochastic process, and the noise term... Independent and identically distributed, expressed using basis function expansion as follows: ,in, It is a mean function. For the first Principal component basis functions (orthogonal). For the first Signal data in the first effective time period The score on each principal component, K is the number of principal components.

[0125] Covariance operator Defined as: ,in, For mathematical expectation operators, For random processes At point random variables at that location, For random processes At point The mean at that point, For random processes At point random variables at that location, For random processes At point The mean at point is used for spectral decomposition using Mercer's theorem: ,in, For eigenvalues, Representing the covariance operator The k-th eigenfunction, the principal component score is calculated by numerical integration: ,in, For time intervals, For the i-th sample at time t, the score variance is the eigenvalue. .

[0126] In this embodiment, B-spline basis functions are used. These functions possess local support and smoothness, better capturing local abrupt changes caused by damage, and are suitable for processing non-periodic and smoothly varying functional data. The mathematical representation of the B-spline basis function is as follows: wherein, is a B-spline basis function, is a B-spline basis function The coefficients of the principal component scores matrix, combined with L2 regularization, can eliminate noise interference and enhance the robustness of the data.

[0127] The objective function F of L2 regularization is: wherein is the parameter set is optimized, is the original data matrix, is the basis function matrix, is the principal component score matrix, is the square of the Frobenius norm, is the coefficient of the regularization strength. Through Cholesky decomposition, the covariance matrix C is decomposed into the product of a lower triangular matrix L and its transpose, and combined with numerical integration to efficiently calculate the signal energy distribution and statistical characteristics, further improving the calculation efficiency, wherein the mathematical expression of the covariance matrix C is: .

[0128] Assuming that the cumulative variance explanation rate of the first two principal components reaches 95%, it means that they have captured most of the information in the data, and the remaining principal components are usually considered as noise or redundant information. In order to comprehensively consider the information of the two principal components, the Euclidean distance is used to calculate the comprehensive score, which reflects the comprehensive change intensity of the signal data of the effective time period in the direction of the two principal components , wherein, and are the scores of the signal data of the effective time period on the first principal component and the second principal component, respectively.

[0129] Based on the absolute value of the comprehensive score, 5% of the significant time points are selected as the data set for subsequent experiments, and the specific selection logic is as follows: Given a time series joint score denoted as wherein denotes the score of the time point, first calculate the 95th percentile (i.e. 95th percentile) of the joint score as the threshold , wherein, is the quantile function, then the index set of significant time points can be represented as: wherein, contains all time point indexes that satisfy ​This allows us to derive an index of significant time points, i.e., those time points where the joint score is greater than the 95th percentile. .

[0130] Specifically, S5 includes the following steps:

[0131] S51. Perform network initialization on the DSCN structure according to the progressive architecture construction method. The network initialization includes setting the DSCN structure as a single hidden layer and setting the hidden layer as a single node, and making all hidden layer nodes fully connected to the output layer of the DSCN structure.

[0132] S52. Construct training samples based on the damage feature screening results, train the DSCN structure based on the training samples, and dynamically adjust the node parameters through a supervised learning mechanism to obtain the optimized node parameters and the corresponding model performance indicators.

[0133] S53. Based on the optimized node parameters and corresponding model performance indicators, the number of hidden layers and nodes is gradually increased along the depth and width directions to obtain the extended DSCN structure.

[0134] S54. Perform output weight analysis on the extended DSCN structure, and calculate the connection weights between all hidden layer nodes and the output layer using the least squares method to establish an initial CFRP wing skin damage localization prediction model.

[0135] In this embodiment, the algorithm description of the DSCN structure can be summarized as follows: It has... The input features of each dimension are represented as follows: , and have The corresponding labels for each dimension are... This indicates. Here, This represents the number of training points. Assuming this is in the hidden layer of DSCN... It was generated in If there are nodes, then the current output can be defined as Therefore, the residual vector can be calculated using the equation. :

[0136] , among which, when hour, These are the input features of the training data, and when hour, , Indicates hidden layer The output.

