Orthogonal three-dimensional braiding analysis method based on CT three-dimensional reconstruction and threshold fitting

Abstract

The application discloses a kind of orthogonal three-dimensional weave analysis methods based on CT three-dimensional reconstruction and threshold fitting, it is related to weave technical field, comprising: while following the physical constraint of X-ray imaging, generate three-dimensional voxel model using latent diffusion priori;Reserve orthogonal three-dimensional weave main body to generate purified three-dimensional voxel model;Get calibrated three-dimensional voxel model;Separate weave main stress area and generate each analysis area three-dimensional voxel model;Set continuous gray threshold interval and with fixed step length scan each analysis area three-dimensional voxel model, form threshold-volume fraction data set;According to threshold-volume fraction data set, construct threshold proportion trend curve, get fitting deviation index;According to fitting deviation index, generate orthogonal three-dimensional weave internal infiltration quality evaluation result.The application improves the imaging quality and authenticity of orthogonal three-dimensional weave microstructure.

Description

Technical Field

[0001] This invention relates to the field of knitted fabric technology, and in particular to an orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting. Background Technology

[0002] With the widespread application of composite materials in aerospace, rail transportation and high-end equipment, higher requirements have been placed on their internal microstructure and quality control. Orthogonal triaxial braided fabric, as a typical high-performance composite material, has a complex internal structure, and the yarn distribution and resin impregnation state directly affect the mechanical properties of the final material.

[0003] Currently, non-destructive testing technology based on X-ray computed tomography (CT) has become a common method for analyzing the internal structure of woven fabrics. However, to maintain the efficiency of industrial production, large inter-layer spacing is commonly used in actual testing, leading to insufficient resolution and blurred microscopic details in the acquired data. Existing 3D reconstruction algorithms struggle to clearly reconstruct the fine structure of woven fabrics under low-resolution conditions, often resulting in partial volumetric effects, grayscale mixing between yarns and the matrix, metallic artifacts, and unclear inter-layer boundaries. While simply increasing the scanning resolution can improve image quality, it significantly increases scanning time and data volume, making it unsuitable for the actual needs of mass production.

[0004] Existing methods for evaluating internal defects and porosity generally rely on a single grayscale threshold for volume segmentation and porosity calculation. However, due to the influence of noise and artifacts on the grayscale of CT images, the fixed threshold method is poorly adaptable to interface blurring, gradient transitions, and local density changes, which can easily lead to subjective errors and misjudgments. It cannot accurately reflect the internal density distribution of the woven fabric and the gradual characteristics of resin impregnation. Furthermore, it lacks a unified spatial registration process and an automatic region segmentation mechanism, resulting in insufficient comparability and repeatability of analysis between samples from the same batch. Summary of the Invention

[0005] One objective of this invention is to propose an orthogonal triaxial woven fabric analysis method based on CT three-dimensional reconstruction and threshold fitting. This invention improves the imaging quality and realism of the microstructure of orthogonal triaxial woven fabrics, breaks through the physical limits of traditional FDK and iterative reconstruction algorithms, and enhances the applicability and universality of CT detection in industrial batch scenarios.

[0006] An orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to an embodiment of the present invention includes:

[0007] The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device. Continuous X-ray projection data is obtained according to the preset scanning interval. The continuous X-ray projection data is input into the physical-guided potential diffusion posterior sampling model. While following the physical constraints of X-ray imaging, the potential diffusion prior is used to generate a three-dimensional voxel model.

[0008] Digital envelope processing is performed on the three-dimensional voxel model to identify and remove the core mold structure, while retaining the orthogonal triaxial woven body to generate a purified three-dimensional voxel model.

[0009] The purification 3D voxel model is rigidly registered with the preset standard part model to establish a unified spatial coordinate system and obtain the calibrated 3D voxel model.

[0010] Based on a unified spatial coordinate system, the calibration three-dimensional voxel model is programmed to cut, the main stress areas of the woven fabric are separated and three-dimensional voxel models of each analysis area are generated.

[0011] Set continuous grayscale threshold intervals and scan the three-dimensional voxel model of each analysis area with a fixed step size. Calculate the volume fraction of the number of voxels in each threshold interval relative to the number of voxels in the effective grayscale range to form a threshold-volume fraction dataset.

[0012] A threshold proportion trend curve is constructed based on the threshold-volume fraction dataset, and the trend curve is fitted and compared with a pre-stored standard trend curve to obtain a fitting deviation index.

[0013] The results of the internal impregnation quality assessment of the orthogonal triaxial woven fabric are generated based on the fitting deviation index, and an analysis report is output.

[0014] Optionally, inputting continuous X-ray projection data into a physically guided potential diffusion posterior sampling model includes:

[0015] The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device, and the scanning interlayer spacing is set to the preset scanning interlayer spacing parameter. Continuous X-ray projection measurement results are obtained at multiple scanning angles to form continuous X-ray projection raw data.

[0016] Physical consistency transformation is performed on the raw data of continuous X-ray projection, and the transmission intensity value of the k-th pixel of the detector and the corresponding incident reference intensity value are collected.

[0017] Based on the transmitted intensity value and the incident reference intensity value, logarithmic projection data is calculated, and all logarithmic projection data are combined to form a continuous X-ray projection dataset.

[0018] The continuous X-ray projection dataset is input into the physical guided potential diffusion model, and the three-dimensional voxel model is defined as a voxel field with linear decay coefficient in the physical guided potential diffusion model.

[0019] In the physical-guided potential diffusion model, a forward projection operator is defined to map the voxel field of linear decay coefficients to a predicted projection dataset.

[0020] Introduce latent variables into the physics-guided latent diffusion model and define a decoding operator;

[0021] A data consistency metric is constructed based on continuous X-ray projection datasets and predicted projection datasets.

[0022] Based on the pre-trained potential diffusion prior density and data consistency metric, reverse stochastic differential equation posterior sampling is performed to obtain the latent variables corresponding to the end of the reverse sampling.

[0023] The decoding operator is executed on the latent variables corresponding to the end of the backsampling to generate a three-dimensional voxel model.

[0024] Optionally, performing digital envelope processing on the three-dimensional voxel model includes:

[0025] In a voxel field with linear decay coefficient, if the linear decay coefficient of a voxel is greater than or equal to the effective voxel determination threshold parameter, the corresponding voxel is determined to be an effective voxel; otherwise, the corresponding voxel is determined not to be an effective voxel, and an effective voxel indicator field is generated for all voxels.

[0026] Based on the effective voxel indicator field, expansion and erosion operations are performed using three-dimensional discrete structural elements to obtain the digital envelope voxel mask.

[0027] Based on the digital envelope voxel mask, the envelope of the linear decay coefficient voxel field is clipped to form an envelope-constrained linear decay coefficient voxel field.

[0028] In the envelope-constrained linear decay coefficient voxel field, by judging whether the linear decay coefficient of the voxel is between the lower threshold and the upper threshold of the linear decay coefficient of the core model, and whether the voxel is within the connected domain determined by the spatial connectivity judgment parameter, all voxels that meet the conditions are selected, and all selected voxels constitute the core model candidate voxel set.

[0029] Generate a core phantom voxel mask based on the core phantom candidate voxel set;

[0030] The core voxel mask is removed from the envelope-constrained linear decay coefficient voxel field to generate a purified three-dimensional voxel model.

[0031] Optionally, the rigid body registration of the purified 3D voxel model with the preset standard part model includes:

[0032] Based on the voxel index of the purified linear decay coefficient voxel field, establish the physical coordinates of each voxel.

[0033] A rigid body transformation is established between the purified linear decay coefficient voxel field and the standard linear decay coefficient voxel field to obtain the corresponding physical coordinates under the standard linear decay coefficient voxel field.

[0034] Based on rigid body transformation, a resampling operation is performed on the voxel field of the purified linear attenuation coefficient to obtain the resampled voxel field of the linear attenuation coefficient.

