CT Image Performance Recognition Method of Composite Materials Based on Deep Learning

Abstract

The present invention relates to a method for identifying the performance of composite material CT images based on deep learning, belonging to the field of industrial CT image processing. It includes: obtaining the CT image data of the composite material sample for gray value calibration, and inputting it into the trained U-net model to extract the images of different semantic regions, and then obtaining the corresponding gray matrix; constructing a finite element model of the composite material sample, and performing linear static analysis on the model to obtain its stiffness matrix; obtaining the gray matrix corresponding to the images of different semantic regions in the CT image of the composite material sample to be tested, inputting the gray matrix into the trained two-tower neural network model to obtain the corresponding stiffness matrix, and calculating the elastic modulus and Poisson's ratio of the composite material sample to be tested based on the stiffness matrix, so as to realize the performance detection of the composite material sample to be tested. This method realizes the detection from the composite material CT image to the composite material performance, and solves the problems of low efficiency and low accuracy in the current composite material performance detection.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial CT image processing, and particularly to a method for identifying the performance of composite material CT images based on deep learning. Background Technique

[0002] Due to their excellent mechanical properties and lightweight characteristics, composite materials are widely used in fields such as aerospace and the automotive industry. During the service process of composite materials, due to the influence of many factors, with the extension of the service time, their own properties will change to a certain extent. Therefore, it is particularly important to detect the properties of composite materials.

[0003] At present, the performance detection of composite materials mostly relies on manual detection methods, which are not only time-consuming and laborious, but also rely on manual experience and subjective judgment, resulting in large errors in the detection results and reducing the accuracy of detection.

[0004] It can be seen that providing a method for identifying the performance of composite materials to improve the detection efficiency and accuracy is an urgent problem to be solved. Summary of the Invention

[0005] In view of the above analysis, the present invention aims to provide a method for identifying the performance of composite material CT images based on deep learning to solve the problems of low efficiency and low accuracy in the current performance detection of composite materials.

[0006] The present invention provides a method for identifying the performance of composite material CT images based on deep learning, and the method includes the following steps:

[0007] Obtain the CT image data of the composite material sample, and calibrate the gray value of the CT image data;

[0008] Input the calibrated CT image data into the trained U-net model, extract the images of different semantic regions in the CT image, and obtain the corresponding gray matrix based on the extracted images of different semantic regions;

[0009] Construct a finite element model of the composite material sample, and perform linear static analysis on the finite element model to obtain its stiffness matrix;

[0010] Construct a two-tower neural network model, and train the two-tower neural network model based on the gray matrix and the stiffness matrix to obtain a trained two-tower neural network model;

[0011] Input the CT image data of the calibrated composite material sample to be measured into the trained U-net model, extract the images of different semantic regions in the CT image, and then obtain the corresponding grayscale matrix; input the grayscale matrix into the trained dual-tower neural network model to obtain the corresponding stiffness matrix, and calculate the elastic modulus and Poisson's ratio of the composite material sample to be measured based on the stiffness matrix, so as to realize the performance detection of the composite material sample to be measured.

[0012] Furthermore, calibrate the grayscale values of the CT image data by means of histogram matching.

[0013] Furthermore, the step of inputting the calibrated CT image data into the trained U-net model and extracting the images of different semantic regions in the CT image includes:

[0014] Input the calibrated CT image data into the trained U-net model to identify the images of different semantic regions in the CT image;

[0015] Establish corresponding masks for the identified images of different semantic regions;

[0016] Extract the corresponding images of different semantic regions by performing bitwise operations on the CT image and the corresponding masks.

[0017] Furthermore, the step of obtaining the corresponding grayscale matrix based on the extracted images of different semantic regions includes:

[0018] For each of the extracted images of different semantic regions: calculate the grayscale value of each pixel point of the image, and arrange the grayscale values of all pixel points of the image in the form of a matrix to obtain the corresponding grayscale matrix of the image.

[0019] Furthermore, use Digimat software to construct the finite element model of the composite material sample, perform linear static analysis on the finite element model through Abaqus software to obtain the stress, strain, reaction force and displacement data of the finite element model, and obtain the stiffness matrix of the finite element model based on the stress, strain, reaction force and displacement data.

