Composite material damage quantitative evaluation method and device based on cross-modal and hybrid sensing network

Abstract

The invention provides a composite material damage quantitative evaluation method and device based on a cross-modal and hybrid sensing network, and relates to the technical field of damage detection, and the method comprises the steps: obtaining computed tomography images, global strain data and local sound wave signals of a composite material in a health state and different damage states; inputting the computed tomography image into the teacher model, and training to obtain a modal weight of the injury area; training a student model based on a hybrid sensing network according to the modal weight of the damaged area, the global strain data and the local acoustic signal to obtain a target student model; inputting the global strain data and the local sound wave signal of the to-be-detected composite material into a target student model, and outputting an attention weight; obtaining fusion data based on the attention weight, the global strain data of the to-be-detected composite material and the local sound wave signal; and performing damage quantitative evaluation on the fused data to obtain a damage result. According to the scheme, the accuracy of composite material damage quantitative identification is improved.

Description

Technical Field

[0001] This invention relates to the field of material damage identification technology, particularly to the field of health monitoring technology for aerospace composite materials, and especially to a method and apparatus for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks. Background Technology

[0002] Composite materials are widely used in aerospace and other fields due to their excellent properties such as lightweight, high strength, and corrosion resistance. However, the real-time and accurate identification and quantification of internal damage in composite material structures remains a significant challenge. Current technologies, which use numerous sensors to locate damage in composite materials, not only have limited data accuracy and a limited data structure, but also exhibit poor quantification of multiple different types of damage and low identification accuracy. Summary of the Invention

[0003] To address the aforementioned issues, this invention provides a method and apparatus for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks, thereby improving the accuracy of quantitative identification of composite material damage.

[0004] In a first aspect, embodiments of the present invention provide a method for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks, comprising: The method acquires computed tomography images, global strain data, and local acoustic signals of composite materials under healthy and different damage conditions. Fiber Bragg grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersed therein, with the number of fiber Bragg grating sensors exceeding the number of piezoelectric sensors. The computed tomography images are input into the teacher model to train and obtain the modal weights of the damaged areas; The target student model is obtained by training a student model based on a hybrid sensing network according to the modal weights of the damaged area, the global strain data, and the local acoustic signal. The global strain data and local acoustic wave signal of the composite material to be tested are input into the target student model, and the attention weights are output. Based on the attention weights, the global strain data and local acoustic signals of the composite material to be tested are fused to obtain fused data; The fused data is subjected to quantitative damage assessment to obtain damage results.

[0005] Optionally, for wing composite materials, fiber optic grating sensors are uniformly arranged according to a sine curve, and piezoelectric sensors are placed between adjacent peaks or troughs of the sine curve, as well as at the intersections of the upper and lower edge strips of the wing spars and the wing ribs.

[0006] Optionally, the computed tomography image is input into the teacher model to train and obtain the modal weights of the damaged area, including: The computed tomography image is reconstructed and segmented in three dimensions to obtain a three-dimensional geometric model composed of several grids; For each of the grids, the following steps are performed: convolution operation is performed on the grid and its neighboring grids to obtain the local features of the grid; and the position encoding of the grid is concatenated with the local features to obtain enhanced features, so as to generate contextual features using a self-attention mechanism. The local energy is obtained by summing the squares of the local features of the neighboring grids of the given grid. A multi-head attention mechanism is adopted to determine the attention weights of other grids to this grid in order to obtain the global energy; The weight allocation is dynamically adjusted based on the local energy and the global energy, and the local weight and global weight of the grid are determined. Based on the local and global weights of each grid within the damaged region, the modal weights of the damaged region are determined; wherein, the modal weights include local weights and global weights.

[0007] Optionally, modal weights include local weights and global weights; Based on the modal weights of the damaged region, the global strain data, and the local acoustic signal, a student model based on a hybrid sensing network is trained to obtain the target student model, including: Several samples, including global strain data and local acoustic signals, are input into the student model, and strain feature vectors and acoustic feature vectors are obtained through feature extraction. The strain feature vector and the acoustic feature vector are input into the attention layer to determine the first attention weight and the second attention weight; The student model is dynamically trained based on the global weights, the local weights, the first attention weights, and the second attention weights to obtain the target student model.