[0035] Based on the resampled linear decay coefficient voxel field, a registration cost function is constructed within the common domain of the standard linear decay coefficient voxel field;

[0036] Under the orthogonal constraint that the orthogonal rotation matrix and translation vector must satisfy, the optimal rigid body transformation parameter set that minimizes the registration cost function is solved. The optimal rigid body transformation parameter set is then applied to the voxel field of the purified linear attenuation coefficient, and the calibration three-dimensional voxel model in a unified spatial coordinate system is obtained by using the interpolation operator.

[0037] Optionally, the programmed cutting of the calibrated 3D voxel model based on a unified spatial coordinate system includes:

[0038] Based on a unified spatial coordinate system, a set of cutting coordinate parameters for programmed cutting is established in the voxel field of the calibration linear attenuation coefficient. The set of cutting coordinate parameters consists of at least one set of cutting plane parameters.

[0039] The equation of the cutting plane is defined based on the normal vector parameters, offset parameters, and voxel physical coordinates of the cutting plane parameters.

[0040] Based on the cutting plane equation, the voxel field of the calibration linear attenuation coefficient is classified into voxels. According to the sign relationship between the physical coordinates of each voxel and the cutting plane equation, a region label is assigned to the voxel, and a region label field is generated.

[0041] Based on the region label field, the voxel field of the calibration linear attenuation coefficient is extracted to generate a region mask for each target region label. For each target region label, the region mask is multiplied with the voxel field of the calibration linear attenuation coefficient in a voxel-by-voxel dimension to obtain the corresponding three-dimensional voxel model of the analysis region.

[0042] Optionally, the step of setting a continuous grayscale threshold range and scanning the three-dimensional voxel model of each analysis area with a fixed step size includes:

[0043] In the voxel field of the linear decay coefficient in each analysis region, a continuous gray-scale threshold sequence is constructed by gradually increasing the effective gray-scale range lower limit threshold parameter with a fixed step size parameter.

[0044] In each analysis region, all voxels are traversed one by one within the voxel field of the linear decay coefficient, and an effective gray voxel indicator field is generated within the voxel field of the linear decay coefficient of the analysis region.

[0045] Based on the effective gray voxel indicator field, the number of all voxels belonging to the effective gray range within the region linear decay coefficient voxel field is statistically analyzed to obtain the total number of effective voxels.

[0046] Using a fixed step size parameter, each gray level threshold in the continuous gray level threshold sequence is used as the current upper limit of the threshold to construct the cumulative threshold interval, and a cumulative threshold voxel indicator field is generated in the voxel field of the linear decay coefficient in the analysis region.

[0047] For each cumulative threshold interval, the sum of all elements with a value of one in the cumulative threshold voxel indicator field is counted to obtain the number of cumulative threshold voxels in the current cumulative threshold interval.

[0048] For each cumulative threshold interval, the volume fraction corresponding to the current cumulative threshold interval is obtained by dividing the number of cumulative threshold voxels in the current cumulative threshold interval by the total number of effective voxels.

[0049] The continuous grayscale threshold sequence is paired with the corresponding volume fraction sequence in a one-to-one correspondence to form a threshold-volume fraction dataset.

[0050] Optionally, constructing the threshold proportion trend curve based on the threshold-volume fraction dataset includes:

[0051] Based on the threshold-volume fraction dataset corresponding to the target region label, construct a threshold proportion trend curve;

[0052] Under the condition that the grayscale threshold sampling points are aligned one by one, the threshold proportion trend curve is fitted and compared with the pre-stored standard trend curve. All discrete grayscale threshold sampling points are used as the set of sampling points for fitting and comparison, and a fitting residual sequence is constructed.

[0053] The fitting deviation index is calculated based on the fitted residual sequence.

[0054] Optionally, generating the internal impregnation quality assessment result of the orthogonal triaxial knitted fabric based on the fitting deviation index includes:

[0055] Construct a set of internal infiltration quality level rules for classifying fitting deviation indices, and define the internal infiltration quality level range of orthogonal triaxial knitted fabrics by combining preset evaluation experience;

[0056] The fitting deviation index corresponding to each analysis region is matched with the internal wetting quality grade range of the orthogonal triaxial woven fabric to determine the internal wetting quality grade label of the current analysis region.

[0057] Based on the internal infiltration quality level label of each analysis area, construct an internal infiltration quality level distribution map under a unified spatial coordinate system;

[0058] Summarize the fit deviation index, internal infiltration quality level label, threshold-volume fraction dataset, and threshold proportion trend curve for all analysis regions, and output a standardized analysis report.

[0059] Optionally, the internal impregnation quality grade range and internal impregnation quality grade label of the orthogonal triaxial woven fabric include:

[0060] When the fitting deviation index is less than or equal to the first fitting deviation classification threshold parameter, it is judged as excellent infiltration.

[0061] When the fitting deviation index is greater than the first fitting deviation classification threshold parameter and less than or equal to the second fitting deviation classification threshold parameter, it is judged as good immersion.

[0062] When the fitting deviation index is greater than the second fitting deviation classification threshold parameter and less than or equal to the third fitting deviation classification threshold parameter, it is determined to be the infiltration critical point;

[0063] When the fitting deviation index is greater than the third fitting deviation classification threshold parameter, it is judged as poor immersion level.

[0064] The beneficial effects of this invention are:

[0065] The physical-guided latent diffusion model of this invention is used for high-precision 3D reconstruction of sparse projection data in CT, achieving a balance between high efficiency and high fidelity. By introducing the physical-guided latent diffusion model, the latent diffusion prior is combined with the physical constraints of X-ray imaging. Through a posterior sampling process driven by inverse stochastic differential equations, it can complete the interlayer structural details under sparse or low-resolution scanning conditions, effectively suppressing some volume effects, metal artifacts, and yarn / matrix gray-scale mixing problems, and obtaining a 3D voxel model close to high-resolution scanning. It balances scanning efficiency and structural resolution, and can improve the imaging quality and realism of orthogonal triaxial woven fabric microstructures without increasing the number of scanning layers. It breaks through the physical limits of traditional FDK and iterative reconstruction algorithms, and improves the applicability and universality of CT inspection in industrial batch scenarios.

[0066] This invention comprehensively characterizes the internal density distribution and resin impregnation state of materials through multi-level progressive grayscale threshold ratio fitting, improving the objectivity and robustness of evaluation. It proposes to construct a continuous threshold-volume fraction dataset by scanning grayscale intervals with a fixed step size, and generate a threshold ratio trend curve based on it. By fitting and comparing with a standard trend curve, it quantitatively analyzes the continuous distribution of internal density and impregnation state in different regions, avoiding the sensitivity of traditional single threshold methods to noise and grayscale fluctuations. It can effectively capture the density gradient and transition characteristics from resin-rich areas to fiber-dense areas, achieve statistical smoothing of local grayscale changes in CT images, and achieve quantitative evaluation of process consistency through fitting deviation index, improving the resolution and automation level of fabric defects and quality criteria.

[0067] This invention unifies the spatial coordinates of the purification voxel model and standard parts based on three-dimensional rigid body registration. Combined with programmed cutting technology using multi-plane parameters, it can automatically separate the main stress areas of orthogonal triaxial woven fabrics. Each area independently generates threshold-volume fraction data, trend curves, and internal impregnation quality grade labels. It also supports the automatic generation and structured export of three-dimensional grade distribution maps and standardized analysis reports, improving the batch comparability of analysis, on-site process traceability, and quality feedback efficiency. Attached Figure Description

[0068] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0069] Figure 1 This is a flowchart of an orthogonal triaxial woven fabric analysis method based on CT three-dimensional reconstruction and threshold fitting proposed in this invention;

[0070] Figure 2 This is a single CT scan result image of continuous X-ray projection data of an orthogonal triaxial woven fabric analysis method based on CT three-dimensional reconstruction and threshold fitting proposed in this invention.

[0071] Figure 3 This is an overall topographic image of a three-dimensional voxel model of an orthogonal triaxial woven fabric analysis method based on CT three-dimensional reconstruction and threshold fitting proposed in this invention.

