CN117907445A - Damage identification method for composite stiffened plates based on ultrasonic guided waves and space-time hybrid network - Google Patents

Abstract

The present invention discloses a composite material stiffened plate damage identification method based on ultrasonic guided wave and spatiotemporal hybrid network, which mainly includes the steps of establishing a data set, establishing a damage identification model based on a spatiotemporal parallel hybrid network model, learning the temporal characteristics and spatial characteristics of the signal at the same time, learning the high-dimensional characteristics of the guided wave through the residual block, improving the information sensitivity of the model by using the convolution attention module, and finally establishing the mapping relationship between the guided wave signal and the damage coordinates. The test results show that this method can learn more guided wave signal features, and the addition of the loss function can better improve the generalization ability of the model, indicating that the composite material damage identification technology combining deep learning and ultrasonic guided wave has certain advantages.

Description

Damage identification method for composite stiffened plates based on ultrasonic guided waves and space-time hybrid network

Technical Field

The present invention relates to the field of structural health monitoring, and in particular to a damage identification method and system for a composite material stiffened plate.

Background technique

With the continuous advancement of science and technology, structural health monitoring technology has become a key tool in many fields. This technology can monitor the quality and damage of structural materials such as aircraft, railway tracks, bridges, etc. in real time, providing an important basis for the use and maintenance of these structures. Especially in the aviation field, carbon fiber reinforced composite materials are widely used in the main load-bearing components of aircraft due to their high specific strength and specific stiffness, excellent fatigue resistance, customizable performance design, and easy overall molding. However, composite structures face problems such as external impact, fatigue load and environmental erosion during the service of aircraft, which may lead to a decrease in the bearing capacity of the structure, degradation of material properties and potential invisible damage. In order to ensure the safety of the structure while minimizing the waste of resources, timely and effective health monitoring technology becomes crucial.

Structural health monitoring systems use a variety of physical principles to detect damage, including ultrasonic guided waves, acoustic emission, fiber Bragg gratings, electromechanical impedance measurement, etc. Among these technologies, ultrasonic guided waves are a particularly suitable method for damage monitoring because the guided waves are affected by the medium they travel through, which means that even invisible damage can be detected. Traditional damage identification methods often rely on techniques such as damage factors and probabilistic imaging, but their identification accuracy is limited due to the multimodal complexity of the signal. For complex structures such as composite stiffened panels, it is often difficult to determine the threshold of the damage factor, and accurate identification of damage is also challenging.

Summary of the invention

The purpose of this application is to provide an improved composite material stiffened plate damage identification method to address the problems existing in the prior art.

A composite stiffened plate damage identification method based on ultrasonic guided waves and time-space hybrid network, characterized in that it includes the steps of: collecting ultrasonic guided wave signals set at simulated damage positions of the composite stiffened plate based on a piezoelectric sensor network arranged on the composite stiffened plate to form ultrasonic guided wave simulated damage data simulating the damage of the stiffened plate; wherein the piezoelectric sensor network forms a path, and the ultrasonic guided wave signal is a multi-frequency guided wave time series signal containing multiple paths; marking the simulated damage position of the simulated damage for the ultrasonic guided wave simulated damage data, establishing a database and dividing the training set and the test set in proportion; establishing a time-space hybrid network model; wherein the time-space hybrid network model includes parallel recurrent neural networks and convolutional neural networks; training the recurrent neural network with the training set to obtain a first output including the time series features of the ultrasonic guided wave simulated damage data as global dependency information, and training the convolutional neural network with the training set to obtain the local spatial features including the ultrasonic guided wave simulated damage data a second output of the ultrasonic guided wave simulation damage data; a third output including local spatial features and global spatiotemporal features formed at least based on the sum of the first output and the second output to establish a mapping relationship between the ultrasonic guided wave simulation damage data and the simulated damage position; and based on the mapping relationship, the damage of the composite material stiffened plate is identified; wherein, a sinusoidal time domain signal modulated by a Hanning window at multiple different center frequencies is used as an excitation; the ultrasonic guided wave simulation damage data is the ultrasonic guided wave simulation damage data after the signals of the multiple different center frequencies are fused; the data dimensions corresponding to the ultrasonic guided wave data include the data channels of the multiple different center frequencies, the waveguide multipath height and the waveguide signal length; wherein, the recurrent neural network includes a multi-layer gated recurrent unit, the multi-layer gated recurrent unit is used to generate the first output, the first output includes the time series features of a single path, and the convolutional neural network is a multi-layer convolutional neural network for generating the second output of the local spatial features; the second output includes the waveguide features between paths.

