Precision evaluation method applied to hybrid composite reverse microstructure design

Abstract

The application provides a precision evaluation method applied to reverse microstructure design of mixed composite materials, relates to the technical field of artificial intelligence, and comprises the following steps: based on finite element analysis technology, a two-dimensional finite element model of the mixed composite material is established, finite element simulation is performed on structure design parameters, and corresponding paired data sets are formed; based on a conditional generative adversarial network and a long short-term memory network, an initial microstructure image generation model is constructed and trained, and then a microstructure image generation model is obtained; a target whole stress-strain curve is input into the microstructure image generation model, a target microstructure image of the corresponding mixed composite material is generated, and target structure design parameters are extracted; the target structure design parameters are introduced into the two-dimensional finite element model for finite element simulation, a simulation whole stress-strain curve of the corresponding target microstructure image is generated, and the simulation whole stress-strain curve is compared with the target whole stress-strain curve, and then the precision of the reverse microstructure design is evaluated.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method for evaluating the accuracy of reverse microstructure design of hybrid composite materials. Background Technology

[0002] The microstructure and mechanical properties of hybrid composites are closely related. Reverse microstructure design is a key technology for customizing composite material properties. Its core is to deduce the corresponding microstructure parameters based on the target mechanical properties, while accuracy assessment is the crucial step in ensuring the reliability of reverse design. Currently, existing accuracy assessment methods for reverse microstructure design mostly rely on single-index comparisons, lacking a systematic parameter extraction and curve comparison system. This leads to significant deviations and fails to meet the practical needs of high-precision design for hybrid composites, thus limiting the application and development of hybrid composites in high-end fields.

[0003] Therefore, a method for accuracy evaluation applied to reverse microstructure design of hybrid composite materials is now provided. Summary of the Invention

[0004] To address the technical challenge of high-precision evaluation of hybrid composite materials, the present invention aims to provide a precision evaluation method applicable to reverse microstructure design of hybrid composite materials.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for accuracy evaluation applied to reverse microstructure design of hybrid composite materials, the method comprising: Based on finite element analysis technology, a two-dimensional finite element model of hybrid composite material is established; the structural design parameters of the hybrid composite material are corresponding to the system changes, and finite element simulation is performed on each set of structural design parameters to extract the corresponding microstructure images and stress-strain curves throughout the process, thereby forming corresponding paired datasets. An initial microstructure image generation model is constructed based on conditional generative adversarial networks and long short-term memory networks; the initial microstructure image generation model is trained on the paired dataset to obtain the final microstructure image generation model. The target stress-strain curve throughout the entire process is input into the microstructure image generation model to generate the target microstructure image of the corresponding hybrid composite material; The target structure design parameters are extracted from the target microstructure image, and the target structure design parameters are imported into a two-dimensional finite element model for finite element simulation. The simulation full-process stress-strain curve of the corresponding target microstructure image is generated and compared with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design.

[0006] Furthermore, the process of establishing a two-dimensional finite element model of hybrid composite materials based on finite element analysis technology includes: Based on finite element analysis technology, a two-dimensional finite element model of the hybrid composite material is created. The two-dimensional finite element model includes a matrix finite element model and a reinforcement finite element model. The matrix finite element model is composed of two-dimensional solid elements in the matrix region. The reinforcement finite element model is composed of two-dimensional solid elements in the fiber region and two-dimensional solid elements in the particle region. The matrix region, fiber region, and particle region are meshed using equal spacing, and boundary conditions and failure behavior are set.

[0007] Furthermore, the process of obtaining the paired dataset includes: The structural design parameters include fiber length, fiber width, fiber orientation angle, fiber spacing, fiber volume fraction, particle radius, particle volume fraction, and particle spacing. Several sets of structural design parameters are obtained, and each set of structural design parameters is input into a two-dimensional finite element model for finite element simulation. The microstructure image and the stress-strain curve of the corresponding set of structural design parameters are output. The stress-strain curve is represented by a continuous coordinate point sequence, and the sequence length is several stress-strain coordinate point pairs. The microstructure images and stress-strain curves of the corresponding group of structural design parameters are matched one by one to form the corresponding paired datasets. Then, the paired datasets of all groups of structural design parameters are obtained, and the paired datasets are divided into training set, validation set and test set according to the proportion.

