Modeling analysis method, device and system for water-rich sand layer permeation grouting model test

Abstract

The application provides a modeling analysis method, device and system for a water-rich sand layer permeation grouting model test, belongs to the technical field of grouting model tests, and comprises the following steps: constructing a water-rich sand layer permeation grouting visual simulation device; acquiring three-dimensional structure information inside the simulation device in a water-rich sand layer permeation grouting test process, and generating time sequence voxel data; extracting features in the voxel data by using a convolutional neural network; reconstructing a three-dimensional model for each time sequence of voxel data according to the extracted features; quantitatively analyzing a slurry diffusion process based on the three-dimensional model of each time sequence, and calculating test parameters.

Description

Technical Field

[0001] This invention belongs to the field of grouting model test technology, and particularly relates to the modeling and analysis method, device and system for permeable grouting model test of water-rich sand layer. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In the field of engineering construction, water-rich sand layers pose significant challenges to construction due to their complex characteristics. Water-rich sand layers typically have poor cementation, low strength, and high permeability. The characteristics of water-rich sand layers vary greatly in different regions, making them prone to geological disasters such as water inrush, sand surge, quicksand, surrounding rock instability, tunnel collapse, surface subsidence, and groundwater level drop during tunnel engineering, foundation reinforcement, and other construction processes, seriously threatening the safety of the project.

[0004] Current grouting technology theories have limitations in their application to water-rich sand layers. The diffusion pattern of grout in water-rich sand layers is influenced by multiple factors and cannot be clearly defined at present, making it difficult to accurately predict the reinforcement effect. This hinders the direct application of existing grouting theories to practical engineering projects. Existing grouting simulation test devices also have limitations, offering poor visual observation and often exhibiting "black box problems," making it difficult to directly observe the grout diffusion process and the movement of wetting peaks, thus hindering the understanding of the physical phenomena during the grouting process.

[0005] Therefore, engineering practice urgently needs a more efficient and precise research method to study the grouting process in water-rich sand layers. With the continuous development of civil engineering construction, the requirements for grouting technology in water-rich sand layers are becoming increasingly stringent. Accurately grasping the diffusion of grout in water-rich sand layers is crucial for determining reasonable grouting parameters and ensuring the construction safety and quality of tunnels and other projects. Summary of the Invention

[0006] To overcome the shortcomings of the existing technology, this invention provides a modeling and analysis method for permeable grouting model tests in water-rich sand layers. A three-dimensional model is established for quantitative analysis, and the entire grouting process in the test device is reproduced, providing a powerful tool for the research of grouting technology in water-rich sand layers.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] Firstly, the modeling and analysis method for permeable grouting model tests in water-rich sand layers is disclosed, including:

[0009] Construct a visual simulation device for infiltration grouting in water-rich sand layers;

[0010] To obtain the three-dimensional structural information of the simulation device inside the water-rich sand layer permeation grouting test and generate time-series voxel data;

[0011] Features are extracted from voxel data using convolutional neural networks, and a 3D model is reconstructed from the voxel data of each time series based on the extracted features.

[0012] Based on the three-dimensional model of each time series, the slurry diffusion process is quantitatively analyzed and experimental parameters are calculated.

[0013] As a further technical solution, the three-dimensional structural information of the simulation device inside the grouting test of the water-rich sand layer is obtained, including:

[0014] An initial scan is performed before the sand layer is filled and before the grout is injected.

[0015] Select an appropriate scanning mode, set the scanning parameters to be consistent with the parameters determined during the debugging phase, and record the initial state of the water-rich sand layer;

[0016] Grouting begins, and continuous scanning is performed simultaneously;

[0017] Set a scanning time interval, and scan the sand layer and grouting conditions in the test chamber once at a fixed time interval.

[0018] As a further technical solution, the process of generating time-series voxel data is as follows:

[0019] Denoising the scanned 2D image data;

[0020] Perform image enhancement operations on the denoised 2D image data;

[0021] Based on the differences in X-ray absorption by the water-rich sand layer, slurry, and test chamber, a threshold segmentation method was used to segment the enhanced image.