In some embodiments, the recurrent neural network includes a multi-layer gated recurrent unit, and the multi-layer gated recurrent unit is used to generate the first output, the first output includes the timing characteristics of a single path, and the convolutional neural network is used to generate the second output of local spatial characteristics; the second output includes the waveguide characteristics between paths.

In some embodiments, batch normalization is performed after each convolution calculation of the convolutional neural network and an activation function is added after each layer of batch normalization to introduce nonlinear factors.

In some embodiments, the method further includes allowing the third output to pass through at least one layer of residual blocks for learning high-dimensional features of the ultrasonic guided wave simulation damage data.

In some embodiments, a convolutional attention mechanism module is also included to improve the information sensitivity of the model so as to enhance the modeling capability of the convolutional neural network model for data features.

In some embodiments, the piezoelectric sensor network includes a plurality of grids consisting of four adjacent piezoelectric sensors, the simulated damage is sequentially arranged on each of the grids, the ultrasonic guided wave signal is sequentially stimulated by the piezoelectric sensor in each grid, and is received by the remaining piezoelectric sensors in the grid to collect the ultrasonic guided wave simulated damage data.

The benefits of the composite material structure damage location method based on ultrasonic guided waves and time-space hybrid network provided in the present application are: on the one hand, by using deep learning technology for damage identification, there is no need to consider the multimodal complex characteristics of the signal, and the deep intrinsic characteristics of the data can be mined and learned. On the other hand, the time-space hybrid network can simultaneously capture the global dependency information and local spatial characteristics of the signal, and more fully learn the abstract characteristics of the signal. On the other hand, the method in the present application is suitable for identifying ultrasonic guided waves after multi-frequency signal fusion to determine the damage location, because the ultrasonic guided waves after multi-frequency signal fusion contain more damage information than a single frequency, so the recognition accuracy is higher. Other beneficial effects of each embodiment of the present application will be described in the specific implementation section below.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

FIG1 is a schematic diagram of a test piece and a sensor network in an embodiment of a composite material stiffened plate damage identification method based on ultrasonic guided waves and a spatiotemporal hybrid network according to the present application;

FIG2 is an excitation diagram of ultrasonic guided wave signals at five frequencies in an embodiment of the identification method of the present application;

FIG3 is a diagram of a network model of a damage location algorithm in an embodiment of an identification method according to the present application;

FIG4A , FIG4B , and FIG4C are respectively activation function diagrams in an embodiment of a recognition method according to the present application;

FIG5 is an iterative graph of a loss function of a test set in an embodiment of the recognition method of the present application;

FIG6 is an iteration curve diagram of evaluation indicators of a test set in an embodiment of the recognition method according to the present application;

7A, 7B, 7C and 7D are respectively comparison diagrams of the actual values and predicted results of the damage locations of the test set in an embodiment of the method of the present application.

Detailed ways

In order to make the technical solutions and advantages of the present invention more clear, the technical solutions in the embodiments of the present invention are clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. 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.

It should be understood that when used in this specification and the appended claims, the terms "include" and "comprise" indicate the presence of the described features, wholes, steps, operations, elements and/or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and/or their collections. It should also be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms of "one", "an" and "the" are intended to include plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and/or" used in the present specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

The accompanying drawings show various structural schematic diagrams of the embodiments disclosed in the present invention. These figures are not drawn to scale, and some details are magnified and some details may be omitted for the purpose of clear expression. The shapes of various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are only exemplary, and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may design regions/layers with different shapes, sizes, and relative positions according to actual needs.

According to an embodiment of the present application, a composite material stiffened plate damage identification method based on ultrasonic guided waves and time-space hybrid network specifically includes the following steps:

Step 1: Paste the piezoelectric sensor, conduct a simulated damage experiment, and collect ultrasonic guided wave signals.

