Composite material fatigue damage ultrasonic guided wave monitoring method based on BO-CNN model

Abstract

The invention belongs to the field of composite material structure fatigue damage monitoring, and provides a composite material fatigue damage ultrasonic guided wave monitoring method based on a BO-CNN model, and the method comprises the steps: firstly screening out ultrasonic guided wave signal features which are greatly influenced by fatigue damage, then carrying out the collinearity analysis to remove redundant features, and carrying out the preprocessing to form a sample library; a convolutional neural network (CNN) damage prediction model is constructed by taking the signal feature matrix as input and the damage area as output, model hyper-parameters are adjusted by combining a Bayesian optimization (BO) algorithm, finally, a sample library is randomly divided into a training set, a verification set and a test set according to a preset proportion, hyper-parameter selection is performed by utilizing the verification set, model evaluation is performed by utilizing the test set, and the damage prediction result is obtained. The test result shows that the prediction precision of the monitoring method realized by the constructed BO-CNN damage prediction model is improved from 0.94 to 0.98 by the traditional monitoring method based on the CNN damage prediction model, the dispersibility of the prediction result is smaller, and the reliability is higher in the damage regression prediction task.

Description

Technical Field

[0001] The present invention belongs to the field of fatigue damage monitoring of composite material structures, and particularly relates to a method for monitoring fatigue damage of composite materials based on a deep learning model. Background Art

[0002] Composite materials have the advantages of high specific strength / specific stiffness and strong designability, and are increasingly widely used in the aerospace field. However, due to the fact that composite materials usually exhibit nonlinearity and general anisotropy, the failure mechanisms and damage monitoring of these composite material structures are more complex than those of traditional structures. Aerospace structures are under the action of alternating loads for a long time during their service life, and fatigue damage will continuously accumulate in the structure in the form of fiber fracture, matrix cracking, debonding, transverse layer cracking, and delamination, etc., and will eventually suddenly lead to structural failure. Therefore, it is necessary to develop Structural Health Monitoring (SHM) technology for the early characteristics of composite material fatigue damage, so as to reduce the risk of the structure approaching failure under harsh operating loads and environmental conditions, and reduce the long-term downtime costs caused by regular inspections and maintenance.

[0003] Ultrasonic guided waves have the advantages of being able to propagate long distances in structures and being sensitive to damages such as cracks, debonding, and delamination, etc., and can realize the quantitative identification of internal micro-cracks and debonding, etc. in large plate and shell structures. Therefore, the ultrasonic guided wave damage monitoring technology has received extensive attention in the field of aerospace structure maintenance and reliability research. The principle of the damage monitoring technology based on guided waves is as Figure 1 shown, and it has become a relatively promising monitoring method in the field of current aerospace structure damage monitoring. Figure 1 In which T B is the TOF value in the reference state, and W B is the amplitude in the reference state. The corresponding T D and W D correspond to the TOF value and amplitude in the damaged state.

[0004] In the research of structural health monitoring based on ultrasonic guided waves, machine learning algorithms have powerful data processing and pattern recognition capabilities. By training and optimizing machine learning models, the identification and classification of different types of damages can be realized, and the accuracy and efficiency of damage identification can be improved. Among them, the Convolutional Neural Network (CNN) is superior to traditional algorithms in some challenging tasks, and it can be proved by existing research that CNN is superior to traditional methods in terms of accuracy.

[0005] However, CNN still has certain limitations in practical applications and it is difficult to find the optimal solution efficiently and accurately in terms of hyperparameter selection. Summary of the Invention

[0006] The object of the present invention is to propose a structural fatigue damage monitoring method with optimized selection of hyperparameters to monitor the fatigue damage characteristics in the fatigue loading damage of composite materials for at least one of the above problems.