[0008] Optionally, the student model is dynamically trained based on the global weights, the local weights, the first attention weights, and the second attention weights, including: The first loss value is obtained by calculating the global weight, the local weight, the first attention weight, and the second attention weight; Calculate the first similarity between the strain feature vector and the acoustic feature vector of each sample containing the damaged region, and calculate the second similarity between the strain feature vector of each sample containing the damaged region and the acoustic feature vector of other samples containing the damaged region. Calculate the second loss value based on the first similarity and the second similarity; The first loss value and the second loss value are summed to obtain the loss value; Training is completed when the loss value meets the preset training termination condition.

[0009] Optionally, the attention weights include a first attention weight corresponding to global strain data and a second attention weight corresponding to local acoustic signals; The fused data is determined by the following formula: in, S F The fused data; The first attention weight; S f This provides the global strain data for the composite material to be tested. This is the second attention weight; S p This represents the local acoustic signal of the composite material to be tested. It is a constant.

[0010] Optionally, a quantitative damage assessment is performed on the fused data to obtain damage results, including: The fused data is input into a pre-trained damage assessment model, which outputs key damage features. Based on the aforementioned key damage features, damage parameters including damage size, damage depth, and residual strength are calculated. The damage parameters of the composite material to be tested at different times are obtained, and the damage evolution is predicted by a pre-trained time series model to obtain the predicted damage result; wherein, the damage result includes the predicted damage result and the damage parameters at the current time.

[0011] Secondly, embodiments of the present invention also provide a quantitative assessment device for composite material damage based on cross-modal and hybrid sensing networks, comprising: The acquisition module is used to acquire computed tomography images, global strain data and local acoustic signals of composite materials in healthy state and under different damage states; wherein, fiber optic grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersedly arranged, and the number of fiber optic grating sensors is greater than the number of piezoelectric sensors. The training module is used to input the computed tomography (CT) image into the teacher model to train and obtain the modal weights of the damaged area; and to train a student model based on a hybrid sensing network according to the modal weights of the damaged area, the global strain data and the local acoustic signal to obtain the target student model. The fusion module is used to input the global strain data and local acoustic wave signal of the composite material to be tested into the target student model, output attention weights, and fuse the global strain data and local acoustic wave signal of the composite material to be tested based on the attention weights to obtain fused data. The damage assessment module is used to perform quantitative damage assessment on the fused data to obtain damage results.

[0012] Thirdly, embodiments of the present invention also provide a computing device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the composite material damage quantitative assessment method based on cross-modal and hybrid sensing networks as described above.

[0013] Fourthly, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute the composite material damage quantitative assessment method based on cross-modal and hybrid sensing networks as described above.

[0014] Fifthly, embodiments of the present invention also provide a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the method described in any of the first aspects of this specification.

[0015] This invention provides a method and apparatus for quantitative damage assessment of composite materials based on cross-modal and hybrid sensing networks. The method first trains a teacher model using computed tomography (CT) images of the composite material in healthy and damaged states to obtain modal weights for the damaged region. Then, based on these modal weights, global strain data, and local acoustic signals, a student model based on a hybrid sensing network is trained to obtain a target student model. Thus, by analyzing the global strain data and local acoustic signals of the composite material under test, attention weights are output for each, and these weights are used to fuse the strain data and local acoustic signals to obtain fused data. Finally, quantitative damage assessment is performed on the fused data to obtain the damage result. This invention not only achieves cross-modal data fusion, eliminating modal differences between damage information acquired by different sensing technologies, but also transfers attention information detected by CT images to the student model, improving damage detection accuracy and precision while reducing computational complexity, achieving real-time and accurate damage assessment. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of a method for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks, provided by an embodiment of the present invention. Figure 2 This is a hardware architecture diagram of a computing device provided in an embodiment of the present invention; Figure 3 This is a structural diagram of a composite material damage quantitative assessment device based on a cross-modal and hybrid sensing network, provided by an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0019] The following is the concept of the present invention, such as Figure 1 As shown, this embodiment of the invention provides a method for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks. The method includes: Step 100: Acquire computed tomography images, global strain data and local acoustic signals of the composite material in healthy state and under different damage states; wherein, fiber optic grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersedly arranged, and the number of fiber optic grating sensors is greater than the number of piezoelectric sensors. Step 102: Input the computed tomography image into the teacher model and train it to obtain the modal weights of the damaged area; Step 104: Train a student model based on a hybrid sensing network according to the modal weights of the damaged area, global strain data and local acoustic signals to obtain the target student model; Step 106: Input the global strain data and local acoustic signal of the composite material to be tested into the target student model and output the attention weights; Step 108: Based on attention weights, the global strain data and local acoustic wave signals of the composite material to be tested are fused to obtain fused data; Step 110: Perform a quantitative damage assessment on the fused data to obtain the damage results.

