UHPFRC fiber reconstruction and failure mechanism analysis method based on computer vision
The distribution and orientation of UHPFRC fibers are reconstructed through computer vision technology, and the failure mechanism is analyzed, which solves the problem of fiber distribution and orientation control in the prior art, optimizes the material design and use, and improves the mechanical properties and utilization efficiency of UHPFRC.
Patent Information
- Application Number
- CN202411153667.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-15
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The prior art cannot effectively control the distribution and orientation of steel fibers in ultra-high performance fiber reinforced concrete (UHPFRC), affecting its strength and damage mechanism, making it difficult to optimize material design and use.
Using a computer vision-based method, through image processing and deep learning technology, the geometric model of fiber is reconstructed, the distribution and orientation information of fibers are obtained, the crack position and orientation of the damage area is analyzed, and the damage mechanism of the component is explained in combination with the fiber distribution law.
The visualization and detailed analysis of the three-dimensional distribution rules of UHPFRC fibers are realized, the relationship between fiber distribution and mechanical properties is revealed, the material design and usage method is optimized, and the utilization efficiency of UHPFRC is improved.
Smart Images

Figure CN118967939A_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the technical field of ultra-high performance fiber reinforced concrete, and specifically is a method for analyzing the fiber reconstruction and damage mechanism of UHPFRC based on computer vision. Background Art
[0002] Ultra-high performance fiber reinforced concrete (UHPFRC) is a multiphase heterogeneous composite material composed of a matrix ultra-high performance concrete (UHPC) and randomly distributed fibers. The matrix concrete itself has ultra-high strength that ordinary concrete cannot match, and the fiber's toughening and crack-resistance effects make UHPFRC have very superior mechanical properties. The extremely high strength and toughness can withstand the cyclic load of vehicles, the high density can prevent the infiltration of rainwater, and it also has the ability to self-repair microcracks. Based on these advantages, UHPFRC is particularly suitable for use as a repair material for bridge panels.
[0003] Since it is a self-filling concrete, the steel fibers inside it show significant distribution and orientation characteristics due to their own weight and the flow of the matrix during the pouring process, and this characteristic cannot be artificially controlled at present. It is known that the distribution and orientation of fibers have a great influence on the strength and failure mechanism of UHPFRC. Therefore, it is an urgent problem for technicians in this field to find out the specific situation of the distribution and orientation of fibers and study the relationship between the distribution and orientation characteristics of fibers and mechanical properties and failure mechanisms. Summary of the invention
[0004] In response to the above-mentioned problems existing in the prior art, the present application provides a UHPFRC fiber reconstruction and failure mechanism analysis method based on computer vision, which extracts the position information and orientation information of cracks in the damaged area, and analyzes the failure mechanism of the component in combination with the distribution and orientation rules of the fibers.
[0005] To achieve the above objectives, the present application adopts the following technical solution: a method for analyzing the reconstruction and destruction mechanism of UHPFRC fibers based on computer vision, comprising the following steps:
[0006] Collecting images of the destroyed structure to form a data set;
[0007] Construct a fiber segmentation model to segment the data set, obtain pure fiber images and construct a fiber geometry model;
[0008] Based on the fiber geometry model, the fiber orientation distribution probability density function is obtained and visualized, and the fiber distribution and orientation law are analyzed;
[0009] And, construct a damaged component model based on the data set to obtain the spatial distribution and orientation information of cracks in the damaged area;
[0010] The failure mechanism of the damaged components was obtained through comparative analysis.
[0011] Among them, obtaining the fiber orientation distribution probability density function based on the geometric model includes:
[0012] Extract the fiber area on each slice section of the fiber geometric model along the X, Y, and Z axes and draw it into a line graph;
[0013] The center line of each fiber is extracted, and the three-dimensional coordinates, length, angle with each axis and orientation tensor of the fiber sampling point are obtained by spatial surface analysis;
[0014] Perform tensor decomposition on the orientation tensor and use the eigenvalues and eigenvectors to determine the parameters of the model through maximum likelihood estimation fitting;
[0015] The fiber orientation distribution probability density function is constructed based on parameters and distribution mathematical model.
[0016] Among them, extracting the center line of each fiber includes:
[0017] Preset a cylindrical template with the same specifications as the fiber;
[0018] Cylindrical correlation calculation based on cylindrical template and XFiber method;
[0019] The centerline of each fiber is extracted by correlation line tracing.
[0020] Among them, the visualization of the fiber orientation distribution probability density function is achieved through histograms and density plots to show the probability distribution of different fiber orientation angles.