[0020] Furthermore, the dual-tower neural network model includes an input layer, a dual-tower structure and an output layer;

[0021] Among them, the input layer is used to receive the input grayscale matrix and stiffness matrix;

[0022] The double - tower structure includes two juxtaposed tower structures and an information fusion layer. The tower structures are respectively used to extract the eigenvectors of the grayscale matrix and the eigenvectors of the stiffness matrix. The information fusion layer adopts a cross - attention mechanism to fuse and splice the eigenvectors of the grayscale matrix and the eigenvectors of the stiffness matrix, and sends the fused and spliced eigenvectors to the output layer;

[0023] The output layer includes a fully - connected layer, which is used to map the fused and spliced eigenvectors back to a matrix to generate a predicted stiffness matrix and output the predicted stiffness matrix.

[0024] Further, each tower structure includes two serially connected fully - connected modules, and the fully - connected module includes a serially connected fully - connected layer and a ReLU activation function.

[0025] Further, the expression of the loss function of the double - tower neural network model is as follows:

[0026] ,

[0027] Among them, represents the value of the loss function, represents the element in the \(i\) - th row and \(j\) - th column of the input stiffness matrix, represents the element in the \(i\) - th row and \(j\) - th column of the predicted stiffness matrix, represents the dimension of the stiffness matrix.

[0028] Further, the elastic modulus of the composite material to be measured is calculated by the following formula:

[0029] ,

[0030] Among them, \(E_1\) is the elastic modulus of the composite material to be measured in the \(x\) - direction, \(E_2\) is the elastic modulus of the composite material to be measured in the \(y\) - direction, and \(E_3\) is the elastic modulus of the composite material to be measured in the \(z\) - direction; is the element in the \(i\) - th row and \(i\) - th column of the stiffness matrix, \(A\) 11 represents the positive stiffness of the composite material to be measured in the \(x\) - direction, \(A\) 22 represents the positive stiffness of the composite material to be measured in the \(y\) - direction, \(A\) 33 represents the positive stiffness of the composite material to be measured in the \(z\) - direction.

[0031] Further, the Poisson's ratio of the composite material to be measured is calculated by the following formula:

[0032] ,

[0033] Among them, is the Poisson's ratio of the composite material to be measured, representing the transverse deformation in the j direction when the composite material to be measured is stretched in the i direction. Here, i = 1 represents the x direction, i = 2 represents the y direction, i = 3 represents the z direction, j = 1 represents the transverse deformation in the x direction, j = 2 represents the transverse deformation in the y direction, and j = 3 represents the transverse deformation in the z direction; is the element in the i-th row and j-th column of the stiffness matrix.

[0034] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0035] 1. By extracting the gray matrices corresponding to the images of different semantic regions in the CT image of the composite material sample to be measured, and obtaining the corresponding stiffness matrix through a two-tower neural network model, the present invention realizes the detection of the performance of the composite material sample from the CT image of the composite material, improves the detection efficiency and accuracy, solves the problems of low detection efficiency and low accuracy of the current composite material performance detection, and can be widely applied to the performance detection of most composite materials before and after service.

[0036] 2. By using the U-net model to identify the images of different semantic regions in the CT image of the composite material sample to be measured, the present invention improves the identification efficiency and accuracy.

[0037] 3. The present invention uses histogram matching to calibrate the gray values of the images collected by different CT scanners or the same CT scanner at different times, realizes the unification of the gray values of the CT images, avoids the errors caused by the gray value differences, reduces the noise and non-uniformity of the CT images, and thus improves the quality and efficiency of CT image processing.

[0038] 4. The present invention uses Digimat software and Abaqus software to complete a complete set of processes from the construction of the finite element model of the composite material to the linear static analysis, which is easy to operate and reduces the dependence on the professionalism of R & D personnel.

[0039] In the present invention, the above technical solutions can also be combined with each other to achieve more preferred combination schemes. Other features and advantages of the present invention will be described in the subsequent specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The drawings are only for the purpose of showing specific embodiments, and are not considered as a limitation to the present invention. Throughout the drawings, the same reference signs represent the same components;

[0041] Figure 1Flowchart of the method for identifying the performance of CT images of composite materials based on deep learning according to an embodiment of the present invention;

[0042] Figure 2 CT image of a woven composite material specimen according to an embodiment of the present invention;

[0043] Figure 3(a) is the result graph of the CT image after prediction and recognition according to an embodiment of the present invention;

[0044] Figure 3(b) is the result graph of the CT image after region division according to an embodiment of the present invention;

[0045] Figure 3(c) is the result graph of the bit operation of the CT image according to an embodiment of the present invention;

[0046] Figure 3(d) is the result graph of the extraction of the gray matrix of the CT image according to an embodiment of the present invention. Detailed implementation manners

[0047] The following will specifically describe the preferred embodiments of the present invention with reference to the accompanying drawings. The accompanying drawings form a part of this application and are used together with the embodiments of the present invention to explain the principle of the present invention, and are not used to limit the scope of the present invention.

