Time-varying condition guided wave structure damage monitoring method

Abstract

The invention discloses a time-varying condition guided wave structure damage monitoring method, which comprises the following steps: arranging a piezoelectric patch damage monitoring path on a structural member, and respectively collecting response signals of different damage degrees at different temperatures; classifying according to different damage degrees, and extracting a plurality of characteristic parameters of the response signals one by one; the intra-class distance and the inter-class distance between the feature vectors are calculated through variance calculation and mahalanobis distance, when the intra-class distance is smaller than the inter-class distance, the separability of the vectors is good, and therefore appropriate feature parameters are screened; selecting a Gaussian kernel function to carry out kernel processing on the feature matrix, constructing a feature filtering model by using the first three feature vectors with the maximum feature values of the health kernel matrix, and then projecting the health kernel matrix and the damage kernel matrix onto the feature filtering model at the same time to obtain features with temperature factors filtered out; and finally, inputting the filtered sample into a CNN classification model for training, wherein the trained model can complete the recognition of the damage degree of the structural member. The feature filtering model and the damage degree recognition model are constructed by means of a machine learning algorithm, and structural damage monitoring under the time-varying condition is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of material damage information analysis, and in particular to a method for monitoring damage of a waveguide structure under time-varying conditions. Background Art

[0002] In recent years, as structural health monitoring has been continuously expanded to multiple fields, the elimination of the influence and role of environmental factors has gradually become a new research hotspot for this technology. Structures such as wind turbine blades are usually exposed outdoors, resulting in the monitoring response signal being affected by the coupling of environmental factors such as temperature and damage. This limits structural health monitoring with Lamb waves as the main monitoring method, affecting the judgment of structural safety. Therefore, under the condition of temperature environment changes, in order to ensure the stability and accuracy of the structural health monitoring system, the study of the temperature effect of Lamb wave structural health monitoring is particularly important. The changes in the mechanical properties of structural materials caused by temperature changes lead to changes in the Lamb wave dispersion equation, thereby changing the signal propagation speed and amplitude. This will interfere with or even submerge the actual defect information in the diagnostic signal, and ultimately reduce the reliability of damage monitoring. Therefore, the impact of Lamb waves caused by environmental changes needs to be effectively compensated.

[0003] There have been many studies and methods for temperature compensation of Lamb waves. The most widely used of these methods is signal normalization. The basic concept of signal normalization is to take into account the impact of temperature changes and structural defects on the different trends of Lamb wave signals. By using signal normalization processing methods such as cointegration, independent component analysis (ICA), principal component analysis (PCA) or singular value decomposition (SVD), features that are only sensitive to structural defects can be extracted, thereby separating the benign features of ambient temperature changes. The selection of features in the above methods is directly related to the accuracy of the results. Therefore, how to select effective features in damage monitoring has become a topic that needs to be solved urgently. Summary of the invention

[0004] Purpose of the invention: The purpose of the present invention is to address the deficiencies in the prior art and provide a method for constructing a damage identification model for a waveguide structure under time-varying conditions. The damage to structural parts is monitored and analyzed based on an optimal feature set and a convolutional neural network classification model. The method is capable of identifying the damage state of composite materials of waveguide structures under temperature changes and improving the accuracy of damage monitoring of composite materials under temperature changes.

[0005] Another object of the present invention is to provide a method for monitoring damage of guided wave structures under time-varying conditions.

[0006] Technical solution: In order to solve the above technical problems, the present invention provides a method for constructing a damage identification model of a time-varying guided wave structure, comprising the following steps:

[0007] A monitoring area is divided on the structure to be tested, and a monitoring path is constructed in the area. Then, under temperature change conditions, signal tests of different damage degrees are performed according to the monitoring path to obtain a response signal.

[0008] Extract multiple characteristic parameters of all response signals to form a characteristic vector set, and select the features least affected by temperature, then perform normalization and kernel processing on the selected features to obtain the optimal characteristic vector set;

[0009] The best feature vector set is used as the input of the pre-trained CNN classification model, and the classification results are output. The classification results include four categories: healthy signals, slightly damaged signals, moderately damaged signals, and severely damaged signals. The time-varying conditional guided wave structure damage recognition model is trained.