[0137] From respectively and Initialize candidate weights and biases using a uniform distribution, and select those that meet the conditions. and maximizing the objective function input weights and biases as parameters of the nodes wherein:

[0138] ;

[0139] ; wherein, denotes the output of the node , , .

[0140] The output weight vector can be calculated based on the least squares method as the equation: ; wherein, , , denotes the Moore-Penrose inverse of H.

[0141] Specifically, S6 comprises the following steps:

[0142] S61, determining the input feature dimension and the hyperparameter set of the initial CFRP wing skin damage positioning prediction model according to the Bayesian algorithm;

[0143] S62, determining the test error of the predicted coordinates and the true coordinates of the test set samples based on the input feature dimension and the hyperparameter set, and constructing an objective optimization function with the goal of minimizing the test error;

[0144] S63, performing Bayesian probability modeling analysis on the objective optimization function through the prior distribution and the mixed kernel function, and constructing a Gaussian process model;

[0145] S64, constructing a time-weighted expected improvement acquisition function based on the prediction mean, standard deviation and current optimal observation value of the Gaussian process model, and selecting the hyperparameter set of the next evaluation point through the time-weighted expected improvement acquisition function;

[0146] S65, updating the Gaussian process model according to the hyperparameter set of the next evaluation point combined with the model prediction value and the actual test error until the preset convergence condition is met, and outputting the optimal hyperparameter combination;

[0147] S66, inputting the optimal hyperparameter combination into the initial CFRP wing skin damage positioning prediction model for hyperparameter optimization to obtain the final CFRP wing skin damage positioning prediction model.

[0148] In this embodiment, the initial CFRP wing skin damage positioning prediction model is assumed to be wherein: is the input feature, For the parameter combination to be optimized, its space is defined as: ;

[0149] The optimization objective is: ; where, is the optimal parameter combination, is the test error, where, The mathematical expression of is: ; where, is the predicted value of the DSCN for the th sample for the th coordinate, is the corresponding true coordinate value.

[0150] Gaussian processes model the objective function, providing predictions for any parameter point and their uncertainty estimates. Gaussian process characteristics: non-parametric model, suitable for any complex function form; the output is a probability distribution, containing mean and variance; capture the correlation between parameters through the covariance kernel.

[0151] The mathematical expression of the prior distribution is: ; where, is the objective function, is the Gaussian process, is the mean function, is the covariance function, is the observation noise variance, is the identity matrix.

[0152] The mixed kernel function is designed as: ; where, the continuous parameter kernel : ;

[0153] The integer parameter kernel : ;

[0154] The discrete parameter kernel : ; The time-weighted expected improvement acquisition function is based on the prediction results of the Gaussian process model, defines the strategy for selecting the next evaluation point, and balances exploration and development. The mathematical expression of the time-weighted expected improvement acquisition function is: where, , is the current optimal observation value, is the predicted mean and standard deviation; is the estimated training time cost.

[0155] Specifically, S7 comprises the following steps:

[0156] S71, multi-modal signal acquisition is performed on the to-be-tested CFRP wing skin to obtain a to-be-tested sample, and the to-be-tested sample is preprocessed to obtain a preprocessed to-be-tested sample;

[0157] S72, the preprocessed to-be-tested sample is input into the final CFRP wing skin damage positioning and prediction model for damage positioning and prediction, a spatial comparison graph containing real three-dimensional coordinates and predicted three-dimensional coordinates of the damage is obtained, and spatial constraint analysis is performed on the spatial comparison graph to obtain the corresponding CFRP wing skin damage positioning result.

