A nanofiber network modeling method and system based on image recognition and process mechanism

Abstract

The application discloses a kind of nanofiber network modeling method and system based on image recognition and process mechanism, belong to the intersection technical field of advanced ceramics and computational materials science, including: setting modeling basic parameters;Through image recognition, curved fiber is discretized into first and last connected line segment sequence, and relevant statistical features are extracted;Combining process physical mechanism, with the statistical features as control parameter, generate fiber network through random walk algorithm;Based on fiber network construction, obtain three-dimensional nanofiber network representative volume element, namely RVE model.The method is closer to real preparation process compared with traditional random throwing model, has higher structure fidelity compared with idealized geometric assumption, and the model established can be directly used for mechanical property analysis of ceramic nanofiber aerogel.The method is universal for nanofiber and nanowire constructed aerogel, and provides effective numerical analysis tool for material structure design and performance optimization.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of advanced ceramics and computational materials science, specifically relating to a nanofiber network modeling method and system based on image recognition and process mechanism. Background Technology

[0002] Ceramic aerogels possess excellent properties such as low density and high porosity, making them widely used in industrial and military fields. However, the inherent brittleness of ceramic materials severely limits their practical applications. Current research primarily addresses this issue through structural design strategies, with one-dimensional nanofiber-constructed ceramic aerogels exhibiting the best performance, achieving up to 90% elastic compressive deformation and demonstrating broad application prospects. Existing research mainly employs two strategies to improve performance: optimizing constituent elements and controlling microstructure. However, due to the complex multi-scale and multi-level structural characteristics of nanofiber aerogels, experimental characterization alone is insufficient to fully reveal their deformation mechanisms. Therefore, theoretical modeling has become an important means to deeply understand the mechanical behavior of fiber networks. However, existing fiber network modeling methods suffer from the following key problems: (1) The fiber morphology description is too simplistic. The fiber networks established in current research all use straight lines or simple arcs to represent fibers, rarely considering the tortuous shape of actual fibers, and cannot accurately reflect the complex bending characteristics of real fibers.

[0003] (2) The modeling parameters lack support from real data. Existing methods mainly determine fiber geometric parameters based on mathematical statistical theory or empirical assumptions, which deviate from the statistical characteristics of the actual fiber network.

[0004] (3) The physical mechanism of fiber preparation process is ignored. Existing methods are mostly based on geometric construction or statistical generation strategies, lacking consideration of fiber growth process and network formation mechanism in specific preparation process, and are difficult to reflect the true structural characteristics of fiber network under specific preparation conditions.

[0005] Therefore, there is an urgent need to develop a modeling method that can accurately reflect the microstructural characteristics of real fiber networks. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, the present invention aims to provide a nanofiber network modeling method and system based on image recognition and process mechanism, which aims to solve the problems of existing ceramic nanofiber aerogel fiber network modeling methods that simplify fiber morphology description, distort modeling parameters, and fail to consider the preparation process mechanism, thus failing to accurately reflect the real microstructure.

[0007] To achieve the above objectives, the present invention employs the following technical solution: This invention discloses a nanofiber network modeling method based on image recognition and process mechanisms. It uses a sequence of connected line segments to represent curved fibers, and achieves process-oriented modeling through real data-driven and CVD process physical mechanisms. The method constructs a three-dimensional fiber network representative volume element (RVE) model, including the following steps: S1: Set basic modeling parameters; S2: By using image recognition, the curved fiber is discretized into a sequence of line segments connected end to end, and relevant statistical features are extracted; S3: Combining the physical mechanism of the process, using the statistical characteristics as control parameters, a fiber network is generated through a random walk algorithm; S4: Based on the fiber network, a representative volumetric unit of the three-dimensional nanofiber network, namely the RVE model, is obtained.

[0008] Specifically, in S1, the basic modeling parameters include RVE size and target volume fraction.

[0009] Specifically, in S2, the relevant statistical characteristics include the distribution of line segment length, the range of angle variation, and the distribution of diameter; Specifically, in S3, the fiber network is generated by using a multi-generation random walk algorithm to combine the physical growth law of CVD process with the above control parameters.

