A thermographic sequence processing method and system for hybrid woven composite damage characterization

Abstract

The application provides a thermal imaging sequence processing method and system for hybrid woven composite material damage characterization, and relates to the technical field of composite material damage, and comprises the following steps: acquiring an original thermal image sequence of a hybrid woven composite material to be measured, and obtaining an initial temperature difference sequence based on the original thermal image sequence; preprocessing the initial temperature difference sequence to obtain a preprocessed temperature difference sequence; sparsely decomposing each frame of image of the preprocessed temperature difference sequence into a woven texture component and a damage component based on morphological component analysis, and obtaining a spatially decoupled damage sequence by iterative solving; obtaining a fusion feature sequence by multi-feature thermal signal reconstruction and principal component analysis; converting the fusion feature sequence to obtain a corresponding damage physical area; performing depth layer attribution and multi-layer collaborative clustering based on the temperature difference time sequence curve of each pixel in the damage sequence to obtain each depth layer damage area and a through damage area, and outputting a three-dimensional damage characterization result.

Description

Technical Field

[0001] This invention relates to the field of composite material damage technology, specifically a thermal imaging sequence processing method and system for characterizing damage in hybrid woven composite materials. Background Technology

[0002] Due to their complex microstructure, hybrid braided composite materials exhibit high non-uniformity and multi-mode coupling characteristics during damage evolution, posing a severe challenge to non-destructive testing and damage characterization.

[0003] Infrared thermal imaging is an effective means of monitoring the initiation and propagation of damage in such materials, but existing processing methods still have obvious limitations: on the one hand, thermal image sequences are easily affected by factors such as noise, non-uniform heating and specimen movement, and traditional image enhancement and registration methods are difficult to achieve high-precision alignment while preserving the characteristics of minor damage; on the other hand, the woven texture and the real damage signal are highly mixed in space, and conventional segmentation methods are difficult to decouple effectively, resulting in insufficient accuracy of damage identification.

[0004] Furthermore, existing technologies are mostly focused on qualitative analysis of planar damage, lacking effective mining of damage depth information, making it difficult to achieve a comprehensive characterization from two-dimensional distribution to three-dimensional damage.

[0005] Therefore, a thermal imaging sequence processing method and system for characterizing damage in hybrid braided composite materials are provided. Summary of the Invention

[0006] To address the aforementioned technical problems, the present invention aims to provide a thermal imaging sequence processing method and system for characterizing damage in hybrid woven composite materials.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a thermal imaging sequence processing method for damage characterization of hybrid woven composite materials, the method comprising: The original thermal image sequence of the hybrid braided composite material to be tested is obtained, and the initial temperature difference sequence is obtained based on the original thermal image sequence; The initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain a preprocessed temperature difference sequence. Based on morphological component analysis, each frame of the preprocessed temperature difference sequence is sparsely decomposed into woven texture component and damage component, and spatially decoupled damage sequence is obtained by iterative solution. Based on the damage sequence, a fused feature sequence is obtained through multi-feature thermal signal reconstruction and principal component analysis. Based on the fused feature sequence, the corresponding physical area of ​​the damage is calculated by using damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. Based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain the damage regions of each depth layer and the penetrating damage regions. The output is a three-dimensional damage characterization result that includes planar distribution, depth layer attribution and penetrating markers.

[0008] Preferably, the initial temperature difference sequence obtained based on the original thermal image sequence includes: A transient thermal pulse excitation was applied to the hybrid braided composite material sample under test using a pulsed thermal imaging device, and the original thermal image sequence of the surface of the hybrid braided composite material under test during the cooling stage after thermal excitation was continuously recorded by an infrared thermal imager at a preset acquisition frequency. Extract a predetermined number of cold frames before thermal excitation from the original thermal image sequence, and calculate the pixel-by-pixel time-averaged temperature of the cold frames; The average temperature is subtracted from the original thermal image sequence frame by frame and pixel by pixel to obtain the initial temperature difference sequence.

[0009] Preferably, the initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain a preprocessed temperature difference sequence, including: Based on the initial temperature difference sequence, subpixel-level image registration is performed using optical flow, the inter-frame displacement vector field is calculated, and the image is resampled to eliminate pixel shift caused by vibration, thus obtaining the registered temperature difference sequence. Based on the registered temperature difference sequence, the image is decomposed into wavelets by generalized cross-validation adaptive wavelet threshold denoising. The noise standard deviation is estimated according to the statistical distribution of the high-frequency subband coefficients. The optimal soft threshold is calculated for each subband with the goal of minimizing the generalized cross-validation criterion. The high-frequency coefficients are shrunk to suppress random noise. The image is then reconstructed by inverse wavelet transform to obtain the denoised temperature difference sequence. Based on the denoised temperature difference sequence, temporal non-uniformity correction is performed using Kalman filtering, the pixel response is modeled as a state-space model, and the true response value of each pixel is estimated iteratively using temporal correlation to dynamically eliminate fixed pattern noise, thus obtaining a preprocessed temperature difference sequence.

