A method, system, electronic device, and storage medium for modeling representative volumetric elements with realistic pore characteristics.

By acquiring measured microscopic three-phase data of fiber-reinforced composite materials, and using a collision detection-based random delivery algorithm and feature characterization model to generate pore defect unit clusters, the limitations of pore models in existing technologies are solved, and a high degree of consistency and quantifiable verification between the pore model and measured features are achieved.

CN123116568APending Publication Date: 2026-11-17LUOYANG SUNRUI RUBBER & PLASTIC SCIENCE & TECHNOLOGY CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202611237191.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-14
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to transfer the real-world features of measured pores, such as morphology, connectivity clusters, and spatial relationships, to randomly generated, flexibly adjustable fiber structures, resulting in limitations in the generated pore models.

Method used

By acquiring measured microscopic three-phase data of fiber-reinforced composite materials, a random drop algorithm based on collision detection is used to generate basic representative volumetric units that satisfy the statistical parameters of fiber arrangement. Periodic mapping processing is then performed, and combined with a feature characterization model and a defect probability prediction model, a cluster of pore defect units that meet multiple constraints is selected and formed, generating representative volumetric units containing real pore characteristics.

Benefits of technology

While maintaining the randomness of fiber arrangement, the generated pore model is closer to the measured statistical characteristics in terms of position, size, morphology and connectivity. This reduces the occurrence of pore errors intruding into the fiber entity and can quantitatively verify the consistency of the model, making it suitable for performance prediction of composite materials.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123116568A_ABST
    Figure CN123116568A_ABST
Patent Text Reader

Abstract

This invention provides a method, system, electronic device, and storage medium for modeling representative volumetric units containing real pore features. The method includes: acquiring measured data and determining target parameters; generating basic representative volumetric units with periodic boundaries; discrete modeling and candidate unit determination; isomorphic local feature calculation and standardization; training the model and outputting defect probabilities; selecting defective units under constraints and forming pore entities; and consistency verification and output. This invention decouples the morphology, connectivity clusters, and spatial relationships of measured pores from specific microscopic images through isomorphic local feature transfer. Combined with probabilistic prediction, it accurately identifies high-porosity pore regions in low-proportion defect samples and directionally implants pores under six physical constraints. Consistency closed-loop verification ensures that the generated model matches the measured features, thereby achieving batch generation of representative volumetric units containing real pores with accurate statistical features and reliable mechanical responses.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of numerical simulation modeling technology for materials, and more specifically, to a method, system, electronic device, and storage medium for modeling representative volumetric units with realistic pore characteristics. Background Technology

[0002] Fiber-reinforced composites possess high specific strength, high specific modulus, and designable properties, making them widely used in aerospace, transportation, and high-end equipment manufacturing. For large-tow carbon fiber reinforced composites, the large number of fibers, high fiber volume fraction, and dense local arrangement can hinder resin wetting and flow within the fiber bundles and interfibers, leading to the formation of microscopic defects such as pores within the material. The content, size, morphology, connectivity, and spatial relationship between pores and surrounding fibers all alter the load transfer path within the matrix and influence the local stress distribution, equivalent stiffness, and damage evolution behavior of the composite material.

[0003] Representative volume element modeling is an important method for studying the relationship between the microstructure and macroscopic mechanical properties of composite materials. Existing modeling methods typically generate fiber and matrix structures based on parameters such as fiber size, fiber volume fraction, and fiber orientation, and then introduce pore defects into the model using regular geometry, random distribution functions, or mapping of measured images. However, different modeling methods still have certain limitations in terms of the realism of pore morphology, spatial distribution, model adjustability, and handling of periodic boundaries.

[0004] For example, Chinese Patent CN110674589A discloses a method for simulating the random distribution of pore defects in woven composite materials. It uses an improved three-dimensional random collision algorithm to determine the location distribution of approximately spherical pores and introduces pore defects into the finite element model. While this type of method can control porosity and simulate the randomness of pore locations, the pore morphology usually needs to be pre-defined, and the pore locations are mainly determined by random collisions and the degree of dispersion. This makes it difficult to reflect the irregular boundaries, connected cluster scale, and distribution patterns of pores relative to fiber boundaries and local fiber aggregation regions in actual measurements.

[0005] Chinese patent CN118798005A discloses a global continuous modeling method for the real microstructure of composite materials. It establishes spatial distribution functions for different material parameters based on the brightness information of scanning electron microscopy images and assigns these functions to a numerical model corresponding to the image coverage area. This method can retain microstructural information for a specific microscopic field of view in the model, but the established numerical model has a direct correspondence with the input image coverage area. When it is necessary to change the fiber volume fraction, model size, or random fiber arrangement, or to generate multiple representative volumetric units with similar statistical characteristics but different specific geometric structures, it is difficult to directly transfer the spatial correlation between pores and fibers in the original image.

[0006] Chinese patent CN102819647A discloses a finite element method for modeling the random microstructure of heterogeneous materials. It uses pseudo-random numbers to determine the material properties of each element in the discrete finite element space based on the probability distribution function of the material's constituent phases. While this method can convert continuous probability distributions to finite element meshes, its probability distributions primarily describe the probability of occurrence of constituent phases in the global space, without modeling local conditions such as the distance from pores to the nearest fiber boundary, local fiber fraction, spacing between nearest-neighbor fibers, and heterogeneity of fiber arrangement. When using unconditional random assignment, even if the generated model satisfies the overall porosity, it may still fail to maintain the scale of connected clusters, morphological distribution, and neighboring fiber relationships of the measured pores.

[0007] Furthermore, Chinese Patent CN117789885A discloses a method for establishing a finite element model of short fiber reinforced composite materials. This method parameterizes the geometric and spatial characteristics of the fibers and uses an improved random sequence adsorption method to generate representative volume elements with geometric periodicity. This method primarily addresses the issues of random fiber placement, fiber interference judgment, and periodic boundary generation. Its basic model assumes that the fibers are well-bonded to the matrix and that there are no porosity defects, but it does not involve extracting the porosity occurrence patterns from measured microscopic data and transferring them to the randomly generated fiber structure.

[0008] Furthermore, pore defects typically account for a low proportion in composite materials, and there is a significant class imbalance between defective and non-defective samples. At the same time, existing methods also suffer from the problem of separation between porosity control and pore authenticity evaluation, which may lead to a large difference between the morphology of generated pores and the measured results. Summary of the Invention

[0009] The problem solved by this invention is that existing technologies have difficulty transferring the real features of measured pore morphology, connectivity clusters and spatial relationships to randomly generated and flexibly adjustable fiber structures, resulting in limitations in the generated pore models.

[0010] To address the aforementioned problems, this disclosure provides a method for modeling representative volumetric elements containing real pore characteristics, including:

[0011] S1. Obtain measured microscopic three-phase data of fiber-reinforced composite materials, and determine the basic parameters of the target fiber and the measured statistical characteristics of the target porosity based on the measured microscopic three-phase data;

[0012] S2. Based on the basic parameters of the target fiber, a random placement algorithm based on collision detection is used to generate a basic representative volume unit that satisfies the statistical parameters of fiber arrangement, and periodic mapping processing is performed on the fibers that cross the boundary so that the basic representative volume unit has a periodic geometric boundary.

[0013] S3. Discretize the basic representative volume unit, establish the fiber unit set, matrix unit set, periodic corresponding unit set, and the mapping relationship between the unit physical coordinates and unit numbers, and determine the units in the matrix unit set as candidate matrix units;

[0014] S4. Using the same feature definition and the same standardization parameters, calculate the local relative geometric features that characterize the fiber arrangement state around the measured sample position and the candidate matrix unit respectively, and obtain the measured local feature vector and the model local feature vector and standardize them.

[0015] S5. Train the feature representation model using the standardized measured local feature vectors, and train the defect probability prediction model using the pore defect labels corresponding to the measured sample locations. Input the features corresponding to the candidate matrix units into the trained model to obtain the pore defect probability of each candidate matrix unit.

[0016] S6. Determine the target size of the pores to be implanted based on the target porosity. Use the pore defect probability as the selection criterion for candidate matrix units. Under the constraints of target porosity, fiber non-intrusion, pore morphology, pore connectivity clusters, pore-fiber spatial relationship, and periodicity, select defect units from the candidate matrix units and form at least one defect unit cluster. Based on the formation of pore entities in the basic representative volume unit by the defect unit cluster, obtain the pore-containing representative volume unit.

