Intelligent Detection and Visualization Characterization System for Defects in Solid Propellant Glue

Abstract

This invention provides an intelligent detection and visualization characterization system for defects in solid rocket motor propellant grains. The system includes a two-dimensional slice data preprocessing module, a three-dimensional visualization modeling module, and a model post-processing and analysis module. By integrating multimodal image processing algorithms and three-dimensional modeling technology, the system achieves sub-voxel-level three-dimensional reconstruction and spatial quantification analysis of defects such as porosity and cracks. It can accurately locate the coordinate distribution, geometric dimensions, and quantity statistics of defects within the three-dimensional space of the propellant grain; it can overcome the technical bottlenecks of traditional two-dimensional detection methods in defect depth analysis and multi-parameter coupled analysis. The core modules work synergistically to improve the accuracy and efficiency of detecting porosity defects inside solid rocket motors, accurately detecting the spatial location and shape of pores inside the propellant grain, and ensuring the integrity and safety of the SRM propellant grain structure.

Description

Technical Field

[0001] This invention relates to the field of industrial inspection and computer vision interdisciplinary technology, and relates to, but is not limited to, an intelligent detection and visualization characterization system for defects in solid rocket motor propellant grains. Background Technology

[0002] Solid rocket motor (SRM) propellant grains are subjected to various complex loads during production, storage, transportation, and use. These loads include temperature loads caused by solidification cooling, environmental changes, and aerodynamic heating; gravity loads during long-term storage; high overloads during launch and maneuvering flight; vibration loads during transportation and airborne operations; and impact and internal pressure loads during maintenance and ignition. Under single or combined loads, and constrained by the shell, the propellant grain generates internal stress and strain fields. If these stresses exceed the allowable limits of its mechanical properties, defects such as internal pores, cracks, and debonding can occur, posing significant risks to engine safety and potentially leading to engine test failure. Therefore, propellant grain structural integrity analysis is a crucial aspect of the entire lifespan of a solid rocket motor and a key research area urgently needing to be addressed in practical applications.

[0003] Industrial Computed Tomography (ICT) inspection technology has been gradually applied to SRM quality inspection and has played a significant role. However, the practical application of 3D ICT technology in SRM also faces some challenges, such as the complexity of data processing. The massive amount of data generated by ICT scans requires efficient data processing algorithms for processing and analysis; automated defect identification algorithms may be affected by noise, artifacts, and material properties, leading to false alarms or missed alarms; and the visualization of 3D images may not be intuitive enough for displaying certain shapes and locations, making it difficult to understand complex internal structures and defects. Summary of the Invention

[0004] This invention provides an intelligent detection and visualization characterization system for defects in solid rocket motor propellant grains, which solves the technical problems existing in the current solid rocket motor propellant grain defect detection technology, such as the lack of accurate three-dimensional information, low data processing efficiency, and reliance on manual defect quantification.

[0005] The technical method of this invention is implemented as follows:

[0006] In a first aspect, embodiments of the present invention provide an intelligent detection and visualization characterization system for defects in solid rocket motor propellant grains, the system comprising a two-dimensional slice data preprocessing module, a three-dimensional visualization modeling module, and a model post-processing and analysis module;

[0007] The two-dimensional slice data preprocessing module is used to identify and locate structural components in the two-dimensional ICT image of the acquired solid rocket motor ICT file based on a preset circular structural component target automatic recognition and detection algorithm, and to determine the circular structural region image of the solid rocket motor.

[0008] An improved anisotropic diffusion denoising algorithm based on adaptive median filtering removes impulse interference from the circular structure region image multiple times, resulting in a denoised circular structure region image.

[0009] The robust adaptive Canny algorithm based on defect edges sequentially performs circular structure boundary detail extraction and noise interference removal processing on the circular structure region image after multiple denoising processes to obtain a high-precision edge image.

[0010] The three-dimensional visualization modeling module is used to perform image registration, normalization and ROI extraction processing on the original ICT slice image based on a preset deep learning ICT slice image frame interpolation algorithm to obtain an integrated continuous slice sequence.

[0011] A 3D convolutional neural network is used to preview the deep features of adjacent slices, and the features are aligned by optical flow estimation or deformable convolution.

[0012] Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices through a U-Net generator. The realism of the generated image is evaluated by a PatchGAN discriminator and optimized through a multi-objective loss function.

[0013] Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data.

[0014] The processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model.

[0015] The reconstructed 3D model is subjected to edge smoothing and voxel fusion optimization to generate a high-precision interactive 3D visualization model.

[0016] The model post-processing and analysis module is used to perform three-dimensional reconstruction of pores in the two-dimensional ICT image to obtain a three-dimensional reconstruction model of deformed pores and volume estimation results.

[0017] Based on the volume estimation results, the fault diagnosis results of the solid rocket motor are determined.

[0018] In some embodiments, the two-dimensional slice data preprocessing module further includes a circular region extraction submodule;

[0019] The circular region extraction submodule is used to perform binarized background segmentation, morphological processing, and connected component analysis on the two-dimensional ICT image to obtain multiple candidate regions; each candidate region carries the basic features of the solid rocket motor; the basic features include area features, boundary features, and geometric features.

[0020] Non-target regions with areas smaller than a preset area threshold are removed from the candidate regions to obtain retained regions; the retained regions are numbered and segmented to obtain multiple image subsets;

[0021] For each subset of images, Sobel operator edge extraction is performed to obtain an edge point set; a rectangular coordinate system is established with the centroid of each edge point set as the origin, and the polar angles of the four vertices of the minimum bounding rectangle relative to the center are calculated;

[0022] Based on the set of edge points, the circularity index of each of the image subsets is calculated, and regions that do not meet the circularity feature are removed to obtain regions that meet the circularity feature.

[0023] A subset of edge points is randomly selected in the reserved area for circle fitting. An optimized sampling strategy is used to reduce computational complexity and generate candidate circles containing center coordinates and radius parameters.

[0024] Each candidate circle is verified. If a sufficient number of edge points fall within the preset radius tolerance range of each candidate circle, it is determined to be a valid circle, and the circular structure region image is output.

[0025] In some embodiments, the two-dimensional slice data preprocessing module further includes a filtering and noise reduction submodule;

[0026] The filtering and denoising submodule is used to perform local region division on the circular structure region image, resulting in a circular region image.

[0027] An adaptive median filter is used to filter the divided circular region image to remove impulse noise, resulting in a pre-denoised circular structure region image.

[0028] Calculate the gradient magnitude of the circular structure region image after preliminary denoising, replace the gradient magnitude of the original image with the gradient magnitude, and dynamically adjust the diffusion coefficient in the anisotropic diffusion model.

[0029] Based on the adjusted diffusion coefficient, anisotropic diffusion denoising is performed on the initially denoised circular structure region image to preserve image edge details, resulting in the multiple denoised circular structure region image.

[0030] In some embodiments, the two-dimensional slice data preprocessing module further includes an adaptive Canny submodule;

[0031] The adaptive Canny submodule is used to perform grayscale processing on a selected region in the circular structure region image after multiple denoising steps to obtain a grayscale processed image.

[0032] A Gaussian filter is constructed and a two-dimensional convolution is performed on the grayscale image to reduce local detail fluctuations in the image, resulting in a convolved image.

[0033] The gradient magnitude and direction of the convolved image are calculated using the finite difference of the first-order partial derivative; non-maximum suppression is applied to the gradient magnitude to retain local gradient maximum points;

[0034] A dual-threshold algorithm is used, which uses a high threshold to segment the background and the target, and a low threshold to connect the broken edges;

[0035] The edge region is refined to extract the flaw contour and identify potential defect features, thus obtaining the high-precision edge image.

[0036] In some embodiments, the model post-processing and analysis module includes a stomatal ellipsoid model building module, a stomatal shape parameter extraction module, a 3D display module, and a volume calculation module;

[0037] The pore ellipsoid model building module is used to simulate deformed pores as two triaxial ellipsoids, one above the other. By determining nine parameters, including the three-dimensional coordinates of the ellipsoid center, the lengths of the three semi-axes, and the three-axis rotation angles, a three-dimensional ellipsoid model is constructed. The origin of the ellipsoid model is taken as the center of the solid rocket motor base.

[0038] The stoma shape parameter extraction module is used to preprocess the stoma image to obtain a binary image of the stoma edge. First, it directly calculates the center coordinates of the ellipse, the length of the semi-major axis, and the angle between the major axis and the horizontal axis. Then, it divides the stoma edge image according to the major axis, uses Hough transform to detect the length of the semi-minor axis of the upper and lower halves of the ellipse respectively, and determines the optimal semi-minor axis parameter from multiple solutions using the area minimum error method.

[0039] The three-dimensional display module is used to visualize the upper and lower hemispheres of the stomata based on the constructed three-dimensional regular volume data field and according to the ellipsoid parameters.

[0040] The volume calculation module is used to perform three-dimensional reconstruction on the binary image of the pore edge, calculate the pore volume using the volume calculation formula, and calculate the average value of the pore volume in multiple images.

[0041] In some embodiments, the system further includes a centralized management platform; the centralized management platform includes a front-end interface module, a confidentiality management module, a data security management module, and a flaw detection report management module;

[0042] The front-end interface module is developed based on the PyQt library and QtDesigner to create a cross-platform operation interface. The interface includes a left interactive bar and a right preview box. The left interactive bar is used to receive user operation commands, and the right preview box is used to load CT images and 3D reconstructed images. The front-end interface module supports users to import CT slice images, generate 3D images, analyze and interact with 3D images, and output reconstruction results and a defect statistics list.

[0043] The confidentiality management module is used to encrypt sensitive data using the AES-GCM algorithm of the Crypto.js library, store the encryption key in the backend, and ensure data transmission security through HTTPS; based on the RBAC model, access control is implemented by storing user access control lists in the database, and sensitive data access logs are recorded.