This step, for example, includes pasting piezoelectric sensors on the composite material structure to form a sensor network as shown in FIG1. Every four adjacent sensors can form a sensor grid. In each sensor grid, an ultrasonic guided wave signal is excited, and a 5-cycle Hanning window modulated sinusoidal time domain signal at a center frequency of 100, 150, 200, 250, and 300 kHz can be used as an excitation, as shown in FIG2. Glue is sequentially pasted at different positions on the test piece to simulate damage, and the remaining sensors receive the response to the excited ultrasonic guided wave signal as an ultrasonic guided wave simulated damage signal, and convert it into ultrasonic guided wave simulated damage data.

Step 2: Label the data and establish a database.

For example, the collected ultrasonic guided wave signals are labeled, that is, the damage location is represented by a coordinate pair (x, y), and the training set and the test set are divided into a ratio of 7:3. Here, (x, y) is also called the damage coordinate.

Step 3: Establish a spatiotemporal hybrid network model, which is based on parallel convolutional neural networks and recurrent neural networks. Convolutional neural networks perform well in processing static spatial data. They can effectively capture spatial features and structures in data and extract local information from data through the convolutional layers in convolutional neural networks. Recurrent neural networks are specifically used to process time series data and can establish time correlations. By combining convolutional neural networks and recurrent neural networks, spatiotemporal hybrid networks can effectively process data that contains both spatial and temporal information.

For example, a spatiotemporal hybrid network learning network model is established, and its network model structure is shown in Figure 3. The ultrasonic guided wave data after multi-frequency signal fusion corresponds to a three-dimensional signal. The data dimension is recorded as , for example ( ), where C, H, and W represent data channels respectively, and the data channels include the above five different frequencies, waveguide multipath heights, and waveguide signal lengths.

The ultrasonic guided wave data after the fusion of the multi-frequency signals is used as input data to pass through a five-layer gated recurrent unit (GRU) network for global dependency information learning and a five-layer convolutional neural network for local feature learning.

The number of neurons in the GRU network is 128, and we get The temporal feature of is taken as the first output OP1, and the GRU is shown in FIG3 . Specifically, the calculation process of the GRU is shown in formula (1).

; ; ; (1).

in, is the input at the current moment, Output the hidden state at the current moment. It is the update gate. It is the reset gate. is the candidate hidden state at the current moment, t represents the current time step, and h _(t-1) represents the hidden state output of the previous time step. is the sigmoid activation function, and its specific form is shown in Figure 4A. The specific form of the activation function is shown in FIG4B . , , is the neural network weight matrix during training.

The ultrasonic guided wave data after fusion of the multi-frequency signals is used as input data to pass through the five-layer CNN, where the convolution kernel size is , the stride is The convolutional layer is then and the stride is The maximum pooling operation of The spatial signal features are used as the second output OP2. The convolution form is shown in the CNN part of Figure 3. The convolution process is to multiply the original data value by the corresponding convolution kernel value, and then sum it as the result of the convolution calculation. Then the convolution kernel slides on the original data with a certain step size and performs convolution calculation until the entire original data is slid. The calculation process of two-dimensional convolution is shown in formula (2).

(2)

Among them, k is the size of the convolution kernel, f(i, j) is the original data, and g(k-i, k-j) is the convolution kernel.

The waveguide signal is a time series signal containing multiple paths. The GRU-based one extracts the time series features of a single path; the CNN-based one couples the waveguide features between paths. This can better learn the data features. After the parallel spatiotemporal network output, the first output OP1 and the second output OP2 are matrix-added to obtain a dimension size of The third output OP3 contains local spatial features and global spatiotemporal signal features.

Furthermore, the third output OP3 can be made to pass through at least one residual block (RES), for example, two residual blocks for learning, and the logic of each residual block is shown in the RES part of FIG3 , while retaining the low-dimensional signal features, learning the high-dimensional spatial features, and obtaining the fourth output OP4. The connection method of the linear projection layer is shown in formula (3).

(3)

Among them, y is the output after the residual block, and W ₁ and W ₂ are convolution operations.

Batch normalization (BN) is performed after each convolution calculation. Batch normalization can speed up the training process and reduce internal covariate bias, improve performance and solve problems such as gradient disappearance. The process of BN is to learn parameters ,The batch normalization process in this application is based on formula (4).

(4)

Among them, γ, β are learnable parameters, which are initialized to γ=1, β=0, μ _β is the mean of the data features in the batch, σ _β is the standard deviation, and ε is a constant to prevent the denominator from being zero.