[0008] Furthermore, the process of constructing the initial microstructure image generation model includes: An initial microstructure image generation model is constructed based on conditional generative adversarial networks and long short-term memory networks. The initial microstructure image generation model consists of a generator architecture and a discriminator architecture. The generator architecture consists of a temporal sequence encoding module, a feature fusion module, and a transposed convolutional decoding module. The discriminator architecture consists of a temporal sequence encoding module and a convolutional downsampling module.

[0009] Furthermore, the process of training the initial microstructure image generation model based on the paired dataset to obtain the microstructure image generation model includes: Based on the bulldozer distance and gradient penalty mechanism, and using the full stress-strain curve in the training set as the condition input, and the corresponding microstructure image as the training target, the initial microstructure image generation model is trained. The FID scores in the validation and test sets are used as quality evaluation metrics to assess the quality matching degree between the microstructure images output by the initial microstructure image generation model and the microstructure images in the corresponding paired dataset. If the quality evaluation metrics meet the quality evaluation criteria, training is stopped, the optimal model parameters are determined, and the initial microstructure image generation model is denoted as the microstructure image generation model. If the quality evaluation metrics do not meet the quality evaluation criteria, training continues until they do.

[0010] Furthermore, the target full-process stress-strain curve is input into the trained microstructure image generation model, and the target full-process stress-strain curve is processed through the corresponding generator architecture to output the target microstructure image of the corresponding hybrid composite material.

[0011] Furthermore, the process of extracting target structure design parameters from the target microstructure image includes: Image segmentation is performed on the target microstructure image to distinguish between fiber and particle regions. An adaptive threshold denoising algorithm is used to eliminate edge noise generated during image segmentation. The fiber contour of the fiber region is fitted based on the minimum bounding rectangle, and the particle contour of the particle region is fitted based on the Hough circle transform. Then, the fiber length, fiber width, fiber orientation angle, fiber spacing, and fiber volume fraction of the fiber region and the particle radius, particle volume fraction, and particle spacing of the particle region are extracted. A structural parameter correlation verification mechanism is introduced to remove abnormal extracted parameters based on the matching relationship between the volume fraction of fibers and particles and the spacing coordination constraint.

[0012] Furthermore, the process of comparing the simulated full-process stress-strain curve of the corresponding target microstructure image with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design includes: A multi-dimensional curve comparison index system was constructed, including feature point error, curve shape similarity, slope error in the elastic stage, and decay rate error in the plastic stage. The weight of each comparison index was determined by the analytic hierarchy process (AHP). Combined with the overall deviation of the bulldozer distance quantification curve, a comprehensive accuracy evaluation model was constructed to obtain the corresponding comprehensive deviation value. A graded accuracy standard is set, and the comprehensive deviation value is divided into three levels: excellent, qualified, and unqualified according to the graded accuracy standard, so as to evaluate the accuracy of reverse microstructure design.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention solves the technical problems of low parameter extraction accuracy, incomplete curve comparison, and insufficient reliability of evaluation results in existing evaluation methods by constructing a systematic reverse microstructure design accuracy evaluation process. By optimizing the construction and training method of the microstructure image generation model, the generation quality of the target microstructure image is improved. Combined with a precise structural parameter extraction method and a multi-dimensional curve comparison system, a comprehensive and accurate evaluation of reverse design accuracy is achieved. At the same time, the introduction of parameter correlation verification and closed-loop optimization mechanism further improves the rationality and reliability of reverse design, providing reliable accuracy assurance for the reverse microstructure design of hybrid composite materials, promoting the upgrading of reverse design technology for hybrid composite materials, and broadening its application scope. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a schematic diagram illustrating the steps of a method for evaluating the accuracy of reverse microstructure design in hybrid composite materials.