[0044] The data security management module includes a data encryption submodule and a data backup and recovery submodule. The data encryption submodule uses a database engine that supports transparent data encryption to encrypt database and file system data and implements a key rotation strategy. The data backup and recovery submodule uses the Veeam Backup & Replication tool to perform full backups and incremental backups and manages the storage and access permissions of backup data.

[0045] The flaw detection report management module is used to store document template information based on the Django backend framework and MySQL / PostgreSQL database, and to design the interface using the React / Vue.js frontend framework and Ant Design / Bootstrap. It generates PDF reports using Python's ReportLab library and processes the report data logic through Django. It uses Django's RESTful API to handle report viewing and modification operations, and implements report interaction through state management. The flaw detection report files are stored on a storage server, and report metadata is stored using a MySQL / PostgreSQL database. Report retrieval is achieved using Elasticsearch.

[0046] Secondly, embodiments of the present invention provide a method for intelligent detection and visual characterization of defects in solid rocket motor propellant grains, the method comprising:

[0047] Based on a preset automatic identification and detection algorithm for circular structural components, structural component identification and localization are performed on the two-dimensional ICT image in the acquired solid rocket motor ICT file to determine the circular structural region image of the solid rocket motor.

[0048] An improved anisotropic diffusion denoising algorithm based on adaptive median filtering removes impulse interference from the circular structure region image multiple times, resulting in a denoised circular structure region image.

[0049] The robust adaptive Canny algorithm based on defect edges sequentially performs circular structure boundary detail extraction and noise interference removal processing on the circular structure region image after multiple denoising processes to obtain a high-precision edge image.

[0050] Based on a pre-defined deep learning-based ICT slice image frame interpolation algorithm, image registration, normalization, and ROI extraction are performed on the original ICT slice images to obtain an integrated continuous slice sequence.

[0051] A 3D convolutional neural network is used to preview the deep features of adjacent slices, and the features are aligned by optical flow estimation or deformable convolution.

[0052] Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices through a U-Net generator. The realism of the generated image is evaluated by a PatchGAN discriminator and optimized through a multi-objective loss function.

[0053] Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data.

[0054] The processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model.

[0055] The reconstructed 3D model is subjected to edge smoothing and voxel fusion optimization to generate a high-precision interactive 3D visualization model.

[0056] The pore images in the two-dimensional ICT image are reconstructed into three-dimensional pores to obtain a three-dimensional reconstruction model of deformed pores and volume estimation results.

[0057] Based on the volume estimation results, the fault diagnosis results of the solid rocket motor are determined.

[0058] Compared with the prior art, the beneficial effects of the technical solution of the present invention are as follows:

[0059] The intelligent detection and visualization characterization system for solid rocket motor propellant grain defects provided in this invention has the following advantages: First, this invention achieves a geometric error of ≤5% between the 3D defect model and the actual structure (the international leading level is 8-12%), ensuring that the morphological reproduction of defects such as porosity and cracks meets engineering diagnostic requirements. Furthermore, the defect spatial coordinate positioning error is ≤0.1% (compared to 0.5-1% in traditional methods), accurately identifying the distribution of defects at the shell / propellant grain interface and providing reliable data support for structural integrity assessment. Second, this invention requires only ≤5 ICT tomographic images to complete 3D reconstruction, reducing data acquisition costs and avoiding the risk of radiation accumulation caused by multiple scans. In addition, the automatic fitting time for the 3D defect model is ≤10 minutes, and combined with GPU parallel acceleration technology, it enables the engineering application of SRM batch detection. (3) This invention supports virtual sectioning, material layer peeling, and stress-defect coupling display, enabling inspectors to intuitively analyze the spatial relationship between defects and the shell / insulation layer, thus improving diagnostic efficiency. Furthermore, this invention achieves automatic statistical analysis based on defect volume, equivalent diameter, and spatial density parameters to generate a risk assessment report that conforms to the MIL-STD-3022 standard, thereby reducing the misjudgment rate of defects within the propellant grain of a solid rocket motor.

[0060] Furthermore, this invention can also achieve a security and resource optimization system: This invention adopts encrypted storage and blockchain evidence storage technology, combined with hierarchical permission management, to achieve three-dimensional permission isolation of "personnel-data-operation", and supports real-time audit log backtracking; broad application prospects and market: Relying on the three-in-one technical advantages of "high precision-fast response-full dimension", this invention has built a full-stack software platform of "three-dimensional intelligent detection-multi-field coupled simulation-autonomous decision-making closed loop", forming an independent and controllable industrial software technology ecosystem. Attached Figure Description

[0061] Figure 1 This is a schematic diagram of the structure of the intelligent detection and visualization characterization system for solid rocket motor propellant grain defects provided by the present invention;

[0062] Figure 2 This is a schematic diagram of the ICT file processing process in the sliced ​​data preprocessing module provided by the present invention; wherein, Figure 2 (a) is the original image; Figure 2 (b) is the extracted circular region map; Figure 2 (c) is the shadow removal image; Figure 2 (d) is a multi-layer structure identification diagram; Figure 2 (e) is the edge detection map; Figure 2 (f) is the noise removal diagram;

[0063] Figure 3 This is a three-dimensional visualization result view provided by the present invention; wherein, Figure 3 (a) is a view along the X-axis; Figure 3 (b) is a view along the Y-axis; Figure 3 (c) is a view along the Z-axis;

[0064] Figure 4 This is a detailed defect illustration provided by the present invention; wherein, Figure 4 (a) Detailed view of the flaw below; (b) Detailed view of the flaw from the side; (c) Detailed view of the flaw above;

[0065] Figure 5 This invention provides a hierarchical structure and internal mold diagram; wherein, Figure 5 (a) is a hierarchical structure diagram; Figure 5 (b) is a drawing of the internal mold;

[0066] Figure 6 This is a schematic diagram of the projection of the deformable bubble model and the ellipsoid model in spherical coordinates provided by the present invention; wherein, Figure 6 (a) is a projection of the deformed bubble model. Figure 6 (b) is a diagram of an ellipsoid model in spherical coordinates;

[0067] Figure 7 This is a schematic diagram of the three-dimensional ellipsoid and its parameters provided by the present invention, wherein, Figure 7 (a) is a schematic diagram of a three-dimensional ellipsoid in space; Figure 7 (b) Schematic diagram of three-dimensional ellipsoid parameters;

[0068] Figure 8 This is a schematic diagram of the Hough transform reconstructed elliptical pore model provided by the present invention;

[0069] Figure 9 This is a schematic diagram of the pore visualization obtained by the Hough transform provided by the present invention;

[0070] Figure 10 This is a schematic diagram of the system operation interface provided by the present invention;

[0071] Figure 11 This is a display image of the three-dimensional reconstruction effect provided by the present invention;

[0072] Figure 12 This is a screenshot showing the interface effect of the detailed information of defect 1 provided by the present invention;

[0073] Figure 13 This is a screenshot showing the interface effect of the detailed information on defect 2 provided by the present invention;

[0074] Figure 14 This is a screenshot showing the interface effect of the detailed information on defect 3 provided by the present invention;

[0075] Figure 15 This is a display image of the interface effect of the three-dimensional engine model provided by the present invention;

[0076] Figure 16 This is the confidentiality management interface provided by the present invention;

[0077] Figure 17 This is a schematic diagram of the overall architecture of the data security management submodule provided by the present invention;

[0078] Figure 18 This is a schematic diagram of the overall architecture of the flaw detection report management submodule provided by the present invention. Detailed Implementation

[0079] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limitations on the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0080] In the following description, references to "some embodiments" refer to a subset of all possible embodiments; however, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the invention have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of the invention pertain. The terminology used in the embodiments of the invention is for the purpose of describing the embodiments of the invention only and is not intended to limit the invention.

[0081] This invention provides an intelligent detection and visualization characterization system for defects in solid rocket motor propellant grains. The system includes: a two-dimensional slice data preprocessing module 100, a three-dimensional visualization modeling module 200, and a model post-processing and analysis module 300. See also... Figure 1 , Figure 1 This is a schematic diagram of the structure of the intelligent detection and visualization characterization system for solid rocket motor propellant grain defects provided in an embodiment of the present invention.

[0082] The two-dimensional slice data preprocessing module 100 is used to identify and locate structural components in the two-dimensional ICT image of the acquired solid rocket motor ICT file based on a preset circular structural component target automatic recognition and detection algorithm, and to determine the circular structural region image of the solid rocket motor.

[0083] An improved anisotropic diffusion denoising algorithm based on adaptive median filtering removes impulse interference from the circular structure region image multiple times, resulting in a denoised circular structure region image.

[0084] The robust adaptive Canny algorithm based on defect edges sequentially performs circular structure boundary detail extraction and noise interference removal processing on the circular structure region image after multiple denoising processes to obtain a high-precision edge image.

[0085] The three-dimensional visualization modeling module 200 is used to perform image registration, normalization and ROI extraction processing on the original ICT slice image based on a preset deep learning ICT slice image frame interpolation algorithm to obtain an integrated continuous slice sequence.

[0086] A 3D convolutional neural network is used to preview the deep features of adjacent slices, and the features are aligned by optical flow estimation or deformable convolution.

[0087] Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices through a U-Net generator. The realism of the generated image is evaluated by a PatchGAN discriminator and optimized through a multi-objective loss function.

[0088] Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data.

[0089] The processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model.

[0090] The reconstructed 3D model is subjected to edge smoothing and voxel fusion optimization to generate a high-precision interactive 3D visualization model.

[0091] The model post-processing and analysis module 300 is used to perform three-dimensional reconstruction of pores in the two-dimensional ICT image to obtain a three-dimensional reconstruction model of deformed pores and volume estimation results.