[0010] Preferably, the fused feature sequence obtained based on the damage sequence through multi-feature thermal signal reconstruction and principal component analysis includes: Based on the spatial decoupling sequence, the cooling curve of each pixel is fitted by an adaptive parameter selection thermal signal reconstruction algorithm to obtain the fitted curve and its derivatives. The original temperature difference curve and at least one first derivative curve are stacked along the channel dimension to construct a multi-feature cube; Based on the multi-feature cube, a fusion feature sequence is extracted through weighted principal component analysis guided by a physical model.

[0011] Preferably, based on the spatial decoupling sequence, the cooling curve of each pixel is fitted using an adaptive parameter selection thermal signal reconstruction algorithm to obtain the fitted curve and its derivatives, including: Based on the spatial decoupling sequence, a multi-model competition pool is constructed, which includes 2nd to 6th order polynomial fitting models and physical analytical models derived based on one-dimensional thermal diffusion theory. For the cooling curve of each pixel, each model in the competition pool is used for fitting. The fitting residuals of each model and its time series complexity index and physical bias index are calculated. With the goal of minimizing the joint cost function, the optimal fitting model and its parameters are automatically selected. The complexity index is measured by calculating the sample entropy of the fitting residuals, and the physical bias index is obtained by calculating the dynamic time warping distance between the fitted curve and the theoretical curve of the physical analytical model. Based on the selected best-fit model, the first and second derivatives with respect to the logarithm of time are calculated, and a confidence plot is output.

[0012] Preferably, based on the multi-feature cube, the extraction of fused feature sequences through weighted principal component analysis guided by a physical model includes: Based on the multi-feature cube and confidence map, a spatiotemporal adaptive weight tensor is constructed; Based on the spatiotemporal adaptive weight tensor, a weighted centering process is performed on the multi-feature cube to obtain a weighted covariance matrix. The weighted covariance matrix is ​​decomposed into eigenvalues, and the first principal component corresponding to the largest eigenvalue is extracted. When solving the principal component score, a sparsity constraint is introduced. The L1 norm regularization term forces the principal component to exhibit sparsity in the spatial domain, highlighting the local abnormal features of the damaged area, and thus obtaining the fusion feature sequence of damage enhancement.

[0013] Preferably, based on the fused feature sequence, the corresponding physical area of ​​the damage is calculated using damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds, including: Based on the fused feature sequence, a candidate pool of healthy regions is constructed, and the optimal base healthy reference region is automatically selected through a multi-attribute decision algorithm, and the corresponding time-varying healthy reference baseline is calculated. Based on the time-varying health reference baseline, a dynamic threshold surface is constructed; Based on the dynamic threshold surface, the fused feature sequence is adaptively segmented pixel by pixel and frame by frame to obtain an initial value binary sequence. Based on the initial binary sequence, perform three-dimensional spatiotemporal connected domain analysis and evolution tracing to obtain the optimized binary sequence; Based on the optimized binary sequence, a weighted accumulation along the time axis is performed to obtain a damage binary map; Based on the weighted cumulative binary image, morphological adaptive repair and area quantification are performed to obtain the physical area of ​​the damage.

[0014] Preferably, based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain damage regions at each depth layer and penetrating damage regions. The output includes a three-dimensional damage characterization result containing planar distribution, depth layer attribution, and penetrating markers, including: For each damaged pixel in the damage sequence, extract the corresponding temperature difference time-series curve; Based on the temperature difference time series curve, several core time-domain features corresponding to the temperature difference time series curve are extracted; Based on the number of layers in the composite material, a template library containing the temporal characteristic range of each depth layer is pre-established; The temporal features of each damaged pixel are matched with the template library, the membership degree of the damaged pixel to each depth layer is calculated, and a depth layer priority list of damaged pixels is generated according to the membership degree from high to low. Based on the depth layer priority list of all damaged pixels, and combined with the spatial adjacency relationship between pixels, multi-layer collaborative clustering is performed to obtain the damaged region corresponding to each depth layer. For each damaged area, the pixel area corresponding to that area in the physical area of ​​the damage is corrected according to the thermal diffusivity of the layer at its depth, so as to obtain the corrected physical area. Identify and mark penetrating damage areas, record the number of penetrating layers and spatial extent, and output the corresponding three-dimensional damage characterization results.

[0015] Preferably, based on the depth layer priority list of all damaged pixels and combined with the spatial adjacency relationship between pixels, multi-layer collaborative clustering is performed to obtain the damaged regions corresponding to each depth layer, including: For each damaged pixel, the support of each depth layer is calculated based on the priority list of all pixels in its neighborhood, and the priority list of damaged pixels is reordered according to the support to obtain a smoothed priority list. Based on the smoothed priority list, the hierarchical competitive growth and penetration labeling strategy is executed on each depth layer in the order from the surface layer to the bottom layer, and the initial damage area, the corresponding main depth layer label matrix, and the penetration candidate label matrix are output for each depth layer. Based on the initial damage region corresponding to each depth layer and the corresponding main depth layer label matrix, isolated regions with an area smaller than a preset threshold are removed, and spatially adjacent regions belonging to the same depth layer are merged to obtain the final set of layered damage regions for each depth layer. Based on the penetration candidate marker matrix, spatial connectivity analysis is performed on the pixels marked as penetration candidates to obtain spatially continuous penetration candidate regions; based on the smoothed priority list, the set of depth layers covered by each penetration candidate region is determined to obtain the final set of penetration damage regions. Based on the set of layered damage regions and the set of penetrating damage regions, the final three-dimensional damage characterization result is constructed.