[0021] The steps for drawing the histogram and density map include:
[0022] Determine the numerical range and divide it into several intervals;
[0023] Calculate the number and frequency of data points in each bin and plot a histogram;
[0024] and, determining the bandwidth based on cross-validation or information criterion, with a Gaussian kernel as the kernel function;
[0025] Use kernel density to estimate the density estimate for each data point and draw a density plot.
[0026] Among them, building a fiber segmentation model includes:
[0027] Use deep learning framework to build a network model of deep residual network infrastructure;
[0028] Input the data set and run the code to train and verify the network;
[0029] Evaluate the performance of the model and further optimize and adjust it to find the best hyperparameters of the model and apply it;
[0030] A fiber segmentation model with the best segmentation ability is obtained.
[0031] Among them, the fiber geometry model and the damaged component model are reconstructed based on image stacks and 3D volume rendering.
[0032] Among them, the failure mechanisms of the damaged components obtained through comparative analysis include:
[0033] Based on the spatial location of the crack, the distribution and orientation of the fibers and the stress conditions are compared to explain the cause of the crack.
[0034] And, combining the crack orientation law and the fiber orientation law, explain the mechanism of crack development and propagation;
[0035] Furthermore, according to the distribution and orientation rules of the fibers, the mechanical properties are analyzed in combination with the stress conditions of the components, and then the ideal distribution and orientation of the fibers are inferred in reverse.
[0036] Beneficial effects of this application:
[0037] The present application provides a method for UHPFRC fiber reconstruction and failure mechanism analysis based on computer vision, which realizes the visualization of steel fibers inside UHPFRC and reconstruction of fiber geometric models, extracts the position information and orientation information of cracks in the damaged area, analyzes the failure mechanism of the component in combination with the distribution and orientation law of the fibers, and elaborates on the 3D distribution and orientation law of the fibers and their close relationship with the mechanical properties and failure mechanism of UHPFRC, which can better understand the action mechanism of UHPFRC fibers, optimize the design and use of materials, and realize the efficient use of UHPFRC. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 A schematic diagram of the process of analyzing the UHPFRC fiber reconstruction and destruction mechanism in this application;
[0039] Figure 2 A schematic diagram of a computer tomography scan of a damaged component for the present application of the UHPFRC fiber reconstruction and damage mechanism analysis method;
[0040] Figure 3 A schematic diagram of image data in 8-bit PNG format converted from a DICOM file for the UHPFRC fiber reconstruction and damage mechanism analysis method of the present application;
[0041] Figure 4 Schematic diagram of the fiber pixel-by-pixel annotation of the UHPFRC fiber reconstruction and destruction mechanism analysis method of this application;
[0042] Figure 5 A schematic diagram of a pure fiber image of the method for analyzing the reconstruction and destruction mechanism of UHPFRC fibers in this application;
[0043] Figure 6 Schematic diagram of the fiber three-dimensional geometric model for the UHPFRC fiber reconstruction and damage mechanism analysis method of this application;
[0044] Figure 7 Schematic diagram of the fiber centerline for the analysis method of UHPFRC fiber reconstruction and destruction mechanism in this application;
[0045] Figure 8 Schematic diagram of fiber visualization vector field for the UHPFRC fiber reconstruction and damage mechanism analysis method in this application;
[0046] Fig. 9 Schematic diagram of the three-dimensional geometric model of the damaged component for the UHPFRC fiber reconstruction and damage mechanism analysis method of this application;
[0047] Fig.10 Schematic diagram of the process flow of fiber distribution and orientation processing analysis for the UHPFRC fiber reconstruction and damage mechanism analysis method of this application. DETAILED DESCRIPTION
[0048] The embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although the embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.
[0049] It should be understood that although the terms "first", "second", "third", etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.
[0050] In the description of the present application, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.
[0051] Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0052] like Figure 1 The computer vision-based UHPFRC fiber reconstruction and destruction mechanism analysis method shown includes the following steps:
[0053] S1. Collect images of destroyed structures to form a data set.
