CN-121120992-B - Intelligent recognition and modeling method for prefabricated body in aviation composite special-shaped member

CN121120992BCN 121120992 BCN121120992 BCN 121120992BCN-121120992-B

Abstract

The invention discloses an intelligent recognition and modeling method for a prefabricated body in an aviation composite special-shaped member, which belongs to the technical field of digital reconstruction and modeling of microstructure of an aviation composite, and comprises the following steps of S1, data acquisition; S2, data preprocessing, S3, model training preparation, S4, intelligent segmentation, S5, three-dimensional reconstruction, S6, grid optimization, S7, topology processing, S8, and model output. The intelligent identification and modeling method for the prefabricated body in the aviation composite material abnormal-shaped member aims at the problem of topology errors and grid conformal which are easy to occur in explicit modeling, combines a grid Boolean operation and a post-treatment restoration algorithm, accurately processes grid intersection and discontinuous areas, repairs closed defects such as grid gaps, missing surfaces and the like which are introduced by the Boolean operation, ensures node conformality, topology connectivity and space closure of the grid of the prefabricated body substructure, reduces manual participation, and improves modeling efficiency and stability.

Inventors

GONG ZHENYUAN
ZENG QINGFENG
LIU JIANTAO
RU YI
WANG PINGPING
WANG KEKE

Assignees

天目山实验室

Dates

Publication Date: 20260508
Application Date: 20250904

Claims (6)