[0072] Figure 4 This is an overall topographic image of a purified three-dimensional voxel model based on an orthogonal triaxial woven fabric analysis method based on CT three-dimensional reconstruction and threshold fitting proposed in this invention. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0074] refer to Figures 1-4 As shown in Example 1: An orthogonal triaxial knitted fabric analysis method based on CT 3D reconstruction and threshold fitting, comprising:

[0075] The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device. Continuous X-ray projection data is obtained according to the preset scanning interval. The continuous X-ray projection data is input into the physical-guided potential diffusion posterior sampling model. While following the physical constraints of X-ray imaging, the potential diffusion prior is used to generate a three-dimensional voxel model.

[0076] In this embodiment, inputting continuous X-ray projection data into a physically guided potential diffusion posterior sampling model includes:

[0077] The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device, and the scanning interlayer spacing is set to the preset scanning interlayer spacing parameter. Continuous X-ray projection measurement results are obtained at multiple scanning angles to form continuous X-ray projection raw data.

[0078] Physical consistency transformation is performed on the raw data of continuous X-ray projection, and the transmission intensity value of the k-th pixel of the detector and the corresponding incident reference intensity value are collected.

[0079] Based on the transmitted intensity value and the incident reference intensity value, logarithmic projection data is calculated, and all logarithmic projection data are combined to form a continuous X-ray projection dataset.

[0080] Logarithmic projection data is obtained by taking the logarithm and negative of the ratio of each transmitted intensity value to the corresponding incident reference intensity value. The continuous X-ray projection dataset consists of logarithmic projection data corresponding to all scanning angles and all detector pixels.

[0081] The continuous X-ray projection dataset is input into the physical guided potential diffusion model, and the three-dimensional voxel model is defined as a voxel field with linear decay coefficient in the physical guided potential diffusion model.

[0082] In Example 1, a continuous X-ray projection dataset is input into a physically guided potential diffusion model. Under the combined effect of potential diffusion priors and physical constraints of X-ray imaging, a posteriori sampling inference is performed using inverse stochastic differential equations to obtain a linear attenuation coefficient voxel field. The linear attenuation coefficient voxel field is a three-dimensional array, and each voxel value represents the linear attenuation coefficient at the corresponding spatial location, serving as a three-dimensional voxel model of the orthogonal triaxial weave.

[0083] In the physical-guided potential diffusion model, a forward projection operator is defined to map the voxel field of linear decay coefficients to a predicted projection dataset.

[0084] In Example 1, in the physical-guided potential diffusion model, the forward projection operator is defined as a set of operators based on X-ray imaging geometry and physical parameters. The input is a voxel field with linear attenuation coefficients. The linear attenuation coefficients of each spatial location of the voxel field with linear attenuation coefficients are accumulated and integrated along all the set ray paths. After completing the integration operation for each set of ray paths, the integral value under each ray path is output. The set of integral values ​​under all ray paths constitutes the prediction projection dataset.

[0085] Introduce latent variables into the physics-guided latent diffusion model and define a decoding operator;

[0086] In Example 1, in the physical-guided potential diffusion model, by jointly constraining the continuous X-ray projection dataset and the pre-trained potential diffusion prior density, the potential variables at the current diffusion time are obtained by sequentially sampling using the inverse stochastic differential equation. Each potential variable corresponds to a unique diffusion time marker. The decoding operator is used to map the potential variables to the corresponding intermediate linear decay coefficient voxel field, thereby realizing the reconstruction of the voxel field during the diffusion process.

[0087] A data consistency metric is constructed based on continuous X-ray projection datasets and predicted projection datasets.

[0088] The difference between each predicted projection data in the predicted projection dataset and the corresponding observed projection data in the continuous X-ray projection dataset is calculated. All differences are squared and then summed to obtain the data consistency metric. The data consistency metric is used to measure the consistency between the predicted projection dataset and the continuous X-ray projection dataset.

[0089] Based on the pre-trained potential diffusion prior density and data consistency metric, reverse stochastic differential equation posterior sampling is performed to obtain the latent variables corresponding to the end of the reverse sampling.

[0090] In Example 1, based on the pre-trained latent diffusion prior density and data consistency metric, backward stochastic differential equation posterior sampling is performed on the latent variables. The evolution direction of the latent variables in the latent space is determined by the pre-trained latent diffusion prior density, and the data consistency metric is used to constrain the consistency between the latent variable generation results and the continuous X-ray projection dataset. In each sampling iteration, the latent variables are gradient corrected according to the data consistency metric between the predicted projection dataset corresponding to the current latent variable and the continuous X-ray projection dataset. The latent variables are gradually updated with random perturbation until the backward sampling process from the initial diffusion state to the final diffusion state is completed, and the latent variables corresponding to the end of the backward sampling are obtained.

[0091] ;

[0092] in, To mark the diffusion time as The latent variables at time points represent the random states of the orthogonal triaxial woven fabric 3D voxel model in the latent space. Indicates a continuous diffusion time marker in the potential diffusion process. This indicates that the physical-guided potential diffusion model is time-stamped. The drift term below is used to describe the latent variables when physical measurement constraints are not considered. The deterministic evolutionary trend along the pre-trained prior distribution direction in the latent space. This indicates that the physical-guided potential diffusion model is time-stamped. The diffusion coefficient at that point, Indicates a set of parameters The pre-trained latent diffusion prior probability density function characterized in latent variables The logarithmic value at a given point is used to characterize the statistical distribution characteristics of the orthogonal triaxial weave texture in the latent space. This represents the log-probability density function of the pre-trained latent diffusion prior with respect to the latent variables. The gradient term is used to guide the latent variables to evolve towards a high-probability region that conforms to the prior distribution of orthogonal triaxial woven texture. This represents a data consistency metric function constructed using continuous X-ray projection data as constraints. This represents the data consistency measure function with respect to latent variables. The gradient term is obtained by using the backpropagation algorithm, which calculates the gradient network of the data residuals in the projection domain by sequentially passing them through the adjoint operator and the decoding operator of the forward projection operator. This represents the data consistency weight parameter, used to dynamically adjust the balance between physical constraints and orthogonal triaxial woven texture priors. In the early stages of sampling (t is greater than the threshold), the focus is on recovering low-frequency physical structures, while in the later stages of sampling (t is less than the threshold), the focus is on generating high-frequency texture details. Indicates the reverse Wiener process. The differential increment representing the diffusion time. Indicates the reverse Wiener process in time increment Random increments within.

[0093] The decoding operator is executed on the latent variables corresponding to the end of the backsampling to generate a three-dimensional voxel model.

[0094] In Example 1, the latent variables are mapped using a decoding operator, directly converting the latent variables into a linear attenuation coefficient voxel field, and the linear attenuation coefficient voxel field is output as a three-dimensional voxel model. The three-dimensional voxel model is the orthogonal triaxial woven fabric three-dimensional voxel model obtained under the combined action of X-ray imaging physical constraints and latent diffusion priors.

[0095] Digital envelope processing is performed on the three-dimensional voxel model to identify and remove the core mold structure, while retaining the orthogonal triaxial woven body to generate a purified three-dimensional voxel model.

[0096] In this embodiment, digital envelope processing is performed on the three-dimensional voxel model, including:

[0097] In a voxel field with linear decay coefficient, if the linear decay coefficient of a voxel is greater than or equal to the effective voxel determination threshold parameter, the corresponding voxel is determined to be an effective voxel; otherwise, the corresponding voxel is determined not to be an effective voxel, and an effective voxel indicator field is generated for all voxels.

[0098] In Example 1, a three-dimensional voxel model is used as the linear decay coefficient voxel field. The linear decay coefficient voxel field is a three-dimensional array, and the value of each voxel represents the linear decay coefficient of the spatial position. The effective voxel indicator field is used to identify the position of all effective voxels in the three-dimensional space.

[0099] The effective voxel determination threshold parameter can be set manually or by an expert model. In the orthogonal triaxial woven fabric CT analysis scenario, the effective voxel determination threshold parameter can be selected as 80.