[0020] In this embodiment of the invention, a teacher model is first trained using computed tomography (CT) images of the composite material in a healthy state and under different damage states to obtain modal weights for the damaged region. Then, a student model based on a hybrid sensing network is trained based on these modal weights, global strain data, and local acoustic signals to obtain the target student model. Thus, by analyzing the global strain data and local acoustic signals of the composite material to be detected, attention weights can be output for each, and these weights are used to fuse the strain data and local acoustic signals to obtain fused data. Finally, quantitative damage assessment is performed on the fused data to obtain the damage result. Therefore, this invention not only achieves cross-modal data fusion, overcoming modal differences between damage information acquired by different sensing technologies, but also transfers attention information detected by CT images to the student model, improving the accuracy and precision of damage detection while reducing computational complexity, achieving real-time and accurate damage assessment.

[0021] The following description Figure 1 The execution method for each step is shown.

[0022] In a preferred embodiment, for the wing composite material, fiber optic grating sensors are uniformly arranged according to a sine curve, and piezoelectric sensors are placed between adjacent peaks or troughs of the sine curve, as well as at the intersections of the upper and lower edge strips of the wing spars and the wing ribs. This allows for more accurate monitoring of data at various points on the wing composite material, enabling a comprehensive analysis of its damage state.

[0023] It should be noted that in step 100, the computed tomography images, global strain data, and local acoustic signals in any state are all acquired at the same time. Preferably, the distribution coordinates of the fiber grating sensor are all points on a sine curve. Fiber grating (FBG) sensors mainly rely on optical principles and can measure physical quantities such as temperature, strain, and vibration. Piezoelectric sensors (such as PZT sensors) mainly rely on the piezoelectric effect and can sense signals such as force, strain, vibration, and acoustic waves.

[0024] In this invention, a large number of uniformly distributed fiber optic grating sensors are used to achieve global strain detection of the composite material, while a small number of dispersed piezoelectric sensors are used to achieve local high-precision monitoring of the composite material. By acquiring different data from two sensors with different levels of precision, the health status of the composite material structure can be comprehensively analyzed, thereby improving the accuracy of quantitative damage assessment.

[0025] In step 102, the computed tomography (CT) image is input into the teacher model to train and obtain the modal weights of the damaged area, including: Three-dimensional reconstruction and segmentation of computed tomography images are performed to obtain a three-dimensional geometric model composed of several grids; For each grid cell, the following steps are performed: convolution operation is performed on the grid cell and its neighboring grid cells to obtain the local features of the grid cell; and the position encoding of the grid cell is concatenated with the local features to obtain enhanced features, so as to generate contextual features using a self-attention mechanism. The local energy is obtained by summing the squares of the local features of the neighboring grids of the given grid. A multi-head attention mechanism is adopted to determine the attention weights of other grids to this grid in order to obtain the global energy; The weight allocation is dynamically adjusted based on local and global energy, and the local and global weights of the grid are determined. The modal weights of the damaged region are determined based on the local and global weights of each grid cell included in the damaged region; wherein, the modal weights include local weights and global weights.

[0026] It should be noted that the input to the teacher model is a labeled computed tomography image, which includes information such as the location of the damage, crack length and depth, porosity, and delamination. The output is the modal weights of the damaged area.

