Composite material defect positioning method and system fusing ultrasonic guided waves and spatial information

Abstract

The present application belongs to the technical field of structural health monitoring and nondestructive testing, and particularly relates to a composite material defect positioning method and system fusing ultrasonic guided waves and spatial information, which comprises: arranging a piezoelectric sensor network on the surface of a structure to obtain health state signals, damage state signals and sensor spatial coordinates; constructing a multi-channel contrast input comprising health signals, damage signals and differential signals; screening high-quality propagation paths based on a path quality evaluation function; extracting guided wave depth features from the multi-channel input using a one-dimensional convolutional neural network; encoding the sensor coordinates into a spatial feature vector and fusing the spatial feature vector with the guided wave depth features to obtain spatial fusion features; and predicting two-dimensional coordinates of defects through a regression network. The present application establishes an accurate mapping relationship between signal propagation paths and defect positions through a multi-channel contrast input and spatial information fusion mechanism, effectively improving the precision and robustness of composite material structure defect positioning.

Description

Technical Field

[0001] This invention belongs to the field of structural health monitoring and non-destructive testing technology, and particularly relates to a method and system for locating defects in composite materials that integrates ultrasonic guided waves and spatial information. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Composite materials are widely used in aerospace and other fields due to their advantages such as high specific strength and high specific stiffness. However, they are prone to internal defects such as delamination and cracks during long-term use, which seriously affect structural safety. Therefore, it is of great significance to achieve efficient and accurate defect detection and location.

[0004] Ultrasonic guided wave testing technology is characterized by long propagation distance and sensitivity to damage, and is often used for health monitoring of plate structures. However, guided wave propagation in carbon fiber composites exhibits multimodal phenomena and strong attenuation, resulting in complex signals. Traditional methods that rely on manual interpretation of waveform characteristics are subjective, inefficient, and have limited feature extraction capabilities.

[0005] Furthermore, existing methods mostly utilize only the current detection signal without effectively combining it with the health status baseline signal, making them insensitive to minor defects. Simultaneously, the signal quality varies greatly across multiple sensor paths, and there is a lack of effective screening for high-quality paths, affecting positioning accuracy and stability. While existing deep learning methods can automatically extract temporal features, they often use single-channel or simple concatenation inputs, failing to adequately utilize signal difference information.

[0006] Therefore, there is an urgent need for a method that can effectively utilize multi-source ultrasonic signals to improve the accuracy of locating defects in complex structures. Summary of the Invention To overcome the shortcomings of the prior art, this invention provides a method and system for locating defects in composite materials by integrating ultrasonic guided waves and spatial information. The aim is to enhance the correlation between defect-sensitive features and spatial location by constructing a comparison input mechanism between healthy and damaged signals and combining sensor spatial coordinates to perform spatial fusion of extracted depth waveform features, thereby improving the accuracy and robustness of defect location in complex composite material structures.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions: Firstly, a method for locating defects in composite materials by integrating ultrasonic guided waves and spatial information is disclosed, including: A network of piezoelectric sensors is arranged on the surface of the structure to acquire health status signals, damage status signals, and sensor spatial coordinates. Signal preprocessing, for each excitation The receiving path is constructed using a multi-channel comparison input consisting of a healthy signal, an impaired signal, and their differential signal; a path quality evaluation function is constructed to filter paths that meet the threshold. A one-dimensional convolutional neural network is used to extract waveguide depth features from multi-channel contrast input; Based on the spatial information fusion mechanism, the normalized excitation and receiving sensor coordinates are encoded into spatial coordinate features, and the waveguide depth features are fused to obtain spatial fused features. The spatial fusion features are input into a regression network to predict the two-dimensional coordinates of the defect.

[0008] Furthermore, the step of arranging a piezoelectric sensor network on the structural surface to acquire health status signals, damage status signals, and sensor spatial coordinates specifically includes: An ultrasonic guided wave detection system with a main controller as its core is constructed. The main controller integrates a waveform generation module and a signal acquisition module. The waveform generation module generates an excitation signal, which is amplified and then applied to the piezoelectric sensor at the excitation end to excite the ultrasonic guided wave in the structure. The piezoelectric sensor at the receiving end collects the response signal propagating through the structure and records it by the signal acquisition module. Multiple propagation paths are formed by different sensor combinations, and ultrasonic signals of healthy and defective states under each path are collected and the spatial coordinates of each sensor are recorded.