[0016] A second aspect of the present invention also provides a thermal imaging sequence processing system for damage characterization of hybrid braided composite materials, comprising: The sequence acquisition module is used to acquire the original thermal image sequence of the hybrid braided composite material to be tested, and to obtain the initial temperature difference sequence based on the original thermal image sequence; The preprocessing module is used to sequentially perform optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal nonuniformity correction based on Kalman filtering on the initial temperature difference sequence to obtain a preprocessed temperature difference sequence. The damage decoupling module is used to sparsely decompose each frame of the preprocessed temperature difference sequence into woven texture components and damage components based on morphological component analysis, and obtain the spatially decoupled damage sequence through iterative solution. The sequence fusion module is used to obtain a fused feature sequence based on the damage sequence through multi-feature thermal signal reconstruction and principal component analysis. The area conversion module is used to convert the physical area of ​​the damage based on the fused feature sequence through damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. The damage characterization module is used to perform depth layer attribution and multi-layer collaborative clustering based on the temperature difference time-series curves of each pixel in the damage sequence, to obtain the damage regions at each depth layer and the penetrating damage regions, and outputs a three-dimensional damage characterization result including planar distribution, depth layer attribution and penetrating label.

[0017] Compared with the prior art, the beneficial effects of the present invention are: This invention employs morphological component analysis to sparsely decompose preprocessed images into woven texture components and damage components. Spatial decoupling is achieved through iterative solving, accurately extracting the pure damage sequence. Subsequently, for the damage sequence, a multi-model competitive pool is constructed for adaptive thermal signal reconstruction, extracting multi-dimensional features such as the original temperature difference curve and its derivative. Based on this, a physical model-guided weighted principal component analysis is introduced, utilizing a confidence map to construct an adaptive weight tensor. Sparse constraints are applied during principal component solving to highlight local anomalies in the damage region. Through three-level linkage processing, the signal-to-noise ratio and spatial alignment accuracy of the thermal image sequence are significantly improved. This effectively solves the problems of complex surface textures, susceptibility to noise interference, and jitter during dynamic acquisition of hybrid woven composite materials, providing a high-quality data foundation for subsequent accurate separation of woven textures and minute damage signals, and avoiding false damage or false detections.

[0018] This invention employs morphological component analysis to sparsely decompose preprocessed images into woven texture components and damage components. Spatial decoupling is achieved through iterative solving, accurately extracting the pure damage sequence. Subsequently, for the damage sequence, a multi-model competitive pool is constructed for adaptive thermal signal reconstruction, extracting multi-dimensional features such as the original temperature difference curve and its derivative. Based on this, a physical model-guided weighted principal component analysis is introduced, utilizing a confidence map to construct an adaptive weight tensor. Sparse constraints are applied during principal component solving to highlight local anomalies in the damaged area. This effectively suppresses complex woven backgrounds (strong texture interference), solving the problem of traditional methods' difficulty in distinguishing between inherent material texture and actual damage. Through multi-feature fusion and sparse principal component analysis, the contrast between damaged and healthy areas is further amplified, making minute, shallow, or early-stage damage significantly identifiable in the feature sequence, significantly improving the sensitivity of damage detection.

[0019] This invention constructs a complete damage quantification system from planar to three-dimensional perspectives. In the planar dimension, damage segmentation based on spatiotemporal adaptive thresholds is used, and a healthy reference region is automatically selected using multi-attribute decision-making. A dynamic threshold surface is constructed and optimized by combining three-dimensional spatiotemporal connected domain analysis, ultimately yielding a precise physical area of ​​damage. In the depth dimension, based on the time-series temperature difference curves of damaged pixels, temporal features are extracted and matched with a pre-established depth layer template library. Through multi-layer collaborative clustering (combining spatial adjacency relationships and hierarchical competitive growth strategies), the depth layer attribution of damage is determined, and penetrating damage is identified. This not only achieves precise and automated quantification of damage area (avoiding the subjectivity of fixed thresholds), but also, for the first time, characterizes the three-dimensional spatial distribution (planar location + depth layer + penetration) of damage within a thermal imaging processing framework. The output results include planar distribution, depth layer attribution, and penetration markers, providing a three-dimensional, high-dimensional quantitative basis for assessing the damage severity, failure mode, and remaining life of composite materials. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0021] Figure 1 This is a schematic diagram of a thermal imaging sequence processing method for damage characterization of hybrid braided composite materials.