[0027] Specifically, this also includes preprocessing each mesh to determine its damage attributes, which include: binary labels (0 / 1, where 0 indicates the mesh is undamaged and 1 indicates the mesh belongs to a damaged region), damage degree indices (such as porosity, crack length ratio, etc.), and material parameters. The damage attributes of each mesh are concatenated with its 3D coordinates to form a feature vector. This feature vector is then aggregated using a local convolution kernel (3×3×3) to generate higher-order local features. F local ∈R C (C represents the number of channels). The local features encode detailed information such as microcrack orientation and layered interface morphology. For example, channel 1 represents the damage gradient along the fiber direction; channel 2 represents the interlaminar shear strain estimate; and channel 3 represents the damage edge curvature. The mesh position encoding is a sine-cosine position code generated from the mesh's 3D coordinates. This position code is then concatenated with the local features to obtain the enhanced features. After multi-head self-attention computation, layer normalization, and residual connections are performed on the enhanced features to obtain the global context features.

[0028] Specifically, the neighborhood grid consists of the 26 other grids within a 3×3×3 region centered on the given grid. The local energy is determined using the following formula: in, For grid i Local energy, N (i ) represents a grid i The neighborhood, j For grid i neighborhood grid, ω j For convolution kernel weights, Neighborhood grid j Local features; The global energy is determined by the following formula: in, For grid i global energy, N This represents the total number of grid cells. α i,j For grid i For the grid j Attention weights For grid j Global features; Calculate the ratio of the global energy to the ratio of the global energy to the local energy for each grid cell. β And determine the global weights. W global = β ×Sigmoid( E global Local weights W global =(1- β )×Sigmoid( E local Thus, the energy is mapped to the [0,1] interval using the Sigmoid function to ensure weight normalization. When a composite material contains several independent damaged regions, for each damaged region, the following is performed: the average of the local weights of all meshes within the damaged region is used as the local weight of the damaged region, and the average of the global weights of all meshes within the damaged region is used as the global weight of the damaged region.

[0029] In a more preferred embodiment, if the composite material is an anisotropic composite material (e.g., carbon fiber), a modulating factor is introduced. γ=cos 2 θ Adjust the global weights. θ Let be the angle between the fiber direction and the principal global strain direction. At this point, the corrected global weight... W’ global = γ × W global .

[0030] In a more preferred embodiment, if the composite material is anisotropic, the global weight of the damaged region is corrected, and the corrected global weight is: ;in, For the damaged area before correction R k global weights η The anisotropy coefficient is given by the material (e.g., 0.3 for carbon fiber and 0.1 for glass fiber). θ The angle between the fiber direction and the principal direction of global strain; If the damaged area of ​​the composite material is a long and narrow damaged area, the aspect ratio is... L / W>5 When this happens, the local weights of the damaged area are corrected, and the corrected local weights are: ; For the damaged area before correction R k Local weights, μ Empirical coefficient (0 < μ <1).

[0031] In this embodiment of the invention, the determination of modal weights for damaged regions in computed tomography (CT) images is achieved through a process of region segmentation, grid weight aggregation, and normalization. This method retains the detail sensitivity at the grid level while providing engineering interpretability at the region level. In actual deployment, the weights can be dynamically updated in conjunction with real-time sensor data to form a closed-loop system.

[0032] For step 104, a student model based on a hybrid sensing network is trained according to the modal weights of the damaged area, global strain data, and local acoustic signals to obtain the target student model, including: Several samples, including global strain data and local acoustic signals, are input into the student model, and strain feature vectors and acoustic feature vectors are obtained through feature extraction. The strain feature vector and the acoustic feature vector are input into the attention layer to determine the first attention weight and the second attention weight. The student model is dynamically trained based on global weights, local weights, first attention weights, and second attention weights to obtain the target student model.

[0033] It should be noted that in step 104, the modal weights of the damaged area obtained from computed tomography images under the same damage state (i.e., acquired at the same time) are used to train the student model with global strain data and local acoustic signal data.