[0016] Figure 2 A diagram illustrating the process of constructing a paired dataset for an accuracy evaluation method applied to the inverse microstructure design of hybrid composite materials.

[0017] Figure 3 This is a flowchart for determining the optimal model in a method for evaluating the accuracy of reverse microstructure design of hybrid composite materials.

[0018] Figure 4 The flowchart shows the evaluation process for a method to assess the accuracy of reverse microstructure design of hybrid composite materials. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] Example: like Figure 1 As shown, a method for accuracy evaluation applied to reverse microstructure design of hybrid composite materials is described, the method comprising: Step S1: Based on finite element analysis technology, establish a two-dimensional finite element model of the hybrid composite material; change the structural design parameters of the hybrid composite material corresponding to the system changes, perform finite element simulation on each set of structural design parameters, extract the corresponding microstructure image and the stress-strain curve throughout the process, and then form the corresponding paired dataset. In this embodiment, the process of establishing a two-dimensional finite element model of the hybrid composite material based on finite element analysis technology includes: Based on finite element analysis technology, a two-dimensional finite element model of the hybrid composite material is created. The two-dimensional finite element model includes a matrix finite element model and a reinforcement finite element model. The matrix finite element model is composed of two-dimensional solid elements in the matrix region. The reinforcement finite element model is composed of two-dimensional solid elements in the fiber region and two-dimensional solid elements in the particle region. It should be noted that the matrix material of the matrix finite element model is polydimethylsiloxane (PDMS), the model size is 0.4mm × 0.6mm, and the CPE4R element type is used. The reinforcement of the reinforcement finite element model includes glass fibers (length 0.031–0.215mm, width 0.015mm, using CPE3 element type) and glass particles (radius 0.02–0.049mm, using CPE3 element type).

[0022] The mesh generation uses a two-dimensional solid quadrilateral free mesh with equal spacing and fine size control. The mesh size is 0.005 mm in the granular region, 0.01 mm in the matrix region, and 0.02 mm in the fiber region. The boundary conditions are set as follows: a 2.5 mm displacement load is applied to the top surface, the y-direction displacement is constrained on the bottom surface, and fixed x and y-direction displacement constraints are applied to the reference point of the bottom plate.

[0023] It should be noted that the polydimethylsiloxane body adopts the Neo-Hookean hyperelastic constitutive model, in which the shear stiffness is 0.16 MPa and the volumetric stiffness is 0.585 MPa; the elastic modulus of the glass fiber is 60 GPa and the Poisson's ratio is 0.3; and the elastic modulus of the glass particles is 50 GPa and the Poisson's ratio is 0.45.

[0024] like Figure 2 As shown, in this embodiment, the process of obtaining the paired dataset includes: The structural design parameters include fiber length, fiber width, fiber orientation angle (0°, 30°, 45°, 90° or random), fiber spacing, fiber volume fraction (0–18 vol%), particle radius, particle volume fraction (0–18 vol%), and particle spacing. Several sets of structural design parameters are obtained, and each set of structural design parameters is input into a two-dimensional finite element model for finite element simulation. The microstructure image of the corresponding set of structural design parameters and the stress-strain curve throughout the process are output. Stress-strain data points are extracted from the simulation results, and a linear interpolation method is used to unify the sequence length to 1000 coordinate point pairs. The data of the coordinate point pairs include engineering strain and nominal stress to ensure the continuity and comparability of the curves.