[0022] Based on the size of the test chamber and the scanning resolution, the three-dimensional space of the scanning area is divided into small cubic voxels. For each voxel, corresponding attribute values ​​are assigned according to its position in the image and the image segmentation result.

[0023] All voxel data are stored in the form of a three-dimensional array, where each element represents the attribute value of the corresponding voxel.

[0024] As a further technical solution, a convolutional neural network is used to extract features from voxel data, wherein the convolutional neural network includes:

[0025] The first convolutional layer, the second convolutional layer, the third convolutional layer, the max pooling layer, and the fully connected layer;

[0026] The first convolutional layer is used to perform preliminary feature extraction on the voxel data;

[0027] The second convolutional layer is used to extract more complex features from the voxel data;

[0028] The third convolutional layer is able to capture more subtle and higher-level features in the voxel data;

[0029] A max-pooling layer is connected after each convolutional layer. The max-pooling layer is used to downsample the feature map output by the connected convolutional layer.

[0030] A fully connected layer is connected after the max pooling layer. The fully connected layer is used to integrate the extracted features for the final classification or regression task.

[0031] As a further technical solution, a 3D model is reconstructed from the voxel data of each time series based on the extracted features. The specific process is as follows:

[0032] The three-dimensional space is divided into small cubes, or voxels. For each small cube, the intersection with the isosurface is determined based on the attribute values ​​of its vertices.

[0033] Then, based on a predefined lookup table, the topological structure of the isosurfaces on the small cube is determined;

[0034] Finally, by interpolating and calculating the coordinates of points on the isosurface, the isosurfaces on all the small cubes are stitched together to obtain the three-dimensional model.

[0035] This model can visually demonstrate the diffusion process of slurry in water-rich sand layers, enabling subsequent visualization and quantitative analysis.

[0036] As a further technical solution, the test parameters are calculated, including the diffusion volume, coverage area, and diffusion rate of the slurry.

[0037] Secondly, a modeling and analysis system for permeable grouting tests in water-rich sand layers is disclosed, including:

[0038] The time-series voxel data generation module is configured to: acquire the three-dimensional structural information inside the simulation device during the water-rich sand layer permeation grouting test and generate time-series voxel data;

[0039] The 3D model reconstruction module is configured to: extract features from voxel data using a convolutional neural network, and reconstruct a 3D model for each time-series voxel data based on the extracted features;

[0040] The quantitative analysis module is configured to perform quantitative analysis of the slurry diffusion process based on the three-dimensional model of each time series and calculate the experimental parameters.

[0041] Thirdly, a modeling and analysis device for permeable grouting model test of water-rich sand layer is disclosed, including: a grouting simulation test device for water-rich sand layer and a computer;

[0042] The grouting simulation test device for water-rich sand layers includes: a grouting system, a water pressure control system, a visualization simulation system for water-rich sand layers, and a data acquisition system;

[0043] The grouting system is directly connected to the water-rich sand layer simulation grouting device to meet the needs of different grouting tests.

[0044] The water pressure control system is directly connected to the water-rich sand layer simulated grouting device to provide it with pore water pressure.

[0045] The data acquisition system is used to collect structural information of the water-rich sand layer during the grouting process;

[0046] The data acquisition system communicates with a computer, which is configured to:

[0047] The system receives three-dimensional structural information of the water-rich sand layer inside the grouting process from the data acquisition system and generates time-series voxel data.

[0048] Features are extracted from voxel data using convolutional neural networks, and a 3D model is reconstructed from the voxel data of each time series based on the extracted features.

[0049] Based on the three-dimensional model of each time series, the slurry diffusion process is quantitatively analyzed and experimental parameters are calculated.

[0050] The aforementioned device can accurately simulate actual working conditions, precisely control experimental parameters, and ensure the reliability and accuracy of experimental research.

[0051] The above one or more technical solutions have the following beneficial effects:

[0052] The technical solution of this invention uses an X-ray scanner to scan and record the grouting simulation test device for water-rich sand layers. This allows for the acquisition of three-dimensional structural information within the water-rich sand layer during the grouting process, ensuring comprehensive monitoring of the sand layer structure, grout distribution, and their interaction. Compared to traditional local observation or indirect measurement methods, this approach yields more complete and accurate data, reflecting the dynamic changes in the micro and macroscopic structures within the sand layer during the grouting process.