[0007] To this end, in view of the fatigue damage characteristics of composite materials, the present application proposes a structural fatigue damage ultrasonic guided wave monitoring method based on the BO-CNN model. The method includes extracting damage-related peak values, TOF values, signal ±25 kHz band energy, mean square deviation, scattered signal energy, differential signal energy, and fuzzy entropy from the collected ultrasonic guided wave signals as features; performing collinearity analysis to eliminate redundant features, thereby determining the peak value, TOF value, scattered signal energy, and fuzzy entropy as features to form a signal feature matrix, and combining the signal feature matrix with the damage area label to form a sample library; constructing a convolutional neural network (CNN) damage prediction model with the signal feature matrix as the input and the damage area as the output, establishing a search space for the hyperparameters to be optimized in the Bayesian optimization (BO) algorithm, and performing Bayesian optimization on the hyperparameters of the convolutional neural network (CNN) model to select the optimal hyperparameter combination; after the Bayesian optimization is completed, outputting the predicted value of the damage area through the CNN model; and using evaluation indicators to verify the accuracy of the BO-CNN model.

[0008] In some embodiments, forming the sample library further includes performing data preprocessing on the data in the sample library to reduce the complexity of the input data of the model.

[0009] In some embodiments, the data preprocessing includes using the formula to normalize the data, where x is the measured value, x min is the minimum value of the measured value, x max is the maximum value of the measured value, and y is the normalized data.

[0010] In some embodiments, the step of performing Bayesian optimization includes: evaluating the objective function through Gaussian process regression to obtain the optimal hyperparameter combination within the search space, where the expected improvement is selected as the acquisition function to select the next set of hyperparameter combinations; inputting the data that has undergone data processing into the convolutional layer of the CNN model for calculation to obtain a first output, including obtaining a second output after dimensionality reduction in the pooling layer, obtaining a third output in the fully connected layer, and finally calculating the predicted value of the BO-CNN model in the output layer; and calculating the objective function, where the loss function is calculated from the predicted value and the true value, and the training end condition is determined according to the objective function. If the objective function does not meet the end condition, then return and update the hyperparameters, and retrain the CNN model until the number of training times reaches the set value.

[0011] In some embodiments, the step of constructing a sample library includes: using Torayca T700G unidirectional carbon prepreg with a size of 15.24cm×25.4cm, creating a 5.08mm×19.3mm notch in the specimen to induce stress concentration; testing composite material specimens with three different layup configurations, all tests were performed on an MTS testing machine, and the transmission and reception of ultrasonic guided wave signals were achieved through 12 piezoelectric sensors. The specimen included 6 exciters and 6 receivers to form a total of 36 signal paths.

[0012] In some embodiments, training with the BO-CNN injury prediction model includes: dividing the data set in the sample library into a training set, a validation set, and a test set in a ratio of 8:1:1.

[0013] In some embodiments, hyperparameter selection is performed through Bayesian optimization, and the selected hyperparameter combination and the validation set error of the model are saved during the Bayesian optimization process. Finally, the hyperparameter combination with the minimum error is obtained through comparison and used for model training, thereby obtaining the damage size prediction result on the test set.

[0014] In some embodiments, the error of the BO-CNN prediction model is evaluated using three indicators, including root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (R 2 ).

[0015] The technical effect of the present invention is as follows: In general, the embodiments of the present invention propose an ultrasonic guided wave monitoring method for fatigue damage of composite materials based on the BO-CNN model, which improves the accuracy and reliability of fatigue damage monitoring by combining Bayesian optimization and convolutional neural network; wherein, features closely related to fatigue damage are extracted to form a high-quality sample library; a BO-CNN fatigue damage monitoring model is constructed, and automatic tuning of model hyperparameters is achieved through the Bayesian optimization algorithm, further improving the prediction performance of the model.

[0016] In some embodiments of the present application, the collected data is converted into the frequency domain by using Fourier transform to obtain features related to fatigue damage of composite materials, and the extracted features are subjected to collinearity analysis to eliminate redundant features. In some embodiments of the present application, the hyperparameters of the CNN model are selected by the Bayesian optimization algorithm to improve the accuracy of model prediction, and the feasibility of the method is verified by a composite material fatigue test data set.

[0017] Extract the features related to the fatigue damage of composite materials from the ultrasonic guided wave signals through feature engineering, and perform collinearity analysis on the extracted features to eliminate redundant features; use the Bayesian algorithm to optimize the hyperparameters of the CNN model to improve the accuracy of model prediction, and verify the feasibility of this method through the fatigue test dataset of composite materials. The test results show that the optimized CNN has high accuracy and reliability. Description of the Drawings

[0018] Figure 1 Shows the principle of the damage monitoring technology based on ultrasonic guided waves.