[0010] A further preferred technical solution of the present invention is that a monitoring area is divided on the structure to be tested, and a monitoring path is constructed in the area, and then under temperature change conditions, signal tests of different damage degrees are performed according to the monitoring path to obtain a response signal, specifically including:

[0011] A circular monitoring area is preset on the waveguide structure to be monitored, and the circular monitoring area is evenly divided into a number of equal sector-shaped areas. A piezoelectric sheet is arranged on each equally divided line, and each piezoelectric sheet is electrically connected to another piezoelectric sheet separated by one piezoelectric sheet, and two piezoelectric sheets form a monitoring path;

[0012] Conduct signal tests with different damage degrees under the same temperature interval and time interval conditions, obtain response signals of multiple paths, and obtain the structure temperature through the temperature sensor:

[0013] T = T0 + n × ΔT, n∈[0,N];

[0014] Where T0 is the initial temperature value, the fixed temperature interval ΔT ≥ 20°C, the fixed time interval Δt ≥ 30min, and N is a positive integer;

[0015] When one monitoring path is completed, the monitoring path is replaced and the monitoring process is cyclically performed to collect the response signals of the monitoring path at each set temperature again.

[0016] Preferably, the response signal is collected by:

[0017] One of the piezoelectric sheets of any monitoring path is used as the excitation end, and the other piezoelectric sheet is used as the receiving end. The piezoelectric sheet at the excitation end of each path is controlled one by one to transmit the Lamb wave test signal, and the piezoelectric sheet at the receiving end of the monitoring path receives the corresponding response signal to obtain the response signal.

[0018] Preferably, each group of response signals includes four status signals of one path, namely a healthy signal, a slightly damaged signal, a moderately damaged signal and a severely damaged signal;

[0019] The response signal obtained is M paths, N levels of set temperature, and 4 state signals collected P times for each level of set temperature, obtaining M*N*P*4 groups of combined signals.

[0020] Preferably, the extracting of multiple characteristic parameters of all response signals forms a characteristic vector set, and screening out the features least affected by temperature, and then performing normalization and kernel processing on the screened features to obtain the best characteristic vector set; specifically:

[0021] Four types of state signals are classified, and each set of characteristic parameters is extracted one by one to obtain the healthy signal feature vector set, the slight injury feature vector set, the moderate injury feature vector set and the severe injury feature vector set respectively;

[0022] The variance of the health state feature vectors obtained from the initial health state at different temperatures is calculated, and the feature quantity with the smallest variance is found as the benchmark feature;

[0023] The Mahalanobis distance is calculated between the baseline feature and the features of different damage degrees, and the intra-class distance and inter-class distance are calculated respectively. When the inter-class distance is greater than the intra-class distance, the feature vector has good separability, and the features with good Mahalanobis distance separability are normalized.

[0024] Select the Gaussian kernel function, perform kernel processing on the initial state feature matrix, calculate the eigenvalues ​​and eigenvectors after obtaining the healthy kernel matrix, and select the first three eigenvectors with the largest eigenvalues ​​to construct the temperature space model K. At this time, K is the filtering model K* that can filter out the temperature influencing factors; then perform the same kernel processing on the damage state feature matrix, and simultaneously map the healthy kernel matrix and the damage kernel matrix to the temperature space K composed of the selected principal components, and obtain the optimal feature set through the filtering model K*.

[0025] A time-varying condition guided wave structure damage monitoring method according to the present invention comprises the following steps:

[0026] Acquire a response signal of the waveguide structure to be monitored;

[0027] According to the response signal of the waveguide structure to be monitored, the waveguide structure damage monitoring is performed through the time-varying condition waveguide structure damage identification model constructed above, and the damage monitoring result of the waveguide structure under the time-varying condition is obtained.