[0092] Based on the volume estimation results, the fault diagnosis results of the solid rocket motor are determined.

[0093] The intelligent detection and visualization characterization system for solid rocket motor propellant grain defects includes the following:

[0094] (1) By studying the detection method and imaging principle of ICT, and combining related technologies such as signal processing, the data information of the ICT file of the engine propellant is analyzed. Data is extracted from the ICT file using data processing technology and a two-dimensional sequence tomographic image of the propellant is formed. Based on image segmentation technology, the boundary detection and extraction algorithm of two-dimensional image is studied to realize the automatic identification and extraction of the defect boundary scanned in the two-dimensional tomographic image of the propellant.

[0095] (2) Using ICT two-dimensional tomographic images as the initial data source, and using digital image processing, computer graphics and other technologies, we study the frame interpolation and three-dimensional reconstruction algorithms of tomographic images, and transform the processed two-dimensional or three-dimensional data into a three-dimensional model that can accurately describe the spatial characteristics of regular bodies, so that the reconstructed three-dimensional model can intuitively display the morphology and size of defects inside the drug column.

[0096] (3) Research on the precise visualization of three-dimensional defect models, which can realize the visualization display of three-dimensional defect models at any position and in any direction, with particular emphasis on the precise visualization of defects in the shell, insulation layer, lining and propellant grain. It can realize the rotation, scaling and translation functions of the display model through interactive operation, so as to achieve a multi-angle intuitive display of defects inside the engine.

[0097] (4) To achieve precise location, measurement, and statistics of defects in the 3D model, and to design efficient and accurate measurement methods to accurately measure the spatial distance, angle, and volume of defect structures in the 3D model. Through digital descriptions such as coordinates and volume, the size, shape, and location information of defects can be obtained, and interval statistics of internal defects of the SRM can be achieved based on parameters such as the equivalent diameter or maximum projected area of ​​the defects.

[0098] (5) Management of 3D visualization modeling results. The constructed 3D visualization modeling results are stored uniformly. Management and operation personnel can log in to the system to view and export the modeling results and flaw detection reports. The system can provide modeling result files and flaw detection reports in different formats from any viewpoint. The system has security management functions to protect data security.

[0099] In some embodiments, the two-dimensional slice data preprocessing module 100 further includes a circular region extraction submodule 110; the circular region extraction submodule 110 is used to perform binarized background segmentation, morphological processing, and connected component analysis on the two-dimensional ICT image to obtain multiple candidate regions; each candidate region carries the basic features of the solid rocket motor; the basic features include area features, boundary features, and geometric features; non-target regions with areas smaller than a preset area threshold are removed from the multiple candidate regions to obtain retained regions; the retained regions are numbered and segmented to obtain multiple image subsets; and Sobel edge extraction is performed on each image subset. Obtain an edge point set; establish a rectangular coordinate system with the centroid of each edge point set as the origin, and calculate the polar angles of the four vertices of the minimum bounding rectangle relative to the center; calculate the circularity index of each image subset based on the edge point set, and remove regions that do not meet the circularity feature to obtain regions that meet the circularity feature; randomly select an edge point subset in the retained region for circle fitting, and use an optimized sampling strategy to reduce computational complexity to generate candidate circles containing center coordinates and radius parameters; verify each candidate circle, and if a sufficient number of edge points fall within the preset radius tolerance range of each candidate circle, it is determined to be a valid circle, and the circular structure region image is output.

[0100] In some embodiments, the two-dimensional slice data preprocessing module 100 further includes a filtering and denoising submodule 120; the filtering and denoising submodule 120 is used to perform local region division on the circular structure region image, resulting in a circular region image; an adaptive median filter is used to filter the divided circular region image to remove impulse noise, resulting in a pre-denoised circular structure region image; the gradient magnitude of the pre-denoised circular structure region image is calculated, and the gradient magnitude is used to replace the gradient magnitude of the original image to dynamically adjust the diffusion coefficient in the anisotropic diffusion model; based on the adjusted diffusion coefficient, anisotropic diffusion denoising is performed on the pre-denoised circular structure region image to preserve image edge details, resulting in the pre-denoised circular structure region image after multiple denoising steps.

[0101] In some embodiments, the two-dimensional slice data preprocessing module 100 further includes an adaptive Canny submodule 130; the adaptive Canny submodule 130 is used to perform grayscale processing on a selected region in the circular structure region image after multiple denoising steps to obtain a grayscale processed image; construct a Gaussian filter to perform two-dimensional convolution on the grayscale processed image to reduce local detail fluctuations in the image to obtain a convolved image; calculate the gradient magnitude and direction of the convolved image using the finite difference of the first-order partial derivative; perform non-maximum suppression on the gradient magnitude to retain local gradient maximum points; adopt a dual threshold algorithm, using a high threshold to segment the background and target, and a low threshold to connect broken edges; perform fine processing on the edge region, extract the flaw contour and identify potential defect features to obtain the high-precision edge image.

[0102] In some embodiments, the model post-processing and analysis module 300 includes a pore ellipsoid model building module 310, a pore shape parameter extraction module 320, a 3D display module 330, and a volume calculation module 340. The pore ellipsoid model building module 310 is used to simulate deformed pores as two triaxial ellipsoids, and constructs a 3D ellipsoid model by determining nine parameters: the 3D coordinates of the ellipsoid center, the lengths of the three semi-axes, and the triaxial rotation angles. The origin of the ellipsoid model is taken as the center of the solid rocket motor base. The pore shape parameter extraction module 320 is used to preprocess the pore image to obtain the pore edge dimensions. The image is first calculated by directly calculating the coordinates of the ellipse center, the length of the semi-major axis, and the angle between the major axis and the horizontal axis. Then, the stomatal edge image is divided according to the major axis, and the length of the semi-minor axis of the upper and lower halves of the ellipse is detected by Hough transform. The optimal semi-minor axis parameter is determined from multiple solutions using the minimum area error method. The 3D display module 330 is used to visualize the upper and lower halves of the stomatal ellipsoid based on the constructed 3D regular volume data field and the ellipsoid parameters. The volume calculation module 340 is used to perform 3D reconstruction of the binary image of the stomatal edge, calculate the stomatal volume using the volume calculation formula, and calculate the average value of the stomatal volume in multiple images.

[0103] In some embodiments, the system further includes a centralized management platform 400; the centralized management platform 400 includes a front-end interface module 410, a confidentiality management module 420, a data security management module 430, and a flaw detection report management module 440.

[0104] The front-end interface module 410 is used to develop a cross-platform operation interface based on the PyQt library and QtDesigner. The interface includes a left interactive bar and a right preview box. The left interactive bar is used to receive user operation commands, and the right preview box is used to load CT images and 3D reconstructed images. The front-end interface module supports users to import CT slice images, generate 3D images, analyze and interact with 3D images, and output reconstruction results and defect statistics lists.

[0105] The confidentiality management module 420 is used to encrypt sensitive data using the AES-GCM algorithm of the Crypto.js library, store the encryption key in the backend, and ensure data transmission security through HTTPS; based on the RBAC model, it implements access control by storing user access control lists in the database and records sensitive data access logs.

[0106] The data security management module 430 includes a data encryption submodule and a data backup and recovery submodule. The data encryption submodule uses a database engine that supports transparent data encryption to encrypt database and file system data and implements a key rotation strategy. The data backup and recovery submodule uses the Veeam Backup & Replication tool to perform full backups and incremental backups and manages the storage and access permissions of the backup data.

[0107] The flaw detection report management module 440 is used to store document template information based on the Django backend framework and MySQL / PostgreSQL database, and to design the interface using the React / Vue.js frontend framework and Ant Design / Bootstrap. It uses Python's ReportLab library to generate PDF format reports and processes report data logic through Django. It uses the RESTful API provided by Django to handle report viewing and modification operations, and realizes report interaction through state management. It stores flaw detection report files on a storage server, uses MySQL / PostgreSQL database to store report metadata, and combines Elasticsearch to realize report retrieval.

[0108] This invention provides a method for intelligent detection and visualization characterization of defects in solid rocket motor propellant grains, applied to a system for intelligent detection and visualization characterization of defects in solid rocket motor propellant grains, comprising the following steps:

[0109] Step S210: Based on the preset circular structural component target automatic recognition and detection algorithm, the structural component is identified and located in the two-dimensional ICT image in the acquired solid rocket motor ICT file to determine the circular structural region image of the solid rocket motor.

[0110] In some embodiments, a solid rocket motor ICT file refers to a tomographic image data file generated by Industrial Computed Tomography (ICT) scanning the internal structure of a solid rocket motor. This data, generated by X-rays or gamma rays penetrating the object, forms a grayscale distribution reflecting material density and structural differences. This file contains two-dimensional tomographic image information of the solid rocket motor's internal casing, insulation layer, lining, and other structures, as well as potential defects.

[0111] Step S220: Based on the improved anisotropic diffusion denoising algorithm of adaptive median filtering, the impulse interference in the circular structure region image is removed multiple times to obtain the circular structure region image after multiple denoising.

[0112] In this invention, the improved algorithm incorporates the result of adaptive median filtering, replacing the gradient magnitude of the original image with the filtered gradient magnitude, thereby dynamically adjusting the diffusion coefficient in the anisotropic diffusion model. The principle is to calculate the gradient of each pixel in the image, determine whether the point is in an edge region or a smooth region, reduce diffusion in edge regions to preserve edge details, and enhance diffusion in smooth regions to remove noise. This improved anisotropic diffusion denoising algorithm with adaptive median filtering can more effectively suppress impulse noise such as gunpowder residue commonly found in solid rocket motor ICT images, while preserving edge detail structures and enhancing the expressive power of target boundary information.