[0048] A specific embodiment of the present invention discloses a method for identifying the performance of CT images of composite materials based on deep learning. As Figure 1 shown, the method includes the following steps:

[0049] Step S1: Obtain the CT image data of the composite material sample, and calibrate the gray value of the CT image data;

[0050] Step S2: Input the calibrated CT image data into the trained U-net model, extract the images of different semantic regions in the CT image, and obtain the corresponding gray matrix based on the extracted images of different semantic regions;

[0051] Step S3: Construct a finite element model of the composite material sample, and perform a linear static analysis on the finite element model to obtain its stiffness matrix;

[0052] Step S4: Construct a two-tower neural network model, and train the two-tower neural network model based on the gray matrix and the stiffness matrix to obtain a trained two-tower neural network model;

[0053] Step S5: Input the CT image data of the calibrated composite material sample to be measured into the trained U-net model, extract the images of different semantic regions in the CT image, and then obtain the corresponding gray matrix; input the gray matrix into the trained two-tower neural network model to obtain the corresponding stiffness matrix, and calculate the elastic modulus and Poisson's ratio of the composite material sample to be measured based on the stiffness matrix, so as to realize the performance detection of the composite material sample to be measured.

[0054] Specifically, in step S1, through CT scanning, obtain the CT image data of several composite material samples of the same type as the composite material sample to be measured. Each CT scan can obtain a series of two-dimensional slice CT images of a composite material sample, and all the two-dimensional slice CT images of this composite material sample constitute its CT image data.

[0055] It should be noted that a three-dimensional composite material sample obtains a series of two-dimensional slice CT images through CT scanning, and each two-dimensional slice CT image represents a cross-section of this composite material sample; all the two-dimensional slice CT images of this composite material sample constitute its three-dimensional volume CT image, that is, CT image data.

[0056] Specifically, the resolution of CT scanning is 2 microns, and the format of CT image data is TIFF.

[0057] It should be noted that TIFF (Tagged Image File Format) is a standard format for storing and exchanging raster images. The original image data generated by CT scanning may not be in TIFF format and can be converted to TIFF format through software to facilitate the sharing and processing of CT image data between different platforms and software.

[0058] Exemplarily, perform CT scanning on a certain woven composite material sample, and the resolution of the CT scanner is 2 microns. Obtain its CT image data through CT scanning, as Figure 2 shown, and its CT image data is output in TIFF format.

[0059] Furthermore, calibrate the gray values of the CT image data by using the method of histogram matching.

[0060] Specifically, randomly select a composite material sample, and randomly select a two-dimensional slice CT image from the CT image data of this composite material sample as the reference image, and all the two-dimensional slice CT images in the CT image data of the remaining composite material samples as the source images. Each source image is subjected to histogram matching with the reference image.

[0061] Specifically, each source image is subjected to histogram matching with the reference image through the following method:

[0062] Calculate the grayscale histograms of the reference image and the source image respectively;

[0063] Based on the grayscale histograms of the reference image and the source image, calculate the cumulative distribution functions corresponding to the reference image and the source image respectively;

[0064] For each grayscale value in the source image, find the closest value in the cumulative distribution function corresponding to the reference image through the cumulative distribution function corresponding to the source image, so as to obtain the corresponding grayscale value in the reference image;

[0065] Map each grayscale value in the source image to the corresponding grayscale value in the reference image for histogram matching to obtain the matched source image.

[0066] It should be noted that even when using the same CT scanner for CT scanning, the difference in grayscale values between the two-dimensional slice CT images obtained in one scan is small, while the difference in grayscale values between the two-dimensional slice CT images obtained in different scans is large, and the grayscale values need to be unified. In this application, by writing Python code and adopting the method of histogram matching to adjust the grayscale distribution of the image, that is, calibrating the grayscale values of the image to match the grayscale values of the reference image.