[0028] Beneficial effects: The present invention monitors and analyzes damaged structures under time-varying conditions based on feature filtering of the optimal feature set, which makes up for the shortcomings and has stronger data processing capabilities for unstable feature changes. The present invention monitors and analyzes the optimal feature set response signal based on a trained convolutional neural network classification model, and can identify the response signals of healthy state, slight damage state, moderate damage state, and severe damage state, eliminating damage identification errors and improving damage identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Shown is a schematic diagram of a method for constructing a damage identification model of a guided wave structure under time-varying conditions according to the present invention;

[0030] Figure 2 The figure shows a schematic diagram of the health status of the waveguide structure collected in Example 1;

[0031] Figure 3 The figure shows a schematic diagram of a slightly damaged state of the waveguide structure collected in Example 1;

[0032] Figure 4 The figure shows a schematic diagram of a moderately damaged state of a waveguide structure collected in Example 1;

[0033] Figure 5 The figure shows a schematic diagram of a severely damaged state of the waveguide structure collected in Example 1;

[0034] Figure 6 The figure shows a comparison diagram of four states of the collected signal of Example 1. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0036] Combine the following Figure 1-Figure 6 The present invention describes the time-varying condition guided wave structure damage monitoring method.

[0037] Embodiment 1: This embodiment provides a method for constructing a time-varying conditional guided wave structure damage identification model, comprising the following steps:

[0038] S100, dividing a monitoring area on the structure to be tested, and constructing a monitoring path in the area, and then performing signal tests of different damage degrees according to the monitoring path under temperature change conditions to obtain a response signal;

[0039] S200, extracting multiple characteristic parameters of all response signals to form a characteristic vector set, and screening out the features least affected by temperature, performing normalization and kernel processing on the screened features to obtain an optimal characteristic vector set;

[0040] S300, using the best feature vector set as the input of the pre-trained CNN classification model, outputting the classification results, which include four categories: healthy signal, slightly damaged signal, moderately damaged signal and severely damaged signal, and training to obtain a time-varying conditional guided wave structure damage recognition model.

[0041] To further explain the solution of this embodiment in detail, a monitoring area is divided on the structure to be tested, and a monitoring path is constructed in the area. Then, under the condition of temperature change, signal tests of different damage degrees are performed according to the monitoring path to obtain a response signal. The specific method is:

[0042] S110. To ensure that the monitoring area is evenly covered, average performance data for the entire area is obtained to avoid possible deviations due to local measurements. A circular monitoring area is preset on the structure to be tested, and the circular monitoring area is divided into five equal sectors. At the same time, in order to improve the redundancy of the system, even if some sensors fail, the system can continue to operate and provide valid data. Data from multiple independent paths can also help cross-validate the results and improve the reliability of monitoring. A piezoelectric sheet is arranged on each equally divided line, such as Figure 2 As shown, they are: PZT A, PZT B, PZT C, PZT D, PZT E. Each piezoelectric sheet is electrically connected to another piezoelectric sheet separated by one piezoelectric sheet, and two piezoelectric sheets form a monitoring path, forming five monitoring paths: AC, AD, BD, BE, and CE. The piezoelectric sheet at the starting end of each path serves as an exciter, and the piezoelectric sheet at the other end serves as a receiver. The direction of the arrow represents the propagation direction of the test signal.

[0043] S120. Obtain the temperature information of the structure to be tested through the temperature sensor, and set 4 set temperatures, namely -10℃, 10℃, 30℃ and 50℃. Gradually increase the temperature from -10℃, and collect a group of signals every 30 minutes after heating to stability until the temperature reaches 50℃. The temperature experiment of each path is cycled from -10℃ to 50℃. Each group of response signals includes 4 state signals of a path, namely healthy signal, slightly damaged signal, moderately damaged signal and severely damaged signal. Figure 2 The figure shows the initial health state of the structural parts, and the five paths are collected multiple times in the healthy state. Figure 3-Figure 5 , three test states of slight damage, moderate damage and severe damage are arranged on the tested structural parts in turn. A path collects signals of four states in total, and the signals of the same state are grouped together. For path AC, Figure 6 As shown, four signals were collected, which were recorded as AC1 (initial healthy state), AC2 (slightly damaged state), AC3 (moderately damaged state), and AC4 (severely damaged state).

[0044] The same operation is performed on the remaining four paths AD, BD, BE, and CE in turn, which are recorded as:

[0045] AD1, AD2, AD3, AD4; BD1, BD2, BD3, BD4; BE1, BE2, BE3, BE4;

[0046] CE1, CE2, CE3, CE4.