[0009] Furthermore, the signal preprocessing includes filtering, effective time window truncation, and amplitude normalization.

[0010] Furthermore, the path quality evaluation function is constructed based on the signal-to-noise ratio, energy level, and waveform statistical characteristics of the signal, and quantitatively scores the signal quality of each propagation path.

[0011] Furthermore, the one-dimensional convolutional neural network adopts a multi-layer serial one-dimensional convolutional structure, with batch normalization and nonlinear activation functions configured after each convolutional layer, and the feature size is gradually compressed through pooling operations; at the end of the network, an adaptive pooling layer is used to unify the temporal features of different lengths into a fixed-dimensional guided wave depth feature output.

[0012] Furthermore, the spatial information fusion mechanism specifically involves: normalizing the spatial coordinates of the excitation end and the receiving end based on the overall dimensions of the structure under test; encoding the normalized coordinates into spatial feature vectors via a fully connected network; and then fusing them with the waveguide depth features to form spatial fusion features.

[0013] Furthermore, the regression network uses a smooth absolute value loss function as the optimization objective to reduce the deviation between the predicted defect coordinates and the actual defect coordinates.

[0014] Secondly, a composite material defect localization system integrating ultrasonic guided waves and spatial information is disclosed, including: Signal acquisition module: A piezoelectric sensor network is arranged on the surface of the structure to acquire health status signals, damage status signals and sensor spatial coordinates; Signal preprocessing and multi-channel signal construction module: performs signal preprocessing for each excitation signal. The receiving path is constructed as a multi-channel comparison input consisting of a healthy signal, an impaired signal, and a differential signal between the two; Path quality assessment and screening module: Constructs path quality evaluation functions and screens paths that meet the threshold. Waveform feature extraction module: Employs a one-dimensional convolutional neural network to extract guided wave depth features from multi-channel contrast input; Spatial information fusion module: Based on the spatial information fusion mechanism, the normalized excitation and receiving sensor coordinates are encoded into spatial coordinate features, and the guided wave depth features are fused to obtain spatial fusion features; Defect location regression prediction module: Input the spatial fusion features into the regression network to predict the two-dimensional coordinates of the defect.

[0015] Thirdly, a computer device is disclosed, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to perform the steps of the method described above.

[0016] Fourthly, a computer-readable storage medium is disclosed having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described above.

[0017] The above one or more technical solutions have the following beneficial effects: This invention constructs a multi-channel contrast input mechanism that includes health signals, damage signals, and differential signals, enabling the model to simultaneously learn original propagation characteristics, damage response characteristics, and state change characteristics. This effectively enhances the sensitivity to minute defects and the ability to express features. Compared with the traditional single-channel input method, it significantly improves the utilization efficiency of defect-related information in ultrasonic guided wave signals.

[0018] This invention constructs a spatial information fusion mechanism, encodes sensor spatial coordinates into spatial coordinate features, and performs feature fusion on the depth waveform features extracted by a one-dimensional convolutional neural network. This enables the model to adaptively adjust signal features according to the geometric relationship of the propagation path, establishing an accurate mapping relationship between the signal propagation path and the spatial location of defects, which significantly improves the accuracy and stability of defect localization in complex composite material structures.

[0019] This invention constructs a path quality evaluation function based on signal-to-noise ratio, signal energy, and waveform kurtosis to quantitatively screen multi-path signals, eliminate low-quality paths, and reduce the adverse effects of noise interference on model training, thereby improving the robustness and engineering applicability of the defect localization method under multi-path conditions.

[0020] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0021] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0022] Figure 1 This is a flowchart of the composite material defect localization method integrating ultrasonic guided waves and spatial information according to Embodiment 1 of the present invention; Figure 2 This is a block diagram of the signal feature and spatial coordinate fusion in the composite material defect localization method that integrates ultrasonic guided waves and spatial information according to Embodiment 1 of the present invention. Figure 3 This is a comparison diagram of the actual and predicted defect locations in the composite material defect localization method that integrates ultrasonic guided waves and spatial information according to Embodiment 1 of the present invention. Figure 4 This is a block diagram of a composite material defect localization system that integrates ultrasonic guided waves and spatial information according to Embodiment 2 of the present invention. Detailed Implementation

[0023] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0025] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0026] Example 1 like Figure 1 As shown, this embodiment discloses a method for locating defects in composite materials by integrating ultrasonic guided waves and spatial information, including: Step 1: Arrange a piezoelectric sensor network on the surface of the structure to acquire health status signals, damage status signals, and sensor spatial coordinates.