[0053] The convolutional neural network (CNN) architecture constructed by the technical solution of this invention can automatically extract features from voxel data. Compared with traditional manual feature extraction methods, CNN has stronger adaptability and accuracy. It can identify complex slurry infiltration paths, accurately determine the boundary range of slurry diffusion, and overcome the subjectivity and errors that may exist in manual analysis. Through learning from a large amount of voxel data, CNN can discover hidden patterns and regularities in the data, providing a more reliable basis for 3D model reconstruction, thereby generating a 3D model that is more consistent with the actual situation and more accurately reflects the diffusion state of slurry in sand layers.

[0054] This invention combines X-ray scanning technology with convolutional neural networks for experimental research on permeable grouting models in water-rich sand layers, representing an innovative research method. This interdisciplinary integration breaks through the limitations of traditional grouting research methods, providing new ideas and tools for studying grouting problems in water-rich sand layers. By leveraging the complementary advantages of the two technologies, comprehensive innovation is achieved from experimental data acquisition and processing to analysis, which can promote research progress in the field of grouting technology for water-rich sand layers.

[0055] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0056] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0057] Figure 1 This is a schematic diagram of the overall process of the method in an embodiment of the present invention.

[0058] Figure 2 A schematic diagram of the structure of the modeling and analysis device for the permeation grouting model test of water-rich sand layers.

[0059] Figure 3 This is a schematic diagram of a convolutional neural network structure. Detailed Implementation

[0060] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] It should be noted that the terminology used herein is for the purpose of describing particular implementations only and is not intended to limit the exemplary implementations of the present invention.

[0062] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0063] Example 1

[0064] This embodiment discloses a modeling and analysis method for permeable grouting model tests in water-rich sand layers, based on X-ray scanning and convolutional neural networks, including:

[0065] Step 1: Construct a visual simulation device for permeation grouting in water-rich sand layers. To facilitate direct observation during the grouting test, the device is made of transparent material.

[0066] For details, please see the appendix. Figure 2 As shown, the visualization simulation device for water-rich sand layer infiltration grouting constructed by the sub-technical solution in this embodiment includes a grouting system, a water pressure control system, a visualization simulation system for water-rich sand layers, and a data acquisition system.

[0067] In this embodiment, the grouting system consists of a grout storage tank, a grout delivery pipe, a grouting pump, and an grouting pipe. The grouting pump inlet is connected to the grout storage tank, and one end of the grouting pipe is connected to the grouting pump outlet. The entire grouting system is directly connected to the visualized water-rich sand layer simulation system via the grout delivery pipe, meeting the needs of different grouting tests, such as two-component and single-component grouting tests. The grout storage tank should be equipped with a well-sealed lid to prevent grout evaporation and impurity entry. The grout storage tank should be placed on a stable support, ensuring its height facilitates connection to the grouting pump and that vibration or shaking will not affect grouting stability during the test. A plunger-type grouting pump is selected, with a flow rate adjustment range of 0.1L / min-10L / min and a pressure range of 0MPa-5MPa. The grouting pump is connected to the grout storage tank outlet via a corrosion-resistant high-pressure rubber hose. Valves are installed on the grout delivery pipe to control the grouting flow rate and to start and stop the grouting pump. The grout delivery pipe is used to transport the grout from the grout storage tank to the grouting pump. The grouting pipe should be a high-strength rubber grouting pipe with an inner diameter of 15mm. The length should be determined according to the size of the test chamber, generally 1m-3m. Connect one end of the grouting pipe tightly to the outlet of the grouting pump, and insert the other end into the grouting test chamber of the visual water-rich sand layer simulation system through the grouting hole reserved in the test chamber. The insertion depth can be determined according to the test design, such as inserting it to 1 / 2 or 2 / 3 of the sand layer depth.