[0113] Step S230: Based on the robust adaptive Canny algorithm for defective edges, the circular structure region image after multiple denoising processes is sequentially processed for circular structure boundary detail extraction and noise interference removal to obtain a high-precision edge image.

[0114] In this invention, the robust adaptive Canny algorithm for defect edges first uses a Gaussian filter to smooth the image, reducing local detail fluctuations and highlighting the main structural contours. Then, non-maximum suppression is employed to preserve local maxima in the edge response, refining the edges to a single pixel width to ensure the continuity and accuracy of edge detection results. Finally, a dual-threshold method is used to determine the final edge image. A high threshold is used to segment the image content and background to be detected, identifying each effective line segment. A low threshold is then used to extend these line segments in both directions to find edge breaks and connect discontinuous edges in the image. This algorithm can accurately extract minute changes in the boundaries of circular structures, assisting in the discovery of potential material defects or structural anomalies in the boundary region, and improving the sensitivity and reliability of detection.

[0115] Step S240: Based on the preset deep learning ICT slice image frame interpolation algorithm, the original ICT slice image is processed by image registration, normalization and ROI extraction to obtain the integrated continuous slice sequence.

[0116] In this invention, the frame interpolation algorithm aims to improve the continuity and integrity of the model by inserting new intermediate slices between the original slices. It primarily utilizes deep learning models such as 3D CNN and GAN. 3D CNN is used to extract deep features from adjacent slices and capture spatial context information between slices; the U-Net in the GAN architecture is responsible for generating intermediate interpolated slices from two adjacent frames, while PatchGAN is used to evaluate the realism of the generated image. Through multi-objective optimization such as L1 / L2 loss, adversarial loss, and perceptual loss, it ensures that the generated image achieves a high level of pixel-level similarity, visual quality, and semantic consistency.

[0117] Step S250: Use a 3D convolutional neural network to extract deep features from adjacent slices in advance, and align the features through optical flow estimation or deformable convolution.

[0118] Step S260: Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices using a U-Net generator. The realism of the generated image is evaluated using a PatchGAN discriminator and optimized using a multi-objective loss function.

[0119] Step S270: Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data.

[0120] Step S280: Import the processed slice data into the 3D modeling module, construct a 3D voxel mesh and reconstruct the model; perform edge smoothing and voxel fusion optimization on the reconstructed 3D model to generate a high-precision interactive 3D visualization model.

[0121] Step S290: Perform three-dimensional reconstruction of the pore image in the two-dimensional ICT image to obtain a three-dimensional reconstruction model of the deformed pore and a volume estimation result; based on the volume estimation result, determine the fault diagnosis result of the solid rocket motor.

[0122] The following will describe an exemplary application of the embodiments of the present invention in a practical application scenario.

[0123] The intelligent detection and visualization characterization system for solid rocket motor propellant grain defects constructed in this invention adopts a modular architecture design and achieves accurate detection and quantitative evaluation of internal defects in the propellant grain through a multi-level data processing link. The system consists of four core units: a two-dimensional slice data preprocessing module, a three-dimensional visualization modeling module, a model post-processing and analysis module, and a centralized management platform (system architecture as follows). Figure 1 As shown in the figure, its technical process is as follows:

[0124] In the system operation and management interface, encrypted and integrity-verified 2D ICT scan slices are imported into the system through a secure channel and then transmitted to the defect analysis and visualization platform for standardized preprocessing. The 2D processing stage employs a multi-scale feature extraction algorithm to automatically identify the boundaries of multiple structural layers such as the shell, insulation layer, and lining from sequential tomographic images. Based on an improved Canny operator and morphological gradient fusion technology, defect contours are delineated with sub-pixel precision, laying the data foundation for 3D modeling. The preprocessed 2D dataset, after spatial registration, is input into the 3D visualization modeling module. It is first used to construct a 3D regular volume data field and, combined with surface topology optimization, generates a high-fidelity 3D solid model. Subsequently, the 3D volume data is imported into the model post-processing module to achieve precise defect location and measurement, and to generate a structured flaw detection report containing defect distribution maps, quantitative parameter tables, and safety level assessments, completing the entire closed-loop processing. All modules interact seamlessly through a data bus, forming a closed-loop processing chain of "data input - feature analysis - 3D reconstruction - engineering output".

[0125] 1) Two-dimensional slice data preprocessing module

[0126] In the slice data preprocessing module, the incoming ICT files undergo fine preprocessing at the two-dimensional level. This step involves extracting key information from the ICT files and converting it into a matrix format to accurately identify the shell, insulation layer, liner, and any defects. Subsequently, these preprocessed two-dimensional images and related ICT data are sent to the three-dimensional visualization modeling module, laying the foundation for subsequent steps.

[0127] 1.1) Extraction of circular regions

[0128] This invention proposes an automatic identification and detection algorithm for circular structural components, enabling the automatic identification and precise localization of structural components such as shells, insulation layers, and linings in 2D ICT images. This method, tailored to practical engineering needs, sequentially performs image binarization background segmentation, morphological processing, feature region extraction, and improved circular Hough Transform detection based on the circular geometric features of the target. In engineering applications, due to factors such as camera lens distortion, lighting variations, and noise interference, originally regular circular structures in images often exhibit blurred, incomplete, or distorted edges, and their contours may be further distorted after binary segmentation. To address these issues, this method first utilizes connected component analysis to eliminate obvious non-circular regions, marks connected components that meet certain geometric constraints, and divides the original image into several sub-image sets. Subsequently, circular target detection is performed on these candidate sub-images. The overall process is as follows:

[0129] (1) Image preprocessing and region segmentation: First, the input two-dimensional ICT image (including shell, insulation layer and liner) is binarized and morphologically processed. The image is divided into several candidate regions by connected component analysis, and the area, boundary and geometric features of each region are extracted.

[0130] (2) Region filtering and image subset generation: Non-target regions with an area smaller than a set threshold are removed to eliminate noise interference and regions of no interest. The retained regions are numbered and divided into independent image subsets for subsequent processing.

[0131] (3) Edge detection and coordinate system establishment: Apply the Sobel operator to each image subset to extract edges and obtain the edge point set. Establish a rectangular coordinate system with its centroid as the origin in each region, and calculate the polar angles of the four vertices of the minimum bounding box relative to the centroid to further assist in the circularity judgment.

[0132] (4) Circularity feature extraction and non-circularity removal: Based on the extracted edge point set, calculate the circularity index of the region (such as roundness, boundary symmetry, curvature consistency, etc.) and remove regions that do not meet the circularity feature, thereby improving the accuracy and robustness of circle fitting.

[0133] (5) Circle detection based on improved stochastic Hough transform: A subset of edge points is randomly selected in the retained region for circle fitting to generate candidate circles, and their center coordinates and radius parameters are recorded. This process uses an optimized sampling strategy to reduce computational complexity.

[0134] (6) Candidate circle verification and true circle screening: Each candidate circle is verified. If a sufficient number of edge points fall within its radius tolerance range, it is considered a valid circle; otherwise, false circles are excluded. The final output is the target circle information used for engineering judgment, which serves as the basis for identifying components such as shell, insulation layer, and lining.

[0135] 1.2) An Improved Anisotropic Diffusion Denoising Algorithm Based on Adaptive Median Filtering

[0136] After the circular target is extracted, to further improve recognition accuracy, this invention addresses the problem of abundant gunpowder slag noise commonly found in solid rocket motor ICT images. It designs an improved anisotropic diffusion denoising algorithm based on adaptive median filtering. Specifically, after extracting circular structural regions such as the shell, insulation layer, and liner, local filtering is performed only within these candidate circular regions, avoiding redundant calculations for non-target regions outside the shell, thus effectively improving processing efficiency. This method significantly enhances the suppression of non-Gaussian gunpowder slag noise by introducing an adaptive median filter. Specifically, adaptive median filtering is first used to accurately remove impulse interference (such as gunpowder slag), and the gradient modulus of the original image is replaced with the filtered gradient modulus. This allows for dynamic adjustment of the diffusion coefficient in the anisotropic diffusion model, achieving refined control of the diffusion process. Compared to traditional anisotropic diffusion models that only smooth Gaussian noise, this method maintains the smoothing effect of Gaussian noise while providing more targeted and efficient suppression of impulse interference such as gunpowder slag. The finally constructed anisotropic adaptive median diffusion filter not only effectively removes mixed noise in circular regions but also preserves edge details in the image well, enhancing the expressive power of target boundary information. The mathematical definition of this filter is as follows:

[0137] ;

[0138] In the formula, Indicates the location and diffusion time Image intensity at time, initial image is . Representing an image The gradient of is a two-dimensional vector, defined as: . It is a divergence operator used to calculate the degree of divergence of a vector field. The diffusion process is controlled by partial differential equations, the core idea of ​​which is to guide the image to selectively smooth in a non-uniform space by using the direction and magnitude of the image gradient. Compared with traditional anisotropic diffusion methods, this model introduces adaptive median filtering (AMF) to preprocess the image and modulate the filtered gradient magnitude. The original gradient mode is replaced, thereby dynamically adjusting the diffusion coefficient. This improved strategy can more effectively suppress impulse noise such as gunpowder residue in solid rocket motor ICT images, while preserving key edge structures and details in the images, enhancing boundary representation capabilities, and providing a more reliable data foundation for subsequent target recognition.

[0139] After extracting circular structures such as the shell, insulation layer, and lining, and suppressing gunpowder slag noise, this invention proposes an efficient and robust edge detection method—the robust adaptive Canny algorithm for defect edges—to further extract regional boundary details and identify potential defects. First, the extracted circular structure regions in the denoised image are used as the processing targets to ensure the edge detection process is highly targeted and efficient. Then, the selected regions are converted to grayscale to reduce data dimensionality while preserving key structural information, providing a clear image foundation for subsequent edge extraction.