[0017] S7. Extract the generated pore statistical features from the representative volumetric unit containing pores, and compare the generated pore statistical features with the measured pore target statistical features to obtain the true feature consistency loss; when the true feature consistency loss does not meet the preset consistency condition, adjust the selection parameters of the defective unit and return to step S6; when the true feature consistency loss meets the preset consistency condition, output the representative volumetric unit containing true pore features.

[0018] Optionally, step S1 includes:

[0019] S11. Obtain microscopic images or three-dimensional microscopic tomographic data calibrated by physical scale and corresponding three-phase labels. The microscopic three-phase data includes phase category data and physical coordinate data of fibrous phase, matrix phase and pore defect phase.

[0020] S12. Extract mutually independent pore entities and pore connectivity clusters based on pore defect phase labels and calculate their morphological parameters.

[0021] S13. Based on the fiber phase label, statistically analyze the fiber diameter distribution, fiber area fraction or volume fraction, fiber orientation distribution and fiber spacing distribution to form the basic parameters of the target fiber.

[0022] S14. Calculate the spatial relationship parameters of pores relative to adjacent fibers, and establish the measured pore target statistical features based on the pore entity, pore connectivity clusters and spatial relationship parameters. The measured pore target statistical features include target porosity, pore morphology features, pore connectivity cluster features and spatial relationship features of pores relative to fibers.

[0023] Optionally, step S2 includes:

[0024] S21. Determine the geometric dimensions, target fiber volume fraction, fiber size distribution, fiber orientation distribution, and minimum fiber spacing of the basic representative volume unit based on the target fiber basic parameters.

[0025] S22. According to the fiber size distribution and fiber orientation, randomly generate the position, size and orientation of candidate fibers within the geometric region of the representative volume unit;

[0026] S23. Calculate the distance between the candidate fiber and the accepted fiber and the periodic image of the accepted fiber;

[0027] S24. If the distances calculated in step S23 all meet the minimum fiber spacing, accept the candidate fiber; otherwise, reject it and regenerate.

[0028] S25. Repeat S22-S24 until the deviation of the fiber volume fraction and fiber arrangement statistical parameters from the target value meets the preset parameter tolerance, and obtain the basic representative volume unit with periodic geometric boundary.

[0029] The fiber positions in the basic representative volume unit are generated by a collision detection-based random placement algorithm driven by statistical parameters, rather than by geometrically replicating the absolute fiber coordinates in the measured microscopic three-phase data.

[0030] Optionally, the local relative geometric features in step S4 include at least four of the following features: the distance from the sample location or candidate matrix unit to the nearest fiber boundary, the average distance to a preset number of neighboring fiber centroids, the standard deviation of the distance to the neighboring fiber centroids, the normalized distance to the local fiber geometric center, the normalized orientation angle relative to the local fiber geometric center, the local fiber area fraction or volume fraction in a preset neighborhood, the eigenvalue ratio of the local fiber centroid covariance matrix, and the coefficient of variation of the distance between neighboring fibers.

[0031] Among them, the local relative geometric features do not use the absolute image coordinates in the measured microscopic three-phase data as the pore defect prediction features.

[0032] Optionally, the measured sample locations and candidate matrix units are isomorphically calculated using the same feature definition formula and the same standardized parameters.

[0033] Optionally, step S5 includes:

[0034] S51. Train a feature representation model using standardized measured local feature vectors. The feature representation model is a sparse autoencoder that includes an encoder and a decoder.

[0035] S52. The encoder maps the measured local feature vectors and the model local feature vectors to the latent feature space to obtain the latent feature representation, and the decoder is used to obtain the corresponding reconstructed feature vectors.

[0036] S53. Determine the feature reconstruction difference based on the difference between the input local feature vector and the corresponding reconstructed feature vector;

[0037] S54. Input the latent feature representation, feature reconstruction difference and at least one local relative geometric feature into the defect probability prediction model, and output the porosity defect probability of each candidate matrix unit.

[0038] The training loss of the sparse autoencoder includes feature reconstruction loss, weight decay loss, and sparse activation constraint loss.

[0039] Optionally, the defect probability prediction model in step S54 is a random forest classification model; when training the random forest classification model, at least one of class weights, stratified sampling, or defect sample resampling is used to handle the class imbalance between pore defect samples and non-defect samples, and the out-of-bag error or validation set error is used to evaluate its prediction performance.

[0040] Optionally, step S6 includes:

[0041] S61. Determine the target total volume or number of target units for the defective units to be selected based on the target porosity and the effective volume of the basic representative volume unit.

[0042] S62. Determine at least one defect cluster seed from the candidate matrix units according to the porosity defect probability;

[0043] S63. Determine the comprehensive selection score of the adjacent candidate matrix units based on the pore defect probability, target pore cluster size, target pore morphology and pore-fiber spatial relationship of the candidate matrix units adjacent to the defect cluster seed, and expand the defect cluster seed according to the comprehensive selection score.

[0044] S64. Determine whether the defect unit cluster formed by the expansion crosses the boundary of the basic representative volume unit. If it does, generate the corresponding part of the cycle at the corresponding relative boundary.

[0045] S65. When the defective unit cluster satisfies the target total volume or the target number of units and the corresponding physical constraints, a porous entity is formed in the basic representative volume unit according to the defective unit cluster.

[0046] Optionally, in step S61, when the basic representative volume unit adopts an equal-volume discrete unit, the number of target defect units is determined according to the product of the target porosity and the total number of effective units of the basic representative volume unit; when the basic representative volume unit adopts a non-equal-volume discrete unit, the difference between the ratio of the sum of the volumes of the selected defect units to the effective volume of the basic representative volume unit and the target porosity is not greater than the preset porosity tolerance.

[0047] The selected defect element belongs to the matrix element set and not to the fiber element set.

[0048] Optionally, step S64 includes:

[0049] S641. Based on the physical coordinates of the defective unit and the boundary coordinates of the basic representative volume unit, identify the defective unit that crosses the boundary of the basic representative volume unit.

[0050] S642. Based on the set of periodic corresponding elements, determine the periodic corresponding element of the defect element at the boundary opposite to the boundary it crosses.

[0051] S643. Add the periodic corresponding element to the defect element cluster so that after the basic representative volume element is periodically spliced ​​along the corresponding direction, the porous entity across the boundary remains continuous.

[0052] Optionally, the method of forming a porous entity based on the defective element cluster in step S65 includes any of the following: deleting the matrix element corresponding to the defective element cluster and updating the material element set and the porous surface set; retaining the element topology corresponding to the defective element cluster and setting the corresponding element as a void material element; reconstructing the porous geometric entity based on the outer surface of the defective element cluster, performing Boolean difference operation on the matrix geometry in the basic representative volume element and re-meshing;

[0053] Specifically, at least one of the following is stored for each pore entity: pore defect probability, defect cluster number, defect unit number, and local relative geometric features, to form pore generation traceability data.

[0054] Optionally, step S7 includes:

[0055] S71. Recalculate multiple parameters of porosity, equivalent pore size distribution, aspect ratio distribution, roundness distribution, pore connectivity cluster scale distribution, pore distance to nearest fiber boundary distribution, and local fiber area fraction or volume fraction distribution in the pore neighborhood for the pore entity in the representative volume unit containing pores, and obtain the statistical characteristics of generated pores.

[0056] S72. Calculate the normalization deviation between the generated pore statistical features and the corresponding measured pore target statistical features, and sum the normalization deviations according to their weights to obtain the consistency loss of the true features.

[0057] S73. Compare the consistency loss of the real features with the preset consistency threshold. If it is greater than the preset consistency threshold, adjust at least one of the following: candidate unit range, defect cluster seed, number of target defect clusters, pore morphology weight, and pore-fiber spatial relationship weight, and return to step S6. If it is not greater than the preset consistency threshold, output the representative volume unit containing the real pore features and its pore generation traceability data.

[0058] Optionally, the true feature consistency loss J is calculated using the following formula: J=Σw _m ×|T _m - T _m* | / S _m T _m* Let T be the measured target statistic for the m-th term. _m S is the corresponding statistic for generating the model. _m For positive normalization, w _m The weights are non-negative and the sum of all weights is 1. The value of m is 1-M. When J is not greater than the preset consistency threshold, it is determined that the preset consistency condition is met.

[0059] Optionally, when training the defect probability prediction model, at least one of class weighting, hierarchical sampling, or defect sample resampling is used to address the class imbalance between pore defect samples and non-defect samples.

[0060] Optionally, the porosity defect phase further includes interface debonding and / or microcracks; for interface debonding, an independent set of interface candidate units is established; for microcracks, constraints on direction, length, and connectivity path are added; different defect types are trained with probabilistic prediction models and corresponding morphology and material set constraints are adopted.