In order to avoid pure linear combination, an activation function is added after batch normalization of each layer to introduce nonlinear factors. The LeakyReLU activation function is introduced in each residual block learning as shown in formula (5), and its image is shown in Figure 4C.

(5)

The convolutional attention mechanism module (CBAM) aims to enhance the modeling ability of the convolutional neural network (CNN) model for data features. The main problem it solves is how to automatically learn and focus on important features in the data to improve the representation ability of the CNN model. Its core idea is to introduce two key attention mechanisms: channel attention and spatial attention, such as the channel attention module and spatial attention module in the CBAM part shown in Figure 3.

The goal of channel attention is to assign weights to different channels, that is, different feature dimensions of feature maps, so that the model can automatically pay attention to which channels are more important for specific tasks. In this way, by using the convolutional attention module, the required features can be automatically learned to facilitate the establishment of a mapping relationship with the damage location. The goal of spatial attention is to assign weights to features at different spatial locations so that the model can automatically focus on important areas in the image. By combining channel attention and spatial attention, CBAM can simultaneously focus on important channels and important locations in the image, thereby improving the feature representation ability of the CNN model. After adding the convolutional attention mechanism module to the above-mentioned spatiotemporal hybrid network learning network model, an attention-enhanced spatiotemporal hybrid network learning network model is formed.

After passing through the CBAM module, the fourth output OP4 is converted into a fifth output OP5 of a one-dimensional signal feature. To improve the generalization ability of the model, at least one fully connected layer (FCL) may be provided to perform classification tasks and regression tasks, so that the fifth output OP5 passes through the at least one fully connected layer to obtain a sixth output OP6. A dropout function with a loss rate of 0.3 may be designed in the fully connected layer, and finally a mapping relationship between the ultrasonic guided wave data after multi-frequency signal fusion and the damage coordinates (x, y) is established.

Absolute error loss, also known as L1 Loss, is a loss function used to measure the difference between two vectors. For two vectors x and y, the formula of L1 Loss is shown in formula (6):

(6)

They represent the true value and corresponding predicted value of the i-th sample respectively, and N is the number of samples.

L1 Loss represents the sum of the absolute differences between the elements at corresponding positions in two vectors, so it is sensitive to outliers, i.e., outliers. L1 Loss is used as a loss function in many machine learning tasks, such as regression tasks, where the goal is to minimize the absolute difference between the predicted value and the actual value.

In the process of deep learning back propagation, the optimizer is an optimization algorithm for solving the loss function, guiding the various parameters of the loss function to update appropriate values in the right direction, so that the parameter values of the updated loss function continue to approach the global minimum. The core idea of the optimizer is the gradient descent method, and the main parameters are the gradient and the learning rate. Some embodiments of the present application use the AdaGrad optimizer, which is an adaptive learning rate algorithm. The square root of the sum of all historical square values of the gradient is used to make the step size monotonically decrease, see formula (7). It adjusts the learning rate in each dimension according to the size of the gradient value of the independent variable in each dimension, thereby avoiding the problem that a fixed learning rate is difficult to adapt to all dimensions.

; ; ; (7)

in To calculate the gradient, To calculate the cumulative squared gradient, For element-wise multiplication, is the gradient update parameter, is the global learning rate, A small constant used for numerical stability. The recommended default value is 1e-6. The initial value , is the number of iterations, is the objective function.

Step 4: Evaluation indicators.

In the regression task, the indicators for evaluating model performance can help understand the model's fit and prediction accuracy.

The present invention uses L1 Loss and ^R2 scores as evaluation indicators. L1 Loss calculates the absolute value of the prediction error of each sample and takes the average value. The smaller the value, the better the model fit. R², or R-squared, is a score between 0.1 and 1.0, which is used to measure the degree to which the model explains the data variance. The closer it is to 1.0, the better the effect. The calculation method is as follows:

(8)

Among them, xi, yi represent the target variables, Represents the mean of the target variable. R² equals 1, indicating that the model fits the data perfectly, and N is a natural number.