[0019] Figure 2 Shows the architecture diagram of the damage monitoring model based on CNN.

[0020] Figure 3 Is a schematic diagram of the architecture of the damage diagnosis model based on BO-CNN.

[0021] Figure 4 Shows a schematic diagram of the specimen prepared for testing, on which the signal path is shown.

[0022] Figure 5 Schematic diagram of the Lamb wave signals at different fatigue cycle numbers for the path 6#→7# at 300 kHz.

[0023] Figure 6a Is an X-ray image when the cyclic tension is 20,000 times, showing that the damaged area gradually increases with the increase of the cycle number.

[0024] Figure 6b Is the curve of the damage size versus the cycle number.

[0025] Figure 7a Is the heat map of the collinearity relationship between features before eliminating redundant features.

[0026] Figure 7b Is the heat map of the collinearity relationship between features after eliminating redundant features.

[0027] Figure 8 Shows the data space of the Bayesian optimization algorithm.

[0028] Figure 9a Is a schematic diagram of the comparison result between the predicted values and the true values of the two models.

[0029] Figure 9b Is a schematic diagram of the comparison result of the prediction errors of the two models.

[0030] Figure 10a Is the scatter plot of the test set of the CNN model.

[0031] Figure 10bIt is a scatter plot of the test set of the BO-CNN model.

[0032] Figure 11 It is a schematic flow chart of the ultrasonic guided wave monitoring method for composite material fatigue damage based on BO-CNN according to an embodiment of the present application. Detailed implementation manners

[0033] The following will describe each embodiment of the present invention in detail with reference to the accompanying drawings. The advantages of the features of the present application will become apparent through the following detailed description.

[0034] It should be understood that although the description of the embodiments is carried out in a way of numbering the steps, the order between the steps is not determined by these numbers, but should be determined by the principle of the solution of the present invention and specific descriptions. For example, although the steps are numbered as S100 and S200 respectively, they can be carried out simultaneously, or step S200 can be carried out first and then step S100.

[0035] In an embodiment of the present invention, as Figure 11 shown, the structural fatigue damage monitoring method based on the BO-CNN model proposed in the embodiment of the present invention includes the following steps:

[0036] Step S100, forming a sample library, including screening out features greatly affected by fatigue damage, performing collinearity analysis to eliminate redundant features, and forming a sample library, or forming a sample library after further preprocessing.

[0037] Step S200, constructing a CNN damage prediction model.

[0038] Step S300, configuring the Bayesian optimization method to adjust hyperparameters to obtain a damage monitoring model based on BO-CNN.

[0039] Step S400, using the test set to evaluate the model.

[0040] Specifically, step S100 first conducts feature engineering, through which the monitoring data can be better extracted and analyzed, thereby effectively improving the monitoring accuracy. In the process of model construction, extracting features related to damage to improve the model's ability is an important issue. Too many signal features may increase the training time of the model and introduce some irrelevant information, thus affecting the prediction accuracy of the model. Therefore, feature selection is required in the process of establishing the damage prediction model, and the features should have sufficient representativeness to ensure the operation efficiency of the model.

[0041] Step S100 then includes screening out features greatly affected by fatigue damage.

[0042] Fatigue damage can change parameters such as the material properties and morphological dimensions of the structure, thereby causing changes in the constitutive equation and boundary conditions of guided wave propagation, and further leading to changes in ultrasonic guided wave signals. This change is mainly reflected in aspects such as the amplitude, arrival time, frequency distribution, and signal complexity of the guided wave signal.

[0043] In previous studies, the following 7 signal features were often used to characterize damage parameters. In previous studies, the following 7 signal features were often used to characterize damage parameters.

[0044] Table 1: Expressions and meanings of seven signal features

[0045]

[0046] Meanwhile, in order to improve the generalization ability and prediction accuracy of the model and better identify and process highly correlated features in the dataset, collinearity analysis of signal features is required, and the calculation formulas are shown in Eqs. (1) and (2):

[0047]

[0048]

[0049] where r ij represents the correlation coefficient between features X i and X j , Cov(X i , X j ) is the covariance of X i and X j , and are the standard deviations of X i and X j respectively, is the mean value of X i .

[0050] Step S200, construct a damage monitoring model based on CNN.