[0054] Computer tomography was performed on each of the test damaged components to obtain the original CT data in DICOM format. The DICOM format data was converted into image data in 8-bit PNG format and 16-bit TIFF format using Image J software. The image data in 8-bit PNG format was selected as the research object and numbered according to the scanning order. The specific steps include:
[0055] S111. After completing the self-inspection and calibration of the CT imaging device, the damaged concrete component is placed vertically along the length direction on the scanning table of the CT scanning device;
[0056] S112, set the scanning parameters such as resolution, X-ray voltage and current, start the CT scanning program, and let the scanning device scan from top to bottom along the XY section of the concrete component to obtain a series of raw data saved in DICOM format, such as Figure 2 Shown is a schematic diagram of computer tomography scanning of a damaged component;
[0057] S113, using Image J software to open the DICOM file, extracting the image data of the CT scan from the DICOM file and adjusting the brightness, contrast, size and resolution thereof;
[0058] S114, convert the obtained image data into 8-bit PNG and 16-bit TIFF formats, select the image data in 8-bit PNG format as the research object of this experiment and number them accordingly according to the scanning order, such as Figure 3 The image data is shown in 8-bit PNG format converted from a DICOM file, and is numbered as slice 1600 according to the scanning order.
[0059] Construct a deep learning dataset, randomly set the 8-bit PNG format image data as training set, validation set and test set in proportion, annotate the fibers on the training set and validation set images, cut the original image and the annotated image into a size of 400 pixel × 400 pixel and number them accordingly. The specific steps include:
[0060] S121. Use the Dataset_classification.py code in Visual Studio Code to set the scale and input and output paths, and randomly divide the image data into training sets, validation sets, and test sets to build a deep learning dataset.
[0061] S122. After setting the parameters such as Alpha Mask, Resize factor, Circlesize and Color in PixelAnnotation Tool, open the folders where the training set and validation set are located, and annotate the images pixel by pixel. Annotate the fibers as white ROI areas and the background and concrete mortar as black. Figure 4 Shown is a schematic diagram of pixel-by-pixel annotation of fibers on training set and validation set images;
[0062] S123, after the annotation is completed, click Save, and a marked image corresponding to the original image will be generated at the specified location;
[0063] S124. Use the crop_rename.py code in Visual Studio Code to set the input and output paths and the cutting size of 400 pixels × 400 pixels, and define the file names of the original image blocks and the corresponding annotated image blocks after cutting to be consistent, and run the code to cut and number the original image and the annotated image; the cutting size is consistent with the subsequent network model input image size, and the number of the original image blocks and the annotated image blocks after cutting is strictly equal, and the numbering corresponds one to one to ensure the accuracy of subsequent network training and verification.
[0064] S2. Construct a fiber segmentation model to segment the data set, obtain a pure fiber image and construct a fiber geometry model. Build a deep learning environment on the Ubuntu system, construct a fiber segmentation model, use the training set and the validation set to train and validate, use the corresponding parameters to evaluate the performance of the model and further optimize and adjust it, and use the network model under the best hyperparameters to perform image semantic segmentation on the test set to obtain a pure fiber image. Specifically, the following steps are included:
[0065] S211. Use the PyTorch deep learning framework to build the Fiberseg-DeepLab-V3+ network model with the ResNet deep residual network as the backbone infrastructure;
[0066] S212. Use the Preparedataset.py code in Visual Studio Code to set the file addresses of the training set and validation set as the input path and specify the output path. Run the code to generate the text formats of the training set and validation set, train.txt and val.txt, at the specified location.
[0067] S213, using train.txt and val.txt as data input formats, determine initial hyperparameters such as batch size and epoch period, run the code to train and verify the network, and view numerical indicators such as verification loss, verification accuracy, F1 score, as well as visualization curves and model parameters in TensorBoard; the determination of the initial hyperparameters such as batchsize and epoch can refer to other related network models with higher prediction accuracy, and the current model can be evaluated by viewing the corresponding parameters in the TensorBoard visualization structure management tool;
[0068] S214, evaluating the performance of the model and further optimizing and adjusting it, finding and applying the best hyperparameters of the model, and obtaining a network model with the best segmentation ability as a fiber segmentation model;
[0069] S215. Enter the path of the test set file in the optimized network model and specify the output path, run the code to implement the fiber semantic segmentation of the test set image and generate a pure fiber image at the specified location, such as Figure 5 Shown is a schematic diagram of a pure fiber image obtained after inference by the Fiberseg-DeepLab-V3+ deep learning network model.
[0070] S216, all pure fiber images are organized into image stacks according to the scanning numbers, and imported into the 3D reconstruction and analysis software Avizo, and the fiber is 3D reconstructed using the volume rendering method to obtain the fiber geometric model. After setting input parameters such as resolution and voxel size, the fiber image stack is opened, and the fiber is 3D reconstructed using the volume rendering method to obtain its geometric model, such as Figure 6 Shown is a schematic diagram of the three-dimensional geometric model of the fiber obtained by volume rendering.