[0158] Please refer to Figure 2 Fig. 1 is a structural schematic diagram of a CFRP wing skin damage positioning system based on multi-modal signal processing and Bayesian optimization DSCN according to an embodiment of the present application, which comprises:

[0159] The sensor receives signal data set acquisition module is used for establishing a wing skin undamaged model in the numerical simulation experiment of the CFRP wing skin, setting the CFRP wing skin damage position on a plurality of spatial coordinates of the wing skin undamaged model, and acquiring the corresponding Lamb wave time domain signal data according to the CFRP wing skin damage position, and obtaining the corresponding sensor receives signal data set according to the Lamb wave time domain signal data;

[0160] The sensor signal screening module is used for arranging entropy analysis and Higuchi fractal dimension analysis on the sensor receives signal data set according to the preset double-index sensor signal screening strategy, and obtaining the sensor receives signal data set deviating from the benchmark;

[0161] The time-frequency feature extraction module is used for extracting time-frequency features from the sensor receives signal data set deviating from the benchmark to obtain effective time period signal data;

[0162] The damage feature screening module is used for screening damage features from the effective time period signal data according to the functional principal component analysis strategy to obtain a damage feature screening result;

[0163] The model training module is used for training the DSCN structure according to the damage feature screening result to obtain an initial CFRP wing skin damage positioning and prediction model;

[0164] The model optimization module is used for optimizing the hyperparameters of the initial CFRP wing skin damage positioning and prediction model according to the Bayesian algorithm to obtain a final CFRP wing skin damage positioning and prediction model;

[0165] The damage positioning and prediction module is used for performing damage positioning and prediction on the to-be-tested sample according to the final CFRP wing skin damage positioning and prediction model to obtain the corresponding CFRP wing skin damage positioning result.

[0166] Experimental analysis is conducted according to the above content:

[0167] First, a finite element simulation software is used to establish a non-damaged model of the wing skin, the model size is long , wide , the skin thickness is , and the curvature radius is , , , respectively. The model is constructed by a single-layer modeling method, and the six-layer lamination structure is realized by assembly operation, and the thickness of each layer is uniform . The mesh division is set to 1mm.

[0168] 45 internal crack damages at different positions are set on the upper half of the wing skin surface. The crack damage center coordinates are shown in Table 1, and the crack damage size at each position is long 25mm, wide 4mm, and thick 3 layers (3rd, 4th, 5th layers).

[0169] Table 1 Center coordinates of different crack damage positions

[0170]

[0171] Through the finite element simulation software simulation experiment, the Lamb wave time domain signal data of 26 different receiving sensors of non-damage and 45 different damage positions can be obtained. For the Lamb wave time domain signal data obtained by the finite element simulation software simulation experiment, if all the channel time domain signal data is directly used as the model input, the model data analysis processing pressure will be increased, which will lead to that the model cannot effectively improve the learning ability and generalization performance. Therefore, effective preprocessing of the data before the Lamb wave data is input into the model can improve the learning efficiency of the model and reduce the fitting burden of the model. The signal data preprocessing of the present application includes: permutation entropy and Higuchi fractal dimension sensor sensitivity screening, continuous wavelet and Markov transition field time-frequency feature extraction, functional principal component analysis damage feature compression. The multi-modal signal processing experiment design is as follows.

[0172] Permutation entropy and Higuchi fractal dimension sensor sensitivity screening: the experimental data set includes Lamb time domain data of 1 non-damage and 45 damage positions, and each damage position corresponds to monitoring data of 26 receiving sensors. By setting the embedding dimension , Higuchi fractal dimension parameter and statistical significance threshold , the PE and HFD values of the non-damaged and damaged data sets are calculated. In order to screen the common sensitive sensors, the minimum sensitive case number is set to 35, that is, at least 35 damage cases in which the sensors show significant changes are identified as common sensitive sensors.