[0100] Based on the effective voxel indicator field, expansion and erosion operations are performed using three-dimensional discrete structural elements to obtain the digital envelope voxel mask.

[0101] A preset three-dimensional discrete structuring element is selected, and the effective voxel indicator field is first subjected to a three-dimensional expansion operation according to the three-dimensional discrete structuring element. For each voxel and its neighborhood, the effective voxel region is expanded according to the structuring element template. Then, the same three-dimensional discrete structuring element is applied to the expansion result to perform a three-dimensional erosion operation. Only the voxel region completely covered by the structuring element is retained within the range of the structuring element template, and a spatially connected digital envelope voxel mask is obtained. The digital envelope voxel mask effectively encloses all effective voxel regions and eliminates isolated noise and small voids.

[0102] The three-dimensional discrete structural element adopts a centrally symmetric cube, sphere, or ellipsoid shape. The size of the structural element is based on the diameter of the yarn or the pore size of the orthogonal three-dimensional woven fabric. It is adaptively set by dividing the physical size by the CT voxel size and rounding it to the nearest integer. The major and minor axes of the structural element can be determined according to the actual diameter of the yarn in each direction to achieve optimal spatial connectivity discrimination for different structural features.

[0103] Based on the digital envelope voxel mask, the envelope of the linear decay coefficient voxel field is clipped to form an envelope-constrained linear decay coefficient voxel field.

[0104] Envelope clipping sets the linear decay coefficient values ​​of the voxels not covered in the digital envelope voxel mask to zero, while retaining the linear decay coefficient values ​​of the voxels covered by the digital envelope voxel mask, thus forming an envelope-constrained linear decay coefficient voxel field.

[0105] In the envelope-constrained linear decay coefficient voxel field, by judging whether the linear decay coefficient of the voxel is between the lower threshold and the upper threshold of the linear decay coefficient of the core model, and whether the voxel is within the connected domain determined by the spatial connectivity judgment parameter, all voxels that meet the conditions are selected, and all selected voxels constitute the core model candidate voxel set.

[0106] In Example 1, a lower threshold and an upper threshold for the linear attenuation coefficient of the core mold are set, and spatial connectivity determination parameters are set.

[0107] The lower threshold for the linear decay coefficient of the core mold is 0.3. Used to exclude values ​​below 0.3 For voxels, only those voxels that are greater than or equal to the lower limit threshold of the linear decay coefficient of the core model may belong to the core model.

[0108] The upper limit threshold for the linear decay coefficient of the core mold is 0.6. Used to exclude values ​​higher than 0.6 Voxels (usually corresponding to high-density fibers, impurities, or anomalies) are considered to belong to the core mold only if they are less than or equal to the upper limit threshold of the linear attenuation coefficient of the core mold.

[0109] The spatial connectivity determination parameter is a 26-connected domain (all 26 neighboring voxels around a three-dimensional hexahedron are considered connected). Using the 26-connectivity standard, the set of candidate voxels is determined to be a valid core model connected domain only when the set of voxels is connected in three-dimensional space (each voxel is in contact with at least one neighboring voxel).

[0110] Generate a core phantom voxel mask based on the core phantom candidate voxel set;

[0111] Each voxel element of the core mold voxel mask is used to identify whether the corresponding position belongs to the core mold structure. If the corresponding position belongs to the core mold candidate voxel set, the value of the corresponding voxel element of the core mold voxel mask is one; otherwise, the value is zero.

[0112] The core voxel mask is removed from the envelope-constrained linear decay coefficient voxel field to generate a purified three-dimensional voxel model.

[0113] At each voxel position in the envelope-constrained linear decay coefficient voxel field, the corresponding voxel element value of the core mold voxel mask is read. For voxels identified as core mold structures in the core mold voxel mask, the linear decay coefficient value of the voxel position in the envelope-constrained linear decay coefficient voxel field is set to zero. For voxels identified as non-core mold structures in the core mold voxel mask, the original linear decay coefficient value of the voxel position in the envelope-constrained linear decay coefficient voxel field is retained. By processing all voxels in the envelope-constrained linear decay coefficient voxel field according to the above operation, a purified three-dimensional voxel model is obtained.

[0114] By performing digital envelope processing on the three-dimensional voxel model, accurately identifying and eliminating the core mold structure while retaining the orthogonal triaxial woven body, interference from non-target structures is effectively removed, improving the accuracy of spatial registration and region segmentation, and ensuring the spatial consistency of the analysis results and the structural authenticity of the material itself.

[0115] The purification 3D voxel model is rigidly registered with the preset standard part model to establish a unified spatial coordinate system and obtain the calibrated 3D voxel model.

[0116] In this embodiment, rigid body registration is performed between the purified 3D voxel model and the preset standard part model, including:

[0117] Based on the voxel index of the purified linear decay coefficient voxel field, establish the physical coordinates of each voxel.

[0118] The purification three-dimensional voxel model is used as the purification linear attenuation coefficient voxel field, and the preset standard part model is used as the standard linear attenuation coefficient voxel field. Both the purification linear attenuation coefficient voxel field and the standard linear attenuation coefficient voxel field are three-dimensional arrays, and the value of each voxel represents the linear attenuation coefficient at its corresponding position in space.

[0119] Physical coordinates are obtained by multiplying the integer coordinates of voxel indices in the three-dimensional array by the voxel size parameters in their respective directions. The voxel size parameters in the x-direction are used to calculate the physical coordinates in the x-direction, the voxel size parameters in the y-direction are used to calculate the physical coordinates in the y-direction, and the voxel size parameters in the z-direction are equal to the preset scanning interlayer spacing parameters and are used to calculate the physical coordinates in the z-direction.

[0120] A rigid body transformation is established between the purified linear decay coefficient voxel field and the standard linear decay coefficient voxel field to obtain the corresponding physical coordinates under the standard linear decay coefficient voxel field.

[0121] Rigid body transformation consists of an orthogonal rotation matrix and a translation vector. By rotating the physical coordinates of each voxel in the purified linear decay coefficient voxel field using the orthogonal rotation matrix and then translating them using the translation vector, the corresponding physical coordinates under the standard linear decay coefficient voxel field are obtained. The orthogonal rotation matrix is ​​a three-dimensional orthogonal matrix, and the translation vector is a three-dimensional column vector with the same dimensions as the physical coordinates. The orthogonal rotation matrix must satisfy the condition that its product with its transpose equals the identity matrix.

[0122] Based on rigid body transformation, a resampling operation is performed on the voxel field of the purified linear attenuation coefficient to obtain the resampled voxel field of the linear attenuation coefficient.

[0123] For each physical coordinate of the standard linear attenuation coefficient voxel field, the inverse transformation of the optimal rigid body transformation parameter set is used to transform the physical coordinates back to the physical coordinate system of the purified linear attenuation coefficient voxel field. The voxel value of the purified linear attenuation coefficient voxel field is obtained at the non-integer physical coordinate position using the interpolation operator. The obtained voxel value is assigned to the current physical coordinate position of the standard linear attenuation coefficient voxel field to obtain the resampled linear attenuation coefficient voxel field.

[0124]

[0125] in, This represents the voxel field of the resampling linear attenuation coefficient in physical coordinates. The linear attenuation coefficient at that point is taken as follows: Indicates the interpolation operator. This represents the purified linear decay coefficient voxel field, i.e., the purified three-dimensional voxel model, where each element represents the linear decay coefficient at a spatial location. This represents the inverse matrix of the orthogonal rotation matrix in the optimal rigid body transformation parameter set. The first voxel in the standard linear decay coefficient voxel field represents the... Individual physical coordinates This represents the translation vector in the optimal rigid body transformation parameter set.