[0061] Optionally, the method further includes: S8, performing periodic boundary finite element analysis on representative volume elements containing real pore characteristics.

[0062] Optionally, step S8 includes:

[0063] S81. Establish the periodic correspondence between relative boundary nodes in the representative volume element containing real pore features;

[0064] S82. Apply periodic displacement constraints to the corresponding nodes on the relative boundaries and apply macroscopic average strain or displacement load to the representative volume element containing real pore features.

[0065] S83. Obtain at least one of the following results based on the finite element calculation: volume average stress, equivalent stiffness, stress concentration around pores, and damage evolution.

[0066] This disclosure also provides a representative volume element modeling system containing real pore features, including:

[0067] The measured feature input module is used to acquire phase category data and physical coordinate data of fiber phase, matrix phase and pore defect phase, and to determine the basic parameters of target fiber and the measured pore target statistical characteristics based on the phase category data and physical coordinate data;

[0068] The basic representative volume unit generation module is used to generate basic representative volume units with fiber statistical parameters that meet the preset parameter tolerance and have periodic geometric boundaries based on the basic parameters of the target fiber and a random drop algorithm based on collision detection.

[0069] The candidate element construction module is used to discretize the basic representative volume elements, establish the mapping relationship between the fiber element set, the matrix element set, the periodic corresponding element set, and the element physical coordinates and element numbers, and determine the candidate matrix elements from the matrix element set;

[0070] The isomorphic feature transfer module is used to calculate local relative geometric features of the measured sample locations and candidate matrix units according to the same feature definition and the same standardization parameters, and to standardize the local feature vectors of the model using feature standardization parameters determined based on the measured local feature vectors;

[0071] The defect probability prediction module is used to obtain the potential feature representation and feature reconstruction differences of candidate matrix units using the feature representation model, and to output the porosity defect probability of candidate matrix units using the defect probability prediction model.

[0072] The physical constraint selection module is used to determine the total target volume or number of target units of the pores to be implanted based on the target porosity, and selects defective units and forms at least one defective unit cluster under the constraints of target porosity, non-intrusive fiber constraint, pore morphology constraint, pore connectivity cluster constraint, pore-fiber spatial relationship constraint and periodic constraint.

[0073] The porosity entity generation module is used to generate porosity entities based on the defect element clusters after the physical constraint selection module has formed the defect element clusters.

[0074] The consistency verification module is used to compare the generated pore statistical characteristics of the representative volumetric unit containing pores with the measured target pore statistical characteristics. When the consistency loss of the true features does not meet the preset consistency conditions, it controls the physical constraint selection module to reselect the defective unit, and outputs the representative volumetric unit containing the true pore features when the consistency loss of the true features meets the preset consistency conditions.

[0075] This disclosure also provides an electronic device, including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the above-described method for modeling representative volumetric elements with real pore features.

[0076] This disclosure also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method for modeling representative volumetric elements with real pore features.

[0077] Compared with the prior art, the representative volumetric element modeling method, system, electronic device, and storage medium containing real pore features described in the embodiments of this disclosure have the following beneficial effects:

[0078] 1) Statistical features such as porosity, pore morphology, pore connectivity clusters and pore-fiber spatial relationship are extracted from measured microscopic three-phase data and transferred to independently generated representative volume units. There is no need to directly copy the fiber and pore coordinates of a certain microscopic field of view. Therefore, multiple models with similar statistical features but different specific geometric structures can be generated.

[0079] 2) Based on the measured fiber basic parameters, a basic representative volume unit is generated by a collision detection-based random placement algorithm, and periodic mapping is performed on the cross-boundary fibers so that the generated model can meet the requirements of target fiber volume fraction, fiber size distribution, minimum fiber spacing and periodic geometric boundary while maintaining the randomness of fiber arrangement.

[0080] 3) Calculate local relative geometric features at the measured sample locations and on the candidate matrix units of the model according to the same definition, and use the feature standardization parameters determined by the measured data to make the porosity occurrence law characterized by relative features such as the nearest fiber boundary distance, local fiber fraction, the distance between neighboring fibers and the heterogeneity of arrangement, thereby reducing the influence of absolute image coordinates and specific microscopic fields of view on the migration results.

[0081] 4) The feature representation model is used to obtain the potential feature representation and feature reconstruction difference of local fiber arrangement. Combined with the defect probability prediction model, the pore defect probability of candidate matrix unit is output. This is helpful to identify low-probability structural environments such as local fiber aggregation and abrupt fiber spacing, and to alleviate the impact of the small number of pore defect samples and the imbalance of positive and negative sample categories on the prediction results.

[0082] 5) The probability of pore defects in candidate matrix units is used as the selection prior, and target porosity, fiber non-intrusion, pore morphology, pore connectivity clusters, pore-fiber spatial relationship and periodic constraints are applied at the same time to avoid arranging pores randomly without conditions based solely on the overall porosity, so that the generated pores are closer to the measured statistical characteristics in terms of location, size, morphology and connectivity.

[0083] 6) Confining defective elements within the matrix element set can reduce the occurrence of pores erroneously intruding into the fibrous entity; for pores that cross the boundary of representative volume elements, by generating corresponding periodic parts at the relative boundary, the pore entity after periodic splicing remains continuous, which is beneficial for subsequent application of periodic boundary conditions.

[0084] 7) Recalculate the porosity, equivalent size, aspect ratio, roundness, connected cluster scale, nearest fiber boundary distance, and local fiber fraction of the generated pores, and establish the consistency loss of the true features accordingly; when the consistency loss does not meet the preset conditions, reselect the defective unit, thereby forming a modeling closed loop that can be quantitatively verified and iteratively corrected, reducing the situation where only the porosity is satisfied while other features deviate from the measured results.

[0085] 8) Porous entities can be formed by deleting defective elements, setting void material elements, or performing Boolean difference operations. Tracing data such as porosity defect probability, defect cluster number, defect element number, and local relative geometric features can be saved to facilitate model verification, parameter adjustment, and correspondence analysis between finite element calculation results and porosity generation basis.

[0086] 9) Based on the same set of measured statistical characteristics of micro-defects, multiple sets of statistically equivalent models can be generated by changing the random seed, fiber volume fraction or representative volume element size. These models can be used to analyze the influence of microstructural randomness on equivalent stiffness, local stress concentration and damage evolution, providing a model basis for performance prediction of porous composite materials. Attached Figure Description

[0087] Figure 1 This is a flowchart of the representative volume element modeling method containing real pore features described in this invention;

[0088] Figure 2 This invention relates to the measured microscopic image three-phase identification and three-phase statistical process;

[0089] Figure 3 This is a schematic diagram of the position of the microscopic image slices in this invention;

[0090] Figure 4 This is a flowchart illustrating the generation of a basic representative volume element with periodic boundaries in this invention.

[0091] Figure 5 These are the verification results of the fiber unit set and matrix unit set in this invention;

[0092] Figure 6 This is a flowchart of the measured defect feature migration and physical constraint pore implantation process in this invention;

[0093] Figure 7 This is the training error curve of the autoencoder in this invention;

[0094] Figure 8 This is the out-of-bag error curve of the global random forest model in this invention;

[0095] Figure 9 This is the importance curve of local geometric features in this invention;

[0096] Figure 10 This invention compares the measured target with the porosity prediction results.

[0097] Figure 11 It is a representative volume unit with porous features in this invention;

[0098] Figure 12 This is a flowchart of the periodic boundary finite element analysis and model verification process in this invention;

[0099] Figure 13 These are stress and strain cloud diagrams of representative volume units containing pores in this invention;

[0100] Figure 14 This is a comparison chart of simulation results between the non-porous model and the porous model in this invention. Detailed Implementation

[0101] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Without conflict, the technical features of the embodiments of the present invention can be combined with each other. The finite element platform in this disclosure is not limited to specific commercial software; as long as it can read element numbers, establish periodic correspondences, modify material sets, delete elements, or perform pore geometry Boolean operations, the above methods can be implemented.

[0102] For reinforced composite materials such as large-tow carbon fibers, the large number of fibers, high fiber volume fraction, and dense local arrangement can easily hinder the wetting and flow of resin inside the fiber bundles and between fibers, thus forming micro-defects such as pores inside the material. The content, size, morphology, connectivity, and spatial relationship between the pores and surrounding fibers can all change the load transfer path in the matrix and affect the local stress distribution, equivalent stiffness, and damage evolution behavior of the composite material.