[0034] Specifically, a temporal convolutional network is used to extract the time-series features of the global strain data to obtain a strain feature vector; a short-time Fourier transform and a residual network are used to extract the frequency domain features of the local acoustic signal to obtain an acoustic feature vector. The first attention weight of the time-series features to the damaged area and the second attention weight of the frequency domain features to the damaged area are determined, and the student model is dynamically trained so that the difference between the first attention weight and the global weight, and the difference between the second attention weight and the local weight, are both less than preset thresholds. Training is then completed, and the target student model is obtained.

[0035] In a preferred embodiment, the student model is dynamically trained based on global weights, local weights, first attention weights, and second attention weights, including: The first loss value is obtained by calculating the global weights, local weights, first attention weights, and second attention weights; Calculate the first similarity between the strain feature vector and the acoustic feature vector of each sample containing the damaged region, and calculate the second similarity between the strain feature vector of each sample containing the damaged region and the acoustic feature vector of other samples containing the damaged region. Calculate the second loss value based on the first and second similarities; The first loss value and the second loss value are summed to obtain the total loss value. Training is completed when the loss value meets the preset training termination condition.

[0036] Specifically, the first loss value is determined by the following formula: in, L 1 is the first loss value; p A sample containing the damaged area is considered a positive sample. p ∈(1, P ), P The number of positive samples in a given set of samples; For the sample p global weights; For the sample p First attention weight; Local weights; For the sample p The second attention weight; The second loss value is determined by the following formula: in, L 2 is the second loss value; p, q All are positive samples; p , q ∈(1, P ), PThe number of positive samples in a given set of samples; sim () represents the cosine similarity; τ For temperature parameters; , Samples p The strain characteristic vector and the acoustic characteristic vector; For the sample q The acoustic wave feature vector.

[0037] In this invention, both the teacher model and the student model are designed for the same damage quantification task. However, the teacher model provides more comprehensive supervision signals through high-precision data, while guiding the student model through modal weights. This allows the student model to achieve efficient inference with lightweight inputs, thereby reducing computational complexity and meeting the real-time requirements of aerospace engineering applications. Simultaneously, the student model, through the aforementioned two loss functions, narrows the distance between positive samples, achieving cross-modal data alignment. This enables the student model to learn more detailed detection capabilities from the teacher model, thus improving its ability to identify damage regions.

[0038] It should be noted that in step 106, the global strain data and local acoustic signal of the composite material to be tested were acquired at the same time.

[0039] For step 108, the fused data is determined using the following formula: in, S F To integrate data; The first attention weight; S f This provides the global strain data for the composite material to be tested. This is the second attention weight; S p This represents the local acoustic signal of the composite material to be tested. It is a constant.

[0040] Since the damage information acquired by different sensing technologies has modal differences, it is difficult to directly fuse data from different modalities. This invention uses the attention weights determined by the student model based on a hybrid sensing network to achieve multimodal information fusion.

[0041] In step 110, a quantitative damage assessment is performed on the fused data to obtain damage results, including: The fused data is input into a pre-trained damage assessment model, which outputs key damage features. Based on key damage characteristics, damage parameters including damage size, damage depth, and residual strength are calculated. Damage parameters of the composite material to be tested at different times are obtained, and damage evolution is predicted by a pre-trained time series model to obtain the predicted damage result; wherein, the damage result includes the predicted damage result and the damage parameters at the current time.

[0042] In this embodiment of the invention, the fused data is analyzed by a pre-trained damage assessment model (such as Transformer), and key damage features are extracted to calculate damage parameters (including damage size D, damage depth d, and residual strength R). Then, for the damage parameters of the composite material to be tested at the current time and before the current time, the predicted damage results at future times are predicted by a pre-trained time series model.

[0043] This invention utilizes a hybrid sensing network combining fiber optics and piezoelectricity to quantitatively study damage in composite material structures by transferring the modal weights detected in computed tomography (CT) images to global strain data and local acoustic signals. This enhances the representational power of the cross-modal model, enabling a comprehensive analysis of the structural health status and improving the accuracy of damage quantification. Compared to single-modal detection methods, combining cross-modal methods improves the damage identification accuracy of the hybrid sensing network by at least 30%. Simultaneously, knowledge distillation reduces computational load, allowing low-cost sensors to achieve performance approaching that of high-end detection equipment, thus lowering aviation maintenance costs. Furthermore, the use of lightweight neural networks and optimized data fusion algorithms reduces online damage assessment time to the second level, improving real-time monitoring capabilities.