[0027] To achieve multi-channel synchronous acquisition of ultrasonic guided wave signals under both healthy and damaged structural conditions, this embodiment constructs an ultrasonic guided wave testing experimental system. The system uses an NI main controller as its core data control unit, integrating an arbitrary waveform generation module and a signal acquisition module. The arbitrary waveform generation module generates an excitation signal, which is amplified by a power amplifier and then applied to the piezoelectric sensor at the excitation end, exciting the ultrasonic guided wave signal within the structure. The receiving sensor acquires the response signal propagating through the structure, which is then recorded by the signal acquisition module. This excitation-reception-acquisition link realizes the active excitation and multi-channel synchronous response acquisition of the ultrasonic guided wave signal.

[0028] As a further implementation scheme, a sensor network consisting of multiple piezoelectric ceramic sensors is arranged on the surface of the composite material structure to be tested. The sensors can function as both excitation and reception ends, and are activated by different excitation methods. The receivers combine to form multiple ultrasonic guided wave propagation paths.

[0029] For each propagation path, health status signals under healthy structural conditions and damage status signals when the structure has defects are collected, while the spatial coordinate information of each piezoelectric ceramic sensor is recorded.

[0030] By traversing multiple sets of incentives By combining receiving paths, ultrasound signal data from multiple paths and states are collected to form a complete dataset containing healthy signals, damaged signals, and corresponding sensor coordinates. This dataset provides a multi-source information foundation for subsequent multi-channel comparative input construction, path quality screening, and spatial information fusion, enabling the model to learn the mapping relationship between defect location and signal features from waveform changes along different propagation paths.

[0031] Step 2: Preprocess the signal for each excitation. The receiving path is constructed as a multi-channel contrast input consisting of a healthy signal, an impaired signal, and a differential signal between the two.

[0032] To improve the distinguishability of defect-sensitive features in guided wave signals and reduce the interference of environmental noise and redundant echoes on model training, the acquired raw ultrasonic signals were sequentially subjected to bandpass filtering, time window truncation, and normalization. The raw ultrasonic signals included healthy and damaged signals to improve signal quality and ensure consistency between data from different paths.

[0033] In this embodiment, the original guided wave signal has 10,000 sampling points. Based on power spectral density analysis, the main energy of the guided wave is concentrated around 300 kHz. Therefore, a fourth-order Butterworth bandpass filter is used for frequency domain filtering, with cutoff frequencies set at 200 kHz and 400 kHz, and a sampling frequency of 10 MHz, to effectively suppress low-frequency drift and high-frequency noise interference. Let the acquired original guided wave discrete signal be:

[0034] in, Indicates the sampling point index. This represents the original signal length.

[0035] The filtered signal is represented as follows:

[0036] in: This is the filtered guided wave signal; Indicates bandpass filtering operation Based on the guided wave envelope energy evolution analysis, it was found that defect-related information is mainly concentrated in the main wave packet propagation stage. Therefore, to highlight the defect-sensitive area and reduce the influence of wake waves and boundary reflection signals, the following was extracted from the filtered signal: to The interval serves as the effective time window.

[0037] The signal after window cutoff is represented as follows:

[0038] in:

[0039] Indicates the effective signal length after window truncation:

[0040] In this embodiment, That is, approximately 2100 sampling points are retained after the window is cut off.

[0041] Since different propagation paths and sensor coupling states can cause scale differences in guided wave amplitudes, this embodiment further employs the Z-score normalization method to normalize the windowed signal in order to reduce the adverse effects of amplitude variations on network training.

[0042] in: The signal is normalized; The mean of the clipped window signal; This represents the standard deviation of the windowed signal.

[0043] After the above preprocessing, multi-channel comparative input data based on structural state differences is constructed for each excitation-reception path.