[0103] Since pore defects in composite materials usually account for a low proportion, directly classifying the location based on a small number of defect samples can easily reduce the model's ability to identify low-probability defect environments. If the absolute coordinates in the microscopic image are used directly as prediction features, the prediction results can easily become dependent on a specific microscopic field of view, making it difficult to transfer to representative volumetric units with different sizes, random seeds, or fiber arrangements.

[0104] Furthermore, due to the separation between porosity control and porosity accuracy evaluation in existing technologies, while some models can achieve the preset porosity, the generated pores may differ significantly from measured results in terms of equivalent size, aspect ratio, roundness, connected cluster scale, distance to the nearest fiber boundary, and local fiber density. Additionally, without fiber non-intrusion constraints and periodic constraints, generated pores may penetrate the fiber entity or form discontinuous structures at the relative boundaries of representative volume elements, thus affecting the accuracy of subsequent periodic boundary finite element analysis.

[0105] To address the challenges of how to transform the morphological features, connectivity cluster features, and pore-fiber spatial relationships of measured pores into features that can migrate between randomly generated representative volumetric units; how to identify local fiber arrangement environments that may form pores when the proportion of pore samples is low; how to constrain pore morphology, connectivity, neighboring fiber relationships, and periodic boundaries while meeting the target porosity; and how to quantitatively verify the consistency between the generated model and the measured pore features, the applicant proposes the following technical solutions:

[0106] like Figure 1 As shown, this disclosure provides a method for modeling representative volumetric elements with realistic pore characteristics, including:

[0107] S1. Obtain phase category data and physical coordinate data of fiber phase, matrix phase and pore defect phase, and extract basic parameters of target fiber and statistical characteristics of measured pore target;

[0108] This operation provides a statistically significant reference benchmark for subsequent modeling, ensuring that the generated model maintains consistency with the real material in terms of macroscopic statistical properties. Specifically, it acquires measured microscopic three-phase data of fiber-reinforced composite materials. This data comes from physically scale-calibrated scanning electron microscopy images, optical microscopy images, or X-ray microtomography data. Through image segmentation or semantic recognition, it obtains phase category data and physical coordinate data of the fiber phase, matrix phase, and pore defect phase.

[0109] Based on the above three-phase data, the diameter distribution, area fraction or volume fraction, orientation distribution, and spacing distribution of the fibers are statistically analyzed to form the basic parameters of the target fibers. At the same time, connected domain analysis is performed on the pore defect phase to extract mutually independent pore entities and pore connected clusters. The equivalent size, aspect ratio, roundness, and other morphological parameters of each pore entity are calculated, and the spatial relationship parameters of the pores relative to the adjacent fibers are calculated, such as the distance to the nearest fiber boundary and the local fiber area fraction. Based on this, the statistical characteristics of the measured pore target are established. The characteristics include at least the target porosity, pore morphological characteristics, pore connected cluster characteristics, and spatial relationship characteristics of the pores relative to the fibers.

[0110] S2. Based on the basic parameters of the target fiber, a random delivery algorithm including collision detection is used to generate basic representative volume units with consistent fiber statistical parameters and periodic geometric boundaries.

[0111] This operation ensures the statistical accuracy of fiber arrangement and lays the geometric foundation for subsequent periodic boundary finite element analysis. Specifically, the accepted fiber set is initialized, and the center position, size, and orientation of candidate fibers are randomly generated. The shortest distance between the candidate fiber and each fiber in the accepted fiber set is calculated, along with the shortest distance between the periodic images of the candidate fiber and the accepted fibers. When all shortest distances are not less than a preset minimum fiber spacing (typically the sum of the equivalent radii of two fibers and a preset safety interval), the candidate fiber is added to the accepted fiber set; otherwise, it is rejected and regenerated. This process is repeated until the deviation of the generated fiber volume fraction and fiber arrangement statistical parameters from the target values ​​meets the preset parameter tolerance. Subsequently, periodic image processing is performed on fibers crossing the boundary of the basic representative volume element, ensuring fiber continuity after periodic splicing in all directions, thus obtaining a basic representative volume element with periodic geometric boundaries.

[0112] The random placement algorithm uses the measured fiber diameter distribution, fiber volume fraction, and fiber spacing distribution as statistical control parameters, but does not replicate the absolute coordinates of the fibers in the measured microscopic image.

[0113] S3. Divide the voxel mesh or finite element mesh, and establish the fiber element set, matrix element set, periodic corresponding element set, and centroid coordinate-element number mapping.

[0114] This operation transforms the continuous geometric model into a discrete element system, providing a workable data structure for subsequent local feature calculations and pore selection. Specifically, the generated basic representative volume elements are discretized using either regular voxel meshes or unstructured finite element meshes; fiber element sets, matrix element sets, and periodically corresponding element sets are established, and a mapping relationship between element physical coordinates and element numbers is established. The periodically corresponding element set records element pairing information located on relative boundaries, ensuring the continuity of the cross-boundary structure during subsequent pore implantation; elements in the matrix element set are identified as candidate matrix elements, while elements in the fiber element set are excluded from the candidates, thus ensuring from the set relationship that subsequently implanted pores will not intrude into the fiber entity.

[0115] S4. Calculate the local relative geometric features at the measured sample locations and on the candidate matrix units according to the same definition, and use the standardized parameters determined by the measured data.

[0116] This operation shares feature definitions and standardized parameters between measured samples and candidate matrix units, enabling the porosity patterns learned from measured data to be directly transferred to randomly generated models, thus achieving feature transfer from measured data to randomly generated models. Specifically, for measured sample locations such as matrix and pore locations, and for determined candidate matrix units, local relative geometric features characterizing the surrounding fiber arrangement are calculated according to the exact same mathematical definitions. These local relative geometric features include at least: the distance to the nearest fiber boundary, the average distance to a preset number of nearest fiber centroids, the standard deviation of the nearest neighbor distance, the local fiber area fraction or volume fraction, the eigenvalue ratio of the local fiber centroid covariance matrix, the coefficient of variation of the nearest neighbor distance, and the normalized distance and orientation angle relative to the local fiber geometric center. All of these features are expressed in the form of local relative geometric quantities rather than absolute image coordinates, thereby eliminating dependence on a specific microscopic field of view. After calculation, the mean and standard deviation of each feature are determined as standardized parameters based on the measured local feature vectors, and the same standardized parameters are used to perform Z-score standardization on the measured local feature vectors and the model local feature vectors.

[0117] S5. Train the feature representation model and the defect probability prediction model, and output the porosity defect probability for each candidate matrix unit;

[0118] This operation reflects the tendency to form pores under the current local fiber arrangement environment by outputting the pore defect probability, providing a quantitative basis for subsequent targeted selection of defective units. Specifically, a feature representation model is trained using standardized measured local feature vectors. This model can employ unsupervised models such as sparse autoencoders, variational autoencoders, or robust autoencoders to extract low-dimensional latent representations of local features and calculate feature reconstruction differences, where feature reconstruction differences refer to the differences between the input feature vector and the autoencoder reconstruction output.

[0119] Simultaneously, a defect probability prediction model is trained using the pore defect labels (whether it is a pore or not) corresponding to the measured sample locations. This model can employ classification models such as random forests or gradient boosting trees. During training, to address the class imbalance problem caused by the extremely low proportion of pore defect samples (typically only 0.12%~0.25%), methods such as class weighting, stratified sampling, or defect sample resampling are used to improve the model's ability to identify low-probability defect environments. After training, at least two of the following—local relative geometric features, latent feature representations, and feature reconstruction differences—are input into the trained defect probability prediction model, which outputs the pore defect probability for each candidate matrix unit.

[0120] S6. Under the constraints of porosity, fiber non-intrusion, morphology, interconnected clusters, pore-fiber spatial relationship and periodicity, at least one defective unit cluster is formed, which is a porous entity formed under the control of six physical constraints; the porous entity is implanted into the basic representative volume unit to obtain a representative volume unit containing pores.

[0121] It should be noted that the defect unit cluster refers to a pore geometric object composed of several spatially adjacent defect units, and the pore entity refers to a pore geometric object formed by defect unit clusters under the constraints of target porosity, fiber non-intrusion, pore morphology, pore connectivity clusters, pore-fiber spatial relationship, and periodicity. It is the pore phase part containing the pore representative volume unit. The pore representative volume unit is composed of the pore entity, fiber units in the basic representative volume unit, and unselected matrix units.