[0010] Preferably, in S2, the image is a SEM image, and the SEM image recognition technology includes key steps such as image preprocessing, skeleton extraction, fiber separation, and feature extraction, including the following steps: 1) Perform image enhancement and binarization processing on the SEM image; 2) The fiber centerline is extracted using a skeletonization algorithm; 3) Identify and separate individual fibers based on endpoints and branch points using topology analysis; 4) The curved fiber is discretized into a sequence of line segments using an adaptive dynamic programming (DP) algorithm; 5) Statistical analysis of line segment length distribution, range of angle variation between adjacent line segments, and fiber diameter distribution.

[0011] Specifically, some key processing steps in the above process are described below: Step 1) can be summarized as image preprocessing: brightness and contrast are adjusted on the original SEM image, local enhancement is performed using an adaptive contrast enhancement algorithm, the image is smoothed and noise is reduced using filtering techniques, binarization is performed using an adaptive thresholding method, and image quality is optimized through morphological operations to remove noise and connect disconnected structures.

[0012] Step 2) can be summarized as skeleton extraction: A skeletonization algorithm is used to process the binarized image, refining the fiber region into a single-pixel-width centerline. Defects in the skeleton are eliminated through image processing operations. Key feature points in the skeleton's topology are identified based on pixel neighborhood analysis, including endpoints, ordinary points, and branch points.

[0013] Step 3) can be summarized as single-fiber separation: A graph structure representing the fiber network is constructed based on topological features. Nodes include key feature points, and edges represent the connections between feature points. A single fiber is defined as a continuous path between feature points. A path search algorithm is used to start from a feature point, record the complete path, and calculate the path direction, thereby achieving accurate identification and separation of intersections and branches in the fiber network.

[0014] Step 4) can be summarized as adaptive discretization and feature extraction, including two aspects: diameter recognition and line segment recognition.

[0015] Diameter identification involves sampling on the skeleton, calculating the fiber orientation, measuring the transverse dimension of the fiber, and statistically obtaining the probability density function of the diameter. ; Line segment recognition uses an adaptive discretization algorithm to process curved fibers. This algorithm dynamically adjusts the segmentation accuracy based on the local curvature characteristics of the fiber, thereby achieving accurate discretization of fibers with different degrees of curvature.

[0016] After discretization, the lengths of the discrete line segments and the angles between adjacent line segments are calculated, and the probability density function of the line segment length is obtained statistically. The probability density function of angle change This provides statistical characteristic parameters for subsequent numerical modeling.

[0017] Preferably, in S3, based on the physical growth law of the CVD process, the probability density function of the line segment length extracted in S2 is used. The probability density function of angle change and the probability density function of diameter change As control parameters, a three-dimensional fiber network is generated using a multi-generational random walk algorithm.

[0018] More preferably, the generation of the fiber network using the random walk algorithm specifically includes the following steps: 1) Determine the initial point and initial angle, and determine the position of each subsequent point in sequence according to the probability density function of the angle distribution and the line segment length distribution, forming a sequence of line segments connected end to end, until it grows to the boundary or reaches the predetermined length; 2) The bottom fiber generates the starting point in the xy plane; the near bottom fiber nucleates in the xy plane with probability p, and nucleates on the surface of the existing fiber with probability (1-p); the starting point of the fiber far from the bottom fiber is completely selected from the set of positions on the surface of the previous layer of fiber, and the fiber network is constructed generation by generation.

[0019] More specifically, the algorithm comprises three core components: a single-fiber random growth process, multi-generation fiber network construction, and network optimization. First, the random growth process of a single fiber: A curved fiber is composed of N line segments connected end to end. The mathematical expression of a single fiber is:

[0020] Wherein, the line segment direction angle: Spatial coordinates of the endpoints of the line segment: .

[0021] The fiber growth process is based on random sampling using a probability density function: the line segment length follows a distribution... The angle change between adjacent line segments was obtained through mid-sampling. Probability density function of angle change The data is obtained through sampling. By setting an initial point and initial direction, the spatial positions of subsequent points are determined sequentially, forming a three-dimensional spatial curve that connects end to end, until it grows to the RVE boundary or reaches a preset length. The growth process satisfies physical constraints such as boundary conditions and spatial density control.