[0126] Based on the resampled linear decay coefficient voxel field, a registration cost function is constructed within the common domain of the standard linear decay coefficient voxel field;

[0127] The registration cost function is used to evaluate the sum of squares of the differences in linear attenuation coefficients between the standard linear attenuation coefficient voxel field and the resampled linear attenuation coefficient voxel field at each corresponding physical coordinate within a common domain. The common domain is the overlapping region in space between the standard linear attenuation coefficient voxel field and the resampled linear attenuation coefficient voxel field. The independent variables of the registration cost function are the orthogonal rotation matrix and the translation vector, and the dependent variable is the sum of squares of the orthogonal rotation matrix and the translation vector.

[0128] Under the orthogonal constraint that the orthogonal rotation matrix and translation vector must satisfy, the optimal rigid body transformation parameter set that minimizes the registration cost function is solved. The optimal rigid body transformation parameter set is then applied to the voxel field of the purified linear attenuation coefficient, and the calibration three-dimensional voxel model in a unified spatial coordinate system is obtained by using the interpolation operator.

[0129] In Example 1, the optimal rigid body transformation parameter set is determined by minimizing the registration cost function. The optimal rigid body transformation parameter set includes the optimal orthogonal rotation matrix and the optimal translation vector. Using the optimal rigid body transformation parameter set, the physical coordinates of each voxel in the purified linear attenuation coefficient voxel field are first rotated and then translated sequentially to obtain the target physical coordinates in a unified spatial coordinate system. For each target physical coordinate, an interpolation operator is applied to obtain the corresponding linear attenuation coefficient voxel value based on the position of the inverse rotation and translation transformation results in the purified linear attenuation coefficient voxel field. The linear attenuation coefficient voxel value is then assigned to the voxel in the target physical coordinate system to generate a calibration three-dimensional voxel model in a unified spatial coordinate system.

[0130] Based on a unified spatial coordinate system, the calibration three-dimensional voxel model is programmed to cut, the main stress areas of the woven fabric are separated and three-dimensional voxel models of each analysis area are generated.

[0131] In this embodiment, the calibrated three-dimensional voxel model is programmed to be cut based on a unified spatial coordinate system, including:

[0132] Based on a unified spatial coordinate system, a set of cutting coordinate parameters for programmed cutting is established in the voxel field of the calibration linear attenuation coefficient. The set of cutting coordinate parameters consists of at least one set of cutting plane parameters.

[0133] The calibration three-dimensional voxel model is used as the calibration linear attenuation coefficient voxel field. Based on a unified spatial coordinate system, in the calibration linear attenuation coefficient voxel field, according to the spatial geometric characteristics of the orthogonal triaxial woven structure or the analysis requirements, the user inputs the cutting rules, and each set of cutting plane parameters is automatically or manually set. The cutting plane parameters include the normal vector parameters used to determine the spatial direction of the cutting plane and the offset parameters used to determine the spatial position of the cutting plane. All cutting plane parameters are combined to establish a set of cutting coordinate parameters for programmed cutting. Each set of cutting plane parameters is used to uniquely determine a cutting plane in the unified spatial coordinate system.

[0134] The equation of the cutting plane is defined based on the normal vector parameters, offset parameters, and voxel physical coordinates of the cutting plane parameters.

[0135] The cutting plane equation uniquely determines a cutting plane by determining the position where the product of the normal vector parameter and the voxel physical coordinate plus the offset parameter equals zero. The normal vector parameter represents the spatial direction of the cutting plane, and the offset parameter represents the signed distance of the cutting plane from the origin.

[0136] Based on the cutting plane equation, the voxel field of the calibration linear attenuation coefficient is classified into voxels. According to the sign relationship between the physical coordinates of each voxel and the cutting plane equation, a region label is assigned to the voxel and a region label field is generated.

[0137] ;

[0138] in, Let be a region label field, where label is a mapping function that maps symbol combinations to region labels. For normal vector parameters, The bias parameter is used to represent the spatial partitioning of the main stress-bearing areas of the woven fabric in a unified spatial coordinate system.

[0139] Based on the region label field, the voxel field of the calibration linear attenuation coefficient is extracted to generate a region mask for each target region label. For each target region label, the region mask is multiplied with the voxel field of the calibration linear attenuation coefficient in a voxel-by-voxel dimension to obtain the corresponding three-dimensional voxel model of the analysis region.

[0140] Each element of the region mask is used to indicate whether the corresponding voxel belongs to the analysis region corresponding to the target region label. If it does, the value of the region mask element is one; otherwise, the value is zero. The three-dimensional voxel model of the analysis region is the region voxel field obtained by programmed cutting and separating the voxel field with calibrated linear attenuation coefficient in a unified spatial coordinate system.

[0141] Set continuous grayscale threshold intervals and scan the three-dimensional voxel model of each analysis area with a fixed step size. Calculate the volume fraction of the number of voxels in each threshold interval relative to the number of voxels in the effective grayscale range to form a threshold-volume fraction dataset.

[0142] In this embodiment, a continuous grayscale threshold range is set, and the three-dimensional voxel model of each analysis area is scanned with a fixed step size, including:

[0143] In the voxel field of the linear decay coefficient in each analysis region, a continuous gray-scale threshold sequence is constructed by gradually increasing the effective gray-scale range lower limit threshold parameter with a fixed step size parameter.

[0144] In Example 1, a three-dimensional voxel model of the analysis region is used as the voxel field of the linear attenuation coefficient of the analysis region. The value of each voxel is used to represent the linear attenuation coefficient of the voxel at the corresponding position in the unified spatial coordinate system. Each voxel field of the linear attenuation coefficient of the analysis region corresponds to a unique target region label. The total number of analysis regions is denoted as S.

[0145] Set the lower limit threshold parameter and the upper limit threshold parameter of the effective grayscale range for determining the effective material voxels of orthogonal triaxial weave. Each grayscale threshold in the continuous grayscale threshold sequence is obtained by adding the product of the lower limit threshold parameter of the effective grayscale range and the current step size parameter until it does not exceed the upper limit threshold parameter of the effective grayscale range. The number of steps in the grayscale threshold sequence is equal to the difference between the upper limit threshold parameter of the effective grayscale range and the lower limit threshold parameter of the effective grayscale range divided by the step size parameter and rounded down.

[0146] In this example, the effective grayscale range lower threshold parameter is set to 80. In most cases, areas with CT voxel grayscale values ​​below 80 correspond to air, pores, areas where resin is not fully impregnated, or background noise, indicating extremely low material density or no material at all. Voxels below the effective grayscale range lower threshold parameter are considered to have no practical contribution to structural strength and material analysis, and are therefore used as the lower limit for determining effective voxels. In this example, the effective grayscale range upper threshold parameter is set to 255. High grayscale values ​​of CT reconstructed voxels correspond to fiber-rich areas or dense materials. 255 is the maximum value of 8-bit grayscale in common CT images. There is no effective grayscale above the effective grayscale range upper threshold parameter, so all voxel grayscale values ​​exceeding 255 are considered extremely high density.

[0147] In each analysis region, all voxels are traversed one by one within the voxel field of the linear decay coefficient, and an effective gray voxel indicator field is generated within the voxel field of the linear decay coefficient of the analysis region.

[0148] In Example 1, if the voxel value of a certain voxel is greater than or equal to the lower threshold parameter of the effective gray range and less than or equal to the upper threshold parameter of the effective gray range, then the corresponding voxel is determined to belong to the effective gray range; otherwise, it is determined not to belong to the effective gray range. The determination result of each voxel is stored in a new three-dimensional array, which is the effective gray voxel indicator field.

[0149] Based on the effective gray voxel indicator field, the number of all voxels belonging to the effective gray range within the region linear decay coefficient voxel field is statistically analyzed to obtain the total number of effective voxels.

[0150] The total number of valid voxels is equal to the sum of all elements with a value of one in the valid grayscale voxel indicator field, which is the total number of voxels that are determined to belong to the valid grayscale range.

[0151] Using a fixed step size parameter, each gray level threshold in the continuous gray level threshold sequence is used as the current upper limit of the threshold to construct the cumulative threshold interval, and a cumulative threshold voxel indicator field is generated in the voxel field of the linear decay coefficient in the analysis region.