1. The intelligent recognition and modeling method for the prefabricated body in the aviation composite special-shaped member is characterized by comprising the following steps of: s1, CT scanning is carried out on a target area of a special-shaped aviation composite material component, and original CT digital slice data are obtained; s2, preprocessing original CT digital slice data in the S1 to obtain a preprocessed slice sequence; S3, selecting 20% of slice samples from the slice sequence preprocessed in the S2, and manually labeling to construct a training set; S4, based on a U-Net neural network frame built by Python, realizing automatic instance segmentation of the fiber bundles on all the preprocessed slice sequences in S2, and obtaining a segmentation label diagram; s5, reconstructing a three-dimensional voxel model of the segmentation label graph in the S4 to obtain an STL grid file; S6, importing the STL grid file in the S5 into a surface grid model of the preform, and automatically coating and smoothly reconstructing the surface grid of the substructure in the preform to obtain a smooth surface grid model; s7, based on the surface grid model smoothed in the S6, carrying out Boolean operation on the substructured grid of the preform, detecting and correcting redundant nodes and intersecting patches generated by factor structural interference in real time in the operation process, and processing grid intersecting and discontinuous areas to obtain a topological connected grid model; s8, extracting a co-interface region of each component generated after Boolean operation based on the topological connected grid model in S7, completing space closure of the subareas, and carrying out tetrahedral mesh subdivision on the closed triangular surface grid region to obtain a preform model.
2. The intelligent recognition and modeling method for the inner preform of the special-shaped member of the aviation composite material according to claim 1, wherein the step S2 comprises the following steps: s201, importing original CT digital slice data in batches, setting an XYZ three-axis space range of a target area, and focusing a representative volume element area containing a preform through a cutting formula, wherein the cutting formula is as follows: ; Wherein, the The left upper corner coordinates of the clipping window; Is the height of the window; Is the width of the window; the fiber bundle identification is improved by adopting a self-adaptive histogram equalization method, and the local contrast is enhanced by histogram equalization conversion, wherein the conversion formula is as follows: ; Wherein, the Is the first Gray scale; is of gray scale L is the number of gray level, MN is the image size; s202, smoothing random noise and high-frequency artifacts by adopting Gaussian filtering, wherein the formula is as follows: ; Wherein, the Is a two-dimensional Gaussian kernel function; s203, eliminating tiny isolated noise points and micro holes by adopting morphological opening and closing operation, wherein the formula is as follows: ; ; Wherein, the B is a structural element; is expansion operation; is a corrosion operation.
3. The intelligent recognition and modeling method for the inner preform of the special-shaped member of the aviation composite material according to claim 1, wherein the step S4 comprises the following steps: S401, carrying out normalization processing on the labeling data set in S3 and the corresponding preprocessed slice sequence pairing in S2, and carrying out center cutting and boundary filling on the image and the label, so as to ensure that the input size is matched with the network structure and the network input size; S402, adopting a U-Net coding-decoding structure, extracting multi-scale features by a coding path through convolution and pooling, and outputting a multi-channel prediction graph by a decoding path through up-sampling and jump connection and restoration features and space positioning, wherein the multi-channel prediction graph is expressed as follows: ; Encoder represents extracting high-dimensional semantic features, wherein a Decoder represents gradually restoring a space structure, and Skip Connections represent fusion of guarantee features and position information; S403, taking multi-category cross entropy as a main loss, and introducing a Dice coefficient and Tversky loss indexes, wherein the multi-category cross entropy loss is expressed as: ; Wherein, the A real label for pixel i; Is a predictive probability; Cross entropy loss for multiple categories; the Dice coefficient is expressed as: ; Wherein, the Is a real label; Is a predictive probability; Is a smooth term; S404, training by adopting an Adam optimizer, monitoring the index of the Dice coefficient through a verification set, and utilizing EarlyStopping and ReduceLROnPlateau mechanisms to prevent overfitting and self-adaptive adjustment of the learning rate; s405, carrying out batch reasoning on the full CT slices, and outputting a pixel-level separation label graph; s406, automatically predicting the batch CT slices to obtain a segmentation label map, and storing the segmentation label map in a standard picture format.
4. The intelligent recognition and modeling method for the inner preform of the special-shaped member of the aviation composite material according to claim 2, wherein the step S5 comprises the following steps: S501, importing 2D segmentation masks in batches, identifying materials according to RGB threshold values, and combining the materials into a voxel array in a three-dimensional space Each voxel is given a label to the material Respectively corresponding to the background, the matrix, the fiber bundle A and the fiber bundle B; s502, eliminating isolated noise points through three-dimensional morphological opening and closing operation, wherein the method is expressed as: ; ; The analysis and elimination volume of the combined connected domain is smaller than the minimum connected domain volume threshold value Is a minor component of (a) with a retained main structure, expressed as: ; Wherein, the For voxels in An output value of the location; Is the minimum connected domain volume threshold; is the original value; s503, merging all material areas into a single three-dimensional voxel model according to priority, and adding a background boundary on the outer layer of the voxel; S504, traversing the voxel array, and for each non-background body Sequentially detecting adjacent voxels in six positive and negative axis directions If the material numbers are different, an interface surface patch is generated between the two, the vertex coordinates of the surface patch are determined according to the voxel center and the facing direction, and the surface patch is classified into a corresponding material interface; S505, each interface surface patch is generated in a quadrilateral form and then split into two triangular surfaces, wherein the quadrilateral is decomposed into a triangular formula: ; Wherein, the Is a quadrilateral dough sheet; 4 vertexes of the quadrangular surface piece; Two triangular patches obtained after splitting; S506, integrally translating and recovering the original space position after the voxel grid coordinates are subjected to boundary expansion, and respectively exporting triangular surface grids of each type of interface into standard STL files to obtain STL grid files.
5. The intelligent recognition and modeling method for the inner preform of the special-shaped member of the aviation composite material according to claim 1, wherein the step S8 comprises the following steps: S801, judging whether the vertexes of all the patches of each fiber bundle surface grid are met or not by adopting the minimum outsourcing cube parameters of the matrix surface grids, wherein the judging conditions are as follows: ; Wherein, the The minimum outsourcing cube is the region boundary; If the judging condition is met, reserving the dough sheet; S802, extracting a common interface between the two prefabricated body components after Boolean operation to enable the internal grid to be completely closed, wherein the method specifically comprises the following steps: The method comprises the steps of carrying out vertex space searching by using a KD tree, positioning A, B common nodes, checking whether the common nodes are on the common edge, identifying the boundary annular edge, dividing A, B into a plurality of patches respectively by taking the common boundary annular edge as a boundary based on a connected domain partition, extracting interface patches except a main patch, merging all the boundary patches and deriving, and endowing the patches to a missing surface part to construct a closed internal interface grid; S803, performing common interface identification and patch partition operation of S802 on each substructure of the preform and the outer boundary of the region, extracting all common interface patches related to the boundary, merging and exporting, and completely closing each substructure of the preform with the outside; s804, when all the substructures in the prefabricated body are closed and have continuous topology, taking the closed boundary as a constraint surface, performing tetrahedral mesh subdivision, and generating a prefabricated body mesh model.
6. An intelligent recognition and modeling device for an inner prefabricated body of an aviation composite special-shaped member, which is applied to the intelligent recognition and modeling method for the inner prefabricated body of the aviation composite special-shaped member according to any one of claims 1-5, and is characterized by comprising the following steps: The data acquisition module is used for CT scanning the target area of the special-shaped aviation composite material component to obtain original CT digital slice data; the data preprocessing module is connected with the data acquisition module and is used for preprocessing the original CT digital slice data; the training set construction module is connected with the data preprocessing module and is used for selecting 20% of slice samples from the preprocessed slice sequence and manually labeling the slice samples to construct a training set; The intelligent segmentation module is connected with the training set construction module and is used for realizing automatic instance segmentation of the fiber bundles on the basis of the U-Net neural network framework built by Python and the all preprocessed slice sequences to obtain a segmentation label graph; The three-dimensional reconstruction module is connected with the intelligent segmentation module and is used for carrying out three-dimensional voxel model reconstruction on the segmentation label graph to obtain an STL grid file; The grid optimization module is connected with the three-dimensional reconstruction module and is used for guiding the STL grid file into the surface grid model of the preform, and automatically cladding and smoothly reconstructing the surface grid of the substructure in the preform to obtain a smooth surface grid model; The topology processing module is connected with the grid optimization module, and is used for carrying out Boolean operation on the sub-structure grids of the preform based on the smoothed surface grid model, and processing grid intersection and discontinuous areas to obtain a topology connected grid model; The model output module is connected with the topology processing module, extracts the co-interface areas of all components generated after Boolean operation based on the topology connected grid model, completes the space closure of the subareas, and performs tetrahedral grid subdivision on the closed triangular surface grid area to obtain a preform model.