[0068] In this embodiment, the water pressure control system consists of a high-head water storage tank, a water inlet pipe, and flow control valves. It is connected to the test chamber in the visualized water-rich sand layer simulation system, providing it with pore water pressure. The high-head water storage tank should be made of plastic or stainless steel. An inlet and a water level monitoring device, such as a float-type water level gauge, are installed on the top of the tank. A drain outlet and an interface for connecting to the water inlet pipe are installed at the bottom. The storage tank should be placed on a support 1-2 meters above the test chamber, and the pore water pressure is controlled by adjusting the height of the support. Flow control valves are installed at both ends of the water inlet pipe connecting the high-head water storage tank and the test chamber.

[0069] In this embodiment, the visualized water-rich sand layer simulation system consists of a cube transparent permeation grouting test box and an internally filled sand layer. This system is connected to the grouting system and the water pressure control system. The box body of the grouting test box is made of transparent plexiglass. A grouting hole is reserved at the bottom of one side for inserting a grouting pipe; a water pressure inlet is reserved at the bottom of the other side for connecting a water inlet pipe; an exhaust hole is reserved at the top to discharge the air in the sand layer. A reinforcing ring is used to reinforce around each hole to prevent the hole edge from cracking due to pressure or vibration during the test. High-quality transparent glass sand is selected for the internal sand layer, and the transparency should be as high as possible for easy observation during the test. The particle size of the glass sand can be selected according to the test requirements. The layered filling method is adopted, and the filling thickness of each layer is 5 cm - 10 cm. A small vibrator is used to vibrate each layer of the sand layer to make the sand layer dense and uniform. During the filling process, the density of the sand layer can be checked by inserting thin rods or probes into the sand layer to ensure that the porosity of the sand layer meets the test requirements.

[0070] In this embodiment, the data acquisition system mainly consists of an X-ray scanner. The scanned images are transmitted to the computer supporting the X-ray scanner. What is directly obtained by the X-ray scanner is a two-dimensional image. Three-dimensional structure information is generated in the computer based on the two-dimensional image, which can reflect the internal structure information of the water-rich sand layer during the grouting process. The X-ray scanner should cover the entire water-rich sand layer area and be able to scan the entire test box comprehensively.

[0071] Step 2: Use the X-ray scanner to scan the grouting test process of the visualized simulation device, obtain the three-dimensional structure information inside the model during the test process, reveal and record information such as the penetration change process, diffusion range, path of the grout, and the interaction with the sand layer particles, and generate time-series voxel data.

[0072] Use the X-ray scanner to scan the grouting test process of the visualized simulation device, obtain the three-dimensional structure information inside the model during the test process. First, the X-ray scanner should be debugged, and appropriate parameters such as resolution, scanning range, and exposure time should be selected to make the scanned image as close as possible to the actual size and density distribution of the standard object. The specific steps are as follows:

[0073] Select the resolution: According to the particle size of the sand layer particles and the requirements for observing the details of the grout penetration in the test, select an appropriate scanning resolution. If studying the relationship between the microscopic pore structure of the sand layer and the grout penetration, select a resolution of 0.2 mm; if mainly focusing on the macroscopic grout diffusion range and the overall change of the sand layer, a resolution of 0.5 mm can be selected.

[0074] Determine the scanning range: Measure the actual dimensions of the test chamber and set the scanning range to be slightly larger than the test chamber dimensions, exceeding it by 5cm-10cm in each direction, to ensure complete acquisition of information about the sand layer and grouting area. Input the scanning range parameters into the scanning software and check whether the scanning range covers the entire test area through the pre-scan and image preview functions.

[0075] Adjusting the exposure time: Conduct pre-scanning tests, using different exposure times (e.g., 1s, 2s, 3s, etc.) to scan the ungrouted sand layer inside the test chamber, acquiring multiple sets of images. Analyze the image contrast and sharpness, selecting an exposure time that ensures the sand layer and test chamber structure are clearly distinguishable and has rich grayscale levels. Simultaneously, use a radiation dosimeter to measure the radiation dose at different exposure times. While meeting image quality requirements, choose the lowest possible exposure time to minimize the impact of radiation on test personnel and equipment.