[0022] Second, multi-generation fiber network construction: Based on the mathematical expression of the above-mentioned random growth process, a multi-generation random growth algorithm is developed to generate fiber networks. This algorithm first determines basic parameters such as RVE size, target volume fraction, segment length, and probability density function, and then generates the fiber network generation by generation according to a hierarchical strategy. The hierarchical generation strategy employs different nucleation rules for different levels: the bottom layer fibers originate on the bottom surface, using a larger initial angle range to simulate relatively straight growth characteristics; near-bottom layer fibers generate with probability density functions... Kerneling on the bottom surface with probability Nucleation occurs on the existing fiber surface, reflecting the transition from basal growth to fiber surface growth; the starting point of the fiber far from the bottom layer is selected entirely from the set of positions on the surface of the previous layer of fibers, and all nucleation point selections follow the principle of spatial density control, realizing hierarchical growth between fibers and building a fiber network generation by generation.

[0023] A spatial density control strategy is implemented during the growth process: the RVE space is divided into a regular grid, and the fiber density of each grid cell is tracked in real time. New fiber nucleation points are preferentially selected in areas with lower density to avoid excessive fiber aggregation. Density information is incrementally updated after each generation of fiber generation to ensure a relatively uniform distribution of the network.

[0024] Third, network optimization: The generated initial fiber network is optimized to ensure physical rationality. Boundary condition handling: fiber growth terminates according to predefined rules when it reaches the RVE boundary; Spatial interference elimination: spatial interference between fibers is identified using minimum distance judgment, and fiber paths are regenerated when interference is detected; Connectivity guarantee: isolated parts in the network are identified using graph theory methods, and overall connectivity is achieved by generating connecting fibers. The generation of connecting fibers follows a spatial density control principle, and the connection process is repeated until the entire network is fully connected. The final mathematical expression of the fiber network is:

[0025] in, The total number of fibers, per fiber Depend on It consists of several points.

[0026] Preferably, in S4, the generated three-dimensional nanofiber network RVE model uses a standardized data format to represent fiber geometry, identifies boundary feature points to set periodic boundary conditions, and outputs a model file compatible with mainstream finite element software, suitable for beam element or solid element modeling.

[0027] Specifically, the geometric data conversion transforms the fiber network data generated by S3 into a format suitable for finite element analysis. Fibers are represented as spatial point sequences, and the segment lengths meet mesh quality requirements. Periodic boundary conditions are achieved by identifying fiber endpoints located at the RVE boundary. Boundary points are identified and classified according to preset tolerance standards to ensure that the RVE model meets the periodicity requirements of representative volume elements. The finite element model file contains complete geometric definitions, material properties, and mesh parameters, supporting multiple element types: beam elements can be used for efficient calculations, with each fiber segment corresponding to one beam element; solid elements can also be used for detailed modeling, where fibers can be built as three-dimensional solids with various cross-sectional shapes such as circles, ellipses, and rectangles, and the corresponding cross-sectional properties are automatically set according to the fiber diameter distribution. The generated file is compatible with mainstream finite element software, providing a standardized interface for the mechanical property analysis of fiber network materials, the embedding skeleton modeling of fiber reinforcement phases in composite materials, and solid CAD modeling and structural design optimization.

[0028] This invention also discloses a nanofiber network modeling system based on image recognition and process mechanism, comprising: The modeling parameters module is used to set basic modeling parameters; The statistical feature extraction module is used to discretize curved fibers into a sequence of line segments connected end to end through image recognition and extract relevant statistical features; The fiber network generation module is used to generate a fiber network using the statistical characteristics as control parameters, based on the physical mechanism of the bonding process; The RVE model construction module is used to construct a representative volumetric unit (RVE model) of a three-dimensional nanofiber network based on the fiber network.

[0029] Preferably, the basic modeling parameters set by the modeling parameter module include: the feature size and the ratio of the length of each dimension, and the target volume fraction as the fiber growth termination condition; The statistical feature extraction module is specifically used for: image enhancement and binarization of SEM images, extraction of fiber center lines using skeletonization algorithm, identification of endpoints and branch points based on topological analysis to separate individual fibers, discretization of curved fibers into a sequence of connected line segments using adaptive DP algorithm, and statistical analysis of line segment length distribution, angle variation range of adjacent line segments, and fiber diameter distribution.