[0044] Set in the path In China, the health status signal is The damage state signal is Where i is the excitation end and j is the receiving end. , This represents the effective signal length after windowing. To explicitly characterize the waveform differences caused by changes in structural state, the differential signal is defined as:

[0045] The above-mentioned health signals, damage signals, and differential signals are combined to construct a three-channel input tensor. The number of channels These correspond to the characteristics of healthy guided waves, damaged guided waves, and differential guided waves, respectively. This multi-channel contrastive input mechanism enables the model to simultaneously learn the original propagation characteristics, damage response characteristics, and state change characteristics, effectively enhancing its sensitivity to minute defects and its feature representation ability.

[0046] Step 3: Construct a path quality evaluation function and filter paths that meet the threshold. Under multipath excitation-reception conditions, the signal quality of different propagation paths varies significantly, and some paths are susceptible to noise interference or attenuation. If all paths are included in the model training without differentiation, the positioning accuracy and robustness will be reduced. Therefore, this embodiment constructs a path quality evaluation function Q to quantitatively evaluate the signal of each path and select high-quality paths for subsequent model construction.

[0047] The path quality evaluation function Q is constructed based on the following metrics: signal-to-noise ratio (SNR), signal energy (E), and waveform statistical characteristics (including kurtosis or kurtosis, etc.). In this embodiment, a path quality evaluation function is constructed to comprehensively evaluate the signal quality of each propagation path. Let the first... The guided wave signal corresponding to each propagation path is Then the path quality evaluation function can be expressed as:

[0048] in, For the first Signal-to-noise ratio of each propagation path; For signal energy; For waveform statistical characteristics (such as kurtosis or kurtosis); , , The weight coefficients for each evaluation indicator are given, and they satisfy the following conditions:

[0049] The signal-to-noise ratio can be expressed as:

[0050] in, For the first The average power of the effective signal segment of the propagation path, This represents the average power of the noise segment.

[0051] Signal energy is defined as:

[0052] in, This represents the number of sampling points within the effective time window.

[0053] Waveform kurtosis is defined as:

[0054] in, and They represent the first The mean and standard deviation of the signal along each propagation path.

[0055] After calculating the quality score for each propagation path, a threshold is set. Only retain those that satisfy: Propagation paths with a value ≥ Q0 are used for subsequent model training and prediction. This path selection mechanism eliminates inferior paths with low signal-to-noise ratios, severe energy attenuation, or insignificant damage scattering characteristics, reducing the interference of low-quality data on model training and thus improving the stability and robustness of subsequent defect localization results.

[0056] Step 4: Use a one-dimensional convolutional neural network to extract depth waveform features from the multi-channel contrast input; like Figure 2 As shown, in order to automatically capture the local waveform structure, propagation mode and temporal features in ultrasonic guided wave signals, this embodiment constructs a one-dimensional convolutional neural network model suitable for ultrasonic time-series signals to extract features from multi-channel input signals. Specifically, the one-dimensional convolutional neural network adopts a multi-layer serial one-dimensional convolutional structure, with batch normalization and nonlinear activation functions configured after each convolutional layer, and the feature size is gradually compressed through pooling operations; at the end of the network, an adaptive pooling layer is used to unify the temporal features of different lengths into a fixed-dimensional guided wave depth feature output.

[0057] Let the preprocessed multi-channel guided wave input signal be represented as:

[0058] in, Input the number of channels. The length of the guided wave sequence is C=3, which corresponds to the characteristics of healthy guided wave, damaged guided wave, and differential guided wave, respectively.

[0059] The guided wave signal is input into a one-dimensional convolutional network for feature extraction to obtain the guided wave depth features:

[0060] in, This represents a one-dimensional convolutional feature extraction network. This is the characteristic vector of the guided wave.

[0061] Step 5: Based on the spatial information fusion mechanism, the normalized excitation and receiving sensor coordinates are encoded into spatial coordinate features, and the guided wave depth features are fused to obtain spatial fused features.

[0062] To fully utilize the spatial geometric information in the waveguide propagation path, this embodiment uses the coordinates of the excitation and receiving ends of the piezoelectric sensor as auxiliary spatial features, and fuses them with the waveguide depth features to establish a correlation between signal features and spatial location.

[0063] Define the spatial coordinates of the excitation end and the receiving end of the piezoelectric sensor as follows:

[0064] in: Indicates the coordinates of the excitation end; Indicates the coordinates of the receiving end.