[0044] The composite material damage quantitative assessment method based on cross-modal and hybrid sensing networks proposed in this invention has the following advantages: it is suitable for multiple damage sites and has good quantification effect on multiple different damage sites; it can be used to combine the detection of internal damage and the characterization of external damage; it realizes multimodal data fusion and alignment, solves the time and space alignment problem of cross-modal data, and improves the accuracy of damage assessment; it realizes low-cost and high-precision quantitative damage assessment; it reduces computational complexity, does not require a large amount of computing resources, and can meet the real-time requirements of aerospace engineering applications.

[0045] like Figure 2 , Figure 3 As shown, this invention provides a quantitative damage assessment device for composite materials based on cross-modal and hybrid sensing networks. The device can be implemented in software, hardware, or a combination of both. From a hardware perspective, as... Figure 2 The diagram shown is a hardware architecture diagram of a computing device for a quantitative assessment device for composite material damage based on a cross-modal and hybrid sensing network, provided in an embodiment of the present invention. (Except for...) Figure 2In addition to the processor, memory, network interface, and non-volatile memory shown, the computing device in the embodiment may also include other hardware, such as a forwarding chip responsible for processing packets. Taking software implementation as an example, such as... Figure 3 As shown, a device in a logical sense is formed by the CPU of its computing device reading the corresponding computer program from the non-volatile memory into memory and running it. This embodiment provides a quantitative assessment device for composite material damage based on cross-modal and hybrid sensing networks, comprising: The acquisition module 300 is used to acquire computed tomography images, global strain data and local acoustic signals of composite materials in healthy state and under different damage states; wherein, fiber optic grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersedly arranged, and the number of fiber optic grating sensors is greater than the number of piezoelectric sensors. Training module 302 is used to input the computed tomography image into the teacher model to train and obtain the modal weights of the damaged area; and to train a student model based on a hybrid sensing network according to the modal weights of the damaged area, the global strain data and the local acoustic signal to obtain the target student model. The fusion module 304 is used to input the global strain data and local acoustic wave signal of the composite material to be tested into the target student model, output attention weights, and fuse the global strain data and local acoustic wave signal of the composite material to be tested based on the attention weights to obtain fused data. The damage assessment module 306 is used to perform quantitative damage assessment on the fused data to obtain damage results.

[0046] In some specific implementations, the acquisition module 300 can be used to perform the above step 100, the training module 302 can be used to perform the above steps 102 and 104, the fusion module 304 can be used to perform the above steps 106 and 108, and the damage assessment module 306 can be used to perform the above step 110.

[0047] In some specific implementations, for wing composite materials, fiber optic grating sensors are uniformly arranged according to a sine curve, and piezoelectric sensors are placed between adjacent peaks or troughs of the sine curve, as well as at the intersections of the upper and lower edge strips of the wing spars and the wing ribs.

[0048] In some specific implementations, the training module 302 is also used to perform the following operations: The computed tomography image is reconstructed and segmented in three dimensions to obtain a three-dimensional geometric model composed of several grids; For each of the grids, the following steps are performed: convolution operation is performed on the grid and its neighboring grids to obtain the local features of the grid; and the position encoding of the grid is concatenated with the local features to obtain enhanced features, so as to generate contextual features using a self-attention mechanism. The local energy is obtained by summing the squares of the local features of the neighboring grids of the given grid. A multi-head attention mechanism is adopted to determine the attention weights of other grids to this grid in order to obtain the global energy; The weight allocation is dynamically adjusted based on the local energy and the global energy, and the local weight and global weight of the grid are determined. Based on the local and global weights of each grid within the damaged region, the modal weights of the damaged region are determined; wherein, the modal weights include local weights and global weights.