[0030] Preferably, the fiber network generation module is specifically used for: determining the initial point and initial angle; determining the positions of subsequent points sequentially according to the probability density functions of the angle distribution and the line segment length distribution, forming a sequence of line segments connected end to end until it grows to the boundary or reaches a predetermined length; controlling the bottom fiber to generate the starting point in the xy plane, with near-bottom fiber nucleating in the xy plane with probability p and on the surface of existing fibers with probability (1-p), and far from the starting point of the bottom fiber, selecting entirely from the set of positions on the surface of the previous layer of fibers, and constructing the fiber network generation by generation; and simultaneously performing spatial interference elimination and deletion of isolated fiber groups on the generated fiber network to ensure the connectivity and physical rationality of the fiber network.

[0031] Preferably, the RVE model construction module is specifically used to: represent a single fiber as a sequence of coordinate points and the fiber network as a set of point sequences; identify the fiber endpoints located on the RVE boundary surface to set periodic boundary conditions; and output a script file containing fiber geometric definitions and material properties; the script file is suitable for ABAQUS software and is used for beam element or solid element modeling. Compared with the prior art, the present invention has the following beneficial effects: Compared to existing technologies, this method combines process physical mechanisms with process modeling, resulting in a more realistic representation of the manufacturing process and effectively addressing the shortcomings of existing modeling methods that neglect process mechanisms. Compared to idealized geometric assumptions, this method extracts statistical features of real fibers through image recognition, discretizing curved fibers into a sequence of connected line segments, significantly improving the structural fidelity of the model and solving the problem that existing models cannot accurately reflect the complex bending characteristics of real fibers. The three-dimensional nanofiber network model constructed by this method can be directly used for the mechanical property analysis of ceramic nanofiber (wire) aerogels, supporting various finite element types and cross-sectional shapes, and is compatible with mainstream finite element software. It provides an effective numerical analysis tool for material structure design and performance optimization, overcoming the limitations of relying solely on experimental characterization to fully reveal the deformation mechanism of fiber networks. This method is universal; by inputting a SEM image of any fiber network, a corresponding three-dimensional fiber network RVE model can be generated, realizing a complete process from real samples to high-fidelity theoretical models and broadening the application scope of the modeling method. Attached Figure Description

[0032] Figure 1 The present invention provides an image recognition and fiber separation process; wherein, (a) is the original SEM image; (b) is image enhancement and binarization processing; (c) is skeleton extraction; and (d) is single fiber identification and separation. Figure 2 This is a flowchart illustrating the generation process of the nanofiber network of the present invention. Figure 3 This is a schematic diagram of the three-dimensional fiber network RVE model generated by the present invention; wherein, (a) is the theoretical model; and (b) is the finite element model. Figure 4 These are fiber network models with different bending characteristics generated by the method of the present invention; wherein, (a) is a high-bending spiral fiber network; and (b) is a straight fiber network. Detailed Implementation

[0033] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0034] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0035] The present invention will now be described in further detail with reference to the accompanying drawings: This invention discloses a nanofiber network modeling method based on image recognition and process mechanism. Based on the theoretical foundation of line segment discretization representation of curved fibers and mathematical description of fiber networks, the method automatically extracts the geometric parameter distribution of real fibers through scanning electron microscopy (SEM) image recognition technology, and uses a multi-generation hierarchical growth algorithm to simulate the actual fiber growth process in chemical vapor deposition (CVD) to construct a high-fidelity three-dimensional fiber network theoretical model.

[0036] The method for establishing the model includes: extracting the skeleton and separating individual fibers from the SEM image; using the adaptive Douglas-Peucker algorithm to obtain the fiber segment length and relative angle distribution; based on the real CVD process mechanism, adopting a hierarchical growth strategy, the first generation of fibers nucleates and grows from the graphite paper substrate, and subsequent generations of fibers nucleate randomly on the surface of the existing fibers to form a three-dimensional network with real sample characteristics; finally, a discrete finite element model suitable for ABAQUS finite element analysis is generated.

[0037] Specifically, see Figure 1 The flowchart of image recognition and fiber separation of this invention mainly includes the following four key stages: In the first stage, Figure (a) shows the data input layer: the microstructure of the real ceramic nanofiber aerogel, i.e., the SEM image, which is the real data source for all subsequent modeling.