[0152] In Example 1, for each voxel in the voxel field of the linear decay coefficient of the analysis region, from the lower limit threshold parameter of the effective grayscale range to the upper limit of the current grayscale threshold, if the voxel value is greater than or equal to the lower limit threshold parameter of the effective grayscale range and less than or equal to the upper limit of the current grayscale threshold, it will be determined that it falls into the cumulative threshold range; otherwise, it will be determined that it does not fall into the cumulative threshold range. The determination result of each voxel is constructed into a cumulative threshold voxel indicator field. Each element of the cumulative threshold voxel indicator field is used to identify whether the corresponding voxel belongs to the current cumulative threshold range.

[0153] For each cumulative threshold interval, the sum of all elements with a value of one in the cumulative threshold voxel indicator field is counted to obtain the number of cumulative threshold voxels in the current cumulative threshold interval.

[0154] The cumulative threshold voxel count is equal to the total number of voxels that are determined to belong to the current cumulative threshold interval.

[0155] For each cumulative threshold interval, the volume fraction corresponding to the current cumulative threshold interval is obtained by dividing the number of cumulative threshold voxels in the current cumulative threshold interval by the total number of effective voxels.

[0156] The continuous grayscale threshold sequence is paired with the corresponding volume fraction sequence in a one-to-one correspondence to form a threshold-volume fraction dataset.

[0157] Each pairing yields an ordered pair consisting of a grayscale threshold and a volume fraction. This process iterates through all grayscale thresholds in a continuous sequence, calculating and pairing the corresponding volume fraction for each threshold. Following the increasing order of grayscale thresholds, all ordered pairs are arranged to obtain a dataset composed of a one-to-one correspondence between grayscale thresholds and volume fractions. Each element in the dataset includes a grayscale threshold and the volume fraction of the voxels in the analysis region at that threshold, reflecting the trend of volume fraction variation with the grayscale threshold. This is the threshold-volume fraction dataset, used for analyzing regional density distribution trends and fitting trend curves.

[0158] A threshold proportion trend curve is constructed based on the threshold-volume fraction dataset, and the trend curve is fitted and compared with a pre-stored standard trend curve to obtain a fitting deviation index.

[0159] In this embodiment, a threshold proportion trend curve is constructed based on the threshold-volume fraction dataset, including:

[0160] Based on the threshold-volume fraction dataset corresponding to the target region label, construct a threshold proportion trend curve;

[0161] Based on the threshold-volume fraction dataset, following the increasing order of grayscale thresholds, each grayscale threshold sampling point is used as the abscissa, and the paired volume fraction ratio is used as the ordinate. All ordered pairs of grayscale threshold sampling points and volume fraction ratios are sequentially connected on the coordinate system to form a threshold proportion trend curve with grayscale threshold as the independent variable and volume fraction ratio as the dependent variable. The threshold proportion trend curve reflects the cumulative trend of volume fraction ratio as grayscale threshold changes within the analysis area.

[0162] Under the condition that the grayscale threshold sampling points are aligned one by one, the threshold proportion trend curve is fitted and compared with the pre-stored standard trend curve. All discrete grayscale threshold sampling points are used as the set of sampling points for fitting and comparison, and a fitting residual sequence is constructed.

[0163] The pre-stored standard trend curve uses the grayscale threshold as the independent variable and the volume fraction as the dependent variable. The pre-stored standard trend curve and the threshold proportion trend curve use the same grayscale threshold dimension and volume fraction dimension.

[0164] At each discrete grayscale threshold sampling point, the volume fraction value of the threshold proportion trend curve adjusted by the fitting mapping function is subtracted from the volume fraction value of the pre-stored standard trend curve at the same grayscale threshold sampling point. All differences are arranged in the order of grayscale threshold sampling points to form a fitting residual sequence.

[0165] The fitting deviation index is calculated based on the fitted residual sequence.

[0166] The fitting deviation index is the sum of the squares of all fitting residuals divided by the total number of grayscale threshold sampling points plus one. The fitting deviation index is a unique fitting deviation evaluation value corresponding to the target analysis area.

[0167] The results of the internal impregnation quality assessment of the orthogonal triaxial woven fabric are generated based on the fitting deviation index, and an analysis report is output.

[0168] In this embodiment, the internal impregnation quality assessment result of the orthogonal triaxial knitted fabric is generated based on the fitting deviation index, including:

[0169] Construct a set of internal infiltration quality level rules for classifying fitting deviation indices, and define the internal infiltration quality level range of orthogonal triaxial knitted fabrics by combining preset evaluation experience;

[0170] In Example 1, the fitting deviation index is used as the numerical basis for measuring the fitting result of the threshold proportion trend curve and the standard trend curve. According to the allowable range of fitting deviation of the historical standard curve, the fabric is divided into multiple orthogonal triaxial weaving quality level intervals according to the specific value of the fitting deviation index.

[0171] The fitting deviation index corresponding to each analysis region is matched with the internal wetting quality grade range of the orthogonal triaxial woven fabric to determine the internal wetting quality grade label of the current analysis region.

[0172] Based on the internal infiltration quality level label of each analysis area, construct an internal infiltration quality level distribution map under a unified spatial coordinate system;

[0173] The spatial positions of all analysis areas in a unified spatial coordinate system are mapped one-to-one with their corresponding internal wetting quality level labels. In the three-dimensional voxel visualization system, different internal wetting quality level labels are assigned different visualization colors according to the preset color mapping rules, thereby generating a three-dimensional internal wetting quality level distribution map and showing the spatial distribution characteristics of the internal wetting state of each region of the orthogonal triaxial woven fabric.

[0174] Summarize the fit deviation index, internal infiltration quality level label, threshold-volume fraction dataset, and threshold proportion trend curve for all analysis regions, and output a standardized analysis report.

[0175] In this embodiment, the internal impregnation quality grade range and internal impregnation quality grade label of the orthogonal triaxial woven fabric include:

[0176] When the fitting deviation index is less than or equal to the first fitting deviation classification threshold parameter, it is judged as excellent infiltration.

[0177] When the fitting deviation index is greater than the first fitting deviation classification threshold parameter and less than or equal to the second fitting deviation classification threshold parameter, it is judged as good immersion.

[0178] When the fitting deviation index is greater than the second fitting deviation classification threshold parameter and less than or equal to the third fitting deviation classification threshold parameter, it is determined to be the infiltration critical point;

[0179] When the fitting deviation index is greater than the third fitting deviation classification threshold parameter, it is judged as poor immersion level.

[0180] The first, second, and third fitting deviation classification threshold parameters are all pre-set numerical limits used to distinguish the quality state of resin impregnation inside the woven fabric. They can be set in conjunction with historical verification sample experience data or expert knowledge.

[0181] In Example 1, the first fitting deviation classification threshold parameter is set to the lower quartile of the historical fitting deviation distribution of the standard qualified product or the maximum allowable fitting deviation value corresponding to the absence of obvious wetting defects in expert experience. If the statistical results of a large number of high-quality standard sample samples are used, when the fitting deviation index is less than or equal to 0.02, the internal wetting state of the woven fabric can be considered to have reached the optimal quality level. Therefore, the first fitting deviation classification threshold parameter can be set to 0.02.

[0182] The second fitting deviation classification threshold parameter is set based on the tolerance range of actual process control standards or batch samples. Its value reflects the fitting deviation that is still acceptable in actual application but whose quality is close to the lower limit. Taking the field statistics of woven fabric engineering as an example, when the fitting deviation index is greater than 0.02 and less than or equal to 0.05, the internal impregnation state of the woven fabric is basically good but there is a certain degree of non-uniformity. Therefore, the second fitting deviation classification threshold parameter is set to 0.05.

[0183] The third fitting deviation classification threshold parameter is set in conjunction with failure analysis and product scrapping standards. It represents that when the fitting deviation index exceeds this value, the internal wetting state of the woven fabric has seriously deviated from the standard curve, and there are large wetting defects or structural problems. In actual testing applications, if the fitting deviation index is greater than 0.05 and less than or equal to 0.10, it is considered critical, while if it exceeds 0.10, it is judged as inferior. Therefore, the third fitting deviation classification threshold parameter is set to 0.10.