[0049] In some specific implementations, the training module 302 is also used to perform the following operations: Several samples, including global strain data and local acoustic signals, are input into the student model, and strain feature vectors and acoustic feature vectors are obtained through feature extraction. The strain feature vector and the acoustic feature vector are input into the attention layer to determine the first attention weight and the second attention weight; The student model is dynamically trained based on the global weights, the local weights, the first attention weights, and the second attention weights to obtain the target student model.

[0050] In some specific implementations, the training module 302 is also used to perform the following operations: The first loss value is obtained by calculating the global weight, the local weight, the first attention weight, and the second attention weight; Calculate the first similarity between the strain feature vector and the acoustic feature vector of each sample containing the damaged region, and calculate the second similarity between the strain feature vector of each sample containing the damaged region and the acoustic feature vector of other samples containing the damaged region. Calculate the second loss value based on the first similarity and the second similarity; The first loss value and the second loss value are summed to obtain the loss value; Training is completed when the loss value meets the preset training termination condition.

[0051] In some specific implementations, the fusion module 304 is also used to perform the following operations: The fused data is determined by the following formula: in,S F The fused data; The first attention weight; S f This provides the global strain data for the composite material to be tested. This is the second attention weight; S p This represents the local acoustic signal of the composite material to be tested. It is a constant.

[0052] In some specific implementations, the damage assessment module 306 is also used to perform the following operations: The fused data is input into a pre-trained damage assessment model, which outputs key damage features. Based on the aforementioned key damage features, damage parameters including damage size, damage depth, and residual strength are calculated. The damage parameters of the composite material to be tested at different times are obtained, and the damage evolution is predicted by a pre-trained time series model to obtain the predicted damage result; wherein, the damage result includes the predicted damage result and the damage parameters at the current time.

[0053] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on a quantitative assessment device for composite material damage based on cross-modal and hybrid sensing networks. In other embodiments of the present invention, a quantitative assessment device for composite material damage based on cross-modal and hybrid sensing networks may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0054] The information interaction and execution process between the modules in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0055] This invention also provides a computing device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a method for quantitative assessment of composite material damage based on a cross-modal and hybrid sensing network, according to any embodiment of this invention.

[0056] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program causes the processor to perform a quantitative assessment method for composite material damage based on a cross-modal and hybrid sensing network according to any embodiment of this invention.

[0057] Embodiments of this application also provide a computer program product, which includes a computer program. A processor of a computer device reads the computer program from a computer-readable storage medium and executes the computer program, causing the computer device to perform a quantitative assessment method for composite material damage based on cross-modal and hybrid sensing networks as described in any of the above embodiments.

[0058] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0059] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0060] Storage media embodiments for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0061] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit programs for use by or in conjunction with an instruction execution system, system, or device.

[0062] The program code contained on a computer-readable medium may be transmitted using any suitable medium, including, but not limited to, wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0063] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0064] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0065] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion module connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion module execute some and all of the actual operations, thereby realizing the function of any of the above embodiments.

[0066] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for quantitative assessment of composite material damage based on cross-modal and hybrid sensing networks, characterized in that, include: The method acquires computed tomography images, global strain data, and local acoustic signals of composite materials under healthy and different damage conditions. Fiber Bragg grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersed therein, with the number of fiber Bragg grating sensors exceeding the number of piezoelectric sensors. The computed tomography images are input into the teacher model to train and obtain the modal weights of the damaged areas; The target student model is obtained by training a student model based on a hybrid sensing network according to the modal weights of the damaged area, the global strain data, and the local acoustic signal. The global strain data and local acoustic wave signal of the composite material to be tested are input into the target student model, and the attention weights are output. Based on the attention weights, the global strain data and local acoustic signals of the composite material to be tested are fused to obtain fused data; The fused data is subjected to quantitative damage assessment to obtain damage results.

2. The method according to claim 1, characterized in that, For composite wing materials, fiber optic grating sensors are uniformly arranged according to a sine curve, and piezoelectric sensors are placed between adjacent peaks or troughs of the sine curve, as well as at the intersections of the upper and lower edge strips of the wing spars and the wing ribs.