[0038] The second stage, shown in Figure (b), is the image recognition and feature extraction layer. It can be seen that the SEM image, after processing, becomes a discrete sequence of line segments. This process involves image enhancement, binarization, skeleton extraction, and discretization using the DP algorithm. The core purpose of this step is to extract statistical features such as line segment length distribution and angle variation range, transforming the complex optical image into mathematical coordinates and line segment parameters that a computer can understand.

[0039] The third stage, shown in Figure (c), illustrates the virtual environment construction process, including the CVD (chemical vapor deposition) process and the random walk algorithm. The CVD process introduces real physical growth patterns (such as substrate attachment and fiber growth direction), and then, based on the random walk algorithm and extracted statistical features, automatically generates a large number of fibers to construct a complete three-dimensional network structure.

[0040] The fourth stage, shown in diagram (d) as the model construction and output layer, displays the constructed 3D RVE model and the ABAQUS software interface. It can be seen that the model is no longer an abstract schematic diagram, but a script file that can be directly imported into professional engineering software such as ABAQUS. Users can use it for practical tasks such as stress analysis and material property testing.

[0041] See Figure 2 Here is a flowchart of the nanofiber network generation process of the present invention. The generation process is as follows: The fiber network generation process is as follows: First, initialization is performed to determine the RVE size and target porosity P. Then, the fiber network is constructed generation by generation according to a hierarchical strategy. For the bottom layer fibers, the starting point is generated on the xy plane. Density sampling is performed step by step according to the probability density function of angle change and the probability density function of line segment length distribution to generate complete fibers and return the generated point set. For the upper layer fibers, the starting point is selected from the set of surface positions of existing fibers. Similarly, complete fibers are generated through density sampling and the generated point set is returned. After each generation of fibers is generated, crosslinking judgment and crosslinking point addition are performed. Then, it is checked whether the porosity of the current network has reached the target value P. If the porosity has not yet reached the target value, the next generation of fiber network is generated iteratively. If it has reached the target value, a connectivity test is performed. If the network is completely connected, the generation ends. If the network has multiple connected components, each connected component is identified and the optimal connecting fiber is created. The porosity judgment is re-entered until the network meets the connectivity requirements.

[0042] Example 1 This embodiment uses a silicon carbide (SiC) ceramic nanofiber network as an example to construct a three-dimensional Resonant Vessel (RVE) model using the method of this invention. This embodiment is developed and implemented using the Python programming language, utilizing open-source libraries such as OpenCV, scikit-image, numpy, and matplotlib. However, the method of this invention is not limited to a specific programming language and can also be implemented using any programming language that supports numerical computation and image processing, such as MATLAB, C++, Java, and Fortran. The RVE size is chosen to be 50μm × 50μm × 60μm, the target volume fraction is set to 1%, and the fiber diameter is determined statistically based on SEM images.

[0043] Feature extraction, image recognition, and fiber separation processes are performed on SEM images of SiC fibers as follows: Figure 1 As shown. In image preprocessing, the Gamma correction parameter was set to 0.25, and the Gaussian filter kernel size was set to 7×7. In the adaptive Douglas-Peucker discretization algorithm, the minimum threshold was set to 2.0 pixels, the base threshold was set to 8.0 pixels, and the curvature sensitivity factor was set to 0.5. Statistical analysis revealed that the line segment lengths of SiC fibers are mainly distributed in the range of 15-25 pixels, with an average length of 20 pixels; 60% of the angle variation is concentrated in the range of 0-30 degrees, with an average absolute angle variation of 15-20 degrees.

[0044] Fiber network generation is performed based on extracted statistical features, with the basic line segment length set to 4 μm. According to image recognition results, statistical analysis of SiC fiber angle variations shows that the angle variations between adjacent line segments are mainly concentrated in the range of 15 to 25 degrees. Based on this, the in-plane angle variation range in 3D modeling is determined to be ±15 to ±20 degrees, and the out-of-plane angle variation range is ±10 to ±15 degrees. Fiber growth is performed according to the multi-generation random walk algorithm described in S3 above, the specific process as follows: Figure 2 As shown. Spatial density control employs appropriate mesh generation and density thresholds to ensure the rationality of fiber distribution. The generated theoretical model is as follows. Figure 3 As shown in (a).