[0140] Based on grayscale images, this method employs the Canny edge detection operator for high-precision image processing, extracting fine boundaries at the pixel level. The Canny operator, with its excellent noise suppression capabilities, high edge localization accuracy, and good continuity performance, is particularly suitable for ICT image analysis of solid rocket motors with complex structures and high detail requirements. Using this operator, the system can accurately capture minute changes in the boundaries of circular structures, thereby assisting in the discovery of potential material defects or structural anomalies in the boundary regions, improving the sensitivity and reliability of detection.

[0141] It should be noted that the extracted edge information not only serves defect detection but also provides crucial input for subsequent 3D modeling and visualization analysis. This high-precision edge data enables more accurate geometric constraints and shape contours in 3D space, supporting the three-dimensional reconstruction of complex internal engine structures. This process significantly improves the accuracy and completeness of recovering 3D models from 2D image data, providing solid technical support and data assurance for the digital inspection, structural visualization, and subsequent performance evaluation of solid rocket motors.

[0142] The basic idea of ​​this method is to first smooth the image using a Gaussian filter to reduce local detail fluctuations and highlight the main structural contours. Then, non-maximum suppression (NMS) is employed to accurately preserve local maxima of the edge response, ensuring the continuity and accuracy of edge detection results. Finally, a dual-threshold method is used to determine the final edge image, further enhancing the detection effect on real boundaries. In practical solid rocket motor ICT image analysis, this preliminary smoothing process using Gaussian filtering not only helps improve the accuracy of edge localization but also lays a good foundation for subsequent structural modeling and defect identification. Its Gaussian function... :

[0143] ;

[0144] in, Represents coordinates in two-dimensional space Gaussian function value at that location, It is the standard deviation, which controls the degree of diffusion of the Gaussian distribution; and These represent the horizontal and vertical offsets relative to the center pixel of the filter, respectively. It is a natural exponential function, and the entire function exhibits a centrally symmetrical bell-shaped distribution.

[0145] After constructing the Gaussian function filter, use Gaussian filtering. For images Perform two-dimensional convolution operations:

[0146] ;

[0147] In the formula, It refers to the number of images, specifically the number of images converted to grayscale. ,image This is the filtered image. The magnitude and direction of the gradient are calculated using finite differences of the first-order partial derivatives, where the first-order difference convolution is:

[0148] .

[0149] Next, the two difference convolution templates are convolved with the image respectively to obtain its horizontal and vertical gradients. The magnitude and direction are then calculated using the gradients, as follows:

[0150] ;

[0151] ;

[0152] ;

[0153] ;

[0154] In the formula, and For the horizontal and vertical gradients, It is the gradient magnitude. This refers to the gradient direction. To determine image edges, non-maximum suppression (NMS) needs to be applied to the obtained gradient magnitude. Points with the largest local gradient values ​​are retained instead of being zeroed out, thus achieving a refined edge effect.

[0155] Finally, a dual-threshold algorithm is used to detect edges. Two thresholds are selected. and ( < ), using high threshold Segment the image content to be detected from the background, find each valid line segment; and use a low threshold. Extend these line segments in both directions to find breaks in the edges and connect the discontinuous edges of the image. Use two thresholds. and Two threshold edge images were obtained. and . This method uses a high threshold, which can effectively separate the background from the image content and contains only a few false edges. However, the excessively high threshold results in discontinuous edges. The algorithm connects the edges to form a contour, and when it reaches the endpoint of the contour, it... Find the edges that can be connected to the contour from the A neighboring points, and use the same method to... The process continuously collects and connects edges until... Continue until it becomes uninterrupted.

[0156] After edge detection, this method further refines the edge region, extracting defect contours and identifying potential defect features. This process not only accurately locates potential material defects or structural anomalies but also provides crucial geometric information support for subsequent 3D modeling and visualization analysis.

[0157] To ensure the accuracy of 3D modeling, the system specifically preserves key edge contours as mesh boundaries for geometric reconstruction. This edge-preservation strategy effectively reduces interference from redundant data, ensuring the integrity and accuracy of the reconstructed model at the level of detail. This strategy provides more constrained boundary information in 3D space, offering accurate geometric basis for high-precision 3D reconstruction of the complex internal structure of solid rocket motors.

[0158] This series of processes not only significantly improves the reliability of subsequent digital analysis and diagnosis, but also lays a solid technical foundation for the digital inspection and evaluation of solid rocket motors. Ultimately, it provides more accurate and efficient support for engine performance evaluation, structural safety analysis, and defect location.

[0159] 2) 3D Visualization Modeling Module

[0160] After completing the 2D processing, the processed CT file will be input into the 3D visualization modeling module for 3D modeling to achieve precise visualization of defects. This step is crucial for in-depth investigation of defects inside solid rocket motor fuel.

[0161] 2.1) Construction of a three-dimensional regular volume data field

[0162] In order to perform three-dimensional reconstruction of two-dimensional CT slices of solid rocket motors, this invention needs to use the information given by two-dimensional sequence tomographic images to analyze three-dimensional volume data and construct a three-dimensional regular volume data field.

[0163] Common methods for constructing 3D regular volume data fields include: (1) Surface reconstruction: Extracting surface points from a 2D cross section. Connecting surface points into a mesh using triangulation or other algorithms. Advantages: Can generate high-resolution models. Disadvantages: May contain holes or topological errors. (2) Volume-based reconstruction: Treating a 2D cross section as a projection of a 3D volume. Using a back-projection algorithm to merge the projections into a 3D volume. Advantages: Can generate smooth, continuous data. Disadvantages: High computational cost, may produce artifacts. (3) Deep learning-based reconstruction: Using neural networks to learn 3D representations from 2D cross sections. Generating 3D data through techniques such as Generative Adversarial Networks (GANs) or autoencoders. Advantages: Can generate realistic models. Disadvantages: Requires a large amount of training data, may contain biases.

[0164] Three-dimensional volume datasets are characterized by being multidimensional arrays composed of scalar or vector data. These data are typically defined on a mesh structure, representing values ​​sampled in three-dimensional space. There are two basic types of three-dimensional volume data: (1) scalar three-dimensional volume data, where each point contains one value; and (2) vector three-dimensional volume data, where each point contains two or three values ​​that define the components of a vector. Considering the characteristics of solid rocket motors, this invention chooses to use scalar three-dimensional volume data to represent solid rocket motors. The specific steps are as follows:

[0165] (1) First, for each two-dimensional tomographic image sequence, a regularized grid partitioning method is used to spatially discretize it at the pixel or voxel level, mapping the image pixels to two-dimensional grid points to form a data representation based on the grid structure. This operation provides a uniform and regular spatial sampling basis for subsequent three-dimensional data construction, ensuring the spatial consistency and operability of the data distribution.

[0166] (2) Subsequently, the actual spatial coordinate information of each two-dimensional tomographic image is extracted and integrated into the corresponding two-dimensional grid point data to complete the mapping from pixel plane coordinates to the actual physical space coordinate system. This step is crucial for constructing a high-precision three-dimensional volume dataset, ensuring the accurate positioning and registration of each tomographic image in three-dimensional space.

[0167] (3) By stacking all two-dimensional grid image data with spatial coordinates layer by layer in an orderly manner according to the scanning order or tomographic slice sequence, a three-dimensional array with voxel consistency and spatial topology is formed. This three-dimensional array completely records the geometric structure and physical property distribution of the tomographic image in three-dimensional space, forming the core data framework for subsequent three-dimensional modeling and analysis.

[0168] (4) To overcome the limitations of spatial sampling and achieve continuous reconstruction of the data field, cubic interpolation is used to perform numerical interpolation calculations on the three-dimensional array for the above-mentioned three-dimensional discrete data field. Specifically, cubic interpolation takes the position of the query point in the three-dimensional space as the reference, selects the values ​​of four adjacent grid nodes in each dimension, and performs weighted fitting using cubic convolution kernels or cubic polynomial basis functions. This interpolation strategy can significantly improve the overall reconstruction accuracy of the volume data while ensuring the smoothness of local data, and is particularly suitable for the construction of three-dimensional volume data fields with rich details and high spatial continuity requirements.

[0169] (5) Finally, the continuous three-dimensional data field generated after three interpolation processes is further organized and packaged to form a high-resolution, regularized three-dimensional volume dataset that is suitable for subsequent three-dimensional visualization, geometric modeling, and fine analysis. This dataset not only has good spatial consistency and continuity, but also provides high-quality geometric and physical data support for the three-dimensional morphology reconstruction, defect detection, and digital evaluation of complex structures such as solid rocket motors, ensuring the reliability and accuracy of digital detection and structural analysis results.

[0170] 2.2) 3D Visualization Modeling

[0171] This invention addresses the practical needs of solid rocket motors with complex internal structures and subtle defect distributions. It proposes a 3D visualization modeling method based on ICT images, combining various image processing and 3D modeling optimization techniques to achieve high-precision modeling and defect visualization of structural components such as the casing, insulation layer, and liner. 3D modeling not only realistically reproduces the shape and size of defects inside the propellant grain but also provides a reliable basis for defect identification, performance evaluation, and maintenance decisions.

[0172] The entire 3D visualization process is based on ICT images. First, the ICT slice images undergo 2D processing to extract the areas to be visualized, including: shell edges, shell itself, insulation layer, lining, and potential defect contours. Then, the processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model. During modeling, the actual size differences of the ICT slices in space are fully considered. Frame interpolation technology is used to insert new data layers between the original slices to improve the continuity and integrity of the model. To enhance modeling accuracy and visual quality, this invention proposes several optimization strategies, including a deep learning-based ICT slice image frame interpolation algorithm, edge smoothing processing, and voxel fusion algorithms.