[0025] For example, consider a set of structural design parameters: fiber length 0.105 mm, fiber width 0.015 mm, fiber orientation angle 45°, fiber volume fraction 8 vol%, particle radius 0.03 mm, and particle volume fraction 6 vol%. Input this entire set of parameters into a pre-built two-dimensional finite element model (i.e., the matrix finite element model and the reinforcement finite element model), and perform a uniaxial tensile simulation using simulation software (such as ABAQUS). After the simulation converges without errors or severe mesh distortion, open the .odb result file. Capture the initial microstructure (the initial moment before deformation), taking the first frame of the initial load step (i.e., Time=0, no displacement load applied, no tensile deformation). This frame represents the true microstructure topology, undistorted by large deformation. Do not capture the configuration after deformation in the later stages of the tensile process; otherwise, the fiber and particle positions will be shifted by the stretching, resulting in a different microstructure than the original design. Set the screenshot framing area, with the original model's overall size at 0.4 mm × 0.6 mm, and center the core effective area at 0.2 mm × 0.3 mm. Avoid the model's edge boundary effects, retaining only the evenly distributed fiber and particle areas within. Display style settings: Rendering mode: use filled solid rendering, do not display mesh lines or unit borders; Grayscale color scheme: PDMS matrix area: grayscale 255 (pure white); glass fiber area: grayscale 128 (medium gray); glass particle area: grayscale 0 (pure black); turn off coordinates, turn off legends, and turn off the title bar, retaining only the microstructure itself. Set the output resolution to 512 × 512 pixels and save as a grayscale PNG format; each set of structural design parameters uniquely corresponds to a 512 × 512 grayscale microstructure image.

[0026] In this embodiment, the microstructure images and stress-strain curves of 1000 sets of structural design parameters are matched one by one to form corresponding paired datasets. Then, paired datasets of all sets of structural design parameters are obtained. The paired datasets are divided into training set (700 sets), validation set (200 sets), and test set (100 sets) using a stratified random sampling method of 7:2:1. The stratification is based on the combination of fiber volume fraction and particle volume fraction to ensure that the parameter distribution of each dataset is consistent and to avoid the decline in model generalization ability caused by the distribution deviation between the training set and the test set.

[0027] Step S2: Construct an initial microstructure image generation model based on conditional generative adversarial networks and long short-term memory networks; train the initial microstructure image generation model according to the paired dataset to obtain the microstructure image generation model. like Figure 3 As shown, in this embodiment, the specific process of constructing the initial microstructure image generation model includes: An initial microstructure image generation model is constructed based on conditional generative adversarial networks and long short-term memory networks. The initial microstructure image generation model consists of a generator architecture and a discriminator architecture. The generator architecture consists of a temporal sequence encoding module, a feature fusion module, and a transposed convolutional decoding module. The discriminator architecture consists of a temporal sequence encoding module and a convolutional downsampling module.

[0028] The temporal sequence encoding module of the generator architecture employs a two-layer LSTM network. The input consists of 1000 coordinate point pairs (1×1000×2 dimensions) of the entire stress-strain curve. The first LSTM layer has 256 hidden units, and the second LSTM layer has 128 hidden units. The dropout probability is 0.3 (to prevent overfitting). The output is a 128-dimensional temporal feature vector. This is used to capture the temporal variation patterns of stress-strain curves (such as the characteristic differences between the elastic and plastic stages). The feature fusion module uses a fully connected layer to fuse the temporal feature vectors. The mapping is performed as a 256-dimensional fused feature vector, and a random noise vector is introduced. (With dimensions of 1×128 and following a normal distribution), feature fusion is achieved through element-wise addition to obtain a fused feature vector. The transposed convolutional decoding module consists of four transposed convolutional layers, which progressively decode the fused feature vectors into a 512×512×1 grayscale microstructure image. The parameters of each layer are as follows: Transposed convolutional layer 1: Input dimension 1×1×256, kernel size 4×4, stride 1, padding 0, output dimension 4×4×128, activation function is ReLU; Transposed convolutional layer 2: Input dimension 4×4×128, kernel size 4×4, stride 2, padding 1, output dimension 8×8×64, activation function is ReLU; Transposed convolutional layer 3: Input dimension 8×8×64, kernel size 4×4, stride 4, padding 2, output dimension 32×32×32, activation function is ReLU; Transposed convolutional layer 4: Input dimension 32×32×32, kernel size 4×4, stride 16, padding 8, output dimension 512×512×1, activation function is sigmoid (maps the output grayscale value to the 0~1 range, corresponding to the image grayscale range 0~255).