[0067] It can be understood that by using the method of histogram matching to calibrate the grayscale values of the images collected by different CT scanners or different scans of the same CT scanner, the unification of the grayscale values of the CT images is realized, the error caused by the difference in grayscale values is avoided, the noise and non-uniformity of the CT images are reduced, and thus the quality and efficiency of CT image processing are improved.

[0068] Exemplarily, for the CT image data of the woven composite material sample, the grayscale value calibration of its CT image data is realized by reading the image, calculating the grayscale histogram and the cumulative distribution function, and mapping through the OpenCV and NumPy libraries of Python.

[0069] Specifically, in step S2, the input of the calibrated CT image data into the trained U-net model to extract the images of different semantic regions in the CT image includes:

[0070] Input the calibrated CT image data into the trained U-net model to identify the images of different semantic regions in the CT image;

[0071] Establish corresponding masks for the identified images of different semantic regions;

[0072] Extract the corresponding images of different semantic regions by performing bitwise operations on the CT image and the corresponding mask.

[0073] Specifically, the calibrated CT image data is input into the trained U-net model to identify the images of different semantic regions in each two-dimensional slice CT image, and the images of different semantic regions are represented by different colors. By writing Python code, the color space of the images of different semantic regions is converted into the HSV color space, and a mask corresponding to each semantic region is established according to the HSV color space. For each two-dimensional slice CT image: the AND operation is performed on each pixel in the CT image and the corresponding pixel in the mask, and the images of different semantic regions corresponding to the CT image are extracted.

[0074] It should be noted that the images of different semantic regions are converted from the BGR (or RGB) color space to the HSV color space, and the HSV color space includes hue (H), saturation (S), and value (V). Among them, the hue represents the color information, that is, the position of the spectral color; the saturation represents the purity of the color, that is, the degree to which the color approaches the spectral color; the value represents the brightness of the color. According to the colors of the identified images of different semantic regions, the value range of the HSV color space is defined. For example, assuming that the color of the identified semantic region image is blue, the value range of the hue in the HSV color space is defined as 100-140, the value range of the saturation is 0-1, and the value range of the value is 0-1; a mask corresponding to the semantic region image is established according to the value range of the HSV color space. If the pixel value in the CT image is within the value range of the HSV color space, the corresponding pixel value in the mask is set to the same value, otherwise, the corresponding pixel value in the mask is set to 0, so as to obtain a blue mask. By performing image bitwise operations on the two-dimensional slice CT image and the corresponding mask, the images of different semantic regions corresponding to the two-dimensional slice CT image are extracted.

[0075] It can be understood that the HSV color space includes hue, saturation, and value, which is more conducive to color analysis and processing. In addition, by establishing corresponding masks for the images of different semantic regions, it helps to extract the regions of the specified color in the CT image, that is, the regions of different semantics. The present invention identifies the images of different semantic regions in the CT image of the composite material sample to be tested through the U-net model, improving the efficiency and accuracy of identification.

[0076] Furthermore, the obtaining of the corresponding gray matrix based on the images of different semantic regions extracted includes:

[0077] For each image of different semantic regions extracted: calculate the gray value of each pixel point in the image, and arrange the gray values of all pixel points in the image in the form of a matrix to obtain the gray matrix corresponding to the image.

[0078] Exemplarily, the CT image data of the labeled woven composite material sample is input into the trained U-net model to identify the images of the warp region and the weft region in the CT image. The recognition results are shown in Fig. 3(a); among them, the warp region is represented by red, and the weft region is represented by green. By writing Python code, the color space of the warp region and the weft region is converted into the HSV color space; according to the red representing the warp region, the value range of the HSV color space corresponding to red is defined, and a mask corresponding to the warp region is established according to the HSV color space corresponding to red; similarly, according to the green representing the weft region, the value range of the HSV color space corresponding to green is defined, and a mask corresponding to the weft region is established according to the HSV color space corresponding to green; the masks corresponding to the warp region and the weft region are shown in Fig. 3(b). By performing a bitwise AND operation on the CT image of the woven composite material sample and the mask corresponding to the warp region, the image of the corresponding warp region is extracted; similarly, by performing a bitwise AND operation on the CT image of the woven composite material sample and the mask corresponding to the weft region, the image of the corresponding weft region is extracted; the images of the extracted warp region and weft region are shown in Fig. 3(c). The gray value of each pixel point in the image is calculated for the images of the extracted warp region and weft region respectively, so as to obtain the gray matrix corresponding to the images of the warp region and the weft region; the image represented by the gray matrix is shown in Fig. 3(d).