Verification example:

To verify the effectiveness of the method of the present invention, as shown in FIG1, a simulated damage experiment was carried out on a carbon fiber composite reinforced wall panel with an overall size of 700 mm × 450 mm and containing 4 T-shaped stiffeners. Twelve piezoelectric ceramic sensors (PZTs) are arranged on the surface of the carbon fiber composite material between the ribs of the wall panel to form a PZT network, and the spacing between the PZTs is 160 mm × 130 mm. The PZT network is shown in FIG1, where every four PZTs form a grid, and the grid is, for example, a rectangular grid. Each PZT is at a vertex of the rectangular grid, and each PZT is used as an excitation sensor in turn, and the other three PZTs are used as receiving sensors. Therefore, the PZT network designed by the present invention forms 72 one-transmit-one-receive signal paths. The sampling frequency of the device is 12 MHz, and the sampling time is 330 Grid points with a length of 2 cm were drawn in the test area, and damage was simulated at each grid point of each grid in turn. Five different excitation frequency signals were stimulated, and a total of 375 damage points were collected. The dimension of each collected data was 5×72×4000.

It should be understood that the piezoelectric sensors used in the sensor network in the present application are not limited to piezoelectric ceramic sensors, but may also be piezoelectric sensors made of other materials.

It should be understood that the grid setting in the present application is not limited to the above-mentioned sizes and shapes, and any grid setting that can achieve the purpose of the present application can be adopted.

The spatiotemporal hybrid network learning model was built using the PyTorch framework, and the recognition method of the present invention was implemented using two NVIDIA GeForceRTX3090s. The spatiotemporal hybrid network learning model includes five layers of GRU, five layers of CNN, matrix addition, two layers of RES, one CBAM and two layers of FC as shown in FIG3 .

Finally, AdaGrad is used to optimize the model parameters, with a weight decay of 1e-5 and an initial learning rate of 1e-4. The cosine annealing algorithm is used to gradually reduce the learning rate. The ratio of the training set to the test set is 7:3, the batch size is set to 16, and a total of 80 epochs are iterated. The R² evaluation index and the Loss curve of the test set are shown in Figures 5 and 6, respectively. After the iteration, the L1 Loss value of the test set is 0.1195 and the R² is 0.9084. The ablation experiment is carried out, and the test set results are shown in Table 1. The results show that this method can learn more waveguide signal features.

Table 1:

method R ² Loss The best embodiment of this application (CNN+GRU+RES+CBAM+Dropout): 0.9084 0.1195 Without RES 0.8857 0.1313 Without CBAM 0.8536 0.1503 Without Dropout 0.8988 0.1223 Single frequency 0.8710 0.1333

Using all the data in the training set to fully train the model is called an Epoch, i.e. one generation of training; using a small number of samples in the training set to perform a back-propagation parameter update on the model weights is called a Batch, i.e. a batch of data; the number of samples in this small number is the number of batches, i.e. the Batch size.

The four test results are visualized as shown in Figures 7A, 7B, 7C and 7D, where the circles represent the actual positions and the triangles represent the predicted positions. From the four test results, it can be seen that this parallel network regression positioning algorithm based on multi-frequency signal fusion is highly accurate and intuitive, and the application prospects of composite material damage identification methods based on deep learning are relatively broad.

It can be seen that combining ultrasonic guided wave technology with deep learning algorithms to achieve more accurate identification of damage in composite structures is a promising approach. Deep learning algorithms can learn the abstract intrinsic features of data without manually extracting damage factors or analyzing the multimodal characteristics of signals, which makes it an end-to-end learning method that can directly establish a mapping relationship between guided wave signals and damage information, thereby improving the accuracy and efficiency of damage identification.

In one embodiment of the invention, a terminal device is provided, which includes a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used to execute a composite material stiffened plate damage identification method based on ultrasonic guided waves and time-space hybrid networks.

The present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory (non-volatile memory), such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the above-mentioned embodiment of the method for damage identification of composite material stiffened plates based on ultrasonic guided waves and spatiotemporal hybrid networks; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the steps of the above-mentioned method.