[0022] Figure 2 This is a schematic diagram of a thermal imaging sequence processing system for damage characterization of hybrid braided composite materials. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0024] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Example 1 like Figure 1 As shown in the figure, this embodiment discloses a thermal imaging sequence processing method for damage characterization of hybrid braided composite materials, the method comprising: The original thermal image sequence of the hybrid braided composite material to be tested is obtained, and the initial temperature difference sequence is obtained based on the original thermal image sequence; It should be noted that, based on the original thermal image sequence, the initial temperature difference sequence includes: A transient thermal pulse excitation was applied to the hybrid braided composite material sample under test using a pulsed thermal imaging device, and the original thermal image sequence of the surface of the hybrid braided composite material under test during the cooling stage after thermal excitation was continuously recorded by an infrared thermal imager at a preset acquisition frequency. ;in For pixel coordinates, For time frame index.

[0026] Extract a predetermined number of cold frames before thermal excitation from the original thermal image sequence, and calculate the pixel-by-pixel time-averaged temperature of the cold frames. Preferably, it is the first 20 to 50 frames before thermal excitation.

[0027] The average temperature is subtracted from the original thermal image sequence frame by frame and pixel by pixel to obtain the initial temperature difference sequence. .

[0028] The initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain a preprocessed temperature difference sequence. It should be noted that the initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain the preprocessed temperature difference sequence, which includes: Based on the initial temperature difference sequence, subpixel-level image registration is performed using optical flow. The inter-frame displacement vector field is calculated, and the image is resampled to eliminate pixel shifts caused by vibration, resulting in a registered temperature difference sequence. In this embodiment, subpixel-level image registration is performed using dense optical flow (such as the Farneback algorithm). The number of pyramid layers is set (preferably 3-5 layers) to handle large displacements, and the inter-frame displacement vector field is calculated. Image resampling via bicubic interpolation eliminates pixel shifts caused by excitation or environmental vibrations (typically less than 0.5 pixels), resulting in a registered temperature difference sequence. .

[0029] Based on the registered temperature difference sequence, a generalized cross-validation adaptive wavelet thresholding denoising method is used. Wavelet decomposition is performed on each frame of the image, and the noise standard deviation is estimated based on the statistical distribution of the high-frequency sub-band coefficients. The optimal soft threshold is calculated for each sub-band with the goal of minimizing the generalized cross-validation criterion. High-frequency coefficients are contracted to suppress random noise, and then reconstructed using inverse wavelet transform to obtain the denoised temperature difference sequence. Specifically, a Symlet wavelet basis (preferably order 4-8) is selected for 3-5 levels of wavelet decomposition. For the high-frequency sub-bands of each frame of the image, the noise standard deviation is estimated using robust median estimation.

[0030] Based on the denoised temperature difference sequence, temporal non-uniformity correction is performed using Kalman filtering, the pixel response is modeled as a state-space model, and the true response value of each pixel is estimated iteratively using temporal correlation to dynamically eliminate fixed pattern noise, thus obtaining a preprocessed temperature difference sequence.

[0031] Specifically, the temporal response of each pixel is modeled as a state-space model: Equations of state: The observation equation is: In the formula, where For true pixel response, For the observed values, For process noise (covariance QQ initial value set to 10−410−4). For measuring noise.

[0032] Based on morphological component analysis, each frame of the preprocessed temperature difference sequence is sparsely decomposed into woven texture component and damage component, and spatially decoupled damage sequence is obtained by iterative solution. In this embodiment, in the thermal image of the hybrid woven composite material, the woven texture (warp and weft structure) and damage (such as cracks and delamination) are highly overlapping in space. This step utilizes the morphological differences between the two (the texture is periodic and directional; the damage is localized and isotropic) to decouple them.

[0033] Based on the damage sequence, a fused feature sequence is obtained through multi-feature thermal signal reconstruction and principal component analysis. It should be noted that, based on the damage sequence, the fused feature sequence obtained through multi-feature thermal signal reconstruction and principal component analysis includes: Based on the spatial decoupling sequence, the cooling curve of each pixel is fitted by an adaptive parameter selection thermal signal reconstruction algorithm to obtain the fitted curve and its derivatives. Specifically, based on the spatial decoupling sequence, the cooling curve of each pixel is fitted using an adaptive parameter selection thermal signal reconstruction algorithm, resulting in the fitted curve and its derivatives, including: Based on the spatial decoupling sequence, a multi-model competition pool is constructed, which includes 2nd to 6th order polynomial fitting models and physical analytical models derived based on one-dimensional thermal diffusion theory. For the cooling curve of each pixel, each model in the competition pool is used for fitting. The fitting residuals of each model and its time series complexity index and physical bias index are calculated. With the goal of minimizing the joint cost function, the optimal fitting model and its parameters are automatically selected. The complexity index is measured by calculating the sample entropy of the fitting residuals, and the physical bias index is obtained by calculating the dynamic time warping distance between the fitted curve and the theoretical curve of the physical analytical model. Based on the selected best-fit model, the first and second derivatives with respect to the logarithm of time are calculated, and a confidence plot is output.