[0045] When the generated fiber network is converted into a finite element model, a script file is automatically generated through programming. This script generation process can be implemented using any programming language. In this embodiment, an ABAQUS Journal script file is generated, with material properties set according to the actual performance of SiC ceramic material, and beam elements are used for modeling. The generated script file can be directly imported into ABAQUS software, or it can be adjusted to a format compatible with other finite element software such as ANSYS and LS-DYNA as needed. The generated finite element model is as follows. Figure 3 As shown in (b).

[0046] Example 2 Without relying on image recognition, the fiber morphology can be directly controlled by adjusting angle parameters. Using the single-fiber mathematical description and multi-generation network generation algorithm of this invention, fiber networks with different configurations can be generated. When the angle distribution between adjacent line segments is set to within the range of 30 to 50 degrees, a fiber network with helical bending characteristics is generated, such as... Figure 4 As shown in (a); when the angle between adjacent line segments is set to be within the range of 5 to 15 degrees, a relatively straight fiber network is generated, such as... Figure 4 As shown in (b).

[0047] Example 3 In this embodiment, a nanofiber network modeling system based on image recognition and process mechanism is provided. This system can be used to implement the above-mentioned nanofiber network modeling method based on image recognition and process mechanism. Specifically, the system includes a modeling parameter module, a statistical feature extraction module, a fiber network generation module, and an RVE model construction module.

[0048] Specifically: The modeling parameter module is the parameter support unit for the entire system. It is used to set all the basic parameters required for modeling. In specific implementation, relevant parameters are input or selected through the visual interface of the computer terminal. The basic modeling parameters include: feature size and the ratio of the length of each dimension (for example, setting the feature size of the RVE model to 10μm×10μm×10μm and the ratio of the length of each dimension to 1:1:1), and the target volume fraction as the fiber growth termination condition (for example, setting the target volume fraction to 5%, and stopping fiber growth when the volume fraction of the fiber network reaches this value).

[0049] The statistical feature extraction module is the system's real data extraction unit. Its core function is to obtain the statistical features of real fibers through image recognition technology, specifically including the following operations: The first step is to receive the SEM image of the ceramic nanofiber aerogel input by the user and preprocess the SEM image. First, image enhancement processing (such as grayscale stretching and Gaussian noise reduction filtering) is performed to remove noise interference in the image and improve the contrast between the fiber and the background. Then, an adaptive threshold segmentation algorithm is used to perform binarization processing to convert the SEM image into a black and white binary image, so that the fiber area and the background area are clearly separated.

[0050] The second step involves using a skeletonization algorithm to extract the fiber centerline from the binary image, removing redundant pixels and preserving the core outline of the fiber. Subsequently, a topology analysis algorithm is used to identify the endpoints and branch points of the fiber centerline. Based on the distribution characteristics of the endpoints and branch points, individual fibers are separated to avoid interference from multiple fibers crossing each other.

[0051] The third step involves using the adaptive DP (Douglas-Peucker) algorithm to discretize the centerline of each curved fiber into a sequence of connected line segments. During the discretization process, the number of line segments is adaptively adjusted according to the degree of fiber curvature to ensure that the line segment sequence can accurately fit the true shape of the curved fiber. Finally, the relevant parameters of all line segments are statistically analyzed to obtain the line segment length distribution (e.g., the length range is 0.5μm-5μm, which is normally distributed), the angle variation range of adjacent line segments (e.g., 0°-90°), and the fiber diameter distribution (e.g., the diameter range is 50nm-200nm). These statistical feature data are then synchronized to the fiber network generation module.

[0052] The fiber network generation module is the core generation unit of the system. It combines the physical mechanisms of the process with the statistical features output by the statistical feature extraction module as control parameters to generate fiber networks through a random walk algorithm. During execution, it specifically includes the following operations: The first step is to receive the basic parameters output by the modeling parameter module and the statistical feature data output by the statistical feature extraction module to determine the initial point and initial angle of fiber growth. The initial angle is randomly selected based on the probability density function of the angle variation range of adjacent line segments, and the initial point is determined in combination with the feature size in the modeling parameters. Then, based on the probability density function of the line segment length distribution, the positions of subsequent points are determined in sequence to form a sequence of line segments connected end to end until the line segment sequence grows to the boundary of the RVE model or reaches a predetermined length (the predetermined length is determined based on the statistical characteristics of the line segment length distribution, for example, taking the average line segment length of 2.5μm).