[0071] S3. Based on the fiber geometry model, the fiber orientation distribution probability density function is obtained and visualized, and the distribution and orientation rules of the fibers are analyzed.
[0072] Volume Fraction is used to extract the area of the fibers on each slice section along the X, Y, and Z axes and plot them into a line graph. XFiber calculates the cylinder correlation and performs correlation line tracing to extract the center line of each fiber. The center line is then used to perform spatial surface analysis to obtain the three-dimensional coordinates, length, angle with each axis, orientation tensor, and other data of the fiber sampling point.
[0073] The statistical part of fiber distribution area extraction in each section is used to obtain the fiber area on each slice section along the X, Y, and Z axes and the distribution law of fibers along each axis, on this basis, to have a clear grasp of the overall three-dimensional distribution of fibers inside the component; the fiber centerline extraction part is used to obtain the centerline on the neutral axis of the fiber to replace the original fiber for subsequent orientation data extraction and analysis; the spatial surface analysis part of the centerline is used to realize the visualization of fiber orientation information and the concrete representation of data.
[0074] The specific steps include:
[0075] S311. Use Volume Fraction to extract the fiber area on each slice section along the X, Y, and Z axes and draw them into a line graph, and analyze the distribution of the fibers along each axis; Volume Fraction can extract the fiber area percentage along each axis section, and the obtained percentage can be multiplied by the total area of the section to obtain the fiber area of each section. After drawing the corresponding line graph, the distribution of the fibers along each axis can be analyzed;
[0076] S312, set up a cylindrical template with the same specifications as the fiber, use the template and XFiber method to calculate the cylinder correlation and perform correlation line tracking to extract the center line of each fiber, perform spatial surface analysis on the center line to obtain the fiber direction position information and orientation tensor data, such as Figure 7The figure shows the centerline of each fiber extracted by calculating the cylinder correlation and tracking the correlation line using the XFiber method. The generated centerline will replace the original fiber for subsequent orientation data extraction and analysis.
[0077] S313, the fiber direction information and orientation tensor are respectively generated using Vector Slice and Tensor View to generate corresponding vector fields and tensor graphics to visualize the approximate orientation of the fiber, such as Figure 8 The figure shows a schematic diagram of a visualized vector field obtained by applying vector slicing to the orientation information of the fiber. Through the visualized vector field, the orientation of the fiber in each cross section inside the UHPFRC can be clearly observed;
[0078] S314, preliminarily characterize the fiber orientation law by statistically analyzing the components of 6 independent orientation tensors, wherein the components of the 6 independent orientation tensors are T XX , T YY , T ZZ , T XY , T XZ and T YZ .
[0079] After tensor decomposition of the orientation tensor, three eigenvalues and corresponding eigenvectors are obtained. The maximum eigenvalue and the corresponding eigenvector are determined to characterize the main orientation direction of the fiber. Through exploratory data analysis of the three-dimensional coordinates of the fiber, the angles with each axis and other data, the von Mises distribution mathematical model is selected to describe the orientation distribution of the fiber. The eigenvalues and eigenvectors are used to estimate the parameters in the model and construct a probability density function. The specific steps include:
[0080] S321, in Avizo, the local orientation tensor is decomposed to obtain three eigenvalues and corresponding eigenvectors, and the eigenvector corresponding to the maximum eigenvalue is compared and determined to be the main orientation direction of the fiber in the local area; each orientation tensor is a 3×3 symmetric matrix, and the eigenvalue decomposition is performed to obtain three eigenvalues and corresponding three eigenvectors, the eigenvector corresponding to the maximum eigenvalue is the main orientation of the fiber, and the remaining eigenvectors are the secondary orientations of the fiber;
[0081] S322. Use Python to draw histograms and cumulative distribution function graphs to preliminarily explore the distribution characteristics of data, analyze the distribution shape of data through histograms, and observe the changes in the cumulative probability of data through cumulative distribution function graphs to determine the uniformity or aggregation of data distribution; histograms and cumulative distribution function graphs are both graphical representations, the former is used to show the frequency distribution of data distribution, and the latter shows the cumulative probability of data being less than or equal to a certain value. Through these graphical tools, you can have an intuitive understanding of the distribution characteristics of data and provide preliminary guidance for selecting appropriate statistical models;
[0082] S323. Based on the characteristics of the data, the von Mises orientation distribution model is selected to describe the orientation distribution of the fiber. The parameters of the model are determined by fitting the eigenvalues and eigenvectors through statistical techniques such as maximum likelihood estimation. In the von Mises distribution, the concentration of the distribution can be estimated by the ratio of the eigenvalues.