[0173] 2.2.1) Deep Learning-Based ICT Sliced ​​Image Frame Interpolation Algorithm

[0174] In 3D visualization workflows, ICT slice images often suffer from insufficient inter-layer resolution, leading to staircase artifacts in the reconstructed 3D model. To address this, we propose a deep learning-based ICT slice image interpolation algorithm. This algorithm first preprocesses the original data, including image registration, normalization, and ROI extraction, to ensure the accuracy and consistency of the input data. Then, a 3D convolutional neural network (3D CNN) is used to extract deep features from adjacent slices, and optical flow estimation or deformable convolution is used to align features, ensuring that the interpolated intermediate layers are spatially consistent with the real data. In the frame generation stage, we employ a GAN architecture, where the generator (U-Net) is responsible for generating intermediate interpolated slices from two adjacent frames, while the discriminator PatchGAN is used to evaluate the realism of the generated images. Multi-objective optimization, including L1 / L2 loss, adversarial loss, and perceptual loss, ensures pixel-level similarity, visual quality, and semantic consistency of the generated images. In the post-processing stage, the algorithm further employs anisotropic diffusion filtering for edge smoothing and uses voxel fusion technology to seamlessly integrate interpolated data with the original data to improve the continuity of 3D reconstruction.

[0175] 2.2.2) Edge smoothing and voxel fusion algorithm

[0176] In the 3D visualization process, edge smoothing and voxel fusion algorithms are key steps in improving model quality. Edge smoothing employs anisotropic diffusion filtering technology, dynamically adjusting the smoothing intensity to eliminate noise interference while preserving key structural features such as shell edges, insulation layer boundaries, and defect contours. The algorithm first calculates the 3D gradient of each voxel (…). The algorithm identifies high-gradient regions (such as edges) and low-gradient regions (such as the interior of homogeneous materials), and then performs iterative diffusion based on the Perona-Malik equation. This maintains edge sharpness in high-gradient regions and implements uniform smoothing in low-gradient regions, effectively solving the edge blurring problem caused by traditional Gaussian filtering. For interfaces between different materials (such as metal shells and rubber insulation layers), the algorithm introduces material property constraints and guides the diffusion process through pre-segmented semantic label maps, ensuring natural transitions at material boundaries. In the voxel fusion stage, the algorithm first establishes a multi-resolution voxel pyramid, performs global registration at a coarse-grained level to eliminate minor misalignments between interpolated slices and the original data, and then employs a weighted fusion strategy based on local similarity at a fine-grained level. By calculating the structural similarity and gray-level histogram matching degree within the neighborhood window, the fusion weights are dynamically adjusted to ensure that the transition region maintains geometric coherence while avoiding artifacts. For regions containing potential defects, the algorithm combines morphological opening and closing operations for preprocessing, removing isolated noise points and repairing discontinuous defect contours, and then generates smooth isosurfaces using the MarchingCubes algorithm. Ultimately, the fused voxel data is rendered using GPU-accelerated 3D texture mapping, supporting real-time interactive visualization.

[0177] Figure 2 It is the ICT file processing process in the sliced ​​data preprocessing module. Figure 2 (a) is the original image; Figure 2 (b) involves extracting circular regions, preserving the main structure, and removing background redundancy; Figure 2 (c) is shadow removal, which suppresses image interference factors such as gunpowder residue and improves image uniformity; Figure 2 (d) is the identification of multi-layer structures, highlighting the boundaries between material layers such as shell, insulation layer, and lining; Figure 2 (e) is edge detection, which accurately extracts the structural contour and potential defect edges; Figure 2 (f) is noise removal, which further removes residual irrelevant content, retains effective boundary information, and provides a clear image foundation for subsequent modeling and recognition.

[0178] In terms of visualization module implementation, this invention develops a 3D modeling and display system based on the open-source Mayavi library. Mayavi, built on the Visualization Toolkit (VTK), supports interactive 3D visualization of scalar fields, vector fields, and tensor data. Its powerful rendering capabilities and flexible view control functions allow the generated 3D model to not only be rotated, scaled, and sectioned, but also to clearly observe structural details and defect distribution through transparency adjustments. In practical applications, this visualization system helps domain experts intuitively identify defect types and locations, supports viewing the model from multiple angles, thereby improving the accuracy and depth of defect analysis. The final reconstructed 3D model not only meticulously displays the spatial hierarchy of the engine casing, insulation layer, and lining, but also accurately reflects the distribution characteristics of local defects. Through the graphical interface, any area can be magnified for observation, achieving panoramic scanning and local analysis of defects, greatly enhancing human-computer interaction efficiency and judgment reliability. Figure 3 This is a schematic diagram of the 3D visualization result, in which Figure 3 (a) is a 3D visualization result view from the X-axis direction; Figure 3 (b) is a three-dimensional visualization view from the Y-axis direction; Figure 3 (c) is a three-dimensional visualization view from the Z-axis direction. Figure 4 These are detailed images showcasing the flaws; among them, Figure 4 (a) is a detailed illustration of the defects shown below; Figure 4 (b) A detailed illustration of the lateral defects; Figure 4 (c) is a detailed illustration of the above flaws. Figure 5 This invention provides a hierarchical structure and internal mold diagram; wherein, Figure 5 (a) is a hierarchical structure diagram; Figure 5 (b) is a diagram of the internal mold. Figure 5 The visualization of some reconstructed models and typical defects is shown, verifying the applicability and effectiveness of the method of the present invention in the three-dimensional modeling and defect presentation of complex structures.

[0179] 3) Model Post-processing and Analysis Module

[0180] In the fault diagnosis of solid rocket motors, measuring the distance to different types of defects to further assess their impact on the engine is a crucial aspect of nondestructive testing (NDT) research. Detection of deformable porosity is a paramount fault diagnosis method for solid rocket motors. It leverages the characteristic that the porosity is a rotating body at the moment of formation. Typically, a deformable porosity can be considered to be simulated by two hemispherical structures, each a triaxial ellipsoid (Jacobi ellipsoid). Using porosity images obtained from ICT image detection of the solid rocket motor, a 3D reconstruction of the porosity is performed, allowing for a rough estimation of the volume of the rising porosity.

[0181] 3.1) Establishment of the ellipsoidal model of stomata

[0182] Figure 6 (a) is a projection diagram of the deformable bubble model provided by the present invention; Figure 6 (b) is an ellipsoidal model diagram in spherical coordinates provided by the present invention, as shown in the appendix. Figure 6 As shown, stomata can be simulated using two hemispherical structures, one above the other. The length of the semi-major axis is... The semi-central axis lengths are respectively and The semi-minor axis length is Appendix Figure 6 (a) is the projection of the ellipsoidal model onto the projection plane. This parameter allows us to represent various different pore shapes, such as a hemispherical cap shape with concave textures. ), hemispherical cap shape ( ), deformed sphere ( ),spherical( (Attached) Figure 6 In the spherical coordinate system shown in (b), the parametric equation of the upper semi-ellipse is:

[0183] .

[0184] In the formula, Indicates the polar angle; 'r' represents the azimuth angle; 'r' represents the radial distance.

[0185] Figure 7 (a) is a schematic diagram of a three-dimensional ellipsoid in space; Figure 7 (b) Schematic diagram of three-dimensional ellipsoid parameters, as attached. Figure 7 As shown, at least nine parameters are needed to uniquely determine a three-dimensional ellipsoid in space: the three-dimensional coordinates of the ellipsoid's center (… The lengths of the three semi-axes of the ellipsoid , , and the attitude angles of its three principal axes in space. ,in, express shaft and The angle between the axes, express shaft and The angle between the axes, express shaft and The angle between axes.

[0186] Based on the aforementioned constructed three-dimensional regular volume data field, the specific location of the vents is directly determined by establishing a three-dimensional Cartesian coordinate system. The origin of the three-dimensional coordinate system is taken at the exact center of the solid rocket motor base. According to the numerical results of the three-dimensional modeling, the edge coordinates of the vents in the newly established coordinate system can be directly determined.

[0187] 3.2) Extracting stomata shape parameters using Hough transform

[0188] Typically, the projection of pores in ICT images can be simulated as two semi-ellipses, one above the other, where the semi-major axis is... The semi-minor axes are respectively and . Variation parameters This can represent pores of different shapes. In solid rocket motors, pores are mostly deformed spheres (…). Taking the upper part of the stomata as an example, let the angle between the major axis of the elliptical model of the stomata and the horizontal axis of the image coordinate system be θ. The semi-major axis of the elliptical model is The semi-minor axis length is The center of the elliptical model is Then the equation of this elliptical model is:

[0189] .

[0190] By using the edge image of the stomata and the Hough transform method, the shape parameters of the elliptical model of the stomata can be extracted from the stomata image. As shown in the above formula, determining an ellipse requires identifying at least five parameters from the image: the x and y coordinates of the ellipse center. The angle between the major axis of the ellipse and the horizontal axis of the image coordinate system ellipse semi-major axis length Semi-minor axis length .

[0191] Using the Hough transform to detect 5-dimensional ellipses would consume excessive resources; therefore, it is necessary to reduce the dimensionality and decrease the computational load of the Hough transform. In this invention, the center of the ellipse of the pore is first directly extracted. Semi-major axis length The angle between the major axis of the ellipse and the horizontal axis of the image coordinate system Then, the length of the semi-minor axis of the ellipse is detected using the Hough transform. The specific testing steps are as follows:

[0192] (1) Perform image preprocessing on the image to obtain a binary image of the stomatal edges;

[0193] (2) Among the edge points of the image, find the distance between two edge points. The longest distance is the major axis, and the two points are the two endpoints of the major axis. From these two endpoints, the angle between the major axis and the horizontal axis can be found.