[0076] The X-ray scanner scan of the test apparatus includes: Initial scanning with the X-ray scanner performed after sand filling is complete but before grout injection. A suitable scanning mode (e.g., continuous scanning or layer-by-layer scanning) is selected in the scanning control software, and the scanning parameters (resolution, scanning range, exposure time, etc.) are set to match those determined during the commissioning phase. The initial state of the water-rich sand layer is recorded. Grouting is initiated by starting the grouting pump, and simultaneously, the X-ray scanner is started for continuous scanning. The scanned data is stored as two-dimensional image data. A fixed scanning time interval is set, and the sand layer and grouting status within the test chamber are scanned once at fixed intervals. The time interval can be determined based on the level of detail required for the experimental analysis.

[0077] X-ray scan data is converted into voxel data in a computer, and the two-dimensional image data obtained from the X-ray scan is denoised and enhanced to improve image quality. Based on the differences in X-ray absorption by the water-rich sand layer, slurry, and test chamber, an appropriate method is selected for image segmentation.

[0078] The three-dimensional space of the scanned area is divided into small cubic voxels. Based on the image segmentation results, each voxel is assigned a corresponding attribute value. Specifically, the image segmentation result is based on a grayscale threshold to determine whether this part of the image is sand, slurry, or something else. The main steps include:

[0079] Image Denoising: The median filtering algorithm is used to denoise the scanned 2D image data. A 3×3 pixel filtering window is selected. For each pixel in the image, the pixel values ​​in its surrounding 3×3 neighborhood are sorted, and the median value is taken as the new value of the pixel. The median filtering formula is: g(x,y)=med{f(xi,yj)},(i,j∈[-1,1]), where g(x,y) is the denoised image pixel value, f(x,y) is the original image pixel value, and med represents the median operation.

[0080] Image Enhancement: Image enhancement operations are performed using histogram equalization. The gray-level histogram of the image is calculated, and the frequency of each gray level is statistically analyzed. The cumulative distribution function (CDF) is calculated based on the gray-level histogram, and then normalized to obtain a mapping function. This mapping function transforms the gray-level values ​​of the original image, making the gray-level distribution more uniform and improving image contrast. The formula for calculating the enhanced image pixel value is: g(x,y) = T(f(x,y)), where T is the mapping function, f(x,y) is the original image pixel value, and g(x,y) is the enhanced image pixel value.

[0081] Image segmentation: A threshold segmentation method is used based on the differences in X-ray absorption by the water-rich sand layer, slurry, and test chamber. An appropriate threshold is selected by analyzing the image's gray-level histogram. For example, observation reveals that the sand layer has a lower gray-level value, the slurry has a higher gray-level value, and the test chamber's gray-level value falls between the two. A gray-level threshold T is selected, and pixels with gray-level values ​​less than T are identified as sand layers, while pixels with gray-level values ​​greater than T are identified as slurry or test chamber components. In the segmented image, the pixel values ​​for the sand layer are set to 0, and the pixel values ​​for the slurry and test chamber components are set to 1.

[0082] The threshold segmentation formula is: Where g(x,y) is the pixel value of the segmented image, f(x,y) is the pixel value of the original image, and T is the threshold.

[0083] Voxel data generation: Based on the test chamber size and scanning resolution, the three-dimensional space of the scanned area is divided into small cubic voxels. Whether each small three-dimensional space is defined as a sand layer, slurry, or the test chamber body is determined by the image segmentation results.

[0084] If the test chamber has a side length of 50cm and a scanning resolution of 0.5mm, then it can be divided into 1000 voxels in each direction. For each voxel, a corresponding attribute value is assigned based on its position in the image and the image segmentation result. The corresponding attribute value is the key data describing its characteristics and can indicate the material category of the voxel. All voxel data is stored in the computer's memory in the form of a three-dimensional array, with the array dimensions being (number of voxels x, number of voxels y, number of voxels z). The value of each element is the attribute value of the corresponding voxel, such as its grayscale value and material category.

[0085] Step 3: Construct a convolutional neural network (CNN) architecture suitable for voxel data, extract features from the voxel data, identify the infiltration path of slurry in the sand layer, determine the boundary range of slurry diffusion, and finally reconstruct a three-dimensional model for each time series of voxel data based on the extracted features.