[0184] Example 2: In a batch of composite material production, inspectors needed to perform non-destructive quality analysis on multiple orthogonal triaxial woven fabric samples. In the testing process, an orthogonal triaxial woven fabric sample numbered "TQW-03-15" was fixed on the sample tray of a CT scanning device. The system was set with a scanning layer spacing of 1.5 mm and a rotation step of 1.2 degrees, obtaining a total of 620 sets of X-ray projection data. The average raw signal intensity range of a single projection was [146000, 154000], corresponding to 2097152 pixels on the detector. After physical consistency transformation, each set of projection data yielded a set of logarithmic projection values, with the largest logarithmic projection value being 7.13 and the smallest being 2.84.

[0185] The continuous X-ray projection dataset was input into the physics-guided potential diffusion model. The system then input the projection data into the potential diffusion prior model and sampled through 20 rounds of inverse stochastic differential equations, finally outputting a three-dimensional voxel field with linear attenuation coefficients. This voxel field has a spatial resolution of 1.5 mm × 1.5 mm × 1.5 mm, totaling 1510 × 1400 × 58 voxels. The linear attenuation coefficient for each voxel ranges from 0.13 to 0.92. Reconstruction took approximately 14 minutes.

[0186] The system performs digital envelope processing on the generated 3D voxel model. The system sets the effective voxel detection threshold to 0.18. Of all voxels, 1,342,564 have a linear decay coefficient greater than or equal to 0.18, accounting for 73.1% of all voxels. Three-dimensional expansion and erosion are performed using a centrally symmetric cubic structuring element to eliminate isolated noise points. The system identifies a total of 1,300,934 envelope voxels. Using the core model's linear decay coefficient range [0.34, 0.63] and the 26-connected component standard, the system detects 57,219 suspected core model candidate voxel regions. The envelope voxel field and core model mask are then removed, outputting a cleaned 3D voxel model. Comparison shows that the number of missed voxels in the core model region is 0, and the rate of incorrectly deleted yarn voxels is less than 0.15%.

[0187] In the spatial registration step, the inspector imports the standard part model. The system calculates the optimal rigid body transformation parameter set using a point cloud automatic registration algorithm. The average coordinate overlap error between the purified 3D voxel model and the standard part model is 0.74mm, and the maximum offset is 2.1mm. After spatial registration, the system automatically establishes a unified spatial coordinate system and generates a calibrated 3D voxel model.

[0188] During programmed cutting, the inspector sets three sets of cutting plane parameters according to the analysis requirements: normal vector of the main stress area on surface A (0,1,0), offset 25.5; normal vector of the area on surface B (0,-1,0), offset -28.5; and normal vector of the dense yarn area (1,0,0), offset 13. The system automatically extracts three major analysis regions—surface A, surface B, and the dense yarn area—from the calibrated 3D voxel model. The number of voxels on surface A is 234,151, on surface B it is 201,682, and in the dense yarn area it is 425,033.

[0189] For each analysis region, the effective grayscale range of the linear decay coefficient voxel field is set with a lower threshold of 0.18, an upper threshold of 0.85, and a step size of 0.05, resulting in 14 grayscale threshold sampling points. For example, the total number of effective voxels in the A-side region between 0.18 and 0.85 is 232,419. Taking the current grayscale threshold of 0.33 as an example, the cumulative threshold voxel count for the A-side is 75,223, with a volume fraction of 0.324. All threshold sampling points and volume fractions are paired as shown in Table 1 below:

[0190] Table 1. Threshold-Volume Fraction Dataset Fragment

[0191]

[0192] The dataset is plotted as a threshold proportion trend curve. The system retrieves the corresponding pre-stored standard trend curve and fits it. For all threshold sampling points, the residual between the volume fraction of the trend curve and the volume fraction of the standard trend curve is calculated. The mean of the sum of squares of the residual sequence is taken. The fitting deviation index of the A-side region is 0.013, that of the B-side region is 0.019, and that of the yarn-dense area is 0.009.

[0193] Based on quality grading rules, the system classifies a fitting deviation index of less than 0.015 as "Excellent," 0.015-0.025 as "Good," 0.025-0.035 as "Critical," and greater than 0.035 as "Inferior." Accordingly, surface A and the densely yarn-covered area are rated "Excellent," and surface B as "Good." The system automatically colors the data in the 3D visualization module, displaying surface A in green, surface B in blue, and the densely yarn-covered area in green, outputting a 3D grade distribution map.

[0194] All data is automatically generated into standardized analysis reports, including: spatial coordinates of each region, effective voxel count, volume fraction of each sampling point, fit deviation index and corresponding quality level, trend curves, and 3D distribution visualization. The test reports are output in structured data and PDF format, and the data interface automatically connects to the quality traceability system.

[0195] To verify the beneficial effects of this invention, 10 samples from the same period were selected and analyzed using both the method of this invention and the conventional method. The conventional method, employing FDK + single threshold method (threshold 0.20), yielded a mean porosity of 2.6%, a standard deviation of 0.7%, and a local anomaly detection rate of 38%. The method of this invention showed that the root mean square error of the volume fraction curve fitting for the A-side, B-side, and densely yarn-covered areas was less than 0.015, with a local anomaly detection rate of 92%. In the main stress area of ​​the A-side, the conventional method failed to detect high-density, fiber-rich areas in two samples, while the method of this invention accurately identified all of them. The overall automation rate reached 98%, and the manual verification time was reduced from an average of 18 minutes per sample to 2 minutes.

[0196] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting, characterized in that, include: The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device. Continuous X-ray projection data is obtained according to the preset scanning interval. The continuous X-ray projection data is input into the physical-guided potential diffusion posterior sampling model. While following the physical constraints of X-ray imaging, the potential diffusion prior is used to generate a three-dimensional voxel model. Digital envelope processing is performed on the three-dimensional voxel model to identify and remove the core mold structure, while retaining the orthogonal triaxial woven body to generate a purified three-dimensional voxel model. The purification 3D voxel model is rigidly registered with the preset standard part model to establish a unified spatial coordinate system and obtain the calibrated 3D voxel model. Based on a unified spatial coordinate system, the calibration three-dimensional voxel model is programmed to cut, the main stress areas of the woven fabric are separated and three-dimensional voxel models of each analysis area are generated. Set continuous grayscale threshold intervals and scan the three-dimensional voxel model of each analysis area with a fixed step size. Calculate the volume fraction of the number of voxels in each threshold interval relative to the number of voxels in the effective grayscale range to form a threshold-volume fraction dataset. A threshold proportion trend curve is constructed based on the threshold-volume fraction dataset. The threshold proportion trend curve is then fitted and compared with the pre-stored standard trend curve to obtain the fitting deviation index. The fitting deviation index is the sum of the squares of all fitting residuals divided by the total number of gray threshold sampling points plus one. The pre-stored standard trend curve uses the gray threshold as the independent variable and the volume fraction as the dependent variable. The pre-stored standard trend curve and the threshold proportion trend curve use the same gray threshold dimension and volume fraction dimension. Construct a set of internal infiltration quality level rules for classifying fitting deviation indexes, combine preset evaluation experience, define the internal infiltration quality level range of orthogonal triaxial knitted fabrics, generate internal infiltration quality assessment results of orthogonal triaxial knitted fabrics, and output an analysis report; The internal wetting quality grading rules are set based on historical verification sample experience data or expert knowledge, and are judged as excellent wetting, good wetting, critical wetting, and poor wetting. The construction of the threshold proportion trend curve based on the threshold-volume fraction dataset includes: Based on the threshold-volume fraction dataset corresponding to the target region label, construct a threshold proportion trend curve; Under the condition that the grayscale threshold sampling points are aligned one by one, the threshold proportion trend curve is fitted and compared with the pre-stored standard trend curve. All discrete grayscale threshold sampling points are used as the set of sampling points for fitting and comparison, and a fitting residual sequence is constructed. The fitting deviation index is calculated based on the fitted residual sequence.

2. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The step of inputting continuous X-ray projection data into a physically guided potential diffusion posterior sampling model includes: The orthogonal triaxial woven fabric sample is fixed in the scanning cavity of the computed tomography device, and the scanning interlayer spacing is set to the preset scanning interlayer spacing parameter. Continuous X-ray projection measurement results are obtained at multiple scanning angles to form continuous X-ray projection raw data. Physical consistency transformation is performed on the raw data of continuous X-ray projection, and the transmission intensity value of the k-th pixel of the detector and the corresponding incident reference intensity value are collected. Based on the transmitted intensity value and the incident reference intensity value, logarithmic projection data is calculated, and all logarithmic projection data are combined to form a continuous X-ray projection dataset. The continuous X-ray projection dataset is input into the physical guided potential diffusion model, and the three-dimensional voxel model is defined as a voxel field with linear decay coefficient in the physical guided potential diffusion model. In the physics-guided potential diffusion model, a forward projection operator is defined to map the voxel field of linear decay coefficients to a predicted projection dataset. Introduce latent variables into the physics-guided latent diffusion model and define a decoding operator; A data consistency metric is constructed based on continuous X-ray projection datasets and predicted projection datasets. Based on the pre-trained potential diffusion prior density and data consistency metric, reverse stochastic differential equation posterior sampling is performed to obtain the latent variables corresponding to the end of the reverse sampling. The decoding operator is executed on the latent variables corresponding to the end of the backsampling to generate a three-dimensional voxel model.

3. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The digital envelope processing of the three-dimensional voxel model includes: In a voxel field with linear decay coefficient, if the linear decay coefficient of a voxel is greater than or equal to the effective voxel determination threshold parameter, the corresponding voxel is determined to be an effective voxel; otherwise, the corresponding voxel is determined not to be an effective voxel, and an effective voxel indicator field is generated for all voxels. Based on the effective voxel indicator field, the digital envelope voxel mask is obtained by performing dilation and erosion operations using three-dimensional discrete structuring elements. Based on the digital envelope voxel mask, the envelope of the linear decay coefficient voxel field is clipped to form an envelope-constrained linear decay coefficient voxel field. In the envelope-constrained linear decay coefficient voxel field, by judging whether the linear decay coefficient of the voxel is between the lower threshold and the upper threshold of the linear decay coefficient of the core model, and whether the voxel is within the connected domain determined by the spatial connectivity judgment parameter, all voxels that meet the conditions are selected, and all selected voxels constitute the core model candidate voxel set. Generate a core phantom voxel mask based on the core phantom candidate voxel set; The core voxel mask is removed from the envelope-constrained linear decay coefficient voxel field to generate a purified three-dimensional voxel model.

4. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The process of rigidly registering the purified 3D voxel model with a preset standard part model includes: Based on the voxel index of the purified linear decay coefficient voxel field, establish the physical coordinates of each voxel. A rigid body transformation is established between the purified linear decay coefficient voxel field and the standard linear decay coefficient voxel field to obtain the corresponding physical coordinates under the standard linear decay coefficient voxel field. Based on rigid body transformation, a resampling operation is performed on the voxel field of the purified linear attenuation coefficient to obtain the resampled voxel field of the linear attenuation coefficient. Based on the resampled linear decay coefficient voxel field, a registration cost function is constructed within the common domain of the standard linear decay coefficient voxel field; Under the orthogonal constraint that the orthogonal rotation matrix and translation vector must satisfy, the optimal rigid body transformation parameter set that minimizes the registration cost function is solved. The optimal rigid body transformation parameter set is then applied to the voxel field of the purified linear attenuation coefficient, and the calibration three-dimensional voxel model in a unified spatial coordinate system is obtained by using the interpolation operator.

5. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The programmed cutting of the calibrated 3D voxel model based on a unified spatial coordinate system includes: Based on a unified spatial coordinate system, a set of cutting coordinate parameters for programmed cutting is established in the voxel field of the calibration linear attenuation coefficient. The set of cutting coordinate parameters consists of at least one set of cutting plane parameters. The equation of the cutting plane is defined based on the normal vector parameters, offset parameters, and voxel physical coordinates of the cutting plane parameters. Based on the cutting plane equation, the voxel field of the calibration linear attenuation coefficient is classified into voxels. According to the sign relationship between the physical coordinates of each voxel and the cutting plane equation, a region label is assigned to the voxel, and a region label field is generated. Based on the region label field, the voxel field of the calibration linear attenuation coefficient is extracted to generate a region mask for each target region label. For each target region label, the region mask is multiplied with the voxel field of the calibration linear attenuation coefficient in a voxel-by-voxel dimension to obtain the corresponding three-dimensional voxel model of the analysis region.

6. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The process of setting a continuous grayscale threshold range and scanning the three-dimensional voxel model of each analysis area with a fixed step size includes: In the voxel field of the linear decay coefficient in each analysis region, a continuous gray-scale threshold sequence is constructed by gradually increasing the effective gray-scale range lower limit threshold parameter with a fixed step size parameter. In each analysis region, all voxels are traversed one by one within the voxel field of the linear decay coefficient, and an effective gray voxel indicator field is generated within the voxel field of the linear decay coefficient of the analysis region. Based on the effective gray voxel indicator field, the number of all voxels belonging to the effective gray range within the region linear decay coefficient voxel field is statistically analyzed to obtain the total number of effective voxels. Using a fixed step size parameter, each gray level threshold in the continuous gray level threshold sequence is used as the current upper limit of the threshold to construct the cumulative threshold interval, and a cumulative threshold voxel indicator field is generated in the voxel field of the linear decay coefficient in the analysis region. For each cumulative threshold interval, the sum of all elements with a value of one in the cumulative threshold voxel indicator field is counted to obtain the number of cumulative threshold voxels in the current cumulative threshold interval. For each cumulative threshold interval, the volume fraction corresponding to the current cumulative threshold interval is obtained by dividing the number of cumulative threshold voxels in the current cumulative threshold interval by the total number of effective voxels. The continuous grayscale threshold sequence is paired with the corresponding volume fraction sequence in a one-to-one correspondence to form a threshold-volume fraction dataset.

7. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 1, characterized in that, The method involves constructing a rule set for classifying the internal impregnation quality level of the fitting deviation index, combining it with pre-set evaluation experience, defining the internal impregnation quality level range of orthogonal triaxial knitted fabrics, generating internal impregnation quality assessment results for orthogonal triaxial knitted fabrics, and outputting an analysis report, including: Construct a set of internal infiltration quality level rules for classifying fitting deviation indices, and define the internal infiltration quality level range of orthogonal triaxial knitted fabrics by combining preset evaluation experience; The fitting deviation index corresponding to each analysis region is matched with the internal wetting quality grade range of the orthogonal triaxial woven fabric to determine the internal wetting quality grade label of the current analysis region. Based on the internal infiltration quality level label of each analysis area, construct an internal infiltration quality level distribution map under a unified spatial coordinate system; Summarize the fit deviation index, internal infiltration quality level label, threshold-volume fraction dataset, and threshold proportion trend curve for all analysis regions, and output a standardized analysis report.

8. The orthogonal triaxial knitted fabric analysis method based on CT three-dimensional reconstruction and threshold fitting according to claim 7, characterized in that, The internal impregnation quality grade range and internal impregnation quality grade label of the orthogonal triaxial woven fabric include: When the fitting deviation index is less than or equal to the first fitting deviation classification threshold parameter, it is judged as excellent immersion. When the fitting deviation index is greater than the first fitting deviation classification threshold parameter and less than or equal to the second fitting deviation classification threshold parameter, it is judged as good immersion. When the fitting deviation index is greater than the second fitting deviation classification threshold parameter and less than or equal to the third fitting deviation classification threshold parameter, it is determined to be the infiltration critical point; When the fitting deviation index is greater than the third fitting deviation classification threshold parameter, it is judged as poor immersion level.

Images