[0051] CNN has powerful data processing capabilities. Figure 2 The architecture of the damage monitoring model based on CNN constructed in this embodiment is described in , which is mainly composed of a convolutional layer (CL), a pooling layer (PL), and a fully connected layer (FCL). Since the pooling layer may cause loss of valuable information, pooling operations can be selected after two convolutional layers, and a regression layer can be added after the fully connected layer for regression prediction. CNN can effectively reduce data complexity and avoid overfitting.

[0052] During the damage monitoring process, first, the extracted features are composed into a sample library as the input of the CNN. Convolution operations are performed in the convolutional layer (CL). In this embodiment, the selected convolution operation has a convolution kernel size of 3*3 and a stride of 1. Then, data dimensionality reduction is performed through max pooling with a size of 2*2 and a stride of 2. The FCL expands the data into the form of a one-dimensional vector and outputs the predicted value of the damage size through the FCL combination and activation function.

[0053] Step S300 first includes Bayesian optimization settings.

[0054] During the model construction process, the robustness of the model is an important performance, and the selection of hyperparameters is crucial. There are many hyperparameters involved in the CNN-based damage prediction model, such as the size, stride, and number of convolution kernels, learning rate, regularization parameter, etc. If the hyperparameters are not selected properly, it will lead to a decline in the performance of the model during training and testing. For example, too high a learning rate may cause the model to oscillate during training and fail to converge effectively; while setting the learning rate too low may lead to a slow convergence speed, an extended model training time, and may fall into a local optimal solution. In this paper, the BO algorithm with fast convergence speed, good performance, and strong scalability is used to handle the hyperparameter optimization problem. The basic principle of BO is to use Bayes' theorem to estimate the posterior distribution of the objective function, and then select the next hyperparameter combination for sampling according to this posterior distribution. BO makes full use of the results of the previous few sampling points to improve the objective function and find the global optimal solution.

[0055] Step S300 then includes forming a damage monitoring model based on BO-CNN.

[0056] This embodiment proposes a prediction model architecture based on BO-CNN for fatigue damage diagnosis of composite material structures. This model architecture consists of a convolution module, a perceptron module, and a Bayesian optimization module. The convolution module is mainly used to process input data and capture key feature data that can effectively reflect damage information. The perceptron module is mainly used to establish the mapping relationship between signal features and damage area to achieve the function of structural damage diagnosis. Then, the Bayesian optimization module is used to select appropriate hyperparameters, including the number of training epochs, mini-batch size, and number of convolutional layers. Figure 3 Shows the basic architecture of the proposed prediction model based on BO-CNN.

[0057] Step S400, model evaluation. After constructing the prediction model of BO-CNN, the evaluation metrics (such as RMSE, MAE, and R 2 ) are used to verify the accuracy of the BO-CNN model.

[0058] In summary, the main steps of the monitoring method for the BO-CNN prediction model include:

[0059] Build a sample library. Extract damage-related features, perform collinearity analysis to eliminate redundant features, form a signal feature matrix, and combine it with the damage area labels to form a sample library.

[0060] Data processing. Since the scales of the input data are different, normalization processing needs to be performed before training.

[0061] Establish a hyperparameter search space. In the BO algorithm, establish a search space for the hyperparameters to be optimized.

[0062] Bayesian optimization. The Bayesian optimization algorithm evaluates the objective function through Gaussian process regression to obtain the optimal combination of hyperparameters. In this process, the expected improvement (EI) is selected as the acquisition function to select the next set of hyperparameter combinations.

[0063] Calculate the output layer. Include: input the normalized data into the convolutional layer for calculation to obtain the first output; obtain the second output after dimensionality reduction in the pooling layer; obtain the third output in the fully connected layer; and finally calculate the predicted value of the BO-CNN model in the output layer.

[0064] Calculate the objective function. The loss function is calculated from the predicted value and the measured value. If the end condition is not met, return to step S440 to update the hyperparameters and retrain the model until the number of training times reaches the set value.

[0065] Predictive output. After the Bayesian optimization is completed, the predicted value of the damage area is output through the CNN model.

[0066] Model evaluation. Use RMSE, MAE, and R2 to evaluate the accuracy of the BO-CNN model.