[0086] Building a convolutional neural network (CNN) architecture suitable for voxel data involves designing network layers, selecting activation functions, determining loss functions, and optimizing algorithms, as detailed below:

[0087] The first step is to design the network layer, see appendix. Figure 3 As shown, the main steps for designing the convolutional layers, pooling layers, and fully connected layers of a convolutional neural network are as follows:

[0088] (1) Design convolutional layers

[0089] The design consists of three concatenated convolutional layers. The first convolutional layer uses 32 kernels of size 3×3×3 with a stride of 1 and 'same' padding (i.e., maintaining the same image size after convolution). Its function is to perform preliminary feature extraction from voxel data in three-dimensional space. The convolution operation formula is: Where V o (i,j,k) represents the output voxel value, W(m,n,p) represents the convolution kernel weights, and V i (i,j,k) represents the input voxel values, b is the bias term, and i,j,k are the coordinates of the voxels in 3D space. m, n, and p are used to locate the positions of elements within the convolution kernel. m can represent the index of the element in the x-dimensional region of the convolution kernel, and its value range depends on the size of that dimension of the convolution kernel; n is the y-dimensional index, and p is the z-dimensional index, which together determine the coordinates of each element within the convolution kernel.

[0090] The second convolutional layer uses 64 3×3×3 convolutional kernels with a stride of 1 and 'same' padding. This layer further extracts more complex features, enhancing the network's ability to understand data features.

[0091] The third convolutional layer uses 128 3×3×3 convolutional kernels with a stride of 1 and 'same' padding. By increasing the number of convolutional kernels, the network can capture more subtle and higher-level features from the voxel data, providing richer information for subsequent classification and regression tasks. Each convolutional layer is followed by a ReLU activation function to introduce non-linearity and enhance the network's expressive power. The ReLU activation function formula is: f(x) = max(0, x), where x is the input value and f(x) is the output value.

[0092] (2) Configure the pooling layer

[0093] A max-pooling layer is connected after each convolutional layer, with a pooling window size of 2×2×2 and a stride of 2. The purpose of the max-pooling layer is to downsample the feature map output by the convolutional layer, reducing the amount of data while retaining the main feature information. The formula for max-pooling is: V p (i,j,k)=max{V c(2i+m,2j+n,2k+p)},(m,n,p∈[0,1]), where V p (i,j,k) represents the output voxel value after pooling, V c (i,j,k) represents the output voxel value of the convolutional layer, where i,j,k are the coordinates of the voxel in three-dimensional space.

[0094] (3) Add a fully connected layer

[0095] Two cascaded fully connected layers are connected after the pooling layer. The first fully connected layer has 256 neurons, and the second has 128 neurons. The fully connected layers integrate the features extracted by the preceding network layers for the final classification or regression task. The calculation method for the fully connected layer is: y = Wx + b, where y is the output vector, W is the weight matrix, x is the input vector, and b is the bias vector.

[0096] The second step is to select activation functions. ReLU activation is chosen for both convolutional and fully connected layers to introduce non-linearity and enhance the network's expressive power. The ReLU function outputs the same value as the input when the input is greater than 0, and outputs 0 when the input is less than 0. ReLU activation is computationally simple, converges quickly, and effectively alleviates the vanishing gradient problem, enabling the network to learn features from the data more quickly. It also performs better in deeper networks, improving training efficiency and overall performance.

[0097] The third step is to determine the loss function and optimization algorithm. Based on the task objectives (such as the accuracy of identifying slurry seepage paths and the error in predicting the diffusion boundary range), a suitable loss function is selected. Examples include cross-entropy loss and mean squared error loss. Simultaneously, a suitable optimization algorithm, such as stochastic gradient descent (SGD), Adagrad, Adadelta, or Adam, is selected to optimize and train the network parameters to minimize the loss function and improve network performance. Details are as follows:

[0098] (1) Choosing a loss function:

[0099] For tasks involving identifying slurry seepage paths (a binary classification problem, such as determining whether a voxel belongs to a seepage path), the binary cross-entropy loss function is chosen. The formula is: Where L is the loss value and y is the true label (0 or 1). This is used to predict probabilities. The loss function measures the difference between the predicted result and the true label. Minimizing the loss function optimizes network parameters and improves prediction accuracy.