[0029] The discriminator architecture's temporal sequence encoding module has the same structure as the generator's temporal sequence encoding module. The input is the entire stress-strain curve, and the output is a 128-dimensional temporal feature vector. The convolutional downsampling module consists of four convolutional layers, used to extract spatial features from microstructure images. The parameters of each layer are as follows: Convolutional layer 1: Input dimension 512×512×1, kernel size 4×4, stride 2, padding 1, output dimension 256×256×64, activation function is LeakyReLU (slope 0.2). Convolutional layer 2: Input dimension 256×256×64, kernel size 4×4, stride 2, padding 1, output dimension 128×128×128, activation function is LeakyReLU (slope 0.2). Convolutional layer 3: Input dimension 128×128×128, kernel size 4×4, stride 2, padding 1, output dimension 64×64×256, activation function is LeakyReLU (slope 0.2). Convolutional layer 4: Input dimension 64×64×256, kernel size 4×4, stride 2, padding 1, output dimension 32×32×512, activation function is LeakyReLU (slope 0.2). The spatial features (32×32×512) output by the convolutional downsampling module are subjected to global average pooling to obtain a 512-dimensional spatial feature vector, which is then compared with the temporal feature vector. The features are concatenated to obtain a 640-dimensional joint feature vector; this joint feature vector is then mapped to a 1-dimensional output through a fully connected layer, with the sigmoid activation function. It is used to determine the authenticity of an image.

[0030] In this embodiment, the specific process of training the initial microstructure image generation model based on the paired dataset to obtain the microstructure image generation model includes: It should be noted that, to address the issues of unstable training and significant deviation between generated and real images in traditional CGAN, a bulldozer distance (WD) and gradient penalty (GP) mechanism are introduced to construct an improved loss function, specifically including: Discriminator loss function : ;in Indicates the expected term. For mathematical expectation; Images of actual microstructures; The probability distribution representing the actual microstructure image; This is the corresponding full-process stress-strain curve; Microstructure images generated by the generator; This indicates that the discriminator receives actual microstructure images. and the corresponding full-process stress-strain curve Finally, the output image authenticity judgment value; For discriminator output; Indicates another expected term, This represents a microstructure image generated by the generator; This represents the probability distribution of the images generated by the generator, that is, the statistical distribution pattern of all generated images; (This is the gradient penalty coefficient, which is set to 10 in this embodiment). This is a gradient penalty term. ; A linearly interpolated image representing the real image and the generated image; Represents a linearly interpolated image The probability distribution; Discriminator output For interpolated images The gradient is used to measure the rate at which the discriminator output changes with the input image. The L2 norm (Euclidean norm) of the gradient is used to quantify the magnitude of the gradient, ensuring that the gradient does not exceed 1.

[0031] Where the generator loss function : ;in For loss weights; Mean squared error, used to calculate the generated image Corresponding actual image Pixel-level deviation.

[0032] It should be noted that the Adam optimizer is used, with a learning rate of 0.0002 and momentum... , To avoid oscillations during training; the batch size is set to 32 to balance training efficiency and model convergence speed; training epochs: the maximum number of training epochs is 200, and an early stopping strategy is adopted. If the quality evaluation index of the validation set does not improve for 10 consecutive epochs, training is stopped to avoid overfitting; data augmentation: the microstructure images of the training set are randomly rotated, horizontally flipped, and vertically flipped to enhance the generalization ability of the model.

[0033] The FID scores in the validation and test sets are used as quality evaluation metrics to assess the quality matching degree between the microstructure images output by the initial microstructure image generation model and the microstructure images in the corresponding paired dataset. If the quality evaluation metrics meet the quality evaluation criteria, training is stopped, the optimal model parameters are determined, and the initial microstructure image generation model is denoted as the microstructure image generation model. If the quality evaluation metrics do not meet the quality evaluation criteria, training continues until they do.