[0083] S324. Construct a probability density function based on the selected model and parameters, and verify the validity of the model through cross-validation, goodness-of-fit test and other methods; in von Mises distribution, the probability density function can be expressed as where θ is the angle, I0(k) is the zero-order modified Bessel function, and k is the concentration parameter of the distribution.
[0084] The obtained fiber orientation distribution probability density function is visualized by means of histogram and density map, showing the probability distribution of different orientation angles and summarizing the distribution and orientation law of the fibers. Drawing the probability density function histogram and density map includes the following steps:
[0085] S331, determining the entire numerical range of the probability density function to be displayed, and dividing the entire numerical range into a number of continuous, equal-sized small intervals, which are called "bins" or "classes";
[0086] S332. For each interval in the histogram, calculate the number of data points falling within the interval, i.e., the frequency, divide the frequency in each interval by the total number of data points to obtain the probability of each interval, and then plot the obtained data into a histogram using Python's matplotlib library; a histogram is an approximate representation of a probability density function, which visually discretizes a continuous probability distribution. The resolution of a histogram depends on the size and number of boxes, and these parameters need to be determined based on the characteristics of the actual data and the analysis objectives;
[0087] S333. Determine the appropriate bandwidth based on cross-validation or information criterion, and use Gaussian kernel as kernel function. In kernel density estimation, bandwidth is a key parameter to control the degree of smoothness. Too small bandwidth will lead to overfitting and excessive noise, while too large bandwidth may lead to underfitting and excessive smoothing. The kernel function plays a decisive role in the shape of the density map.
[0088] S334. For each observation in the data set, calculate the weighted average of the kernel function, then sum the results of all observations to obtain the density estimate of each point, and then use Python's matplotlib library to plot the obtained data into a density map. Kernel density estimation is a non-parametric method that does not require any assumptions about the distribution form of the data, so it is very flexible and applicable to various types of data distributions. It can generate smoother and more continuous visualization effects than histograms, thereby better displaying the probability density distribution of data.
[0089] S4. Construct a damaged component model based on the data set to obtain the spatial distribution and orientation information of cracks in the damaged area. All original images are numbered to form an image stack, imported into Avizo, and a three-dimensional geometric model of the damaged component is reconstructed using a volume rendering method to extract the position and orientation information of cracks in the damaged area. Specifically, the following steps are included:
[0090] S411, after setting input parameters such as resolution and voxel size, open the original image stack, and use the 3D volume rendering method to reconstruct the destroyed component to obtain its geometric model, such as Fig. 9 Shown is a schematic diagram of a three-dimensional geometric model of a destroyed component obtained by volume rendering;
[0091] S412. Extracting the spatial position information of the cracks in the damaged area and the corresponding crack orientation information on the component model, and performing statistical analysis to obtain the spatial distribution law and orientation law of the cracks.
[0092] S5. The failure mechanism of the damaged component is obtained by comparative analysis. The failure mechanism of the component is analyzed by combining the distribution and orientation law of the fiber and the location and orientation information of the cracks in the damaged area.
[0093] The spatial distribution and orientation law of cracks are combined with the distribution and orientation law of fibers, and compared to analyze the failure mechanism of components; first find the spatial position of the crack, and explain the cause of the crack here by comparing the overall distribution of fibers in this three-dimensional space and the distribution along each axis and the stress conditions; then combine the orientation law of the crack here with the orientation law of fibers to explain the mechanism of crack development and propagation here; after exploring the specific situation of fiber distribution and orientation, clarifying the relationship between fiber distribution and orientation characteristics and mechanical properties and failure mechanism, the fiber distribution and orientation under ideal conditions can be analyzed according to the stress conditions of UHPFRC components and the mechanical properties to be achieved, and then the ideal fiber distribution and orientation can be achieved by changing the casting method and template style.
[0094] like Fig.10The figure shows the whole process of fiber distribution and orientation analysis in Avizo, which includes three-dimensional reconstruction of fibers and damaged components, statistical extraction of fiber distribution area in each section, fiber centerline extraction and spatial surface analysis of the centerline.
[0095] The 3D reconstruction of fibers and damaged components is used to realize the transition from 2D data to 3D data and obtain a visual geometric model of fibers and components. Specifically, this example is a 3D direct volume rendering method, which generates a 3D model by integrating and sampling the image stack in 3D space and using GPU rendering to reveal the internal structure and 3D shape of the target while retaining all the information of the data.