[0100] For the task of determining the slurry diffusion boundary range (a regression problem), the mean squared error loss function is chosen. The formula is: Where L is the loss value, y iFor the true value, Here, represents the predicted value, and represents the sample size. The mean squared error loss function measures the average error between the predicted and true values, enabling the network to learn a more accurate prediction model.

[0101] (2) Selecting an optimization algorithm

[0102] The Adam optimization algorithm was chosen to optimize the network parameters during training. The Adam algorithm combines the advantages of Adagrad and RMSProp algorithms, adaptively adjusting the learning rate to apply different learning rates to different parameters, thus converging to a better solution more quickly during training. The update formula for the Adam algorithm is:

[0103] m t =β1m t-1 +(1-β1)g t

[0104] v t =β2v t-1 +(1-β2)g t 2

[0105]

[0106] Where m t and v t These are the first-order moment estimate and the second-order moment estimate, respectively. β1 and β2 are the decay rates (typically taken as β1 = 0.9 and β2 = 0.999), g t For gradient, and For the corrected first-order moment estimate and second-order moment estimate, θ t Here are the parameters: α is the learning rate (which can be initially set to 0.001), l is a small constant to prevent division by zero (usually set to a value of ), and is the number of iterations.

[0107] Features are extracted from voxel data to identify the seepage path of slurry in sand layers and determine the boundary range of slurry diffusion. Specifically, based on the features of the voxel data, the material type in each small three-dimensional space can be identified (some are slurry, some are sand layers), thus determining the diffusion boundary. The seepage path can be determined based on the changes in voxel data over time.

[0108] Finally, based on the extracted features, a 3D model is reconstructed from the voxel data of each time series. Specific steps include:

[0109] (1) Data Preprocessing: The generated voxel data is normalized to ensure the data falls within a specific range (e.g., 0-1) to improve network training efficiency and stability. A simple linear normalization method is used, with the following formula: Where Vn (i,j,k) represents the normalized voxel value, and V(i,j,k) represents the original voxel value. min and V max These represent the minimum and maximum values ​​in the voxel data, respectively. Based on the data volume and actual conditions, the preprocessed voxel data is divided into a training set, a validation set, and a test set, with a ratio of 70%:15%:15%.

[0110] (2) Network Training: Input the training set into the constructed convolutional neural network and perform multiple iterations of training. During the training process, adjust the network parameters (such as learning rate, number of convolutional kernels, etc.) based on the loss value and accuracy of the validation set to prevent overfitting. The training process continues until the network's performance on the validation set stabilizes or meets the predetermined stopping conditions (such as reaching the maximum number of training epochs of 100 or the loss value on the validation set no longer decreasing).

[0111] (3) Feature extraction and model reconstruction

[0112] After training, voxel data from the test set or actual experiments are input into the trained convolutional neural network. Through the network's forward propagation process, features are extracted from the voxel data. These features can be connected regions representing slurry infiltration paths, contour features of diffusion boundaries, etc.

[0113] By utilizing the extracted features, a three-dimensional model is reconstructed from the voxel data of each time series using appropriate algorithms (such as voxel-based modeling algorithms, Marching Cubes algorithms, etc.), generating a three-dimensional model that can intuitively show the diffusion process of slurry in water-rich sand layers, so as to facilitate subsequent visualization and quantitative analysis.

[0114] Taking the Marching Cubes algorithm as an example, the basic idea of ​​this algorithm is to construct isosurfaces on the boundaries of voxels by analyzing the attribute values ​​of each voxel in the voxel data (in this case, voxels related to slurry determined after classification or regression). The specific steps are as follows:

[0115] First, the three-dimensional space is divided into small cubes (i.e., voxels). For each small cube, the intersection with the isosurface (a threshold isosurface representing the boundary of the slurry) is determined based on the attribute value of its vertex (0 or 1, indicating whether it belongs to the slurry region).