[0053] The second step involves constructing a fiber network generation by generation according to the layered growth logic: the starting points of the bottom layer fibers are randomly selected in the xy plane to ensure the uniform distribution of the bottom layer fibers; the fibers near the bottom layer nucleate and grow in the xy plane with probability p (p takes the value of 0.5), and nucleate and grow on the surface of the existing bottom layer fibers with probability (1-p)=0.5; the fibers far from the bottom layer (i.e., the upper layer fibers) have their starting points randomly selected from the set of surface positions of the previous layer fibers, realizing the layered growth of fibers, which conforms to the fiber growth law in the actual preparation process.

[0054] The third step involves simultaneously optimizing the generated fiber network during the fiber network generation process: using a spatial interference detection algorithm to identify and eliminate spatial interference between fibers (i.e., avoiding overlapping or cross-interference between two fibers); and using a connectivity detection algorithm to identify and delete isolated fiber groups (i.e., single or multiple fibers that have no connection with other fibers), ensuring that the generated fiber network has good connectivity and physical rationality. After optimization, the fiber network data is synchronized to the RVE model construction module.

[0055] The RVE model construction module is the system's output unit, used to generate a 3D RVE model based on the fiber network and output related files. During execution, it includes the following operations: The first step is to receive the optimized fiber network data output by the fiber network generation module, and construct a three-dimensional nanofiber network RVE model according to the feature size set by the modeling parameter module. During the modeling process, each fiber is represented in the form of a sequence of coordinate points (the two endpoints of each line segment are represented by three-dimensional coordinates). The entire fiber network is represented by the set of coordinate point sequences of all individual fibers to ensure the accuracy of the model.

[0056] The second step involves using a boundary detection algorithm to identify fiber endpoints located on the boundary surface of the RVE model. Periodic boundary conditions are then set for these endpoints to ensure that the RVE model can accurately reflect the mechanical properties and structural characteristics of the overall fiber network, thus avoiding the impact of boundary effects on subsequent analysis.

[0057] The third step is to generate a script file containing fiber geometry definitions (coordinate point sequence, line segment length, diameter, etc.) and material properties (such as the elastic modulus and Poisson's ratio of ceramic nanofibers). This script file is compatible with ABAQUS finite element software and can be directly imported into the software for beam element or solid element modeling, providing direct support for subsequent mechanical property analysis of fiber network materials.

[0058] Specifically, the workflow of the above system during use is as follows: 1. Start the system, set the basic modeling parameters through the modeling parameter module, and complete the parameter initialization; 2. The statistical feature extraction module receives SEM images, performs image preprocessing, fiber discretization, and statistical feature extraction, and outputs statistical feature data; 3. The fiber network generation module combines process physical mechanisms and statistical characteristic data to generate and optimize fiber networks through a random walk algorithm, and outputs optimized fiber network data; 4. The RVE model building module constructs a 3D RVE model based on fiber network data, sets periodic boundary conditions, and outputs a script file; 5. Once the modeling is complete, the user can import the output script file into the ABAQUS software for subsequent finite element mechanical property analysis.

[0059] In summary, this method is closer to the actual preparation process than the traditional random drop model, has higher structural fidelity than idealized geometric assumptions, and the established model can be directly used for the mechanical property analysis of ceramic nanofiber aerogels. This method is universally applicable to aerogels constructed from nanofibers and nanowires, providing an effective numerical analysis tool for material structure design and performance optimization.

[0060] The above content is only for illustrating the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of the claims of this invention.

Claims

1. A nanofiber network modeling method based on image recognition and process mechanism, characterized in that, include: Set the basic modeling parameters; The curved fiber is discretized into a sequence of line segments connected end to end through image recognition, and relevant statistical features are extracted. Combining the physical mechanisms of the process, and using the statistical characteristics as control parameters, a fiber network is generated through a random walk algorithm. Based on the fiber network, a representative volumetric unit of the three-dimensional nanofiber network, namely the RVE model, is obtained.

2. The nanofiber network modeling method based on image recognition and process mechanism according to claim 1, characterized in that, The basic parameters for modeling include: determining the feature size and the proportion of length in each dimension; and setting a target volume fraction as the termination condition for fiber growth.