[0067] The following is a specific composite material fatigue loading damage diagnosis experiment, which includes:

[0068] Experiment setup: The dataset used in this embodiment is the data of the CFRP composite material fatigue aging experiment jointly carried out by the Stanford Structures and Composites Laboratory (SACL) in the United States and the Predictive Center of Excellence (PCoE) of NASA Ames Research Center. In this experiment, a group of specimens were subjected to tensile-tensile fatigue tests under cyclic load control with a frequency of 5.0 Hz and a stress ratio of R = 0.14. The specimen conditions are as Figure 4 shown.

[0069] The specimens were made of Torayca T700G unidirectional carbon prepreg, with dimensions of 15.24 cm × 25.4 cm. A notch of 5.08 mm × 19.3 mm was created in the specimens to induce stress concentration. The tests were conducted on composite specimens with three different ply configurations. All tests were carried out on an MTS testing machine. The transmission and reception of ultrasonic guided wave signals were achieved through 12 PZT sensors (piezoelectric sensors), as Figure 4 shown. There were 6 exciters on the specimens, such as Figure 4 the PZT sensors numbered 1 to 6 in Figure 4 , and 6 receivers, such as Figure 4 the PZT sensors numbered 7 to 12 shown in

[0070] . A total of 36 signal paths were formed, such as 1 L 19 S Figure 5 the path 5-8 shown in Figure 5 . In the range of 150 - 450 KHz, with a step of 50 KHz, 7 excitation frequencies were set, and the average input voltage of the excitation signal was 50 V. The fatigue cycle test stopped at a specific number of fatigue cycles. The PZT sensor data for all paths and frequencies were collected, and at the same time, X-ray imaging was performed on the specimens. The monitored data included Lamb wave signals collected from the PZT sensor network and the fatigue damage propagation observed through X-ray imaging.

[0071] Characterization of specimen fatigue damage: The experimental results of the specimens under different cyclic tensile numbers are shown in Figure 6a , Figure 6b . Among them, Figure 6a is the X-ray image at 20,000 cyclic tensile times, Figure 6b is the curve of the damage size varying with the cyclic number in Table 2. As the cyclic number increases, the damaged area gradually enlarges. It can be seen that the damage size has experienced a typical three-stage evolution process of initiation - propagation - accelerated failure, which is closely related to the fatigue cumulative effect of the specimens.

[0072] Table 2 is a list of the specific values of the damage size varying with the cyclic number in Figure 6b .

[0073]

[0074]

[0075] Sample library construction: This experiment was carried out at 7 different driving frequencies. In order to fully capture the guided wave response of the structure under different conditions, 36 signal transmission paths were designed and constructed using a sensor network. Based on the data collected through these paths, a data set containing 82 working conditions was obtained. Considering that at different frequencies, the guided wave signals obtained by each signal path may carry key information, therefore, during the data processing, the contributions of all signal sources were integrated, and a total of 7 different features were extracted, namely peak value, TOF value, signal energy in the ±25 kHz frequency band, mean square deviation, scattered signal energy, differential signal energy, and fuzzy entropy. However, there may be significant linear relationships among these 7 features, which affect the prediction performance of the model. Therefore, collinearity analysis needs to be performed on the proposed features to eliminate redundant features, improve the model efficiency, and avoid prediction instability caused by multicollinearity. The selected features are as Figure 7a and Figure 7b shown, where Figure 7a is the heat map of the collinearity relationship among features before removing redundant features, Figure 7b is the heat map of the collinearity relationship among features after removing redundant features. Finally, 4 features can be determined as the sample library, namely peak value, TOF value, scattered signal energy, and fuzzy entropy, which will provide rich data support for subsequent fatigue damage analysis.

[0076] A sample library consisting of 4 features was extracted from 82 working conditions, 7 different driving frequencies, and 36 signal paths, as shown in Figure 8 Figure 7a is the heat map of the collinearity relationship among features before removing redundant features.

[0077] Figure 7b is the heat map of the collinearity relationship among features after removing redundant features.

[0078] shown.

[0079]

[0080]

[0081] BO-CNN damage prediction model training: Before model training, the data set needs to be divided first. There are a total of 574 groups of data, which are divided into training set, validation set, and test set according to the ratio of 8:1:1.