[0034] Step S3: Input the target full-process stress-strain curve into the microstructure image generation model to generate the target microstructure image of the corresponding hybrid composite material; In this embodiment, the process of inputting the target full-process stress-strain curve into the trained microstructure image generation model, processing the target full-process stress-strain curve through the corresponding generator architecture, and then outputting the target microstructure image of the corresponding hybrid composite material is as follows: The preprocessed target stress-strain curve is input into the generator of the microstructure image generation model. The generator's two-layer LSTM network extracts temporal features from the target stress-strain curve, outputting a 128-dimensional temporal feature vector. Generate random noise vectors ,and Element-wise addition is performed to obtain the fused feature vector. ; Through 4 transposed convolutional layers, The image is gradually decoded into a 512×512×1 grayscale image. After sigmoid activation, the grayscale values ​​are mapped to the range of 0~255 to obtain the target microstructure image. The generated target microstructure image is initially verified to ensure that the fibers and particles in the image do not overlap or have obvious distortion, and that the grayscale values ​​are clearly distinguishable. If there are any abnormalities, the image is regenerated (by adjusting the random noise vector) until it meets the requirements.

[0035] like Figure 4 As shown, step S4: Extract the target structure design parameters from the target microstructure image, import the target structure design parameters into the two-dimensional finite element model for finite element simulation, generate the simulation full-process stress-strain curve of the corresponding target microstructure image, and compare it with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design.

[0036] In this embodiment, the process of extracting target structure design parameters from the target microstructure image includes: Binarize the target microstructure image and segment it, setting two thresholds. , The image is divided into three regions: gray values ​​≤ 64 represent grain regions, 64 < gray value < 192 represent fiber regions, and gray values ​​≥ 192 represent matrix regions, achieving accurate differentiation between fibers, grains, and the matrix. An adaptive threshold denoising algorithm is used to eliminate edge noise generated during image segmentation. The core formula of the algorithm is: ;in These are the pixel values ​​of the segmented image. An adaptive threshold; These are the pixel values ​​of the denoised image. After denoising, morphological opening operations (erosion followed by dilation) are used to further eliminate minute noise and ensure clear region outlines.

[0037] The fiber profile of the fiber region is fitted using the minimum bounding rectangle (MBR), and the length of the minimum bounding rectangle is the fiber length. Width refers to the fiber width. The angle between the longer side of the smallest bounding rectangle and the x-axis is denoted as the fiber orientation angle. value range accurate to ; Fiber spacing The distance between the centers of adjacent fibers is calculated using Euclidean distance, and the average value of the distances between all adjacent fibers is taken; fiber volume fraction. The ratio of the number of pixels in the fiber region to the total number of pixels in the image.

[0038] The particle contour of the particle region is fitted based on the Hough Circle Transform. The parameters for Hough Circle detection are set as follows: minimum radius. Pixels (corresponding to an actual radius of 0.02mm), maximum radius Pixels (corresponding to an actual radius of 0.049mm), center-to-center spacing threshold ≥ 20 pixels (corresponding to an actual spacing ≥ 0.04mm); particle radius , where is the radius of the corresponding fitted circle; interparticle spacing Calculate the Euclidean distance between the centers of adjacent particles and take the average value; particle volume fraction. The ratio of the number of pixels in the granular region to the total number of pixels in the image.

[0039] A structural parameter correlation verification mechanism is introduced. Based on the matching relationship between fiber and particle volume fractions and the synergistic constraint condition of spacing, abnormal extracted parameters are eliminated. This ensures that the extracted parameters meet the physical logic and preparation process requirements. The verification conditions are as follows: 1. Volume fraction constraint, i.e. vol% vol% 2. Spacing constraints, i.e. , ; 3. Size constraints, i.e. , ; 4. Error constraint: the mean deviation between the extracted parameters and the corresponding parameters in the training set is ≤10%. If the deviation exceeds the range, the parameters are considered abnormal.

[0040] After verification, the final target structure design parameters are obtained.