[0122] Using the probability of pore defects as the selection criterion for candidate matrix units, candidate pools are determined in descending order of probability, and one or more defect cluster seeds are selected from the candidate pools. Subsequently, considering factors such as the probability of pore defects, the size of the target pore cluster, the morphology of the target pore (aspect ratio, roundness, etc.), and the spatial relationship between pores and fibers, the comprehensive selection score of adjacent candidate matrix units is calculated, and each cluster seed is extended to adjacent matrix units to form at least one defect unit cluster.

[0123] Preferably, the following six physical constraints are simultaneously satisfied during the expansion process: 1) Target porosity constraint - the total number of defect units matches the target porosity; 2) Fiber non-intrusion constraint - defect units are selected only from the matrix unit set and must not belong to the fiber unit set; 3) Pore morphology constraint - the equivalent size, aspect ratio, roundness and other morphological parameters of the defect cluster are consistent with the measured pore morphology characteristics; 4) Pore connectivity cluster constraint - the connectivity scale of the defect cluster is consistent with the measured pore connectivity cluster characteristics; 5) Pore-fiber spatial relationship constraint - the distance from the defect cluster to the nearest fiber boundary, the local fiber fraction, etc. are consistent with the measured spatial relationship characteristics; 6) Periodic constraint - when the defect cluster expands to the boundary of the representative volume unit, the corresponding defect unit is added to the relative boundary according to the periodic corresponding unit set to ensure the continuity of the pore entity after periodic splicing.

[0124] After determining the set of defective elements, porous solids are formed by methods such as deleting matrix voxels, retaining mesh topology and setting void material, or reconstructing pore geometry based on the outer surface of defective elements and performing Boolean difference operations, thus obtaining representative volume elements containing pores.

[0125] The target size of the pores to be implanted is determined based on the target porosity. Specifically, for equal-volume voxel meshes, the number of target defect elements is the integer product of the target porosity and the total number of effective elements; for non-equal-volume meshes, the ratio of the sum of the volumes of the selected elements to the volume of the effective model is made such that the difference between the ratio and the target porosity does not exceed a preset tolerance.

[0126] S7. Compare the generated porosity statistical characteristics with the measured target statistical characteristics; if the consistency threshold is not met, reselect the defect element; if it is met, output a representative volume element containing real porosity characteristics.

[0127] This operation constitutes a closed-loop quality control process of generation, verification, feedback, and regeneration, ensuring that the final output model is statistically highly consistent with the real material, thereby providing a reliable microstructure model for subsequent periodic boundary finite element analysis and performance prediction.

[0128] Specifically, statistical features of generated pores are extracted from representative volumetric units containing pores, including porosity, pore equivalent size distribution, aspect ratio distribution, roundness distribution, pore connectivity cluster scale distribution, pore distance to the nearest fiber boundary distribution, and local fiber area fraction or volume fraction distribution within the pore neighborhood. These generated statistical features are compared with the measured target pore statistical features established in step S1, and the consistency loss of the true features is calculated using the following formula:

[0129] J=Σw _m ×|T _m - T _m* | / S _m ,

[0130] Where T_m* Let T be the measured target statistic for the m-th term. _m S is the corresponding statistic for generating the model. _m For positive normalization, w _m The weights are non-negative and the sum of all weights is 1. The value of m is 1-M. If J is greater than the preset consistency threshold, it indicates that there is a significant deviation between the statistical characteristics of the generated pores and the measured results. Then, the parameters such as the candidate pool range, defect cluster seed, number of target clusters, pore morphology weight or pore-fiber spatial relationship weight are adjusted, and the process returns to step S6 to re-execute defect unit selection and pore implantation. If J is not greater than the preset consistency threshold, it indicates that the statistical characteristics of the generated pores are consistent with the measured results within the allowable error range. Then, a representative volume unit containing real pore characteristics is output.

[0131] This application decouples the morphology, connectivity clusters, and spatial relationships of measured pores from specific microscopic images through isomorphic local feature transfer. Combined with probabilistic prediction, it accurately identifies high-incidence pore regions in low-proportion defect samples and directionally implants pores under physical constraints. A consistency closed-loop verification ensures that the generated model matches the measured features, achieving batch generation of representative volumetric units containing real pores with accurate statistical features and reliable mechanical responses. If the consistency verification fails in step S7, the process returns to step S6, adjusting the candidate pool, defect cluster seeds, or constraint weights, and reforming the pore entity until a preset consistency threshold is met.

[0132] Example 1: Construction of Measured Microscopic Three-Phase Data and Target Feature Library:

[0133] The cross-sectional microscopic image of a large-tow carbon fiber reinforced resin matrix composite material is used as input. The sample is cut, mounted, ground and polished to obtain the test section. The image is acquired using a scanning electron microscope or a microscopic imaging device with corresponding resolution, and the proportional relationship between pixel coordinates and physical coordinates is established based on the image scale.

[0134] Grayscale correction, contrast enhancement, and three-phase recognition were performed on the image, and the fibrous phase, matrix phase, and porosity defect phase were encoded separately, see [link to documentation]. Figure 2 . Figure 2 The original microscopic images, identification results, and the three-phase ratio statistical process are shown to demonstrate the source and processing results of the measured data.

[0135] In an example dataset, the fiber proportion, porosity defect proportion, and matrix proportion of nine groups of samples are shown in Table 1. As can be seen from this data, the porosity defect proportion is significantly lower than that of the fiber phase and the matrix phase, belonging to the low proportion category. The following values ​​are only used to illustrate a feasible implementation and do not constitute a limitation of the present invention.

[0136] Table 1. Three-phase area ratio of nine groups of test samples

[0137] Sample Fiber percentage / % Pore ​​percentage Matrix percentage 1 71.88 0.17 27.95 2 71.05 0.23 28.71 3 69.96 0.15 29.89 4 71.75 0.13 28.12 5 71.23 0.12 28.65 6 71.59 0.14 28.27 7 72.39 0.21 27.4 8 69.5 0.17 30.33 9 67.49 0.25 32.27

[0138] Connectivity analysis is performed on the pore defect phase mask to group defect pixels or voxels with connectivity into pore entities. For each pore entity, the area or volume, equivalent diameter, aspect ratio, roundness, orientation, boundary coordinates, and number of cluster units are calculated; at the same time, the distance from the pore centroid or boundary to the nearest fiber boundary, the local fiber area fraction or volume fraction in the preset neighborhood, the average spacing between nearest-neighbor fibers, and their standard deviation are also calculated.

[0139] To avoid the model learning the absolute position of a specific microscopic field of view, absolute coordinates are mainly used for scale registration and tracing, while defect prediction adopts local relative geometric features. For two-dimensional data, the local neighborhood radius can be set to 1 to 5 times the average fiber diameter, and the number of nearest neighbor fibers k can be selected in the range of 3 to 20; for three-dimensional tomographic data, area, boundary, and two-dimensional distance are replaced with volume, surface, and three-dimensional distance, respectively.

[0140] To expand the sample size for low-percentage porosity defects, an original image can be divided into local sub-images with a preset overlap ratio. For example, by setting the sub-image size to 300×300 pixels and the minimum overlap ratio to 0.1, a single original image can yield 15 local sub-images, and each sample group can generate 240 effective training images. The segmentation positions are shown in the figure. Figure 3 As shown.

[0141] The measured data can also be from scanning electron microscopy images, optical microscopy images, X-ray microtomography data, or other data that can distinguish between fibrous phase, matrix phase, and pore defect phase. Two-dimensional cross-sectional data can be used for quasi-two-dimensional representative volume elements, and three-dimensional tomographic data can be directly used for three-dimensional representative volume elements.

[0142] Example 2: Generation of a basic representative volume element with periodic boundaries

[0143] The measured fiber diameter distribution, fiber volume fraction, and fiber spacing distribution were used as statistical control parameters, rather than replicating the fiber geometry according to the absolute coordinates of the fibers in the measured microscopic images. A collision detection-based random sequence adsorption algorithm was used to randomly generate candidate fibers one by one within a defined region.

[0144] The process of generating a representative volume element of periodic fundamental structure is as follows: Figure 4 As shown. If the candidate fiber does not meet the no-interference condition, return to step S23 to re-submit; if the fiber arrangement statistical deviation has not yet entered the parameter tolerance, also return to step S23 to continue generating candidate fibers.

[0145] For each candidate fiber, its center position, size, and orientation are randomly generated. The shortest distance between the candidate fiber and accepted fibers is calculated, as well as the shortest distance between the candidate fiber and fibers in adjacent periodic images. If all shortest distances are not less than the sum of the equivalent radii of the two fibers and the preset safety interval, the candidate fiber is accepted; otherwise, the candidate fiber is rejected and redeployed. This process is repeated until the deviation of the fiber volume fraction and fiber arrangement statistical parameters from the target values ​​is not greater than the preset parameter tolerance.