[0065] To eliminate the influence of dimensional dimensions, this embodiment uses plate dimensions for normalization:

[0066] in, and These represent the width and height of the composite material plate, respectively.

[0067] The normalized spatial coordinates are input into a fully connected network for spatial feature encoding:

[0068] in, For spatial coordinate eigenvectors, Represents the spatial feature dimension. This represents a fully connected network used to map low-dimensional spatial coordinates to high-dimensional spatial feature representations.

[0069] Subsequently, the waveguide depth features and spatial coordinate features are spliced ​​and fused at the feature level to obtain the spatial fused features:

[0070] in, This indicates a vector concatenation operation along the feature dimension. Specifically, it involves concatenating the depth feature vectors extracted from the guided wave signal branch. Spatial feature vectors obtained from spatial coordinate encoding branch By connecting along the feature dimension, two types of feature information with different physical meanings can be jointly represented:

[0071] in:

[0072] Therefore, the fused feature vector is represented as:

[0073] This fusion method does not change the physical meaning of the original guided wave characteristics and spatial coordinate characteristics. By jointly encoding, it retains both the guided wave propagation response information and the sensor spatial distribution information, providing a comprehensive feature representation for subsequent defect coordinate regression.

[0074] Step 6: Input the spatial fusion features into the regression network to predict the two-dimensional coordinates of the defect.

[0075] The spatial fusion feature Inputting the data into a regression network predicts the two-dimensional coordinates of the defect. The regression network consists of multiple fully connected layers, used to achieve a non-linear mapping from fused features to defect location coordinates. Its calculation process is as follows:

[0076] in: The spatial fusion characteristics after fusion; These represent the weight matrices of each fully connected layer; Indicates the corresponding bias parameter; Represents a non-linear activation function; This represents the two-dimensional coordinates of the defect output by the model.

[0077] The final defect location prediction results are as follows:

[0078] in, and These represent the x-coordinate and y-coordinate of the predicted defect location in the structural coordinate system, respectively.

[0079] During training, Smooth L1 Loss is used as the regression loss function to reduce the error between the predicted coordinates and the true coordinates. Its definition is:

[0080] in, The coordinates of the actual defects. For predicted coordinates.

[0081] By training and learning the aforementioned fusion features, the model can simultaneously learn the ultrasonic wave shape characteristics under different propagation paths and their mapping relationship with the spatial location of defects. In practical applications, when the ultrasonic signal data and corresponding path information of the structure under test are input, the trained model can directly output the spatial coordinates of the defect, achieving rapid and accurate prediction of the defect location.

[0082] Specific application examples: To verify the effectiveness of the composite material defect localization method that integrates ultrasonic guided waves and spatial information proposed in this embodiment, a carbon fiber composite plate was selected as the experimental object. A piezoelectric ceramic sensor array was arranged on the plate surface, and an ultrasonic guided wave excitation and response acquisition system was constructed through an arbitrary waveform generation module, a power amplifier, and a signal acquisition module.

[0083] During the experiment, mortar was used to simulate structural defects. Different damage conditions were constructed by changing the arrangement of the mortar on the plate surface. At the same time, different combinations of piezoelectric ceramic sensors were switched as excitation and receiving ends to form multiple ultrasonic guided wave propagation paths, thereby obtaining health and damage state signals under multi-path conditions.

[0084] In the data processing stage, the acquired signals are first preprocessed, and multi-channel comparative input data based on structural state differences are constructed for each propagation path, including healthy signals, damaged signals and differential signals, to characterize the structural state change features. Furthermore, path evaluation indicators are constructed based on signal quality, and different propagation paths are screened to retain high-quality path data for subsequent modeling, so as to reduce the impact of noisy paths on the results.

[0085] In the feature modeling process, the aforementioned multi-channel comparative input data is fed into the feature extraction model for temporal feature learning. Simultaneously, sensor spatial coordinate information is introduced, and a spatial information fusion mechanism is used to fuse signal features, thereby establishing the correlation between propagation path features and the spatial location of the defect. Based on this, a regression model is constructed to predict the defect location coordinates.