[0041] In this embodiment, the process of comparing the simulated full-process stress-strain curve of the corresponding target microstructure image with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design includes: Construct a multi-dimensional curve comparison index system, including feature point error, curve shape similarity, slope error in the elastic stage, and decay rate error in the plastic stage; Feature point error Three key characteristic points of the stress-strain curve are selected: the yield point, the peak stress point, and the fracture point. The stress and strain errors at each characteristic point are calculated, and a weighted average is used to obtain the error at each characteristic point. : ; in , , ; , , These represent the strains corresponding to the yield point, peak stress point, and fracture point of the simulated stress-strain curve throughout the entire process. , , These represent the stresses corresponding to the yield point, peak stress point, and fracture point of the stress-strain curve throughout the simulation process. , , These represent the strains corresponding to the yield point, peak stress point, and fracture point of the target's entire stress-strain curve, respectively. , , These represent the stresses corresponding to the yield point, peak stress point, and fracture point of the target's entire stress-strain curve. The smaller the value, the smaller the deviation of the feature point.

[0042] Curve shape similarity The Dynamic Time Warping (DTW) algorithm is used to calculate the overall similarity between the two curves. The smaller the DTW distance, the higher the similarity of the curve shapes. The DTW distance is normalized to obtain the shape similarity error, and the calculation formula is as follows: ; DTW distance; This represents the number of curve coordinate points; in this embodiment, . The smaller the value, the more closely the curve shape matches.

[0043] Slope error in the elastic phase Calculate the slope of the elastic phase of the two curves. and The formula for calculating the slope error is: . The smaller the value, the smaller the deviation in elastic modulus.

[0044] Error in the rate of decay during the plastic stage The decay rate of the two curves during the plastic stage was calculated using linear fitting. and The formula for calculating the attenuation rate error is: . The smaller the value, the smaller the deviation in plastic deformation capacity.

[0045] The weights of each comparison index are determined by the analytic hierarchy process (AHP), and a comprehensive accuracy evaluation model is constructed by combining the overall deviation of the bulldozer distance quantification curve, thereby obtaining the corresponding comprehensive deviation value. A graded accuracy standard is set, and the comprehensive deviation value is divided into three levels: excellent, qualified, and unqualified according to the graded accuracy standard, so as to evaluate the accuracy of reverse microstructure design.

[0046] If the overall deviation value is less than or equal to 0.1, the reverse design accuracy is high, the deviation between the simulation curve and the target curve is small, and the microstructure perfectly matches the target mechanical properties, allowing direct application in actual production. If the overall deviation value is greater than 0.1 but less than or equal to 0.3, the reverse design accuracy meets basic requirements, and the deviation between the simulation curve and the target curve is within acceptable limits, requiring fine-tuning of the structural parameters. If the overall deviation value is greater than 0.3, the reverse design accuracy is insufficient, and the deviation between the simulation curve and the target curve is large, requiring regeneration of the microstructure image and extraction of parameters.

[0047] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0048] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0049] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0050] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0052] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0053] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0054] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for evaluating the accuracy of reverse microstructure design of hybrid composite materials, characterized in that, The method includes: Based on finite element analysis technology, a two-dimensional finite element model of hybrid composite material is established; the structural design parameters of the hybrid composite material are corresponding to the system changes, and finite element simulation is performed on each set of structural design parameters to extract the corresponding microstructure images and stress-strain curves throughout the process, thereby forming corresponding paired datasets. An initial microstructure image generation model is constructed based on conditional generative adversarial networks and long short-term memory networks; the initial microstructure image generation model is trained on the paired dataset to obtain the final microstructure image generation model. The target stress-strain curve throughout its entire lifecycle is input into the microstructure image generation model to generate the target microstructure image of the corresponding hybrid composite material. The target structure design parameters are extracted from the target microstructure image, and the target structure design parameters are imported into a two-dimensional finite element model for finite element simulation. The simulation full-process stress-strain curve of the corresponding target microstructure image is generated and compared with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design.

2. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 1, characterized in that, The process of establishing a two-dimensional finite element model of hybrid composite materials based on finite element analysis technology includes: Based on finite element analysis technology, a two-dimensional finite element model of the hybrid composite material is created. The two-dimensional finite element model includes a matrix finite element model and a reinforcement finite element model. The matrix finite element model is composed of two-dimensional solid elements in the matrix region. The reinforcement finite element model is composed of two-dimensional solid elements in the fiber region and two-dimensional solid elements in the particle region. The matrix region, fiber region, and particle region are meshed using equal spacing, and boundary conditions and failure behavior are set.

3. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 2, characterized in that, The process of obtaining the paired dataset include: The structural design parameters include fiber length, fiber width, fiber orientation angle, fiber spacing, fiber volume fraction, particle radius, particle volume fraction, and particle spacing. Several sets of structural design parameters are obtained, and each set of structural design parameters is input into a two-dimensional finite element model for finite element simulation. The microstructure image and the stress-strain curve of the corresponding set of structural design parameters are output. The stress-strain curve is represented by a continuous coordinate point sequence, and the sequence length is several stress-strain coordinate point pairs. The microstructure images and stress-strain curves of the corresponding group of structural design parameters are matched one by one to form the corresponding paired datasets. Then, the paired datasets of all groups of structural design parameters are obtained, and the paired datasets are divided into training set, validation set and test set according to the proportion.

4. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 3, characterized in that, The process of constructing an initial microstructure image generation model includes: An initial microstructure image generation model is constructed based on conditional generative adversarial networks and long short-term memory networks. The initial microstructure image generation model consists of a generator architecture and a discriminator architecture. The generator architecture consists of a temporal sequence encoding module, a feature fusion module, and a transposed convolutional decoding module. The discriminator architecture consists of a temporal sequence encoding module and a convolutional downsampling module.

5. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 4, characterized in that, The process of training the initial microstructure image generation model based on the paired dataset to obtain the microstructure image generation model includes: Based on the bulldozer distance and gradient penalty mechanism, and using the full stress-strain curve in the training set as the condition input, and the corresponding microstructure image as the training target, the initial microstructure image generation model is trained. The FID scores in the validation and test sets are used as quality evaluation metrics to assess the quality matching degree between the microstructure images output by the initial microstructure image generation model and the microstructure images in the corresponding paired dataset. If the quality evaluation metrics meet the quality evaluation criteria, training is stopped, the optimal model parameters are determined, and the initial microstructure image generation model is denoted as the microstructure image generation model. If the quality evaluation metrics do not meet the quality evaluation criteria, training continues until they do.

6. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 5, characterized in that, The target stress-strain curve is input into the trained microstructure image generation model. The target stress-strain curve is processed by the corresponding generator architecture, and then the target microstructure image of the corresponding hybrid composite material is output.

7. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 6, characterized in that, The process of extracting target structure design parameters from the target microstructure image includes: Image segmentation is performed on the target microstructure image to distinguish between fiber and particle regions. An adaptive threshold denoising algorithm is used to eliminate edge noise generated during image segmentation. The fiber contour of the fiber region is fitted based on the minimum bounding rectangle, and the particle contour of the particle region is fitted based on the Hough circle transform. Then, the fiber length, fiber width, fiber orientation angle, fiber spacing, and fiber volume fraction of the fiber region and the particle radius, particle volume fraction, and particle spacing of the particle region are extracted. A structural parameter correlation verification mechanism is introduced to remove abnormal extracted parameters based on the matching relationship between the volume fraction of fibers and particles and the spacing coordination constraint.

8. The accuracy evaluation method for reverse microstructure design of hybrid composite materials according to claim 7, characterized in that, The process of comparing the simulated full-process stress-strain curve of the corresponding target microstructure image with the target full-process stress-strain curve to evaluate the accuracy of the reverse microstructure design includes: A multi-dimensional curve comparison index system was constructed, including feature point error, curve shape similarity, slope error in the elastic stage, and decay rate error in the plastic stage. The weight of each comparison index was determined by the analytic hierarchy process (AHP). Combined with the overall deviation of the bulldozer distance quantification curve, a comprehensive accuracy evaluation model was constructed to obtain the corresponding comprehensive deviation value. A graded accuracy standard is set, and the comprehensive deviation value is divided into three levels: excellent, qualified, and unqualified according to the graded accuracy standard, so as to evaluate the accuracy of reverse microstructure design.

Images