3. The method according to claim 1, characterized in that, The computed tomography (CT) images are input into the teacher model to train and obtain the modal weights of the damaged areas, including: The computed tomography image is reconstructed and segmented in three dimensions to obtain a three-dimensional geometric model composed of several grids; For each of the grids, the following steps are performed: convolution operation is performed on the grid and its neighboring grids to obtain the local features of the grid; and the position encoding of the grid is concatenated with the local features to obtain enhanced features, so as to generate contextual features using a self-attention mechanism. The local energy is obtained by summing the squares of the local features of the neighboring grids of the given grid. A multi-head attention mechanism is adopted to determine the attention weights of other grids to this grid in order to obtain the global energy; The weight allocation is dynamically adjusted based on the local energy and the global energy, and the local weight and global weight of the grid are determined. Based on the local and global weights of each grid within the damaged region, the modal weights of the damaged region are determined; wherein, the modal weights include local weights and global weights.

4. The method according to claim 1, characterized in that, Modal weights include local weights and global weights; Based on the modal weights of the damaged region, the global strain data, and the local acoustic signal, a student model based on a hybrid sensing network is trained to obtain the target student model, including: Several samples, including global strain data and local acoustic signals, are input into the student model, and strain feature vectors and acoustic feature vectors are obtained through feature extraction. The strain feature vector and the acoustic feature vector are input into the attention layer to determine the first attention weight and the second attention weight; The student model is dynamically trained based on the global weights, the local weights, the first attention weights, and the second attention weights to obtain the target student model.

5. The method according to claim 4, characterized in that, The student model is dynamically trained based on the global weights, the local weights, the first attention weights, and the second attention weights, including: The first loss value is obtained by calculating the global weight, the local weight, the first attention weight, and the second attention weight; Calculate the first similarity between the strain feature vector and the acoustic feature vector of each sample containing the damaged region, and calculate the second similarity between the strain feature vector of each sample containing the damaged region and the acoustic feature vector of other samples containing the damaged region. Calculate the second loss value based on the first similarity and the second similarity; The first loss value and the second loss value are summed to obtain the loss value; Training is completed when the loss value meets the preset training termination condition.

6. The method according to claim 1, characterized in that, The attention weights include a first attention weight corresponding to global strain data and a second attention weight corresponding to local acoustic signals; The fused data is determined by the following formula: in, S F The fused data; The first attention weight; S f This provides the global strain data for the composite material to be tested. This is the second attention weight; S p This represents the local acoustic signal of the composite material to be tested. It is a constant.

7. The method according to any one of claims 1 to 6, characterized in that, The fused data is subjected to quantitative damage assessment to obtain damage results, including: The fused data is input into a pre-trained damage assessment model, which outputs key damage features. Based on the aforementioned key damage features, damage parameters including damage size, damage depth, and residual strength are calculated. The damage parameters of the composite material to be tested at different times are obtained, and the damage evolution is predicted by a pre-trained time series model to obtain the predicted damage result; wherein, the damage result includes the predicted damage result and the damage parameters at the current time.

8. A quantitative assessment device for composite material damage based on cross-modal and hybrid sensing networks, characterized in that, include: The acquisition module is used to acquire computed tomography images, global strain data and local acoustic signals of composite materials in healthy state and under different damage states; wherein, fiber optic grating sensors are uniformly arranged on the surface of the composite material to be tested, and piezoelectric sensors are dispersedly arranged, and the number of fiber optic grating sensors is greater than the number of piezoelectric sensors. The training module is used to input the computed tomography (CT) image into the teacher model to train and obtain the modal weights of the damaged area; and to train a student model based on a hybrid sensing network according to the modal weights of the damaged area, the global strain data and the local acoustic signal to obtain the target student model. The fusion module is used to input the global strain data and local acoustic wave signal of the composite material to be tested into the target student model, output attention weights, and fuse the global strain data and local acoustic wave signal of the composite material to be tested based on the attention weights to obtain fused data. The damage assessment module is used to perform quantitative damage assessment on the fused data to obtain damage results.

9. A computing device comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of any one of claims 1-7.

Images