[0096] Finally, it should be noted that, in this article, relationships such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms include, include or any other variations are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0097] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0098] The embodiments of the present application have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.
Claims
1. A computer vision-based method for analyzing the fiber reconstruction and destruction mechanism of UHPFRC. It is characterized in that The following steps are involved: Collecting images of the destroyed structure to form a data set; Constructing a fiber segmentation model to segment the data set, obtaining a pure fiber image to construct a fiber geometry model; Based on the fiber geometric model, a fiber orientation distribution probability density function is obtained and visualized, and the distribution and orientation law of the fibers are analyzed; Building a damaged component model based on the data set to obtain spatial distribution and orientation information of cracks in the damaged area; The destruction mechanism of the destruction component is obtained through comparative analysis.
2. The method for analyzing the UHPFRC fiber reconstruction and destruction mechanism based on computer vision as claimed in claim 1, It is characterized in that The obtaining of the fiber orientation distribution probability density function based on the geometric model comprises: Extracting the area of the fiber on each slice section of the fiber geometric model along the X, Y, and Z axes and drawing it into a line graph; The center line of each fiber is extracted, and the three-dimensional coordinates, length, angle with each axis and orientation tensor of the fiber sampling point are obtained by spatial surface analysis; Performing tensor decomposition on the orientation tensor, and determining the parameters of the model by maximum likelihood estimation fitting using eigenvalues and eigenvectors; The fiber orientation distribution probability density function is constructed based on the parameters and the distribution mathematical model.
3. The computer vision-based UHPFRC fiber reconstruction and destruction mechanism analysis method as claimed in claim 2, It is characterized in that The extracting of the center line of each fiber comprises: Preset a cylindrical template with the same specifications as the fiber; Performing cylinder correlation calculation based on the cylindrical template and XFiber method; The center line of each of the fibers is extracted by correlation line tracing.
4. The computer vision-based UHPFRC fiber reconstruction and destruction mechanism analysis method as claimed in claim 1, It is characterized in that The fiber orientation distribution probability density function is visualized through a histogram and a density plot to show the probability distribution of different orientation angles of the fiber.
5. The computer vision-based UHPFRC fiber reconstruction and destruction mechanism analysis method as claimed in claim 4, It is characterized in that The steps of drawing the histogram and the density map include: Determine the numerical range and divide it into several intervals; Calculate the number and frequency of data points in each of the intervals and plot a histogram; and, determining the bandwidth based on cross-validation or information criterion, with a Gaussian kernel as the kernel function; Use kernel density estimation to estimate the density of each data point and draw a density plot.
6. The computer vision-based UHPFRC fiber reconstruction and destruction mechanism analysis method as claimed in claim 1, It is characterized in that The fiber segmentation model is constructed as follows: Use deep learning framework to build a network model of deep residual network infrastructure; Input the data set and run the code to train and verify the network; Evaluate the performance of the model and further optimize and adjust it to find the best hyperparameters of the model and apply it; The fiber segmentation model with the best segmentation ability is obtained.
7. The method for analyzing the UHPFRC fiber reconstruction and destruction mechanism based on computer vision as claimed in claim 1, It is characterized in that The fiber geometry model and the damaged component model are reconstructed based on image stacks and three-dimensional volume rendering.
8. The method for analyzing the UHPFRC fiber reconstruction and destruction mechanism based on computer vision as claimed in claim 1, It is characterized in that The destruction mechanism of the destruction component obtained by the comparative analysis includes: Based on the spatial position of the crack, and in comparison with the distribution and orientation of the fibers and the stress conditions, explain the cause of the crack; And, combining the orientation law of the crack and the orientation law of the fiber, explaining the mechanism of the development and propagation of the crack; Furthermore, according to the distribution and orientation rules of the fibers, the mechanical properties are analyzed in combination with the stress conditions of the components, and then the ideal distribution and orientation of the fibers are inferred in reverse.
Citation Information
Patent Citations
Method for quickly identifying chopped fiber in GFRC image based on deep learning
CN112489019A
Concrete evaluation method based on fiber bridging analysis and matrix softening model
CN115711803A
Method for calculating orientation distribution of steel fibers in fiber reinforced concrete
CN115795910A
Explosion pit-forming calculation method for steel fiber reinforced concrete protection structure
CN118194647A
Cited By
Concrete crack design method, device, equipment, medium and product
CN122310641A