[0034] The original temperature difference curve and at least one first derivative curve are stacked along the channel dimension to construct a multi-feature cube; Based on the multi-feature cube, a fusion feature sequence is extracted through weighted principal component analysis guided by a physical model.

[0035] Specifically, based on the multi-feature cube, the extraction of fused feature sequences through weighted principal component analysis guided by a physical model includes: Based on the multi-feature cube and confidence map, a spatiotemporal adaptive weight tensor is constructed; in this embodiment, the spatiotemporal adaptive weight tensor is: Spatial dimension weights are generated based on confidence maps and local gradient features, with regions of low confidence assigned lower weights; The time dimension weights are generated based on the theoretical signal-to-noise ratio curve derived from the thermal diffusion theory, with earlier frames being assigned higher weights due to their higher signal-to-noise ratios. The channel dimension weights are generated based on the prior distribution of the sensitivity of each feature channel to damage, with the second derivative channel being more sensitive to deep damage. Based on the spatiotemporal adaptive weight tensor, a weighted centering process is performed on the multi-feature cube to obtain a weighted covariance matrix. The weighted covariance matrix is ​​decomposed into eigenvalues, and the first principal component corresponding to the largest eigenvalue is extracted. When solving the principal component score, a sparsity constraint is introduced. The L1 norm regularization term forces the principal component to exhibit sparsity in the spatial domain, highlighting the local abnormal features of the damaged area, and thus obtaining the fusion feature sequence of damage enhancement.

[0036] Based on the fused feature sequence, the corresponding physical area of ​​the damage is calculated by using damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. It should be noted that, based on the fused feature sequence, and through damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds, the corresponding physical damage area is calculated as follows: Based on the fused feature sequence, a candidate pool of healthy regions is constructed, and the optimal base healthy reference region is automatically selected through a multi-attribute decision algorithm, and the corresponding time-varying healthy reference baseline is calculated. In this embodiment, the sequence is processed by spatiotemporal block segmentation, and the energy decay curve of each spatiotemporal block is extracted. The curve smoothness, similarity with the theoretical heat diffusion curve, and local spatial consistency are used as multi-dimensional screening indicators. The optimal healthy reference region is automatically selected through the TOPSIS multi-attribute decision algorithm, and the signal mean and standard deviation of all pixels in the region at each time frame are calculated to obtain the time-varying healthy reference baseline. Based on the time-varying health reference baseline, a dynamic threshold surface is constructed. In this embodiment, based on the time-varying law of signal-to-noise ratio derived from the heat diffusion theory, a time evolution model of the threshold coefficient is established, and a dynamic threshold surface is generated by combining the time-varying standard deviation.

[0037] Based on the dynamic threshold surface, the fused feature sequence is adaptively segmented pixel by pixel and frame by frame to obtain an initial value binary sequence; specifically, the signal value of each pixel at each time step is compared with the threshold at the corresponding time step to obtain the initial value binary sequence.

[0038] Based on the initial binary sequence, three-dimensional spatiotemporal connected domain analysis and evolution tracking are performed to obtain the optimized binary sequence. In this embodiment, the binary sequence is regarded as a three-dimensional spatiotemporal volume, and the connected domain is extracted using the 26-neighborhood connectivity criterion. For each connected domain, its morphological evolution trajectory is tracked along the time axis, and dynamic features such as growth rate, duration, and area change rate are calculated. False damages with excessively fast growth rate (suspected noise) or excessively short duration (suspected transient interference) are removed to obtain the optimized binary sequence.

[0039] Based on the optimized binary sequence, a weighted accumulation is performed along the time axis to obtain a damage binary map. Specifically, the signal-to-noise ratio weight of each time frame is considered during accumulation, and higher accumulation weights are given to earlier frames with high signal-to-noise ratios to obtain a weighted accumulated binary map.

[0040] Based on the weighted cumulative binary image, morphological adaptive repair and area quantification are performed to obtain the physical area of ​​the damage. Specifically, according to the thermal diffusion characteristics of the damaged area, anisotropic structural elements are used for morphological closing operations, with the major axis of the structural elements aligned with the main thermal diffusion direction; for the repaired connected regions, the physical area is calculated based on pixel area.

[0041] Based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain the damage regions of each depth layer and the penetrating damage regions. The output is a three-dimensional damage characterization result that includes planar distribution, depth layer attribution and penetrating markers.

[0042] It should be noted that, based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain the damage regions at each depth layer and the penetrating damage regions. The output includes a three-dimensional damage characterization result containing planar distribution, depth layer attribution, and penetrating markers, including: For each damaged pixel in the damage sequence, extract the corresponding temperature difference time-series curve; Based on the temperature difference time series curve, several core time-domain features corresponding to the temperature difference time series curve are extracted; In this embodiment, the core time-domain features are shown in Table 1: Table 1. Timing Characteristics Based on the number of layers in the composite material, a template library containing the temporal characteristic range of each depth layer is pre-established. In this embodiment, based on the number of layers in the composite material, a template library containing the temporal characteristic range of each depth layer is pre-established through calibration experiments. The calibration experiments use artificial defect specimens with known depths, and their temperature difference time series curves are obtained under the same thermal imaging detection conditions to statistically analyze the temporal characteristic distribution range of each depth layer.