[0173] Time-frequency feature extraction using continuous wavelet and Markov transform fields: The experimental dataset includes monitoring data from sensors 6, 22, 24, 25, and 26 at 45 damage locations selected jointly by permutation entropy and Higuchi fractal dimension. Key parameters were set as follows: sampling frequency of 100 kHz, scale range of continuous wavelet transform analysis of [1, 128], state fraction bin of the Markov transform field of 6, feature scale of [32, 64], and smoothing factor of [missing value]. .

[0174] Damage feature compression using functional principal component analysis: The experimental data, after continuous wavelet and Markov transition fields, mainly shows the distribution of features within the time range of 3~5×10⁻⁶. -4 The number of sampling points ranges from 3000 to 5000. It involves 45 different damage locations, with data from 5 receiving sensors at each location. Therefore, the total dataset is 2000 rows × 225 columns, which means 2000 time points and 225 samples.

[0175] The damage localization experimental dataset consists of preprocessed multimodal signal data, comprising data from 45 different damage locations, with each location containing 5 samples, totaling 225 samples. Each sample sequence is 100 bytes long. The dataset was split by randomly selecting one damage location as the test set and the remaining 44 damage locations as the training set.

[0176] In this embodiment, the proposed application is defined as the BODSCN model. To demonstrate the damage localization performance of the proposed BODSCN model, ELM (Extreme Learning Machine), RVFLN (Random Vector Functional-Link Network), and DSCN (Dynamic Stochastic Configuration Network) are selected as comparison algorithms. ELM and RVFLN use one hidden layer, while DSCN uses a series of hidden layers. Select from the nodes of each layer. from Select from; DSCN and BODSCN Set to 20, To avoid overfitting in random neural networks, L2 regularization is used to calculate the output weights. , BODSCN, on the other hand, is in , , , Combinatorial optimization is then performed.

[0177] This application uses positioning error ( ), 3D root mean square error ( ), three-dimensional mean absolute error ( ) and three-dimensional coefficient of determination ( To evaluate the performance and localization capabilities of different models. Essentially, it is the spatial vector difference between the measured value and the true value. and The spatial prediction bias of the model is quantified by the sum of squared errors and the average of the absolute errors, respectively. More sensitive to larger errors This directly reflects the average distance deviation, providing a more intuitive representation of the model's overall accuracy. and The smaller the value, the stronger the model's localization ability. The ability of a model to explain the joint variation of the true coordinates is measured. The closer the value is to 1, the higher the spatial fit between the predicted value and the true value.

[0178] Table 2. Damage coordinate prediction results for different models

[0179]

[0180] Table 3 Comparison of Evaluation Indicators for Different Models

[0181]

[0182] As shown in Tables 2 and 3, the BODSCN model exhibits the best overall performance. Its positioning errors for all coordinate systems are significantly lower than other models, with RMSE=4.14 and MAE=3.78 being the lowest among the four methods. The accuracy of 0.9910 is close to the ideal value of 1, indicating that it has the best prediction accuracy and stability. In contrast, the error indices of other models are significantly higher. These data fully demonstrate the superior performance of the BODSCN model in spatial positioning tasks.

[0183] like Figure 3As shown, the real coordinates and the four methods predicted coordinate results clearly show the performance difference of each prediction model. The coordinate system covers the X axis (0-140mm), Y axis (0-10mm), and Z axis (260-350mm) range. The red circle in the figure marks the real coordinate point, the yellow triangle represents the ELM predicted coordinate, the blue square corresponds to the RVFLN predicted coordinate, the cyan diamond displays the DSCN predicted coordinate, and the green star is the BODSCN predicted coordinate. By observing the distribution characteristics in the three-dimensional space, it can be clearly observed that the blue square RVFLN prediction point is the most dispersed, and multiple samples have a significant deviation of more than 10mm, especially in the Z axis direction; the yellow triangle ELM prediction point is improved compared with RVFLN, but there is still a large deviation; the cyan diamond DSCN prediction point has significantly improved prediction stability, but there are still 3-5mm errors in some areas; in comparison, all green star BODSCN prediction points are closely clustered around the red real coordinate point, maintaining the smallest deviation in the XYZ three dimensions, with no significant abnormal points, verifying the positioning accuracy advantage of BODSCN.