[0079] Further, the trained U-net model is obtained by the following method:

[0080] Construct a U-net model, based on step S1, obtain the CT image data of the composite material sample, calibrate the gray value of the CT image data, and label the images of different semantic regions in the calibrated CT image data to construct a first data set;

[0081] Input the first data set into the U-net model for training, and the U-net model identifies the images of different semantic regions;

[0082] Calculate the value of the loss function of the U-net model through the error between the labeled and recognized images of different semantic regions. When the value of the loss function of the U-net model reaches the threshold, the trained U-net model is obtained.

[0083] Further, divide the images of different semantic regions in the calibrated CT image data, name and label the divided images, and use the labeled CT image data as the first data set.

[0084] Specifically, the images of different semantic regions in the calibrated CT image data are segmented. Each region represents a material with a specific feature. Each region is named in the form of a serial number and the corresponding material name label is marked.

[0085] Exemplarily, the images of the warp region and the weft region in the calibrated CT image data are divided. The named and marked warp images and weft images after division are used as the first data set.

[0086] Specifically, a sample in the first data set is a two-dimensional slice CT image in the CT image data of the calibrated composite material sample. The first data set includes samples corresponding to multiple composite material samples of the same type, and each composite material sample corresponds to multiple samples. When training the U-net model, the training step size is set to 64, the number of classifications is 3, the batch size is 50 or 100 or 200, and the Adam optimizer is used for the learning rate decay. The parameters of the model are iteratively adjusted by calculating the loss function value.

[0087] Specifically, in step S3, the Digimat software is used to construct the finite element model of the composite material sample. The linear static analysis of the finite element model is carried out by the Abaqus software to obtain the stress, strain, reaction force and displacement data of the finite element model. The stiffness matrix of the finite element model is obtained based on the stress, strain, reaction force and displacement data.

[0088] Specifically, the Digimat software is used to construct the geometric model of the composite material sample. The geometric model is meshed and a file in the.inp format is exported as the finite element model of the composite material sample. Subsequently, the Abaqus software is used to read the.inp file, apply the set boundary conditions and loads, and solve the problem to obtain and output the stress and strain nephograms as well as the stress, strain, reaction force and displacement data of the finite element model. By writing Python code to parse the stress, strain, reaction force and displacement data output by the Abaqus software, the nodes and elements in the finite element model are traversed to extract the stiffness matrix of the finite element model.

[0089] Exemplarily, using Digimat software, the matrix and fibers of the woven composite material sample are established, and the elastic modulus and Poisson's ratio of the matrix and fibers are respectively input; for the fibers, parameters such as fiber density, fiber diameter, and fiber length are input, and the tight arrangement degree of warp and weft yarns is selected, so as to establish the geometric model of the woven composite material sample. A hexahedral mesh is selected to mesh the geometric model, and a file in the.inp format is exported as the finite element model of the woven composite material sample. The.inp file is read using Abaqus software, the analysis type is set, boundary conditions and the magnitude of the applied load are applied to the finite element model, and Abaqus software is submitted for solution to obtain the stress and strain contour maps and stress, strain, reaction force, and displacement data of the finite element model. Python code is written to traverse the nodes and elements in the finite element model to extract the stiffness matrix of the finite element model.

[0090] It can be understood that the present invention uses Digimat software and Abaqus software to complete a complete process from the construction of the finite element model of the composite material to the linear static analysis, with simple operation and reduced dependence on the professionalism of R & D personnel.

[0091] Specifically, in step S4, the dual - tower neural network model includes an input layer, a dual - tower structure, and an output layer;

[0092] Among them, the input layer is used to receive the input grayscale matrix and stiffness matrix;

[0093] The dual - tower structure includes two parallel tower structures and an information fusion layer. The tower structures are respectively used to extract the feature vectors of the grayscale matrix and the stiffness matrix; the information fusion layer uses a cross - attention mechanism to fuse and splice the feature vectors of the grayscale matrix and the stiffness matrix, and sends the fused and spliced feature vectors to the output layer;

[0094] The output layer includes a fully - connected layer, which is used to map the fused and spliced feature vectors back to a matrix to generate a predicted stiffness matrix and output the predicted stiffness matrix.