[0043] The temporal features of each damaged pixel are matched with the template library to calculate the membership degree of the damaged pixel to each depth layer, and a depth layer priority list of damaged pixels is generated according to the membership degree from high to low. In this embodiment, the membership degree of each depth layer is calculated using the reciprocal of the weighted Euclidean distance.

[0044] Based on the depth layer priority list of all damaged pixels, and combined with the spatial adjacency relationship between pixels, multi-layer collaborative clustering is performed to obtain the damaged region corresponding to each depth layer. Specifically, based on the depth layer priority list of all damaged pixels and combined with the spatial adjacency relationships between pixels, multi-layer collaborative clustering is performed to obtain the damaged regions corresponding to each depth layer, including: For each damaged pixel, the support of each depth layer is calculated based on the priority list of all pixels in its neighborhood, and the priority list of damaged pixels is reordered according to the support to obtain a smoothed priority list. Based on the smoothed priority list, the hierarchical competitive growth and penetration labeling strategy is executed on each depth layer in the order from the surface layer to the bottom layer, and the initial damage area, the corresponding main depth layer label matrix, and the penetration candidate label matrix are output for each depth layer. In this embodiment, the hierarchical competitive growth and penetration marking strategy includes: Based on the ranking threshold of the current depth layer, select pixels from all pixels that are not assigned to the main depth layer that satisfy the rank less than or equal to the ranking threshold to obtain the seed point set of the current depth layer. Based on the seed point set, region growth is performed using four-neighbor or eight-neighbor connectivity. During the growth process, the initial damaged region of the current depth layer is obtained based on the constraint that adjacent pixels also meet the ranking conditions and are not assigned to the main depth layer. Based on the growth results, all pixels within the initial damage area are assigned to the main depth layer to obtain the updated main depth layer label matrix; Based on the adjacent pixels of the assigned main depth layer encountered during the growth process, they are marked as penetration candidates, resulting in an updated penetration candidate flag matrix.

[0045] Based on the initial damage region corresponding to each depth layer and the corresponding main depth layer label matrix, isolated regions with an area smaller than a preset threshold are removed, and spatially adjacent regions belonging to the same depth layer are merged to obtain the final set of layered damage regions for each depth layer. Based on the penetration candidate marker matrix, spatial connectivity analysis is performed on the pixels marked as penetration candidates to obtain spatially continuous penetration candidate regions; based on the smoothed priority list, the depth layer set covered by each penetration candidate region is determined to obtain the final set of penetration damage regions; in this embodiment, based on the penetration candidate marker matrix, all pixels marked as penetration candidates are extracted to obtain a set of penetration candidate pixels. Spatial connectivity analysis is performed on the set of through-hole candidate pixels to obtain spatially continuous through-hole candidate regions; Based on the smoothed priority list, the set of depth layers covered by each penetration candidate region (i.e., the depth layers in which all pixels in the region meet the ranking conditions) is determined, and the final set of penetration damage regions is obtained. Each penetration region records the depth layer number and spatial range it covers. Based on the set of layered damage regions and the set of penetrating damage regions, the final three-dimensional damage characterization result is constructed.

[0046] For each damaged area, the pixel area corresponding to that area in the physical area of ​​the damage is corrected according to the thermal diffusivity of the layer at its depth, so as to obtain the corrected physical area. Identify and mark penetrating damage areas, record the number of penetrating layers and spatial extent, and output the corresponding three-dimensional damage characterization results.

[0047] This embodiment also discloses a thermal imaging sequence processing system for damage characterization of hybrid braided composite materials, including: The sequence acquisition module is used to acquire the original thermal image sequence of the hybrid braided composite material to be tested, and to obtain the initial temperature difference sequence based on the original thermal image sequence; The preprocessing module is used to sequentially perform optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal nonuniformity correction based on Kalman filtering on the initial temperature difference sequence to obtain a preprocessed temperature difference sequence. The damage decoupling module is used to sparsely decompose each frame of the preprocessed temperature difference sequence into woven texture components and damage components based on morphological component analysis, and obtain the spatially decoupled damage sequence through iterative solution. The sequence fusion module is used to obtain a fused feature sequence based on the damage sequence through multi-feature thermal signal reconstruction and principal component analysis. The area conversion module is used to convert the physical area of ​​the damage based on the fused feature sequence through damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. The damage characterization module is used to perform depth layer attribution and multi-layer collaborative clustering based on the temperature difference time-series curves of each pixel in the damage sequence, to obtain the damage regions at each depth layer and the penetrating damage regions, and outputs a three-dimensional damage characterization result including planar distribution, depth layer attribution and penetrating label.