[0082] Next is data preprocessing. Since the selected feature scales are different, the data can be normalized using Equation (5) before predicting the fatigue damage size of the composite material. Then, hyperparameter selection is performed through Bayesian optimization. During the Bayesian optimization process, the selected hyperparameter combinations and the validation set error of the model under this combination will be saved. Finally, the hyperparameter combination with the smallest error is obtained through comparison and used for model training, so as to obtain the damage size prediction results on the test set.

[0083]

[0084] where x is the measured value, x min is the minimum value of the measured value, x max is the maximum value of the measured value, and y is the normalized data.

[0085] Results and Discussion: Test results and performance evaluation of the damage prediction model.

[0086] The error of the BO-CNN prediction model is evaluated using three metrics, including root mean square error (RMSE), mean absolute error (MAE), and correlation coefficient (R 2 ). The definitions of these metrics are as follows:

[0087]

[0088]

[0089] where, y i is the measured value, is the predicted value, and n is the number of samples. RMSE comprehensively considers the error magnitude and the influence of outliers. The smaller the value, the more accurate the model prediction; MAE is not sensitive to outliers and pays more attention to the average deviation. The smaller the MAE, the higher the accuracy of the model prediction; R 2 is used to evaluate the interpretability of the model prediction for the data; R 2 ranges from 0 to 1, and the closer it is to 1, the better the fitting effect of the model. These metrics have different focuses and can reflect the prediction performance of the model from different aspects. RMSE pays more attention to the influence of large errors, MAE is suitable for measuring the overall error level, while R 2 emphasizes the goodness of fit of the model. The use of these multiple metrics can more comprehensively evaluate the model performance. Figure 9a ; Figure 9b are the predicted values and measured values of the two models respectively, as well as the comparison of the prediction errors of the two models.

[0090] From Figure 10a Figure 10b it can be seen that the BO-CNN model shows higher prediction accuracy and a smaller error range. The performance evaluation results of the two models are shown in Table 3:

[0091] Table 3 Performance Evaluation of CNN Model and BO-CNN Model

[0092]

[0093] Figure 10a and Figure 10b respectively show the scatter plots of the test sets of the CNN model and the BO-CNN model, which can more intuitively compare the prediction performances of the two models. It can be seen from the figure that the BO-CNN model proposed in this embodiment has a significant improvement in prediction performance compared with the CNN model, where the correlation coefficient R 2 increases from 0.94 to 0.98, indicating that the BO-CNN model is significantly superior to the CNN model in fitting accuracy. By comparing the confidence band and the prediction band, it can be seen that the prediction error range of the CNN model is larger, and the data points are distributed within a wider prediction band, indicating a higher degree of fluctuation in the prediction results; the dispersion of the prediction results of the BO-CNN model is smaller, and the prediction band and the confidence band are significantly narrower, indicating that the prediction accuracy and stability of this model are stronger.

[0094] The test results show that the optimized CNN has high accuracy and low dependence on experience, reducing the time required to train the hyperparameters of the model.

[0095] In particular, according to the embodiments of the present disclosure, the process described by the provided flowchart can be implemented as a computer software program. For example, the embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes program codes for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by a processing device, it executes the above-mentioned functions defined in the method of the embodiments of the present disclosure.

[0096] It should be noted that the above-mentioned computer-readable medium of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium can, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0097] In the embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. In the embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable signal medium may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, radio frequency (RF), etc., or any suitable combination of the above.

[0098] The above computer-readable medium may be included in the above server; or it may exist separately without being assembled into the server.

[0099] The above computer-readable medium carries one or more programs, and when the one or more programs are executed by the server, the server is caused to execute the processing method provided in the embodiments of the present disclosure.

[0100] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++, and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) and a wide area network (WAN), or may be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0101] The units and / or modules described in the embodiments of the present disclosure may be implemented in software or in hardware.

[0102] For the hardware approach, the units and / or modules for implementing the apparatus of the embodiments of the present disclosure may be implemented by one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs) or other electronic components, and are used to execute the methods provided by the embodiments of the present disclosure.

[0103] The preferred embodiments of the present invention are described above, but the spirit and scope of the present invention are not limited to the specific content disclosed herein. Those skilled in the art can arbitrarily combine and expand the above embodiments according to the teachings of the present invention and make more embodiments and applications within the spirit and scope of the present invention. The spirit and scope of the present invention are not defined by the specific embodiments, but by the claims.