3. The nanofiber network modeling method based on image recognition and process mechanism according to claim 1, characterized in that, The curved fiber is discretized into a sequence of line segments connected end to end through image recognition, and relevant statistical features are extracted. The specific steps are as follows: 1) Perform image enhancement and binarization processing on the SEM image; 2) The fiber centerline is extracted using a skeletonization algorithm; 3) Identify and separate individual fibers based on endpoints and branch points using topology analysis; 4) The curved fiber is discretized into a sequence of line segments using an adaptive dynamic programming (DP) algorithm; 5) Statistical analysis of line segment length distribution, range of angle variation between adjacent line segments, and fiber diameter distribution.

4. The nanofiber network modeling method based on image recognition and process mechanism according to claim 1, characterized in that, The fiber network is generated using a random walk algorithm, which includes the following steps: 1) Determine the initial point and initial angle, and determine the position of each subsequent point in sequence according to the probability density function of the angle distribution and the line segment length distribution, forming a sequence of line segments connected end to end, until it grows to the boundary or reaches the predetermined length; 2) The bottom fiber generates the starting point in the xy plane; the near bottom fiber nucleates in the xy plane with probability p, and nucleates on the surface of the existing fiber with probability (1-p); the starting point of the fiber far from the bottom fiber is completely selected from the set of positions on the surface of the previous layer of fiber, and the fiber network is constructed generation by generation.

5. The nanofiber network modeling method based on image recognition and process mechanism according to claim 4, characterized in that, It also includes operations to optimize the generated fiber network to ensure its connectivity and physical rationality, including: eliminating spatial interference and removing isolated fiber clusters.

6. The nanofiber network modeling method based on image recognition and process mechanism according to claim 1, characterized in that, The constructed three-dimensional nanofiber network RVE model represents individual fibers as a sequence of coordinate points, and the fiber network as a set of point sequences; the fiber endpoints located on the RVE boundary surface are identified for setting periodic boundary conditions; and a script file containing fiber geometry definitions and material properties is output.

7. A nanofiber network modeling system based on image recognition and process mechanism, characterized in that, include: The modeling parameters module is used to set basic modeling parameters; The statistical feature extraction module is used to discretize curved fibers into a sequence of line segments connected end to end through image recognition and extract relevant statistical features; The fiber network generation module is used to generate a fiber network using the statistical characteristics as control parameters, based on the physical mechanism of the bonding process; The RVE model construction module is used to construct a representative volumetric unit (RVE model) of a three-dimensional nanofiber network based on the fiber network.

8. The nanofiber network modeling system based on image recognition and process mechanism according to claim 7, characterized in that, The basic modeling parameters set by the modeling parameter module include: the feature size and the proportion of length of each dimension, and the target volume fraction as the fiber growth termination condition. The statistical feature extraction module is specifically used for: image enhancement and binarization of SEM images, extraction of fiber center lines using skeletonization algorithm, identification of endpoints and branch points based on topological analysis to separate individual fibers, discretization of curved fibers into a sequence of connected line segments using adaptive DP algorithm, and statistical analysis of line segment length distribution, angle variation range of adjacent line segments, and fiber diameter distribution.

9. The nanofiber network modeling system based on image recognition and process mechanism according to claim 7, characterized in that, The fiber network generation module is specifically used for: determining the initial point and initial angle; determining the positions of subsequent points sequentially according to the probability density functions of the angle distribution and line segment length distribution, forming a sequence of line segments connected end to end until it grows to the boundary or reaches a predetermined length; controlling the generation starting point of the bottom layer fibers in the xy plane, with near-bottom layer fibers nucleating in the xy plane with probability p and on the surface of existing fibers with probability (1-p), and far from the starting point of the bottom layer fibers, selecting positions from the set of positions on the surface of the previous layer fibers, and building the fiber network generation by generation; at the same time, performing spatial interference elimination and deleting isolated fiber groups on the generated fiber network to ensure the connectivity and physical rationality of the fiber network.

10. The nanofiber network modeling system based on image recognition and process mechanism according to claim 7, characterized in that, The RVE model construction module is specifically used to: represent a single fiber as a sequence of coordinate points, represent the fiber network as a set of point sequences, identify the fiber endpoints located on the RVE boundary surface to set periodic boundary conditions, and output a script file containing fiber geometric definitions and material properties; the script file is applicable to ABAQUS software and is used for beam element or solid element modeling.

Images