[0048] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0049] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0050] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0051] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0052] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0053] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0054] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0055] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A thermal imaging sequence processing method for damage characterization of hybrid braided composite materials, characterized in that, The method includes: The original thermal image sequence of the hybrid braided composite material to be tested is obtained, and the initial temperature difference sequence is obtained based on the original thermal image sequence; The initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain a preprocessed temperature difference sequence. Based on morphological component analysis, each frame of the preprocessed temperature difference sequence is sparsely decomposed into woven texture component and damage component, and spatially decoupled damage sequence is obtained by iterative solution. Based on the damage sequence, a fused feature sequence is obtained through multi-feature thermal signal reconstruction and principal component analysis. Based on the fused feature sequence, the corresponding physical area of ​​the damage is calculated by using damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. Based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain the damage regions of each depth layer and the penetrating damage regions. The output is a three-dimensional damage characterization result that includes planar distribution, depth layer attribution and penetrating markers.

2. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 1, characterized in that, Based on the original thermal image sequence, the initial temperature difference sequence is obtained as follows: A transient thermal pulse excitation was applied to the hybrid braided composite material sample under test using a pulsed thermal imaging device, and the original thermal image sequence of the surface of the hybrid braided composite material under test during the cooling stage after thermal excitation was continuously recorded by an infrared thermal imager at a preset acquisition frequency. Extract a predetermined number of cold frames before thermal excitation from the original thermal image sequence, and calculate the pixel-by-pixel time-averaged temperature of the cold frames; The average temperature is subtracted from the original thermal image sequence frame by frame and pixel by pixel to obtain the initial temperature difference sequence.

3. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 2, characterized in that, The initial temperature difference sequence is sequentially subjected to optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal non-uniformity correction based on Kalman filtering to obtain a preprocessed temperature difference sequence including: Based on the initial temperature difference sequence, subpixel-level image registration is performed using optical flow, the inter-frame displacement vector field is calculated, and the image is resampled to eliminate pixel shift caused by vibration, thus obtaining the registered temperature difference sequence. Based on the registered temperature difference sequence, the image is decomposed into wavelets by generalized cross-validation adaptive wavelet threshold denoising. The noise standard deviation is estimated according to the statistical distribution of the high-frequency subband coefficients. The optimal soft threshold is calculated for each subband with the goal of minimizing the generalized cross-validation criterion. The high-frequency coefficients are shrunk to suppress random noise. The image is then reconstructed by inverse wavelet transform to obtain the denoised temperature difference sequence. Based on the denoised temperature difference sequence, temporal non-uniformity correction is performed using Kalman filtering, the pixel response is modeled as a state-space model, and the true response value of each pixel is estimated iteratively using temporal correlation to dynamically eliminate fixed pattern noise, thus obtaining a preprocessed temperature difference sequence.

4. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 1, characterized in that, Based on the damage sequence, the fused feature sequence obtained through multi-feature thermal signal reconstruction and principal component analysis includes: Based on the spatial decoupling sequence, the cooling curve of each pixel is fitted by an adaptive parameter selection thermal signal reconstruction algorithm to obtain the fitted curve and its derivatives. The original temperature difference curve and at least one first derivative curve are stacked along the channel dimension to construct a multi-feature cube; Based on the multi-feature cube, a fusion feature sequence is extracted through weighted principal component analysis guided by a physical model.

5. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 4, characterized in that, Based on the spatial decoupling sequence, the cooling curve of each pixel is fitted using an adaptive parameter selection thermal signal reconstruction algorithm, resulting in the fitted curve and its derivatives, including: Based on the spatial decoupling sequence, a multi-model competition pool is constructed, which includes 2nd to 6th order polynomial fitting models and physical analytical models derived based on one-dimensional thermal diffusion theory. For the cooling curve of each pixel, each model in the competition pool is used for fitting. The fitting residuals of each model and its time series complexity index and physical bias index are calculated. With the goal of minimizing the joint cost function, the optimal fitting model and its parameters are automatically selected. The complexity index is measured by calculating the sample entropy of the fitting residuals, and the physical bias index is obtained by calculating the dynamic time warping distance between the fitted curve and the theoretical curve of the physical analytical model. Based on the selected best-fit model, the first and second derivatives with respect to the logarithm of time are calculated, and a confidence plot is output.

6. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 5, characterized in that, Based on the multi-feature cube, the extraction of fused feature sequences through weighted principal component analysis guided by a physical model includes: Based on the multi-feature cube and confidence map, a spatiotemporal adaptive weight tensor is constructed; Based on the spatiotemporal adaptive weight tensor, a weighted centering process is performed on the multi-feature cube to obtain a weighted covariance matrix. The weighted covariance matrix is ​​decomposed into eigenvalues, and the first principal component corresponding to the largest eigenvalue is extracted. When solving the principal component score, a sparsity constraint is introduced. The L1 norm regularization term forces the principal component to exhibit sparsity in the spatial domain, highlighting the local abnormal features of the damaged area, and thus obtaining the fusion feature sequence of damage enhancement.