[0095] Specifically, each tower structure includes two serially connected fully - connected modules. The fully - connected module includes a serially connected fully - connected layer and a ReLU activation function. The information fusion layer uses a cross - attention mechanism. Among them, the cross - attention mechanism uses the feature vector of the grayscale matrix as the query vector, and the feature vectors of the stiffness matrix as the key vector and value vector, enabling the dual - tower neural network model to dynamically focus on the features of the stiffness matrix based on the features of the grayscale matrix, thereby strengthening the correlation between the two features. The fully - connected layer in the output layer maps the fused and spliced feature vectors back to a two - dimensional matrix to achieve the output of the predicted stiffness matrix.

[0096] Specifically, the expression of the ReLU activation function is as follows:

[0097] ,

[0098] where represents the output of the activation function, represents the value input to the activation function, represents taking and the maximum value of the two.

[0099] Furthermore, a trained two-tower neural network model is obtained through the following method:

[0100] Construct a second data set based on the grayscale matrix obtained from steps S1 - S2 and the stiffness matrix obtained from step S3;

[0101] Input the second data set into the two-tower neural network model for training, and the two-tower neural network model predicts the stiffness matrix;

[0102] Calculate the value of the loss function of the two-tower neural network model through the error between the input and predicted stiffness matrices. When the value of the loss function of the two-tower neural network model reaches the threshold, a trained two-tower neural network model is obtained.

[0103] Specifically, a sample in the second data set includes a grayscale matrix obtained from the CT image data of a labeled composite material sample and its corresponding stiffness matrix. The second data set includes samples corresponding to multiple composite material samples of the same type, and each composite material sample corresponds to multiple samples. When training the two-tower neural network model, set the batch size to 50 or 100 or 200, adopt the Adam optimizer for learning rate decay, optimize the parameters of the model by calculating the value of the loss function of the two-tower neural network model, and through continuous iteration until the loss function converges, completing the training of the two-tower neural network model.

[0104] Specifically, the expression of the loss function of the two-tower neural network model is as follows:

[0105] ,

[0106] where represents the value of the loss function, represents the element in the i-th row and j-th column of the input stiffness matrix, represents the element in the i-th row and j-th column of the predicted stiffness matrix, represents the dimension of the stiffness matrix.

[0107] It should be noted that the dimensions of the input grayscale matrix, stiffness matrix, and the predicted stiffness matrix are the same. By training the two-tower neural network model, the trained two-tower neural network model establishes a mapping relationship between the grayscale matrix and the stiffness matrix, and saves this mapping relationship in the model through model parameters to achieve the prediction from the grayscale matrix to the stiffness matrix.

[0108] Specifically, in step S5, based on the CT image data of the composite material sample to be measured obtained in step S1, gray value calibration is performed; the CT image data of the calibrated composite material sample to be measured is input into the trained U-net model to extract images of different semantic regions in the CT image, and a corresponding grayscale matrix is obtained based on the extracted images of different semantic regions; by writing code to call the trained two-tower neural network model, the grayscale matrix is input into the trained two-tower neural network model to obtain the corresponding stiffness matrix.

[0109] It should be noted that several corresponding grayscale matrices are obtained from the CT image data of the composite material sample to be measured. Each grayscale matrix obtains the predicted stiffness matrix of this grayscale matrix through the trained two-tower neural network model, and the arithmetic mean of the predicted stiffness matrices of all grayscale matrices is taken to obtain the corresponding stiffness matrix of this composite material sample to be measured. When calling the trained two-tower neural network model to predict the stiffness matrix, the tower structure and information fusion layer for extracting the eigenvector of the stiffness matrix are shielded, the eigenvector of the input grayscale matrix is extracted through the tower structure for extracting the eigenvector of the grayscale matrix, and the corresponding stiffness matrix is generated through the mapping relationship between the grayscale matrix and the stiffness matrix represented by the fully connected layer.

[0110] Specifically, the elastic modulus of the composite material to be measured is calculated by the following formula:

[0111] ,

[0112] where E1 is the elastic modulus of the composite material to be measured in the x direction, E2 is the elastic modulus of the composite material to be measured in the y direction, and E3 is the elastic modulus of the composite material to be measured in the z direction; is the element in the i-th row and i-th column of the stiffness matrix, A 11 represents the positive stiffness of the composite material to be measured in the x direction, A 22 represents the positive stiffness of the composite material to be measured in the y direction, A 33 represents the positive stiffness of the composite material to be measured in the z direction.