The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

The present invention discloses a composite material stiffened plate damage identification method based on ultrasonic guided wave and spatiotemporal hybrid network, which mainly includes the steps of establishing a data set, establishing a damage identification model based on a spatiotemporal parallel hybrid network model, learning the temporal characteristics and spatial characteristics of the signal at the same time, learning the high-dimensional characteristics of the guided wave through the residual block, improving the information sensitivity of the model by using the convolution attention module, and finally establishing the mapping relationship between the guided wave signal and the damage coordinates. The test results show that this method can learn more guided wave signal features, and the addition of the loss function can better improve the generalization ability of the model, indicating that the composite material damage identification technology combining deep learning and ultrasonic guided wave has certain advantages.

Claims (6)

1. A damage identification method for composite stiffened plates based on ultrasonic guided waves and time-space hybrid networks, characterized in that it comprises the following steps: Based on the piezoelectric sensor network arranged on the composite stiffened plate, an ultrasonic guided wave signal set at a simulated damage position of the composite stiffened plate is collected to form ultrasonic guided wave simulated damage data simulating damage of the stiffened plate; wherein the piezoelectric sensor network forms a path, and the ultrasonic guided wave signal is a multi-frequency guided wave timing signal containing multiple paths; Annotating the simulated damage position of the simulated damage for the ultrasonic guided wave simulated damage data, establishing a database and dividing the training set and the test set in proportion; Establishing a spatiotemporal hybrid network model; wherein the spatiotemporal hybrid network model includes a parallel recurrent neural network and a convolutional neural network; Training the recurrent neural network with the training set to obtain a first output including temporal features of ultrasonic guided wave simulated damage data as global dependency information, and training the convolutional neural network with the training set to obtain a second output including local spatial features of the ultrasonic guided wave simulated damage data; Establishing a mapping relationship between the ultrasonic guided wave simulated damage data and the simulated damage position based on at least a third output including local spatial features and global spatiotemporal features formed by the sum of the first output and the second output; and Identifying damage to the composite stiffened plate based on the mapping relationship; Among them, a sinusoidal time domain signal modulated by a Hanning window at multiple different center frequencies is used as an excitation; the ultrasonic guided wave simulation damage data is the ultrasonic guided wave simulation damage data after the signals of the multiple different center frequencies are fused; the data dimensions corresponding to the ultrasonic guided wave data include the data channels of the multiple different center frequencies, the waveguide multipath height and the waveguide signal length; In which, the recurrent neural network includes a multi-layer gated recurrent unit, and the multi-layer gated recurrent unit is used to generate the first output, the first output includes the timing characteristics of a single path, and the convolutional neural network is a multi-layer convolutional neural network used to generate the second output of local spatial characteristics; the second output includes the waveguide characteristics between paths. 2. According to claim 1, the damage identification method for composite reinforced plates based on ultrasonic guided waves and spatiotemporal hybrid networks is characterized in that batch normalization is performed after each convolution calculation of the convolutional neural network and an activation function is added after each layer of batch normalization to introduce nonlinear factors. 3. According to claim 1, the composite material stiffened plate damage identification method based on ultrasonic guided waves and spatiotemporal hybrid network is characterized in that it also includes allowing the third output to pass through at least one layer of residual blocks for learning the high-dimensional features of the ultrasonic guided wave simulated damage data. 4. According to claim 1, the composite material stiffened plate damage identification method based on ultrasonic guided waves and spatiotemporal hybrid network is characterized in that it also includes a convolutional attention mechanism module for improving the model information sensitivity to enhance the convolutional neural network model's modeling ability for data features. 5. According to claim 1, the damage identification method for composite reinforced plates based on ultrasonic guided waves and time-space hybrid networks is characterized in that: wherein the piezoelectric sensor network includes a plurality of grids composed of four adjacent piezoelectric sensors, the simulated damage is sequentially arranged on each of the grids, and the ultrasonic guided wave signal is sequentially stimulated by the piezoelectric sensor in each grid and received by the remaining piezoelectric sensors in the grid to collect the ultrasonic guided wave simulated damage data. 6. The damage identification method for composite stiffened plates based on ultrasonic guided waves and spatiotemporal hybrid networks according to claim 1 is characterized by: L1 Loss and ^R2 scores are used as evaluation indexes for regression tasks, wherein L1 Loss calculates the absolute value of the prediction error of each sample and calculates the average value; R2 is used to measure the degree to which the model explains the data variance, and its calculation method is: ; Among them, _yi , _yi represents the target variable,/> represents the mean of the target variable, R² equal to 1 means that the model fits the data perfectly, and N represents the number of samples.