[0047] The response signal acquisition method is:

[0048] One of the piezoelectric sheets of any monitoring path is used as the excitation end, and the other piezoelectric sheet is used as the receiving end. The piezoelectric sheets of the excitation end of each path are controlled one by one to transmit the Lamb wave test signal, and the piezoelectric sheets of the receiving end of the monitoring path receive the corresponding response signal to obtain the response signal. The Lamb wave test excitation signal used in this embodiment is an S0 mode Lamb wave test signal with a center frequency of 60KHz, wherein the excitation signal adopts a 5-peak Hanning window sine modulation signal.

[0049] There are 5 paths, 4 different temperatures, and 4 state signals. Each group of signals is collected 10 times, and 5*4*4*10, a total of 800 groups of combined signals, are obtained.

[0050] Four state signals are classified, and each group of characteristic parameters is extracted one by one to obtain the healthy signal feature vector set, i.e., the signal feature vectors of paths AC1, AD1, BD1, BE1, and CE1; the slightly injured feature vector set, i.e., the signal feature vectors of paths AC2, AD2, BD2, BE2, and CE2; the moderately injured feature vector set, i.e., the signal feature vectors of paths AC3, AD3, BD3, BE3, and CE3; and the severely injured feature vector set, i.e., the signal feature vectors of paths AC4, AD4, BD4, BE4, and CE4.

[0051] Each feature vector set includes 13 feature selection parameters, and the 13 feature selection parameters include time domain feature selection parameters and frequency domain feature selection parameters;

[0052] The time domain characteristic parameters include root mean square, variance, waveform factor, kurtosis coefficient, K factor, impulse factor, margin factor, peak factor and skew;

[0053] The frequency domain characteristic parameters include frequency mean, frequency variance, mean square frequency and centroid frequency.

[0054] In order to find a feature quantity that changes slightly with temperature among many features, in step S200 of this embodiment, multiple feature parameters of all response signals are extracted to form a feature vector set, and the feature least affected by temperature is screened out, and then the screened features are normalized and kernelized to obtain the best feature vector set.

[0055] The specific method is:

[0056] S210, calculating the variance of the initial health states at the four temperatures, and finding the feature with the smallest variance as the benchmark feature. When the variance is the smallest, it means that the feature is weakly affected by the temperature.

[0057] S220, perform Mahalanobis distance calculation on the features after variance calculation and the features of different damage degrees, calculate the intra-class distance and the inter-class distance respectively, and define the intra-class distance and the inter-class distance. When the inter-class distance is greater than the intra-class distance, the feature vector has good separability and is regarded as a feature that changes slightly with temperature. The specific calculation method is:

[0058] (1) Calculate the intra-class distance:

[0059]

[0060] (2) Calculate the distance between classes

[0061]

[0062] is the benchmark feature obtained after variance calculation. Where u2 is the mean of the sample distribution, and S2 is the covariance matrix of the sample distribution.

[0063] S230, construct a feature filtering model and establish an optimal feature set:

[0064] Principal component analysis (PCA) is a commonly used data dimensionality reduction technique used to simplify data sets while retaining as much information as possible. It reduces the complexity of data by converting the original high-dimensional data into principal components of lower dimensions. Another feature of the PCA algorithm is that it retains the main features, removes noise from the data, and improves data quality. However, considering that PCA is not effective for nonlinear data and the principal components have poor interpretability, the kernel function is introduced to enhance the processing of nonlinear data. The KPCA method is to reduce the dimension of the feature space, use the centralized kernel matrix to perform eigenvalue decomposition, and then obtain a set of eigenvectors and corresponding eigenvalues. The size of the eigenvalue at this time reflects the proportion of each eigenvector in the data change. The first three eigenvectors with the largest eigenvalues ​​are selected, which constitute the basis of the environmental feature space, that is, the main filtering feature direction of this method, and all samples are mapped to the low-dimensional feature space composed of the selected principal components, which intuitively reflects the discrimination between the healthy and damaged states after filtering out the temperature factor.

[0065] a. Normalize the feature vector and select the kernel function

[0066] In order to avoid bias in the analysis results caused by features with a large numerical range, the feature matrix is ​​normalized and all features are scaled to the interval [0,1] using the maximum-minimum normalization method.

[0067] The kernel method is used to reduce the dimension of the normalized features. The kernel method captures nonlinear relationships in a high-dimensional feature space without directly calculating in the high-dimensional space. The Gaussian kernel function is selected as the basis of the kernel method. It is a commonly used radial basis function that can effectively measure the similarity between samples.