Claims

1. Ultrasonic guided wave monitoring method for composite material fatigue damage based on BO-CNN model, characterized by: Included steps The damage-related peak value, TOF value, signal ±25kHz frequency band energy, mean square deviation, scattered signal energy, differential signal energy, and fuzzy entropy are extracted from the collected ultrasonic guided wave signals as features; Perform collinearity analysis to eliminate redundant features, thereby determining peak value, TOF value, scattered signal energy, and fuzzy entropy as features to form a signal feature matrix, and then combine the signal feature matrix and damage area label to form a sample library; A convolutional neural network (CNN) damage prediction model is constructed using the signal feature matrix as input and the damage area as output; In the Bayesian Optimization (BO) algorithm, a search space is established for the hyperparameters to be optimized and the hyperparameters of the Convolutional Neural Network (CNN) model are optimized by Bayesian optimization to select the optimal hyperparameter combination. After the Bayesian optimization is completed, the predicted value of the damage area is output through the CNN model; Evaluation metrics are used to verify the accuracy of the BO-CNN model.

2. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: The collinearity analysis includes calculation using the formula: Among them, r ij Represents feature X i and X j The correlation coefficient between i ,X j ) is X i and X j The covariance of and They are X i and X j The standard deviation of Yes X i The mean of .

3. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 2 is characterized in that: The data preprocessing includes using the formula Normalize the data, where x is the measured value, x min is the minimum value of the measured value, x max is the maximum value of the measured value, and y is the normalized data.

4. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: The steps of performing Bayesian optimization include: evaluating the objective function through Gaussian process regression to obtain the optimal hyperparameter combination in the search space, wherein the expected improvement is selected as the acquisition function to select the next set of hyperparameter combinations; inputting the data processed into the convolution layer of the CNN model to calculate the first output, including obtaining the second output after dimensionality reduction in the pooling layer, obtaining the third output in the fully connected layer, and finally calculating the predicted value of the BO-CNN model in the output layer; and calculating the objective function, wherein the loss function is calculated by the predicted value and the true value, and the training end condition is determined according to the objective function. If the objective function does not meet the end condition, the hyperparameters are returned and updated, and the CNN model is retrained until the number of training times reaches the set value.

5. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: The steps of constructing the sample library include: using Torayca T700G unidirectional carbon prepreg with a size of 15.24cm×25.4cm, creating a 5.08mm×19.3mm notch in the specimen to induce stress concentration; testing composite material specimens with three different layup configurations, all tests were carried out on an MTS testing machine, and the transmission and reception of ultrasonic guided wave signals were achieved through 12 piezoelectric sensors. The specimen included 6 exciters and 6 receivers to form a total of 36 signal paths.

6. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: The step of constructing a sample library also includes: performing fatigue damage test experiments at different driving frequencies, using multiple signal transmission paths designed and constructed by the sensor network; obtaining data from these transmission paths to obtain a data set containing multiple working conditions, and extracting the peak value, TOF value, signal ±25kHz frequency band energy, mean square deviation, scattered signal energy, differential signal energy, and fuzzy entropy from the data set.

7. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: Training with the BO-CNN damage prediction model includes: dividing the data set in the sample library into a training set, a validation set and a test set in a ratio of 8:1:

1.

8. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1 is characterized in that: Hyperparameter selection is performed through Bayesian optimization. The selected hyperparameter combination and the validation set error of the model are saved during the Bayesian optimization process. Finally, the hyperparameter combination with the minimum error is obtained through comparison and used for model training, thereby obtaining the damage size prediction result on the test set.

9. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1, characterized in that: Calculate the objective function. The loss function is calculated by the predicted value and the measured value. If the end condition is not met, return to the step of updating the hyperparameters and retrain the model until the number of training times reaches the set value.

10. The method for monitoring composite material fatigue damage by ultrasonic guided waves based on the BO-CNN model according to claim 1, characterized in that: The error of the BO-CNN prediction model was evaluated using three indicators, including root mean square error (RMSE), mean absolute error (MAE) and correlation coefficient (R 2 ), are defined as follows: Among them, y i is the measured value, is the predicted value and n is the number of samples.

Images