[0184] As shown in Figure 4 , the error distribution of each model on different test samples is shown. The abscissa axis represents the 1-16 coordinate point sequence, and the ordinate axis is the positioning error value. From the overall trend, the positioning error of the green star marked BODSCN model on all test points is significantly lower than that of the other three models, showing better positioning accuracy, and the error fluctuation of the other models is obvious, and the BODSCN model is more stable.

[0185] The above is only an embodiment of the present application, and the common knowledge of specific structures and characteristics in the scheme is not described in detail here. The person skilled in the art knows all the ordinary technical knowledge in the field of the application before the application date or the priority date, can know all the prior art in this field, and has the ability to apply conventional experimental means before that date. The person skilled in the art can improve and implement the present scheme based on their own ability under the guidance of the present application, and some typical known structures or known methods should not be an obstacle for the person skilled in the art to implement the present application. It should be noted that for those skilled in the art, without departing from the structure of the present application, a number of modifications and improvements can be made, which should be considered as the protection scope of the present application, and these will not affect the effect and practicality of the patent. The protection scope claimed in the present application should be subject to the content of its claims, and the specific implementation mode and the like in the specification can be used to explain the content of the claims.

Claims

1. A method for CFRP wing skin damage localization based on multimodal signal processing and Bayesian optimized DSCN, characterized in that: Includes the following steps: S1. In the numerical simulation experiment of CFRP wing skin, a damage-free model of wing skin is established, and the damage location of CFRP wing skin is set on multiple spatial coordinates of the damage-free model of wing skin. The corresponding Lamb wave time domain signal data is obtained according to the damage location of CFRP wing skin, and the corresponding sensor received signal dataset is obtained according to the Lamb wave time domain signal data. S2. Based on the preset dual-index sensor signal filtering strategy, perform permutation entropy analysis and Higuchi fractal dimension analysis on the sensor received signal dataset to obtain the sensor received signal dataset that deviates from the benchmark. S3. Extract time-frequency features from the sensor received signal dataset that deviates from the reference to obtain signal data for the effective time period; S4. Based on the functional principal component analysis strategy, damage features are screened for signal data within the effective time period to obtain the damage feature screening results. S5. Based on the damage feature screening results, the DSCN structure is trained to obtain the initial CFRP wing skin damage localization prediction model. S6. Based on the Bayesian algorithm, the hyperparameters of the initial CFRP wing skin damage localization prediction model are optimized to obtain the final CFRP wing skin damage localization prediction model. S7. Based on the final CFRP wing skin damage localization prediction model, perform damage localization prediction on the test sample to obtain the corresponding CFRP wing skin damage localization results. S2 includes the following steps: S21. Based on the permutation entropy analysis algorithm in the preset dual-index sensor signal filtering strategy, perform permutation entropy analysis on the undamaged state dataset and the damaged state dataset in the sensor received signal dataset, respectively, to obtain the permutation entropy values ​​of each sensor in the undamaged state and the damaged state of the CFRP wing skin. The mathematical expression for the permutation entropy value is: , This represents the probability of the i-th permutation pattern occurring. Indicates the embedding dimension. Representing the embedding dimension factorial; S22. Based on the Higuchi fractal dimension algorithm in the preset dual-index sensor signal filtering strategy, Higuchi fractal dimension analysis is performed on the undamaged state dataset and the damaged state dataset respectively to obtain the Higuchi fractal dimension values ​​of each sensor in the undamaged state and the damaged state of the CFRP wing skin. S23. Integrate the permutation entropy value and the Higuchi fractal dimension value to obtain a dual-index feature matrix; S24. Based on the undamaged state dataset, set the permutation entropy change benchmark and Higuchi fractal dimension change benchmark for each sensor, and determine the corresponding permutation entropy change rate and Higuchi fractal dimension change rate based on the permutation entropy value and Higuchi fractal dimension value of each sensor under the CFRP wing skin damage state. S25. Perform joint analysis on the dual-index feature matrix based on the permutation entropy change benchmark and the Higuchi fractal dimension change benchmark to obtain a list of candidate sensitive sensors that meet the preset sensitive sensor conditions. The preset sensitive sensor conditions include the permutation entropy change rate exceeding the permutation entropy change benchmark, or the Higuchi fractal dimension change rate exceeding the Higuchi fractal dimension change benchmark. S26. Add the sensor received signal corresponding to each candidate sensitive sensor in the candidate sensitive sensor list to the sensor received signal dataset that deviates from the baseline.

2. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN as described in claim 1, characterized in that: S1 includes the following steps: S11. Establish a non-damaged model of the wing skin based on finite element simulation software, and determine multiple internal crack damage locations in the non-damaged model of the wing skin based on preset crack damage rules, forming a CFRP wing skin simulation structure that includes a non-damaged state and different damage states. The preset crack damage rules include preset internal crack damage locations in the high-stress area of ​​the CFRP wing skin, and set according to the typical damage mode of the CFRP wing skin. S12. Simulate the excitation and propagation process of Lamb waves in the CFRP wing skin simulation structure to obtain the original time-domain signal dataset; S13. Locate each preset sensor receiving point from the original time-domain signal dataset, and analyze the fluctuation response characteristics of each sensor receiving point in the time-domain dimension to obtain the sensor receiving signal dataset.

3. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN as described in claim 1, characterized in that: S3 includes the following steps: S31. Time-frequency localization analysis is performed on the sensor received signal dataset deviating from the reference using continuous wavelet transform to obtain a time-frequency feature dataset, wherein the time-frequency plane energy distribution during the time-frequency localization analysis is... The mathematical expression is: ,in, Indicates the scale parameter. Indicates the translation parameter. Describe the wavelet basis functions. Represents a time-domain signal. Represents a time variable; S32. Input the time-frequency feature dataset into the Markov transformation field for state transition analysis, and construct the state transition probability matrix, where the state transition probabilities are... The mathematical expression is: , This represents the number of transitions from state i to state j. This represents the number of transitions from state i to state k. represents the smoothing factor, and k represents the number of states; S33. By analyzing the signal evolution patterns of the state transition probability matrix and filtering characteristic time periods, effective time period signal data is obtained.

4. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN as described in claim 1, characterized in that: S4 includes the following steps: S41. Based on the functional principal component analysis strategy, the effective time period signal data is standardized to obtain the standardized effective time period signal data, and the mean function is constructed based on the standardized effective time period signal data. S42. The standardized effective time period signal data is centered according to the mean function to obtain the centered effective time period signal data. S43. Expand the effective time period signal data after centering processing using B-spline basis functions to obtain the basis function matrix, and construct the projection matrix based on the basis function matrix. S44. Project the signal data of the effective time period after the centering process according to the projection matrix to obtain the projected signal data. S45. Calculate the covariance of the projected signal data to obtain the covariance matrix, and process the covariance matrix according to the L2 regularization method to obtain the regularized covariance matrix. S46. Perform Cholesky decomposition on the regularized covariance matrix to obtain intermediate results containing basis coefficients. Based on the intermediate results containing basis coefficients, perform damage feature screening on the projected signal data to obtain damage feature screening results.

5. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN according to claim 1, characterized in that: S5 includes the following steps: S51. Perform network initialization on the DSCN structure according to the progressive architecture construction method. The network initialization includes setting the DSCN structure as a single hidden layer and setting the hidden layer as a single node, and making all hidden layer nodes fully connected to the output layer of the DSCN structure. S52. Construct training samples based on the damage feature screening results, train the DSCN structure based on the training samples, and dynamically adjust the node parameters through a supervised learning mechanism to obtain the optimized node parameters and the corresponding model performance indicators. S53. Based on the optimized node parameters and corresponding model performance indicators, the number of hidden layers and nodes is gradually increased along the depth and width directions to obtain the extended DSCN structure. S54. Perform output weight analysis on the extended DSCN structure, and calculate the connection weights between all hidden layer nodes and the output layer using the least squares method to establish an initial CFRP wing skin damage localization prediction model.

6. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN according to claim 1, characterized in that: S6 includes the following steps: S61. Determine the input feature dimension and hyperparameter set of the initial CFRP wing skin damage localization prediction model based on the Bayesian algorithm; S62. Based on the input feature dimension and hyperparameter set, determine the test error between the predicted coordinates and the true coordinates of the test set samples, and construct an objective optimization function with the goal of minimizing the test error. S63. By performing Bayesian probabilistic modeling analysis on the objective optimization function through prior distribution and mixture kernel function, a Gaussian process model is constructed. S64. Based on the predicted mean, standard deviation and current best observation value of the Gaussian process model, construct the time-weighted expectation improved acquisition function, and select the hyperparameter set of the next evaluation point through the time-weighted expectation improved acquisition function; S65. Update the Gaussian process model based on the hyperparameter set of the next evaluation point, combined with the model prediction and actual test error, until the preset convergence condition is met, and output the optimal hyperparameter combination. S66. Input the optimal hyperparameter combination into the initial CFRP wing skin damage localization prediction model for hyperparameter optimization to obtain the final CFRP wing skin damage localization prediction model.

7. The CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN according to claim 1, characterized in that: S7 includes the following steps: S71. Multimodal signal acquisition is performed on the CFRP wing skin under test to obtain the test sample, and the test sample is preprocessed to obtain the preprocessed test sample. S72. Input the preprocessed test sample into the final CFRP wing skin damage localization prediction model for damage localization prediction, obtain a spatial comparison map containing the actual three-dimensional coordinates of the damage and the predicted three-dimensional coordinates, and perform spatial constraint analysis on the spatial comparison map to obtain the corresponding CFRP wing skin damage localization result.

8. A CFRP wing skin damage localization system based on multimodal signal processing and Bayesian optimized DSCN, used to implement the CFRP wing skin damage localization method based on multimodal signal processing and Bayesian optimized DSCN as described in any one of claims 1-7, characterized in that: include: The sensor received signal dataset acquisition module is used to establish a non-damaged model of CFRP wing skin in the numerical simulation experiment of CFRP wing skin, set the damage location of CFRP wing skin on multiple spatial coordinates of the non-damaged wing skin model, acquire the corresponding Lamb wave time domain signal data according to the damage location of CFRP wing skin, and obtain the corresponding sensor received signal dataset according to the Lamb wave time domain signal data. The sensor signal filtering module is used to perform permutation entropy analysis and Higuchi fractal dimension analysis on the sensor received signal dataset according to the preset dual-index sensor signal filtering strategy, so as to obtain the sensor received signal dataset that deviates from the benchmark. The time-frequency feature extraction module is used to extract time-frequency features from the sensor received signal dataset that deviates from the reference, so as to obtain signal data for the effective time period. The damage feature screening module is used to screen damage features of signal data within an effective time period based on a functional principal component analysis strategy, and to obtain the damage feature screening results. The model training module is used to train the DSCN structure based on the damage feature screening results to obtain the initial CFRP wing skin damage localization prediction model. The model optimization module is used to optimize the hyperparameters of the initial CFRP wing skin damage localization prediction model according to the Bayesian algorithm, so as to obtain the final CFRP wing skin damage localization prediction model. The damage localization prediction module is used to predict the damage localization of the test sample based on the final CFRP wing skin damage localization prediction model, and obtain the corresponding CFRP wing skin damage localization results.

Images