[0068] b. Construct KPCA feature filtering model

[0069] For the health features that are only affected by temperature, namely the AC1, AD1, BD1, BE1, and CE1 path response signal features, the feature matrices from AC, AD, BD, BE, and CE are kernel processed in order, and the eigenvalues ​​and eigenvectors of the kernel matrices are calculated. h Represents a kernel matrix containing only the initial healthy features, and performs KPCA operation:

[0070]

[0071]

[0072] Where V h Contains the eigenvector, ^ h Contains eigenvalues.

[0073] Select the first k eigenvectors (corresponding to the largest eigenvalue). Sort the calculated eigenvalues ​​by size, and the largest eigenvalue corresponds to the direction with the largest data change. Since the input features at this time come from the initial health states of the five paths at four temperatures of -10℃, 10℃, 30℃ and 50℃ (AC1, AD1, BD1, BE1, CE1). Select the first three largest eigenvalues ​​and their corresponding eigenvectors, and use these three eigenvectors to form a new coordinate system, that is, the principal component space. This is the temperature filtering model.

[0074] Construct the temperature filter part:

[0075]

[0076] Verifying the feature filter model

[0077] The same processing method is used for the damage features affected by temperature, namely AC2, AD2, BD2, BE2, CE2; AC3, AD3, BD3, BE3, CE3; AC4, AD4, BD4, BE4, CE4. The feature matrices from AC, AD, BD, BE, and CE are processed in sequence. At this time, X d is the kernel matrix of the temperature-dependent damage features:

[0078]

[0079]

[0080] Where V damage Contains the eigenvector, ^ damage Contains eigenvalues.

[0081] The feature samples of four states AC1, AD1, BD1, BE1, CE1; AC2, AD2, BD2, BE2, CE2; AC3, AD3, BD3, BE3, CE3; AC4, AD4, BD4, BE4, CE4 at four temperatures of -10℃, 10℃, 30℃ and 50℃ are mapped to the principal component space K composed of the selected principal components in sequence (AC, AD, BD, BE, CE). The filter model K* constructed in the health feature is used to process the damage feature affected by temperature.

[0082] This step projects the temperature-affected part of the damage feature into the principal component space, i.e., the filter model part:

[0083]

[0084] The whole step is to input the initial health characteristics into the KPCA model, extract the eigenvalues ​​and eigenvectors to construct the temperature feature space, that is, the feature filter model, and then input the four damage state characteristics to project them into the temperature space by calculating the eigenvalues ​​and eigenvectors, and remove the temperature factors in the damage characteristics affected by temperature, so as to obtain the part containing only the initial health factors and damage factors, that is, the optimal feature set.

[0085] The best feature set is used as the input of the pre-trained convolutional neural network classification model to train the convolutional neural network classification model, and a time-varying conditional guided wave structure damage recognition model is obtained.

[0086] During model training, 80% of the feature vectors in the best feature set are used as training sample sets, and 20% are divided into test sample sets. A convolutional neural network classification model is built, and the training samples are input into the model for model training. The model after training is the optimal recognition model for the current structural component. The test samples are input into the trained model to test the classification accuracy, and finally the highest recognition rate under the model is obtained. This method can greatly improve the recognition accuracy of damage under time-varying conditions while being able to monitor the health of the structure. After testing the test set, the damage degree classification recognition rate was 91.4%.

[0087] Embodiment 2: This embodiment provides a method for monitoring damage of a guided wave structure under time-varying conditions, comprising the following steps:

[0088] Acquire a response signal of the waveguide structure to be monitored;

[0089] According to the response signal of the waveguide structure to be monitored, the waveguide structure damage monitoring is performed through the time-varying condition waveguide structure damage identification model constructed above, and the damage monitoring result of the waveguide structure under the time-varying condition is obtained.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a time-varying conditional guided wave structure damage identification model, characterized in that: The steps include: A monitoring area is divided on the structure to be tested, and a monitoring path is constructed in the area. Then, under temperature change conditions, signal tests of different damage degrees are performed according to the monitoring path to obtain a response signal. Extract multiple characteristic parameters of all response signals to form a characteristic vector set, and select the features least affected by temperature, then perform normalization and kernel processing on the selected features to obtain the optimal characteristic vector set; The best feature vector set is used as the input of the pre-trained CNN classification model, and the classification results are output. The classification results include four categories: healthy signals, slightly damaged signals, moderately damaged signals, and severely damaged signals. The time-varying conditional guided wave structure damage recognition model is trained.