[0086] The one-dimensional convolutional feature extraction network used in this embodiment adopts a three-layer serial one-dimensional convolutional structure, and the detailed configuration of each layer is as follows: The first 1D convolutional layer has a kernel size of 9, padding of 4, stride of 1, and the number of output channels is increased from 5 to 16. After convolution, a batch normalization layer, ReLU activation function, and 1D max pooling are applied to complete downsampling. The second 1D convolutional layer has a kernel size of 7, padding of 3, stride of 1, and the number of output channels is increased from 16 to 32. Batch normalization, ReLU activation function, and 1D max pooling are applied sequentially. The third 1D convolutional layer has a kernel size of 5, padding of 2, stride of 1, and the number of output channels is increased from 32 to 64. Batch normalization and ReLU activation function are applied. At the network end, one-dimensional adaptive average pooling is used to unify the feature sequence length, thereby completing the output of fixed-dimensional deep features.

[0087] The entire convolutional network uses ReLU as the activation function, employs batch normalization to calibrate the feature distribution, uses max pooling for downsampling, and uses adaptive average pooling for terminal pooling.

[0088] To comprehensively evaluate the defect localization performance of the model, mean absolute error (MAE), root mean square error (RMSE), and coefficient of determination (CQD) were used. The evaluation metrics include Euclidean distance error and Euclidean distance error.

[0089] The mean absolute error in the X and Y directions is defined as follows:

[0090]

[0091] MAE is used to measure the average deviation between predicted coordinates and actual coordinates, and the unit is mm.

[0092] The root mean square error is defined as:

[0093]

[0094] RMSE is more sensitive to larger prediction errors and can be used to evaluate the overall regression stability of a model.

[0095] The coefficient of determination is defined as:

[0096] in: The mean of the true values. , The closer the value is to 1, the better the model fits.

[0097] Since defect localization is a two-dimensional spatial regression problem, this embodiment further uses Euclidean distance error to evaluate the actual localization accuracy:

[0098] The mean Euclidean distance error is defined as:

[0099] This indicator can directly reflect the spatial distance deviation between the predicted defect location and the actual location, and is the most intuitive indicator for evaluating defect location performance.

[0100] like Figure 3 As shown, based on the final plate positioning results, the model's predicted position and the actual defect position generally match well. The predicted points are mainly distributed in the area near the actual defect position, and the prediction error vector is relatively small overall. This indicates that the constructed model can effectively learn the mapping relationship between the ultrasonic guided wave propagation characteristics and the spatial position of the defect.

[0101] Furthermore, with a total sample size of 144 groups, 108 groups were selected as the training set and 36 groups as the test set for validation. Experimental results show that the mean absolute error in the x-direction is 5.109 mm, the mean absolute error in the y-direction is 5.1945 mm, the root mean square error in the x-direction is 6.0347 mm, the root mean square error in the y-direction is 6.0649 mm, the mean Euclidean positioning error is 7.7663 mm, the median Euclidean positioning error is 7.2504 mm, the coefficient of determination R² in the x-direction is 0.9886, and the coefficient of determination R² in the y-direction is 0.9857.

[0102] The experimental results above show that this embodiment achieves collaborative modeling of multi-source information by constructing a multi-channel comparison input mechanism, a path quality screening mechanism, and a spatial information fusion mechanism. It has high positioning accuracy and good stability within the effective detection area of ​​composite material plates and can accurately identify the location of defects under multi-path excitation receiving conditions. This verifies the feasibility and engineering application value of the method in composite material structure health monitoring.

[0103] Example 2 Based on the method described in Embodiment 1, such as Figure 4 As shown, the purpose of this embodiment is to provide a composite material defect localization system that integrates ultrasonic guided waves and spatial information, including: Signal acquisition module: A piezoelectric sensor network is arranged on the surface of the structure to acquire health status signals, damage status signals and sensor spatial coordinates; Signal preprocessing and multi-channel signal construction module: performs signal preprocessing for each excitation signal. The receiving path is constructed as a multi-channel comparison input consisting of a healthy signal, an impaired signal, and a differential signal between the two; Path quality assessment and screening module: Constructs path quality evaluation functions and screens paths that meet the threshold. Waveform feature extraction module: Employs a one-dimensional convolutional neural network to extract guided wave depth features from multi-channel contrast input; Spatial information fusion module: Constructs a spatial information fusion mechanism, encodes the normalized excitation and receiving sensor coordinates into spatial coordinate features, performs feature fusion on the depth waveform features, and obtains spatial fusion features; Defect location regression prediction module: Input the spatial fusion features into the regression network to predict the two-dimensional coordinates of the defect.