[0146] In a quasi-two-dimensional embodiment, the region corresponding to the microscopic image is 65μm × 55μm, and the measured actual fiber diameter is approximately 7μm. To adapt to a computational region of 0.05mm × 15mm × 15mm, the equivalent fiber diameter is set to 1.6mm; one unit is divided in the thickness direction, and 300 units are divided in each of the other two directions, forming 1 × 300 × 300, a total of 90,000 equal-volume voxel units. The above scaling is used to maintain the relative relationship between the fiber diameter and the computational region size.

[0147] For the discretized model, a fiber element set E_f and a matrix element set E_m are established, and the element number, centroid coordinates, material phase, and periodic element number of each element are saved. The verification results of the fiber element set and the matrix element set are shown in [link to relevant documentation]. Figure 5 .

[0148] Only elements in the matrix element set E_m are considered as defect candidate elements, ensuring from the set relationship that subsequent defect elements do not belong to the fiber element set. For the quasi-two-dimensional model, the constant thickness direction coordinates in the centroid coordinates can be removed to obtain the planar coordinates corresponding to the two-dimensional measured sample; for the three-dimensional model, the complete three-dimensional coordinates are retained.

[0149] The fiber generation algorithm for the basic representative volume element can employ a random sequence adsorption algorithm, an improved random collision algorithm, or other random placement algorithms that include fiber interference detection and periodic image detection. The discretization method can use a regular voxel mesh or an unstructured finite element mesh capable of establishing a mapping between the element centroid and the material set.

[0150] Example 3: Isomorphic Local Feature Transfer and Defect Probability Prediction

[0151] For the matrix location, pore location, and candidate matrix elements in the representative volume element of the foundation in the measured data, local relative geometric features are calculated according to the same mathematical definition. Explicit features that can be used include: the distance to the nearest fiber boundary d_min, the average distance to the centroids of the k nearest neighbors d_k, the standard deviation of the nearest neighbor distance s_k, the local fiber area fraction ρ_f, the eigenvalue ratio of the local fiber centroid covariance matrix, the coefficient of variation of the nearest neighbor distance H_f, and the normalized distance and orientation angle relative to the local fiber geometric center.

[0152] The process from isomorphic local feature calculation to physically constrained pore implantation is as follows: Figure 6 As shown. The measured samples and candidate matrix units share the same feature definitions and standardized parameters; if the consistency loss does not meet the requirements, the candidate pool, defect cluster seed or constraint weight is adjusted, and the process returns to step S61 to select again.

[0153] H_f=s_k / d_k

[0154] η=(λ1-λ2) / (λ1+λ2)

[0155] In the formula, λ1 and λ2 are two eigenvalues ​​of the local fibrous centroid covariance matrix, and λ1 ≥ λ2. When λ1 + λ2 equals zero, η is set to zero. All explicit features are Z-score standardized using the mean and standard deviation calculated from the measured training data, and the same standardization parameters are used in the basic representative volume unit inference stage.

[0156] This embodiment uses a sparse autoencoder as the feature representation model. The encoder maps the standardized explicit feature vector z_i to a latent vector h_i, and the decoder obtains the reconstructed vector ẑ_i based on the latent vector, using the squared Euclidean distance between the two as the reconstruction difference e_i.

[0157] h_i=f_enc(z_i),ẑ_i=f_dec(h_i),e_i=||z_i-ẑ_i||²2

[0158] The loss function of the sparse autoencoder consists of the reconstruction mean squared error, the L2 weight decay term, and the sparse activation constraint term. In a specific training instance, the latent space dimension is 5, the L2 regularization coefficient is 0.001, the sparse penalty coefficient is 4, and the maximum number of training epochs is 1500. Figure 7 The training error curve for the autoencoder is shown; the global minimum reconstruction error for this example is approximately 0.044, and the training error is controlled within 5%.

[0159] The latent vectors, reconstructed discriminants, and selected explicit features are concatenated and input into a random forest classifier, with the pore defect labels corresponding to the measured locations used as supervision signals. During training, class weights or hierarchical sampling can be used to address the issue of a low proportion of defect classes. Each training instance uses 100 decision trees, and the out-of-bag error is used to evaluate the model's generalization ability; the corresponding error curve is shown below. Figure 8 As shown, the out-of-bag error is 0.0038 when the number of trees reaches 100.

[0160] By statistically analyzing the feature importance of random forests, we can determine the contribution of features such as local fiber density, minimum fiber distance, standard deviation of nearest neighbor distance, center distance, and orientation angle to the defect probability. Figure 9 The curve represents the characteristic importance. In this example, local fiber density has a high weight in most samples, indicating that high fiber density and abrupt changes in local spacing can serve as important criteria for determining candidate pore locations.

[0161] The feature representation model can be a sparse autoencoder, variational autoencoder, robust autoencoder, or other unsupervised model capable of outputting latent features and reconstructed differences; the defect probability prediction model can be a random forest, gradient boosting tree, or other classification model capable of outputting the defect probability of candidate units. Regardless of the model used, the measured samples and candidate matrix units should use the same defined local relative geometric features and the same standardized parameters.

[0162] Example 4: Porous solid implantation under physical constraints

[0163] Porous solid implantation under physical constraints belongs to Figure 6 The closed-loop process shown in steps S61 to S67. To facilitate understanding of the pore generation stage separately, its core logic can also be summarized as: candidate pool sorting, cluster seed selection, constraint expansion, periodic completion, pore entity formation, and consistency feedback.

[0164] The standardized local features of each candidate matrix unit in the basic representative volume element are input into the trained feature representation model and defect probability prediction model to obtain the probability P_d,i that the candidate matrix unit becomes a defect unit. The number of target defect units is determined based on the measured target porosity φ_d*. For equal-volume voxels, the number of targets can be expressed as:

[0165] N_d=round(φ_d*×N_Ω)

[0166] In the formula, N_Ω represents the total number of effective voxels in the representative volume element. Taking 90,000 voxels and a target porosity of 0.17% as an example, the number of target defect elements is approximately 153. For non-uniform volume meshes, a set of defect elements is selected such that the difference between the ratio of the sum of the volumes of the selected elements to the effective model volume and the target porosity is no greater than the preset porosity tolerance.

[0167] Candidate pools are determined based on the probability of pore defects from high to low, and one or more defect cluster seeds are selected from the candidate pools. A weighted score based on the pore defect probability, target cluster size, target aspect ratio, target roundness, nearest fiber boundary distance, and local fiber fraction is used as the comprehensive selection score for adjacent units. This score is then used to expand from each cluster seed to adjacent matrix units until each defect cluster reaches the corresponding number of target units. In one embodiment, the micropore cluster consists of 1 to 5 adjacent voxels and is preferentially distributed at fiber edges and in locally high fiber density regions.

[0168] If the defect cluster extends to the boundary of the representative volume element, then corresponding defect elements are added to the relative boundary according to the periodic element set, so that the porosity remains continuous after the model is periodically spliced ​​along the corresponding direction. If a candidate element belongs to the fiber element set, then it is excluded and a matrix element is reselected from the candidate pool.

[0169] Figure 10 The comparison chart of porosity prediction results shows the target porosity ratio, predicted location, and three-phase distribution in multiple samples. It can be used to check whether the predicted porosity has a tendency to adhere to the fiber edge and aggregate in local high-density areas.

[0170] After determining the defect element set, a porous solid can be formed using any of the following methods: deleting the selected matrix voxels and updating the surface set; preserving the mesh topology but setting the selected elements as void material elements; or reconstructing the pore geometry based on the outer surface of the selected elements, performing Boolean difference operations on the matrix geometry, and re-meshing. This embodiment uses the method of deleting matrix voxels and saves the defect cluster number, element number, pore defect probability, and the local features used. The formed representative volume element containing pores is shown below. Figure 11 As shown.

[0171] After completing the pore implantation, the porosity, equivalent size, aspect ratio, roundness, connected cluster scale, nearest fiber boundary distance, and local fiber fraction in the generated model are recalculated, and the true feature consistency loss J is calculated according to the following formula:

[0172] J=Σ(m=1…M)w_m ×|T_m -T_m*| / S_m

[0173] In the formula, T_m* is the measured target statistic of the m-th term, T_m is the corresponding statistic of the generated model, S_m is the positive normalization scale, and w_m is the non-negative weight with the sum of all weights being 1. When J is greater than the preset consistency threshold, the candidate pool range, defect cluster seed, number of target clusters, pore morphology weight, or pore-fiber spatial relationship weight are adjusted, and defect cell selection is re-executed; when J is not greater than the threshold, representative volumetric cells containing real pore features are output.