[0113] Specifically, the Poisson's ratio of the composite material to be measured is calculated by the following formula:

[0114] ,

[0115] where, is the Poisson's ratio of the composite material to be measured, representing the transverse deformation in the j direction when the composite material to be measured is stretched in the i direction. Here, when i = 1, it represents the x direction; when i = 2, it represents the y direction; when i = 3, it represents the z direction. When j = 1, it represents the transverse deformation in the x direction; when j = 2, it represents the transverse deformation in the y direction; when j = 3, it represents the transverse deformation in the z direction; is the element in the i-th row and j-th column of the stiffness matrix.

[0116] It should be noted that a composite material is composed of multiple materials, and the mechanical properties of different materials are usually anisotropic; for an anisotropic composite material, its stiffness matrix can be expressed as: , where is the volume fraction of the first material in the composite material, , is the stiffness matrix of the first material in the composite material, and m is the number of different materials in the composite material. Therefore, based on the stiffness matrix of the composite material sample to be measured, the elastic modulus and Poisson's ratio of the composite material sample to be measured can be calculated. Based on the elastic modulus and Poisson's ratio of the composite material sample to be measured, performance tests such as tensile, compressive, and fatigue resistance can be performed on the composite material sample to be measured, so as to realize performance detection from the CT image of the composite material sample to be measured. By performing CT scanning on the composite material and using the above method, the performance of the composite material can be detected, improving the detection efficiency and accuracy.

[0117] It can be understood that the present invention extracts the gray matrix corresponding to the images of different semantic regions in the CT image of the composite material sample to be measured, obtains the corresponding stiffness matrix through a dual - tower neural network model, and then performs performance detection on the composite material sample to be measured, realizing the detection from the CT image of the composite material to the performance of the composite material, improving the detection efficiency and accuracy, solving the problems of low detection efficiency and low accuracy in the current performance detection of composite materials, and being able to be widely applied to the performance detection of most composite materials before and after service.

[0118] Compared with the prior art, the beneficial effects of the method for identifying the performance of a composite material CT image based on deep learning provided by the present invention are as follows:

[0119] 1. The present invention extracts the gray matrix corresponding to the images of different semantic regions in the CT image of the composite material sample to be measured, obtains the corresponding stiffness matrix through a dual - tower neural network model, and then performs performance detection on the composite material sample to be measured, realizing the detection from the CT image of the composite material to the performance of the composite material, improving the detection efficiency and accuracy, solving the problems of low detection efficiency and low accuracy in the current performance detection of composite materials, and being able to be widely applied to the performance detection of most composite materials before and after service.

[0120] 2. The present invention identifies images of different semantic regions in the CT image of the composite material sample to be measured through the U-net model, improving the efficiency and accuracy of identification.

[0121] 3. The present invention uses histogram matching to calibrate the gray values of images collected by different CT scanners or different scans of the same CT scanner, achieving the unification of the gray values of CT images, avoiding errors caused by gray value differences, reducing the noise and non-uniformity of CT images, and thus improving the quality and efficiency of CT image processing.

[0122] 4. The present invention uses Digimat software and Abaqus software to complete a complete process from the construction of a finite element model to linear static analysis of the composite material, with simple operation and reduced dependence on the professionalism of R & D personnel.

[0123] Those skilled in the art can understand that all or part of the processes for implementing the methods of the above embodiments can be completed by instructing relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium. Among them, the computer-readable storage medium is a magnetic disk, an optical disc, a read-only memory, or a random access memory, etc.