[0104] Example 3 The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0105] Example 4 The purpose of this embodiment is to provide a computer-readable storage medium.

[0106] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above method.

[0107] The steps and methods involved in the apparatus of the above embodiments correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0108] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0109] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for locating defects in composite materials by integrating ultrasonic guided waves and spatial information, characterized in that, include: A network of piezoelectric sensors is arranged on the surface of the structure to acquire health status signals, damage status signals, and sensor spatial coordinates. Signal preprocessing, for each excitation The receiving path is constructed using a multi-channel comparative input consisting of a healthy signal, an impaired signal, and their differential signal; a path quality evaluation function is constructed to filter paths that meet the threshold. A one-dimensional convolutional neural network is used to extract waveguide depth features from multi-channel contrast input; Based on the spatial information fusion mechanism, the normalized excitation and receiving sensor coordinates are encoded into spatial coordinate features, and the waveguide depth features are fused to obtain spatial fused features. The spatial fusion features are input into a regression network to predict the two-dimensional coordinates of the defect.

2. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The step of arranging a piezoelectric sensor network on the structural surface to acquire health status signals, damage status signals, and sensor spatial coordinates specifically includes: An ultrasonic guided wave detection system with a main controller as its core is constructed. The main controller integrates a waveform generation module and a signal acquisition module. The waveform generation module generates an excitation signal, which is amplified and then applied to the piezoelectric sensor at the excitation end to excite the ultrasonic guided wave in the structure. The piezoelectric sensor at the receiving end collects the response signal propagating through the structure and records it by the signal acquisition module. Multiple propagation paths are formed by different sensor combinations, and ultrasonic signals of healthy and defective states under each path are collected and the spatial coordinates of each sensor are recorded.

3. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The signal preprocessing includes filtering, effective time window truncation, and amplitude normalization.

4. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The path quality evaluation function is constructed based on the signal-to-noise ratio, energy level, and waveform statistical characteristics of the signal, and quantitatively scores the signal quality of each propagation path.

5. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The one-dimensional convolutional neural network adopts a multi-layer serial one-dimensional convolutional structure. Each convolutional layer is configured with batch normalization and non-linear activation functions, and the feature size is gradually compressed through pooling operations. At the end of the network, an adaptive pooling layer is used to unify the temporal features of different lengths into a fixed-dimensional waveguide depth feature output.

6. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The spatial information fusion mechanism is as follows: the spatial coordinates of the excitation end and the receiving end are normalized based on the overall size of the structure under test. The normalized coordinates are encoded into spatial feature vectors by a fully connected network, and then fused with the waveguide depth features to form spatial fusion features.

7. The composite material defect localization method integrating ultrasonic guided waves and spatial information as described in claim 1, characterized in that, The regression network uses a smooth absolute value loss function as the optimization objective to reduce the deviation between the predicted defect coordinates and the actual defect coordinates.

8. A composite material defect localization system integrating ultrasonic guided waves and spatial information, characterized in that, include: Signal acquisition module: A piezoelectric sensor network is arranged on the surface of the structure to acquire health status signals, damage status signals and sensor spatial coordinates; Signal preprocessing and multi-channel signal construction module: performs signal preprocessing for each excitation signal. The receiving path is constructed as a multi-channel comparison input consisting of a healthy signal, an impaired signal, and a differential signal between the two; Path quality assessment and screening module: Constructs path quality evaluation functions and screens paths that meet the threshold. Waveform feature extraction module: Employs a one-dimensional convolutional neural network to extract guided wave depth features from multi-channel contrast input; Spatial information fusion module: Based on the spatial information fusion mechanism, the normalized excitation and receiving sensor coordinates are encoded into spatial coordinate features, and the guided wave depth features are fused to obtain spatial fusion features; Defect location regression prediction module: Input the spatial fusion features into the regression network to predict the two-dimensional coordinates of the defect.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the composite material defect localization method that integrates ultrasonic guided waves and spatial information as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it performs the steps of the composite material defect localization method that integrates ultrasonic guided waves and spatial information as described in any one of claims 1-7.

Images