[0194] (3) To ensure that the center of the ellipse lies on the major axis, the geometric center of the stomata pattern is calculated as the center of the ellipse. The closed stomata edges are scanned row by row, and the two boundary points that are far apart in the i-th row are selected. and Take the horizontally symmetrical point ; Scan the edges of the closed pores column by column, and at the first... Take the two boundary points that are farthest apart within the column. and Take the vertically symmetrical point ;when When, take the coordinates of the ellipse center. ;

[0195] (4) Using the major axis as the dividing line, the stomatal edge image is divided into upper and lower parts. The Hough transform is used to detect ellipses in the upper and lower parts respectively, and the upper and lower semi-minor axes of the ellipses are extracted. and For parameters and exist Within the range, according to the model equations established above, with the center of the ellipse... semi-major axis and included angle Draw an ellipse on the image plane and count the number of edge points that fall on the ellipse. Finally, find the largest one. The parameter corresponding to the value and The value of corresponds to the length of the semi-minor axis of the ellipse. When extracting the semi-minor axis parameter of an ellipse using the Hough transform, multiple solutions may be obtained (with the same number of edge points falling on the ellipse). In this case, the minimum area error method is needed to find the optimal parameter from the multiple solutions.

[0196] (5) The parameters are obtained through the above calculations: This allows us to reconstruct the elliptical model of the pores, as shown in the attached figure. Figure 8 As shown. To clearly observe the reconstruction effect, the binary image of the stomatal edges is displayed in reverse colors, from the attached... Figure 8As can be seen, the ellipse detected by the Hough transform method can reconstruct the true shape of the stomata relatively accurately.

[0197] 3.3) Ventilation based on three-dimensional regular volume data field

[0198] Since the 3D ellipsoidal model of the bubble is composed of two different hemispheres, and the reconstruction methods for the two hemispheres are the same, the above method is used to visualize the upper and lower hemispheres of the ellipsoid based on the obtained bubble feature parameters. (See attached image.) Figure 9 Taking the bubble image in the image as an example, the visualization result of the ellipsoid is shown in the attached figure. Figure 9 As shown.

[0199] 3.4) Calculation of pore volume

[0200] The volume of the pores can be calculated using the following volume calculation formula by sequentially reconstructing the ICT image sequence of the solid rocket motor: In the formula, The length of the semi-major axis is the semi-minor axis length, and c is the shortest principal axis length.

[0201] After calculating the pore volume from multiple images, the average pore volume is calculated. The accuracy of the pore volume calculation method based on the ellipsoidal reconstruction algorithm depends on many factors, including the shape and size of the defective bubble itself, the accuracy of the two-dimensional tomographic images, the accuracy of automatic defect contour feature recognition, and the effect of three-dimensional reconstruction. Although the accuracy of this three-dimensional reconstructed pore volume needs further verification, the pore volume calculation method based on the ellipsoidal reconstruction algorithm has a significant advantage in computational efficiency and is expected to be applied in the fault diagnosis process of solid rocket motors, thereby further evaluating the impact of specific defects on the engine.

[0202] 4) Centralized management platform

[0203] 4.1) Operating System Management Interface

[0204] This invention uses the Qt framework to develop the front-end software interface. PyQt is a Python library for creating desktop applications, offering rich features and tools that make front-end software interface development simpler and more efficient. PyQt can run on multiple platforms, including Windows, macOS, and Linux, so the same codebase can be used to create applications suitable for different operating systems. PyQt provides a rich widget library, including buttons, text boxes, list boxes, menus, toolbars, etc., and various types of interface elements can be quickly created using Qt Designer.

[0205] The overall interface for system operation. (See attached image) Figure 10As shown, the left side is the interaction panel, where users can interact with the software using designated buttons. The right side is the preview box, used to load CT images and the reconstructed 3D images. After opening the software, clicking the "Import File" button allows users to select and import saved CT slice images. Clicking the "Import" button will redirect the user to the file management system for further selection.

[0206] After importing, click the "Generate 3D" button. The system will generate a 3D image based on the imported 2D CT scan, and the result will be displayed on the right-hand side of the interface, as shown in the attached image. Figure 11 After generation, you can preview and interact with the mouse on the right. Clicking to enter the analysis mode allows you to slice and analyze the 3D image by controlling the analysis angle and height with the mouse. Selecting a defect point will display the defect parameters (location, volume, etc.) in the defect parameter display module below. You can also selectively display these parameters by checking boxes. You can zoom, rotate, and translate the 3D image using the buttons on the left. Finally, clicking the "Output Reconstruction Results" button will output the 3D reconstructed image and a defect statistics list. Furthermore, the system supports displaying individual defects for visualization, as shown in the attached image. Figure 12 , 13 As shown in Figure 14, users can scale, rotate, and translate defects, and can output detailed information such as the defect's volume, location, and spatial parameters.

[0207] For multiple slices, the system supports direct import of 3D engine models and allows rotating partial slices for 3D reconstruction. Furthermore, the system supports marking defect locations on the 3D engine model and can directly export 3D engine models with defect markings. Users can perform scaling, rotation, and translation operations on the 3D engine model, as shown in the attached figure. Figure 15 As shown.

[0208] 4.2) Confidentiality Management Interface

[0209] To achieve robust data security management, this invention uses the Crypto.js library to implement AES encryption for sensitive data on the front end. Encryption keys are stored on the back end, and HTTPS ensures data security during transmission. Access control: User access control lists (ACLs) are stored in the database to restrict user access to sensitive data. Access auditing policies are implemented, and access logs for sensitive data are recorded. Technical solution: Encryption and decryption: The AES-GCM algorithm in the Crypto.js library is used for data encryption and decryption. Encryption keys are updated regularly, and key distribution is performed securely. Access control: A user role table is designed in the database, defining the association between roles and resources. RBAC (Restricted Access Control) model is used for permission management to ensure that users can only access data they are authorized to access. The interface is attached. Figure 16 As shown.

[0210] 4.3) Data Security Management Submodule

[0211] Data security is a crucial task in software design. This invention's data security management submodule design includes two important submodules: data encryption and backup and recovery. We will discuss the technical approach, implementation scheme, and possible best practices for these submodules. The architecture of the entire data security management submodule is shown in the attached figure. Figure 17 As shown.

[0212] 4.3.1) Data Encryption

[0213] Data encryption is a key measure for maintaining data security. Encrypting data in databases and file systems effectively prevents unauthorized access and data leakage. To achieve this, this invention uses a database engine that supports Transparent Data Encryption (TDE), such as MySQL or Microsoft SQL Server. These database engines have built-in strong encryption capabilities, allowing data encryption without modifying application code. When enabling TDE in the database, appropriate key management policies must be set up, including key generation, storage, and access control. Enabling TDE on the database server and configuring the database instance to support transparent data encryption ensures that data is encrypted when stored on disk, thereby improving overall data security. Regularly implementing a key rotation strategy can prevent potential data impacts from key breaches or leaks.

[0214] 4.3.2) Data Backup and Recovery

[0215] Backup and recovery are critical measures to ensure data integrity and recoverability. By regularly backing up data and designing disaster recovery plans, the risk of data loss and business interruption can be minimized.

[0216] When selecting a backup tool, this invention employs the professional backup tool Veeam Backup & Replication, which supports both full and incremental backups. A regularly executed backup schedule is established to ensure data integrity and recoverability. When configuring Veeam Backup & Replication, full and incremental backup plans are set to ensure data is backed up regularly, minimizing the risk of data loss.

[0217] Managing the storage and access permissions of backup data can be achieved by using a backup repository, including storage quotas and access controls. This allows for effective management of backup data and ensures its security.

[0218] 4.4) Flaw Detection Report Management Submodule

[0219] The entire flaw detection report management submodule is shown in the attached document. Figure 18 As shown, generating a large number of flaw detection reports is common during solid rocket motor defect detection. To standardize and simplify the report generation process, this invention first designs a document template management system.

[0220] 4.4.1) Document Template Management

[0221] This invention uses Django as the backend framework, combined with MySQL or PostgreSQL databases to store document template information. On the frontend, this invention uses React or Vue.js as the frontend framework, along with Ant Design or Bootstrap for interface design. By establishing a Django-based backend system, the template information of flaw detection reports can be effectively managed and stored. On the frontend, this invention utilizes React or Vue.js to build the user interface, providing an intuitive and user-friendly experience.

[0222] 4.4.2) Report Generation

[0223] Report generation is one of the core steps in the entire process. To efficiently generate reports and ensure accuracy, this invention designs a report generation system. It uses Python libraries, such as ReportLab, to generate PDF reports and leverages Django to handle the data logic during report generation. This invention develops a backend service that uses the Python ReportLab library to generate PDF reports. Simultaneously, the Django framework handles the data logic during report generation to ensure the accuracy and completeness of the report content.

[0224] 4.4.3) Viewing and Modifying

[0225] The generated flaw detection reports need to be viewed and modified at any time to meet user needs. This invention uses the RESTful API provided by Django to handle viewing and modification operations, and processes the report's data logic. It uses the React or Vue.js framework, combined with Redux or Vuex for state management, to implement the report viewing and modification functionality. This invention establishes a RESTful API, using the Django framework to handle user viewing and modification requests, and ensures the consistency of report data. On the front end, this invention uses React or Vue.js to build the user interface, and combines it with Redux or Vuex for state management to enable report viewing and modification.

[0226] 4.4.4) Storage and Retrieval

[0227] To efficiently manage flaw detection report data, a suitable storage and retrieval scheme is required. This invention plans to store flaw detection report files and related data on a storage server, such as AWS S3, to ensure data security and reliability. Simultaneously, this invention utilizes MySQL or PostgreSQL databases to store report metadata and integrates Elasticsearch to achieve more efficient report retrieval functionality, thereby improving user experience. Through this storage and retrieval scheme, this invention can effectively manage large amounts of report data, ensuring its security and reliability.