7. The thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 1, characterized in that, Based on the fused feature sequence, the corresponding physical area of ​​the damage is calculated using damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds, including: Based on the fused feature sequence, a candidate pool of healthy regions is constructed, and the optimal base healthy reference region is automatically selected through a multi-attribute decision algorithm, and the corresponding time-varying healthy reference baseline is calculated. Based on the time-varying health reference baseline, a dynamic threshold surface is constructed; Based on the dynamic threshold surface, the fused feature sequence is adaptively segmented pixel by pixel and frame by frame to obtain an initial value binary sequence. Based on the initial binary sequence, three-dimensional spatiotemporal connected domain analysis and evolution tracing are performed to obtain the optimized binary sequence. Based on the optimized binary sequence, a weighted accumulation along the time axis is performed to obtain a damage binary map; Based on the weighted cumulative binary image, morphological adaptive repair and area quantification are performed to obtain the physical area of ​​the damage.

8. A thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 7, characterized in that, Based on the time-series temperature difference curves of each pixel in the damage sequence, depth layer attribution and multi-layer collaborative clustering are performed to obtain damage regions at each depth layer and through-damage regions. The output includes a three-dimensional damage characterization result containing planar distribution, depth layer attribution, and through-damage markers, including: For each damaged pixel in the damage sequence, extract the corresponding temperature difference time-series curve; Based on the temperature difference time series curve, several core time-domain features corresponding to the temperature difference time series curve are extracted; Based on the number of layers in the composite material, a template library containing the temporal characteristic range of each depth layer is pre-established; The temporal features of each damaged pixel are matched with the template library to calculate the membership degree of the damaged pixel to each depth layer, and a depth layer priority list of damaged pixels is generated according to the membership degree from high to low. Based on the depth layer priority list of all damaged pixels, and combined with the spatial adjacency relationship between pixels, multi-layer collaborative clustering is performed to obtain the damaged region corresponding to each depth layer. For each damaged area, the pixel area corresponding to that area in the physical area of ​​the damage is corrected according to the thermal diffusivity of the layer at its depth, so as to obtain the corrected physical area. Identify and mark penetrating damage areas, record the number of penetrating layers and spatial extent, and output the corresponding three-dimensional damage characterization results.

9. A thermal imaging sequence processing method for damage characterization of hybrid woven composite materials according to claim 8, characterized in that, Based on the depth layer priority list of all damaged pixels, and combined with the spatial adjacency relationship between pixels, multi-layer collaborative clustering is performed to obtain the damaged regions corresponding to each depth layer, including: For each damaged pixel, the support of each depth layer is calculated based on the priority list of all pixels in its neighborhood, and the priority list of damaged pixels is reordered according to the support to obtain a smoothed priority list. Based on the smoothed priority list, the hierarchical competitive growth and penetration labeling strategy is executed on each depth layer in the order from the surface layer to the bottom layer, and the initial damage area, the corresponding main depth layer label matrix, and the penetration candidate label matrix are output for each depth layer. Based on the initial damage region corresponding to each depth layer and the corresponding main depth layer label matrix, isolated regions with an area smaller than a preset threshold are removed, and spatially adjacent regions belonging to the same depth layer are merged to obtain the final set of layered damage regions for each depth layer. Based on the penetration candidate marker matrix, spatial connectivity analysis is performed on the pixels marked as penetration candidates to obtain spatially continuous penetration candidate regions; based on the smoothed priority list, the set of depth layers covered by each penetration candidate region is determined to obtain the final set of penetration damage regions. Based on the set of layered damage regions and the set of penetrating damage regions, the final three-dimensional damage characterization result is constructed.

10. A thermal imaging sequence processing system for damage characterization of hybrid braided composite materials, implementing the thermal imaging sequence processing method for damage characterization of hybrid braided composite materials as described in any one of claims 1 to 9, characterized in that, include: The sequence acquisition module is used to acquire the original thermal image sequence of the hybrid braided composite material to be tested, and to obtain the initial temperature difference sequence based on the original thermal image sequence; The preprocessing module is used to sequentially perform optical flow subpixel registration, adaptive wavelet threshold denoising based on generalized cross-validation, and temporal nonuniformity correction based on Kalman filtering on the initial temperature difference sequence to obtain a preprocessed temperature difference sequence. The damage decoupling module is used to sparsely decompose each frame of the preprocessed temperature difference sequence into woven texture components and damage components based on morphological component analysis, and obtain the spatially decoupled damage sequence through iterative solution. The sequence fusion module is used to obtain a fused feature sequence based on the damage sequence through multi-feature thermal signal reconstruction and principal component analysis. The area conversion module is used to convert the physical area of ​​the damage based on the fused feature sequence through damage segmentation and area quantification rules based on spatiotemporal adaptive thresholds. The damage characterization module is used to perform depth layer attribution and multi-layer collaborative clustering based on the temperature difference time-series curves of each pixel in the damage sequence, to obtain the damage regions at each depth layer and the penetrating damage regions, and outputs a three-dimensional damage characterization result including planar distribution, depth layer attribution and penetrating label.

Images