[0124] As described above, only the specific preferred embodiments of the present invention are provided, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A composite material CT image performance recognition method based on deep learning, characterized in that: The method comprises the following steps: Acquiring CT image data of the composite material sample, and performing grayscale value calibration on the CT image data; The calibrated CT image data is input into the trained U-net model to extract images of different semantic areas in the CT image, and the corresponding grayscale matrix is ​​obtained based on the extracted images of different semantic areas; Constructing a finite element model of the composite material sample, and performing a linear static analysis on the finite element model to obtain its stiffness matrix; Constructing a twin-tower neural network model, and training the twin-tower neural network model based on the grayscale matrix and the stiffness matrix to obtain a trained twin-tower neural network model; Input the calibrated CT image data of the composite material sample to be tested into the trained U-net model, extract the images of different semantic areas in the CT image, and then obtain the corresponding grayscale matrix; input the grayscale matrix into the trained double-tower neural network model to obtain the corresponding stiffness matrix, and calculate the elastic modulus and Poisson's ratio of the composite material sample to be tested based on the stiffness matrix, so as to realize the performance detection of the composite material sample to be tested; The step of inputting the calibrated CT image data into the trained U-net model and extracting images of different semantic areas in the CT image comprises: The calibrated CT image data is input into the trained U-net model to identify images of different semantic areas in the CT image; Create corresponding masks for images of different semantic areas identified; By performing bitwise operations between the CT image and the corresponding mask, the corresponding images of different semantic areas are extracted; Calibrate the grayscale value of the CT image data by using a histogram matching method; A composite material sample is randomly selected, and a two-dimensional slice CT image is randomly selected from the CT image data of the composite material sample as a reference image, and all the two-dimensional slice CT images in the CT image data of the remaining composite material samples are used as source images, and each source image is histogram matched with the reference image.

2. The method for composite material CT image performance recognition based on deep learning according to claim 1, characterized in that: The method of obtaining a corresponding grayscale matrix based on the extracted images of different semantic areas includes: For each image of different semantic areas extracted: calculate the grayscale value of each pixel of the image, and arrange the grayscale values ​​of all pixels of the image in the form of a matrix to obtain the grayscale matrix corresponding to the image.

3. The composite material CT image performance recognition method based on deep learning according to claim 1 is characterized in that: Digimat software is used to construct a finite element model of the composite material sample, and Abaqus software is used to perform linear static analysis on the finite element model to obtain stress, strain, reaction force and displacement data of the finite element model. Based on the stress, strain, reaction force and displacement data, the stiffness matrix of the finite element model is obtained.

4. The method for composite material CT image performance recognition based on deep learning according to claim 1, characterized in that: The double-tower neural network model includes an input layer, a double-tower structure and an output layer; Wherein, the input layer is used to receive the input grayscale matrix and stiffness matrix; The dual-tower structure includes two parallel tower structures and an information fusion layer, wherein the tower structures are used to extract the eigenvector of the grayscale matrix and the eigenvector of the stiffness matrix respectively; the information fusion layer adopts a cross attention mechanism to fuse and splice the eigenvector of the grayscale matrix with the eigenvector of the stiffness matrix, and sends the fused and spliced ​​eigenvector to the output layer; The output layer includes a fully connected layer, which is used to map the fused and concatenated feature vectors back to a matrix to generate a predicted stiffness matrix, and output the predicted stiffness matrix.

5. The method for composite material CT image performance recognition based on deep learning according to claim 4, characterized in that: Each tower structure includes two fully connected modules connected in series, and the fully connected module includes a fully connected layer and a ReLU activation function connected in series.

6. The method for composite material CT image performance recognition based on deep learning according to claim 1, characterized in that: The expression of the loss function of the twin-tower neural network model is as follows: ， in, represents the value of the loss function, represents the element in the i-th row and j-th column of the input stiffness matrix, represents the element in the i-th row and j-th column of the predicted stiffness matrix, Represents the dimension of the stiffness matrix.

7. The method for composite material CT image performance recognition based on deep learning according to claim 1, characterized in that: The elastic modulus of the composite material to be tested is calculated by the following formula: ， Wherein, E1 is the elastic modulus of the composite material to be tested in the x direction, E2 is the elastic modulus of the composite material to be tested in the y direction, and E3 is the elastic modulus of the composite material to be tested in the z direction; is the element in the i-th row and i-th column of the stiffness matrix, A 11 Represents the positive stiffness of the composite material to be tested in the x direction, A 22 Represents the positive stiffness of the composite material to be tested in the y direction, A 33 Represents the positive stiffness of the composite material to be tested in the z direction.

8. The method for composite material CT image performance recognition based on deep learning according to claim 7, characterized in that: The Poisson's ratio of the composite material to be tested is calculated by the following formula: ， in, is the Poisson's ratio of the composite material to be tested, indicating the transverse deformation in the j direction when the composite material to be tested is stretched in the i direction, wherein i is 1 for the x direction, i is 2 for the y direction, i is 3 for the z direction, j is 1 for the transverse deformation in the x direction, j is 2 for the transverse deformation in the y direction, and j is 3 for the transverse deformation in the z direction; is the element in the i-th row and j-th column of the stiffness matrix.

Images