[0228] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the invention. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of the invention, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the invention. The sequence numbers of the above-described embodiments of the invention are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0229] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A smart detection and visualization characterization system for defects in solid rocket motor propellant grains, characterized in that, The system includes: a two-dimensional slice data preprocessing module, a three-dimensional visualization modeling module, and a model post-processing and analysis module; The two-dimensional slice data preprocessing module is used to identify and locate structural components in the two-dimensional ICT image of the acquired solid rocket motor ICT file based on a preset circular structural component target automatic recognition and detection algorithm, and to determine the circular structural region image of the solid rocket motor. An improved anisotropic diffusion denoising algorithm based on adaptive median filtering removes impulse interference from the circular structure region image multiple times, resulting in a denoised circular structure region image. The robust adaptive Canny algorithm based on defect edges sequentially performs circular structure boundary detail extraction and noise interference removal processing on the circular structure region image after multiple denoising processes to obtain a high-precision edge image. The three-dimensional visualization modeling module is used to perform image registration, normalization and ROI extraction processing on the original ICT slice image based on a preset deep learning ICT slice image frame interpolation algorithm to obtain an integrated continuous slice sequence. A 3D convolutional neural network is used to preview the deep features of adjacent slices, and the features are aligned by optical flow estimation or deformable convolution. Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices through a U-Net generator. The realism of the generated image is evaluated by a PatchGAN discriminator and optimized through a multi-objective loss function. Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data. The processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model. The reconstructed 3D model is subjected to edge smoothing and voxel fusion optimization to generate a high-precision interactive 3D visualization model. The model post-processing and analysis module is used to perform three-dimensional reconstruction of pores in the two-dimensional ICT image to obtain a three-dimensional reconstruction model of deformed pores and volume estimation results. Based on the volume estimation results, the fault diagnosis results of the solid rocket motor are determined.

2. The system according to claim 1, characterized in that, The two-dimensional slice data preprocessing module also includes a circular region extraction submodule; The circular region extraction submodule is used to perform binarized background segmentation, morphological processing, and connected component analysis on the two-dimensional ICT image to obtain multiple candidate regions; each candidate region carries the basic features of the solid rocket motor; the basic features include area features, boundary features, and geometric features. Non-target regions with areas smaller than a preset area threshold are removed from the candidate regions to obtain the retained regions; The reserved regions are numbered and segmented to obtain multiple image subsets; For each subset of images, Sobel operator edge extraction is performed to obtain an edge point set; a rectangular coordinate system is established with the centroid of each edge point set as the origin, and the polar angles of the four vertices of the minimum bounding rectangle relative to the center are calculated; Based on the set of edge points, the circularity index of each of the image subsets is calculated, and regions that do not meet the circularity feature are removed to obtain regions that meet the circularity feature. A subset of edge points is randomly selected in the reserved area for circle fitting. An optimized sampling strategy is used to reduce computational complexity and generate candidate circles containing center coordinates and radius parameters. Each candidate circle is verified. If a sufficient number of edge points fall within the preset radius tolerance range of each candidate circle, it is determined to be a valid circle, and the circular structure region image is output.

3. The system according to claim 1, characterized in that, The two-dimensional slice data preprocessing module also includes a filtering and noise reduction submodule; The filtering and denoising submodule is used to perform local region division on the circular structure region image, resulting in a circular region image. An adaptive median filter is used to filter the divided circular region image to remove impulse noise, resulting in a pre-denoised circular structure region image. Calculate the gradient magnitude of the circular structure region image after preliminary denoising, replace the gradient magnitude of the original image with the gradient magnitude, and dynamically adjust the diffusion coefficient in the anisotropic diffusion model. Based on the adjusted diffusion coefficient, anisotropic diffusion denoising is performed on the initially denoised circular structure region image to preserve image edge details, resulting in the multiple denoised circular structure region image.

4. The system according to claim 1, characterized in that, The two-dimensional slice data preprocessing module also includes an adaptive Canny submodule; The adaptive Canny submodule is used to perform grayscale processing on a selected region in the circular structure region image after multiple denoising steps to obtain a grayscale processed image. A Gaussian filter is constructed and a two-dimensional convolution is performed on the grayscale image to reduce local detail fluctuations in the image, resulting in a convolved image. The gradient magnitude and direction of the convolved image are calculated using the finite difference of the first-order partial derivative; non-maximum suppression is applied to the gradient magnitude to retain local gradient maximum points; A dual-threshold algorithm is used, which uses a high threshold to segment the background and the target, and a low threshold to connect the broken edges; The edge region is refined to extract the flaw contour and identify potential defect features, thus obtaining the high-precision edge image.

5. The system according to claim 1, characterized in that, The model post-processing and analysis module includes a stomatal ellipsoid model establishment module, a stomatal shape parameter extraction module, a 3D display module, and a volume calculation module; The pore ellipsoid model building module is used to simulate deformed pores as two triaxial ellipsoids, one above the other. By determining nine parameters, including the three-dimensional coordinates of the ellipsoid center, the lengths of the three semi-axes, and the three-axis rotation angles, a three-dimensional ellipsoid model is constructed. The origin of the ellipsoid model is taken as the center of the solid rocket motor base. The stoma shape parameter extraction module is used to preprocess the stoma image to obtain a binary image of the stoma edge. First, it directly calculates the center coordinates of the ellipse, the length of the semi-major axis, and the angle between the major axis and the horizontal axis. Then, it divides the stoma edge image according to the major axis, uses Hough transform to detect the length of the semi-minor axis of the upper and lower halves of the ellipse respectively, and determines the optimal semi-minor axis parameter from multiple solutions using the area minimum error method. The three-dimensional display module is used to visualize the upper and lower hemispheres of the stomata based on the constructed three-dimensional regular volume data field and according to the ellipsoid parameters. The volume calculation module is used to perform three-dimensional reconstruction on the binary image of the pore edge, calculate the pore volume using the volume calculation formula, and calculate the average value of the pore volume in multiple images.

6. The system according to claim 1, characterized in that, The system also includes a centralized management platform; the centralized management platform includes a front-end interface module, a confidentiality management module, a data security management module, and a flaw detection report management module; The front-end interface module is developed based on the PyQt library and QtDesigner to create a cross-platform operation interface. The interface includes a left interactive bar and a right preview box. The left interactive bar is used to receive user operation commands, and the right preview box is used to load CT images and 3D reconstructed images. The front-end interface module supports users to import CT slice images, generate 3D images, analyze and interact with 3D images, and output reconstruction results and a defect statistics list. The confidentiality management module is used to encrypt sensitive data using the AES-GCM algorithm of the Crypto.js library, store the encryption key in the backend, and ensure data transmission security through HTTPS; based on the RBAC model, access control is implemented by storing user access control lists in the database, and sensitive data access logs are recorded. The data security management module includes a data encryption submodule and a data backup and recovery submodule. The data encryption submodule uses a database engine that supports transparent data encryption to encrypt database and file system data and implements a key rotation strategy. The data backup and recovery submodule uses the Veeam Backup & Replication tool to perform full backups and incremental backups and manages the storage and access permissions of backup data. The flaw detection report management module is used to store document template information based on the Django backend framework and MySQL / PostgreSQL database, and to design the interface using the React / Vue.js frontend framework and Ant Design / Bootstrap. It generates PDF reports using Python's ReportLab library and processes the report data logic through Django. It uses Django's RESTful API to handle report viewing and modification operations, and implements report interaction through state management. The flaw detection report files are stored on a storage server, and report metadata is stored using a MySQL / PostgreSQL database. Report retrieval is achieved using Elasticsearch.

7. A method for intelligent detection and visual characterization of defects in solid rocket motor propellant grains, applied to a system for intelligent detection and visual characterization of defects in solid rocket motor propellant grains, characterized in that, The method includes: Based on a preset automatic identification and detection algorithm for circular structural components, structural component identification and localization are performed on the two-dimensional ICT image in the acquired solid rocket motor ICT file to determine the circular structural region image of the solid rocket motor. An improved anisotropic diffusion denoising algorithm based on adaptive median filtering removes impulse interference from the circular structure region image multiple times, resulting in a denoised circular structure region image. The robust adaptive Canny algorithm based on defect edges sequentially performs circular structure boundary detail extraction and noise interference removal processing on the circular structure region image after multiple denoising processes to obtain a high-precision edge image. Based on a pre-defined deep learning-based ICT slice image frame interpolation algorithm, image registration, normalization, and ROI extraction are performed on the original ICT slice images to obtain an integrated continuous slice sequence. A 3D convolutional neural network is used to preview the deep features of adjacent slices, and the features are aligned by optical flow estimation or deformable convolution. Using a generative adversarial network architecture, an intermediate interpolated slice is generated from two adjacent frame slices through a U-Net generator. The realism of the generated image is evaluated by a PatchGAN discriminator and optimized through a multi-objective loss function. Anisotropic diffusion filtering is used to smooth the edges of the interpolated slices, and then the interpolated data and the original data are integrated by voxel fusion technology to obtain the processed slice data. The processed slice data is imported into the 3D modeling module to construct a 3D voxel mesh and reconstruct the model; the reconstructed 3D model is then subjected to edge smoothing and voxel fusion optimization to generate a high-precision interactive 3D visualization model. The pore images in the two-dimensional ICT image are reconstructed into three dimensions to obtain a three-dimensional reconstruction model of the deformed pores and a volume estimation result; based on the volume estimation result, the fault diagnosis result of the solid rocket motor is determined.

Images