Description

Intelligent recognition and modeling method for prefabricated body in aviation composite special-shaped member Technical Field The invention belongs to the technical field of digital reconstruction and modeling of microstructure of an aviation composite material, and particularly relates to an intelligent recognition and modeling method for a prefabricated body in an aviation composite material abnormal-shaped component. Background The digital reconstruction and modeling of the microstructure of the aviation composite material means that the internal microstructure of the aviation composite material, such as a continuous fiber toughened ceramic matrix composite material, is converted into a digital form by a technical means. The microstructure of the preform in the special-shaped aviation composite component is automatically identified, and a foundation can be provided for subsequent three-dimensional voxel reconstruction and refined modeling. However, the prior art has the defects of complex realization, large calculation amount, high requirements on programming and mathematical foundation, high engineering application threshold, limited Boolean operation support on complex microstructures, difficult effective extraction and treatment of multi-component coplanar areas, difficult complete closure of fiber bundle area grids, influence on the accuracy of subsequent simulation analysis, and a large amount of manual participation in part of processes, lack of integrated and automatic treatment means, and reduction of modeling efficiency and stability. Thus, a new method is needed. Disclosure of Invention The invention aims to provide an intelligent recognition and modeling method for a prefabricated body in an aviation composite material abnormal-shaped member, which aims at the problem of topology errors and grid conformal which are easy to occur in explicit modeling, combines a grid Boolean operation and a post-treatment repair algorithm, accurately processes grid intersection and discontinuous areas, repairs closed defects such as grid gaps, missing surfaces and the like which are introduced by the Boolean operation, ensures node conformality, topology connectivity and space closure of a prefabricated body substructure grid, reduces manual participation, and improves modeling efficiency and stability. In order to achieve the above purpose, the invention provides an intelligent recognition and modeling method for a prefabricated body in an aviation composite material abnormal-shaped member, which comprises the following steps: s1, CT scanning is carried out on a target area of a special-shaped aviation composite material component, and original CT digital slice data are obtained; s2, preprocessing original CT digital slice data in the S1 to obtain a preprocessed slice sequence; S3, selecting 20% of slice samples from the slice sequence preprocessed in the S2, and manually labeling to construct a training set; S4, based on a U-Net neural network frame built by Python, realizing automatic instance segmentation of the fiber bundles on all the preprocessed slice sequences in S2, and obtaining a segmentation label diagram; s5, reconstructing a three-dimensional voxel model of the segmentation label graph in the S4 to obtain an STL grid file; S6, importing the STL grid file in the S5 into a surface grid model of the preform, and automatically coating and smoothly reconstructing the surface grid of the substructure in the preform to obtain a smooth surface grid model; s7, based on the surface grid model smoothed in the S6, performing Boolean operation on the substructure grid of the preform, and processing grid intersection and discontinuous areas to obtain a topological connected grid model; s8, extracting a co-interface region of each component generated after Boolean operation based on the topological connected grid model in S7, completing space closure of the subareas, and carrying out tetrahedral mesh subdivision on the closed triangular surface grid region to obtain a preform model. Preferably, S2 comprises the steps of: s201, importing original CT digital slice data in batches, setting an XYZ three-axis space range of a target area, and focusing a representative volume element area containing a preform through a cutting formula, wherein the cutting formula is as follows: Icrop(x,y)＝I(x0+x,y0+y),x∈[0,w],y∈[[0,h]; wherein, (x 0,y0) is the left upper corner coordinate of the clipping window, h is the height of the window, and w is the width of the window; the fiber bundle identification is improved by adopting a self-adaptive histogram equalization method, and the local contrast is enhanced by histogram equalization conversion, wherein the conversion formula is as follows: wherein r k is the k-th gray level, n j is the pixel number of the gray level j, L is the gray level number, and MN is the image size; s202, smoothing random noise and high-frequency artifacts by adopting Gaussian filtering, wherein the formul