[0174] The pore types can be further expanded to include interface debonding, microcracks, or combinations of multiple defect types. For interface debonding, an independent set of candidate interface units should be established; for microcracks, constraints on direction, length, and connectivity paths should be added. Different defect types can be trained with probabilistic prediction models, and corresponding morphology and material set constraints should be applied.

[0175] Example 5: Periodic Boundary Finite Element Analysis and Model Verification

[0176] Figure 12 The flowchart for periodic boundary finite element analysis and model verification includes establishing periodic correspondence, assigning material properties, applying periodic constraints and loads, solving stress and strain, calculating equivalent performance, and comparative verification.

[0177] For representative porous volumetric elements that have passed consistency verification, establish the relative boundary node correspondence and apply periodic displacement constraints:

[0178] u⁺-u⁻=ε̄·(x⁺-x⁻)

[0179] In the formula, u⁺ and u⁻ represent the displacements of the corresponding nodes relative to the boundary, x⁺ and x⁻ represent the coordinates of the corresponding nodes, and ε̄ represents the macroscopic average strain applied to the representative volume element. This constraint ensures that adjacent periodic elements maintain displacement continuity and satisfy surface force equilibrium after deformation.

[0180] This embodiment uses T700 grade carbon fiber and vinyl epoxy resin. The longitudinal elastic modulus of the carbon fiber is 230 GPa, the transverse elastic modulus is 18.2 GPa, and the shear modulus is 36.6 GPa; the resin has an elastic modulus of 3.7 GPa, a shear modulus of 1.37 GPa, and a Poisson's ratio of 0.35. The above parameters are only a specific material example.

[0181] In the transverse static tension example, a uniform displacement load of 3 mm was applied along the Y direction, and stress and strain contour plots were obtained, see [reference needed]. Figure 13 .Depend on Figure 13 It can be seen that the continuous matrix load path around the visible pores is truncated, resulting in local stress concentration and strain gradient.

[0182] Calculate the transverse tensile modulus E based on the stress S22 and strain E22 of the uniformly loaded segment, and calculate the transverse equivalent stiffness K based on the model's cross-sectional area A and length L:

[0183] E=S²² / E²², K=E·A / L

[0184] In the example with a cross-sectional area of ​​0.75 mm² and a model length of 15 mm, the transverse tensile modulus of the non-porous foundation model is 9.09 GPa. The calculation results for nine groups of models containing pores are shown in Table 2, with transverse tensile moduli ranging from 7.71 to 8.59 GPa. These results are used to illustrate that the porous entities generated by this method can produce identifiable performance effects in finite element calculations, and do not constitute a numerical limitation on the scope of protection.

[0185] Table 2 Calculation results of nine groups of representative volume elements containing pores

[0186] Model Number 1 2 3 4 5 6 7 8 9 Transverse modulus / GPa 8.47 8.34 8.52 8.56 8.59 8.54 8.38 8.4 7.71 Lateral stiffness / MPa 423.5 417.0 426.0 428.0 429.5 427.0 419.0 419.5 385.5

[0187] Figure 14 The simulation results are shown in the comparison chart. Within this example data range, the transverse modulus of the porous model is reduced by approximately 5.50% to 15.18% compared to the non-porous basic model. By changing the random seed, multiple models with different geometries can be generated while maintaining the basic parameters of the target fiber and the statistical characteristics of the measured porosity. These models can be used to evaluate the performance discrepancies caused by the randomness of the microstructure.

[0188] This invention also provides an electronic device, comprising at least a processor and a memory. The processor may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Various parts of the embodiments of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0189] A processor can also include a main processor and coprocessors. The main processor, also known as the central processing unit (CPU), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some examples, the processor may integrate a graphics processing unit (GPU), which is responsible for rendering and drawing the content that the display screen needs to show. In some examples, the processor may also include an artificial intelligence (AI) processor, which handles computational operations related to machine learning.

[0190] The memory may include one or more computer-readable storage media, which may be non-transitory. The memory may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices.

[0191] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, is used to implement the representative volumetric element modeling method for real pore features as described above.

[0192] This application also provides a computer product including a computer-readable storage medium storing a program that is loaded and executed by a processor to implement the representative volumetric element modeling method for real pore features of the above-described method embodiments.

[0193] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0194] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0195] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for modeling representative volumetric elements containing realistic pore features, characterized in that, include: S1. Obtain measured microscopic three-phase data of fiber-reinforced composite materials, and determine the basic parameters of the target fiber and the measured statistical characteristics of the target porosity based on the measured microscopic three-phase data; S2. Based on the basic parameters of the target fiber, a random placement algorithm based on collision detection is used to generate a basic representative volume unit that satisfies the statistical parameters of fiber arrangement, and periodic mapping processing is performed on the fibers that cross the boundary so that the basic representative volume unit has a periodic geometric boundary. S3. Discretize the basic representative volume unit, establish the fiber unit set, matrix unit set, periodic corresponding unit set, and the mapping relationship between the unit physical coordinates and unit numbers, and determine the units in the matrix unit set as candidate matrix units; S4. Using the same feature definition and the same standardization parameters, calculate the local relative geometric features that characterize the fiber arrangement state around the measured sample position and the candidate matrix unit respectively, and obtain the measured local feature vector and the model local feature vector and standardize them. S5. Train the feature representation model using the standardized measured local feature vectors, and train the defect probability prediction model using the pore defect labels corresponding to the measured sample locations. Input the features corresponding to the candidate matrix units into the trained model to obtain the pore defect probability of each candidate matrix unit. S6. Determine the target size of the pores to be implanted based on the target porosity. Use the pore defect probability as the selection criterion for candidate matrix units. Under the constraints of target porosity, fiber non-intrusion, pore morphology, pore connectivity clusters, pore-fiber spatial relationship, and periodicity, select defect units from the candidate matrix units and form at least one defect unit cluster. Based on the formation of pore entities in the basic representative volume unit by the defect unit cluster, obtain the pore-containing representative volume unit. S7. Extract the generated pore statistical features from the representative volumetric unit containing pores, and compare the generated pore statistical features with the measured pore target statistical features to obtain the true feature consistency loss; when the true feature consistency loss does not meet the preset consistency condition, adjust the selection parameters of the defective unit and return to step S6; when the true feature consistency loss meets the preset consistency condition, output the representative volumetric unit containing true pore features.

2. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, Step S1 includes: S11. Obtain microscopic images or three-dimensional microscopic tomographic data calibrated by physical scale and corresponding three-phase labels. The microscopic three-phase data includes phase category data and physical coordinate data of fibrous phase, matrix phase and pore defect phase. S12. Extract mutually independent pore entities and pore connectivity clusters based on pore defect phase labels and calculate their morphological parameters. S13. Based on the fiber phase label, statistically analyze the fiber diameter distribution, fiber area fraction or volume fraction, fiber orientation distribution and fiber spacing distribution to form the basic parameters of the target fiber. S14. Calculate the spatial relationship parameters of pores relative to adjacent fibers, and establish the measured pore target statistical features based on the pore entity, pore connectivity clusters and spatial relationship parameters. The measured pore target statistical features include target porosity, pore morphology features, pore connectivity cluster features and spatial relationship features of pores relative to fibers.

3. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, Step S2 includes: S21. Determine the geometric dimensions, target fiber volume fraction, fiber size distribution, fiber orientation distribution, and minimum fiber spacing of the basic representative volume unit based on the target fiber basic parameters. S22. According to the fiber size distribution and fiber orientation, randomly generate the position, size and orientation of candidate fibers within the geometric region of the representative volume unit; S23. Calculate the distance between the candidate fiber and the accepted fiber and the periodic image of the accepted fiber; S24. If the distances calculated in step S23 all meet the minimum fiber spacing, accept the candidate fiber; otherwise, reject it and regenerate. S25. Repeat S22-S24 until the deviation of the fiber volume fraction and fiber arrangement statistical parameters from the target value meets the preset parameter tolerance, and obtain the basic representative volume unit with periodic geometric boundary. The fiber positions in the basic representative volume unit are generated by a collision detection-based random placement algorithm driven by statistical parameters, rather than by geometrically replicating the absolute fiber coordinates in the measured microscopic three-phase data.

4. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, The local relative geometric features in step S4 include at least four of the following features: the distance from the sample location or candidate matrix unit to the nearest fiber boundary, the average distance to a preset number of nearest fiber centroids, the standard deviation of the distance to the nearest fiber centroids, the normalized distance to the local fiber geometric center, the normalized orientation angle relative to the local fiber geometric center, the local fiber area fraction or volume fraction in a preset neighborhood, the eigenvalue ratio of the local fiber centroid covariance matrix, and the coefficient of variation of the distance between nearest fibers. Among them, the local relative geometric features do not use the absolute image coordinates in the measured microscopic three-phase data as the pore defect prediction features.

5. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, Step S5 includes: S51. Train a feature representation model using standardized measured local feature vectors. The feature representation model includes a sparse autoencoder of an encoder and a decoder. S52. The encoder maps the measured local feature vectors and the model local feature vectors to the latent feature space to obtain the latent feature representation, and the decoder is used to obtain the corresponding reconstructed feature vectors. S53. Determine the feature reconstruction difference based on the difference between the input local feature vector and the corresponding reconstructed feature vector; S54. Input the latent feature representation, feature reconstruction difference and at least one local relative geometric feature into the defect probability prediction model, and output the porosity defect probability of each candidate matrix unit. The training loss of the sparse autoencoder includes feature reconstruction loss, weight decay loss, and sparse activation constraint loss.

6. The method for modeling representative volumetric elements containing real pore features according to claim 5, characterized in that, The defect probability prediction model in step S54 is a random forest classification model. When training the random forest classification model, at least one of the following methods is used to handle the class imbalance between pore defect samples and non-defect samples: class weighting, stratified sampling, or defect sample resampling. Its prediction performance is evaluated using out-of-bag error or validation set error.

7. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, Step S6 includes: S61. Determine the target total volume or number of target units for the defective units to be selected based on the target porosity and the effective volume of the basic representative volume unit. S62. Determine at least one defect cluster seed from the candidate matrix units according to the porosity defect probability; S63. Determine the comprehensive selection score of the adjacent candidate matrix units based on the pore defect probability, target pore cluster size, target pore morphology and pore-fiber spatial relationship of the candidate matrix units adjacent to the defect cluster seed, and expand the defect cluster seed according to the comprehensive selection score. S64. Determine whether the defect unit cluster formed by the expansion crosses the boundary of the basic representative volume unit. If it does, generate the corresponding part of the cycle at the corresponding relative boundary. S65. When the defective unit cluster satisfies the target total volume or the target number of units and the corresponding physical constraints, a porous entity is formed in the basic representative volume unit according to the defective unit cluster.

8. The method for modeling representative volumetric elements containing real pore features according to claim 7, characterized in that, In step S61, when the basic representative volume unit adopts an equal volume discrete unit, the number of target defect units is determined according to the product of the target porosity and the total number of effective units of the basic representative volume unit. When the basic representative volume unit adopts a non-uniform volume discrete unit, the difference between the ratio of the sum of the volumes of the selected defect units to the effective volume of the basic representative volume unit and the target porosity is not greater than the preset porosity tolerance. The selected defect element belongs to the matrix element set and not to the fiber element set.

9. The method for modeling representative volumetric elements containing real pore features according to claim 7, characterized in that, Step S64 includes: S641. Based on the physical coordinates of the defective unit and the boundary coordinates of the basic representative volume unit, identify the defective unit that crosses the boundary of the basic representative volume unit. S642. Based on the set of periodic corresponding elements, determine the periodic corresponding element of the defect element at the boundary opposite to the boundary it crosses. S643. Add the periodic corresponding element to the defect element cluster so that after the basic representative volume element is periodically spliced ​​along the corresponding direction, the porous entity across the boundary remains continuous.

10. The method for modeling representative volumetric elements containing real pore features according to claim 7, characterized in that, In step S65, the method of forming a porous entity based on the defect unit cluster includes any of the following: deleting the matrix unit corresponding to the defect unit cluster and updating the material unit set and the porous surface set; Preserve the element topology corresponding to the defect element cluster and set the corresponding element as a void material element; reconstruct the pore geometry based on the outer surface of the defect element cluster, perform Boolean difference operation on the matrix geometry in the basic representative volume element and re-mesh; Specifically, at least one of the following is stored for each pore entity: pore defect probability, defect cluster number, defect unit number, and local relative geometric features, to form pore generation traceability data.

11. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, Step S7 includes: S71. Recalculate multiple parameters of porosity, equivalent pore size distribution, aspect ratio distribution, roundness distribution, pore connectivity cluster scale distribution, pore distance to nearest fiber boundary distribution, and local fiber area fraction or volume fraction distribution in the pore neighborhood for the pore entity in the representative volume unit containing pores, and obtain the statistical characteristics of generated pores. S72. Calculate the normalization deviation between the generated pore statistical features and the corresponding measured pore target statistical features, and sum the normalization deviations according to their weights to obtain the consistency loss of the true features. S73. Compare the consistency loss of the real features with the preset consistency threshold. If it is greater than the preset consistency threshold, adjust at least one of the following: candidate unit range, defect cluster seed, number of target defect clusters, pore morphology weight, and pore-fiber spatial relationship weight, and return to step S6. If it is not greater than the preset consistency threshold, output the representative volume unit containing the real pore features and its pore generation traceability data.

12. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, The porosity defect phase also includes interface debonding and / or microcracks; for interface debonding, an independent set of interface candidate units is established; for microcracks, constraints on direction, length and connectivity path are added; different defect types are trained with probability prediction models and corresponding morphology and material set constraints are adopted.

13. The method for modeling representative volumetric elements containing real pore features according to claim 1, characterized in that, The method further includes: S8, performing periodic boundary finite element analysis on representative volume elements containing real pore features.

14. The method for modeling representative volumetric elements containing real pore features according to claim 13, characterized in that, Step S8 includes: S81. Establish the periodic correspondence between relative boundary nodes in the representative volume element containing real pore features; S82. Apply periodic displacement constraints to the corresponding nodes on the relative boundaries and apply macroscopic average strain or displacement load to the representative volume element containing real pore features. S83. Obtain at least one of the following results based on the finite element calculation: volume average stress, equivalent stiffness, stress concentration around pores, and damage evolution.

15. A representative volume element modeling system containing realistic pore features, characterized in that, include: The measured feature input module is used to acquire phase category data and physical coordinate data of fiber phase, matrix phase and pore defect phase, and to determine the basic parameters of target fiber and the measured pore target statistical characteristics based on the phase category data and physical coordinate data; The basic representative volume unit generation module is used to generate basic representative volume units with fiber statistical parameters that meet the preset parameter tolerance and have periodic geometric boundaries based on the basic parameters of the target fiber and a random drop algorithm based on collision detection. The candidate element construction module is used to discretize the basic representative volume elements, establish the mapping relationship between the fiber element set, the matrix element set, the periodic corresponding element set, and the element physical coordinates and element numbers, and determine the candidate matrix elements from the matrix element set; The isomorphic feature transfer module is used to calculate local relative geometric features of the measured sample locations and candidate matrix units according to the same feature definition and the same standardization parameters, and to standardize the local feature vectors of the model using feature standardization parameters determined based on the measured local feature vectors; The defect probability prediction module is used to obtain the potential feature representation and feature reconstruction differences of candidate matrix units using the feature representation model, and to output the porosity defect probability of candidate matrix units using the defect probability prediction model. The physical constraint selection module is used to determine the total target volume or number of target units of the pores to be implanted based on the target porosity, and selects defective units and forms at least one defective unit cluster under the constraints of target porosity, non-intrusive fiber constraint, pore morphology constraint, pore connectivity cluster constraint, pore-fiber spatial relationship constraint and periodic constraint. The porosity entity generation module is used to generate porosity entities based on the defect element clusters after the physical constraint selection module has formed the defect element clusters. The consistency verification module is used to compare the generated pore statistical characteristics of the representative volumetric unit containing pores with the measured target pore statistical characteristics. When the consistency loss of the true features does not meet the preset consistency conditions, it controls the physical constraint selection module to reselect the defective unit, and outputs the representative volumetric unit containing the true pore features when the consistency loss of the true features meets the preset consistency conditions.

16. An electronic device, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program, which, when executed by the processor, causes the processor to perform the representative volumetric element modeling method containing real pore features as described in any one of claims 1-14.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the representative volumetric element modeling method with real pore features as described in any one of claims 1-14.

Citation Information

Patent Citations

  • Finite element modeling method of random microstructure of heterogeneous material

    CN102819647A

  • Method for simulating random distribution of pore defects of woven composite material

    CN110674589A

  • Finite element model establishment method of short fiber reinforced composite material

    CN117789885A

  • Global continuous modeling method for real microstructure of composite material

    CN118798005A