2. The method for constructing a damage identification model for a time-varying waveguide structure according to claim 1, characterized in that: The method of dividing a monitoring area on the structure to be tested, constructing a monitoring path in the area, and then performing signal tests of different damage degrees according to the monitoring path under temperature change conditions to obtain a response signal specifically includes: A circular monitoring area is preset on the waveguide structure to be monitored, and the circular monitoring area is evenly divided into a number of equal sector-shaped areas. A piezoelectric sheet is arranged on each equally divided line, and each piezoelectric sheet is electrically connected to another piezoelectric sheet separated by one piezoelectric sheet, and two piezoelectric sheets form a monitoring path; Conduct signal tests with different damage degrees under the same temperature interval and time interval conditions, obtain response signals of multiple paths, and obtain the structure temperature through the temperature sensor: T = T0 + n × ΔT, n∈[0,N]; Where T0 is the initial temperature value, the fixed temperature interval ΔT ≥ 20°C, the fixed time interval Δt ≥ 30min, and N is a positive integer; When the acquisition of one monitoring path is completed, the monitoring path is replaced and the monitoring process is cyclically performed to collect the response signals of the monitoring path at each set temperature again.

3. The method for constructing a time-varying conditional waveguide structure damage identification model according to claim 2, characterized in that: The response signal collection method is: One of the piezoelectric sheets of any monitoring path is used as the excitation end, and the other piezoelectric sheet is used as the receiving end. The piezoelectric sheet at the excitation end of each path is controlled one by one to transmit the Lamb wave test signal, and the piezoelectric sheet at the receiving end of the monitoring path receives the corresponding response signal to obtain the response signal.

4. The method for constructing a damage identification model for a time-varying waveguide structure according to claim 2, characterized in that: Each group of response signals includes four status signals of a path, namely, a healthy signal, a slightly damaged signal, a moderately damaged signal, and a severely damaged signal; The response signal obtained is M paths, N levels of set temperature, and 4 state signals collected P times for each level of set temperature, obtaining M*N*P*4 groups of combined signals.

5. The method for constructing a damage identification model for a time-varying waveguide structure according to claim 1, characterized in that: The method extracts multiple characteristic parameters of all response signals to form a characteristic vector set, and selects the features least affected by temperature, and then performs normalization and kernel processing on the selected features to obtain the optimal characteristic vector set; specifically: Four types of state signals are classified, and each set of characteristic parameters is extracted one by one to obtain the healthy signal feature vector set, the slight injury feature vector set, the moderate injury feature vector set and the severe injury feature vector set respectively; The variance of the health state feature vectors obtained from the initial health state at different temperatures is calculated, and the feature quantity with the smallest variance is found as the benchmark feature; The Mahalanobis distance is calculated between the baseline feature and the features of different damage degrees, and the intra-class distance and inter-class distance are calculated respectively. When the inter-class distance is greater than the intra-class distance, the feature vector has good separability, and the features with good Mahalanobis distance separability are normalized. Select the Gaussian kernel function, perform kernel processing on the initial state feature matrix, calculate the eigenvalues ​​and eigenvectors after obtaining the healthy kernel matrix, and select the first three eigenvectors with the largest eigenvalues ​​to construct the temperature space model K. At this time, K is the filtering model K* that can filter out the temperature influencing factors; then perform the same kernel processing on the damage state feature matrix, and simultaneously map the healthy kernel matrix and the damage kernel matrix to the temperature space K composed of the selected principal components, and obtain the optimal feature set through the filtering model K*.

6. A time-varying condition guided wave structure damage monitoring method, characterized in that: The steps include: Acquire a response signal of the waveguide structure to be monitored; According to the response signal of the waveguide structure to be monitored, the waveguide structure damage identification model under time-varying conditions constructed by any one of claims 1 to 5 is used to perform the waveguide structure damage monitoring, and obtain the waveguide structure damage monitoring result under time-varying conditions.

Images