[0116] Then, based on a predefined lookup table, the topological structure of the isosurfaces on the small cube (i.e., which edges constitute the isosurfaces) is determined. The lookup table is pre-calculated based on all possible combinations of vertex attribute values ​​(256 cases in total).

[0117] Finally, by interpolating and calculating the coordinates of points on the isosurfaces, the isosurfaces on all the small cubes are pieced together to obtain the three-dimensional model. For example, for a small cube with vertex attribute values ​​of (0,0,0,0,1,1,1,1) (representing the 8 vertices of the small cube in a certain order), according to the lookup table, its isosurface is composed of three edges. The coordinates of the isosurface points on these three edges are calculated by linear interpolation, and then these isosurface points are connected to form the isosurface part on the small cube. This process is repeated for all small cubes to obtain the entire three-dimensional model. This model can intuitively show the diffusion process of slurry in water-rich sand layers, so as to facilitate subsequent visualization and quantitative analysis.

[0118] Step 4: Based on the three-dimensional model of each time series, conduct a quantitative analysis of the slurry diffusion process and calculate parameters such as the diffusion volume, coverage area, and diffusion rate of the slurry.

[0119] Based on the established time-series three-dimensional model, the above parameters can be calculated to conduct grout diffusion analysis. These data are accurate and can be used for quantitative comparative analysis of the grouting effect under different grouting parameters, providing a basis for optimizing grouting parameters.

[0120] The diffusion volume is calculated by counting the number of voxels within the slurry diffusion region and multiplying this number by the volume of each voxel. For 3D models at different time points, the slurry diffusion volume is calculated separately to analyze its variation over time. The specific calculation process is as follows:

[0121] (1) Fundamentals of Voxel Counting and Volume Calculation: First, based on the voxel dimensions determined when generating the voxel data, calculate the volume of a single voxel. Assuming the scan resolution is r (unit: cm), the volume V of a single voxel is... v =r 3 (Unit: cm) 3 ).

[0122] (2) Determining Voxels in the Slurry Diffusion Region: In the 3D model, voxels belonging to the slurry diffusion region are determined based on their attribute values. As in the previous steps, the attribute values ​​of voxels in the slurry and test chamber sections are set to [value missing]. Here, it is necessary to further distinguish the slurry section. This can be achieved by analyzing features such as voxel connectivity. For example, starting with voxels near the grouting inlet that are known to be slurry, a region growing algorithm is used to mark adjacent voxels with the same attribute value as slurry voxels, until no new voxels meet the criteria. The basic principle of the region growing algorithm is to start from one or more seed points and, based on certain similarity criteria (here, voxel attribute values ​​are [value missing] and adjacent voxels), continuously add adjacent pixels or voxels to the growing region.

[0123] (3) Calculate the diffusion volume: Count the number of voxels N of the slurry, then the diffusion volume V of the slurry = N × V v For three-dimensional models at different time points, repeat the above steps to calculate the slurry diffusion volume, thereby obtaining the variation law of slurry diffusion volume with time. For example, the slurry diffusion volume is calculated as V1 at time t1 and V2 at time t2. The change of V2-V1 can be analyzed to understand the diffusion of slurry during this period.

[0124] The coverage area is calculated for a selected calculation section (such as the sand layer surface or a specific vertical section) based on the distribution of slurry on that section. This can be achieved by identifying and extracting the boundaries and contours of the slurry elements on the section, and then using polygon area calculation methods (such as Green's formula, integral methods, etc.) to calculate the coverage area. For models with different time series, the coverage area is calculated separately, and its changing trend over time is observed to assess the lateral diffusion range of the slurry in the sand layer. The specific calculation process is as follows:

[0125] (1) Selecting the calculation section: Select a suitable calculation section according to the experimental requirements and research focus. For example, the surface section of the sand layer can reflect the lateral diffusion range of the grout at the top of the sand layer, and a specific vertical section (such as a section parallel to the grouting direction) can observe the diffusion of the grout at different depths. For the surface section of the sand layer, its plane coordinates can be determined according to the size of the test chamber and the coordinate system.

[0126] (2) Slurry Voxel Boundary Identification and Contour Extraction: On the selected cross section, the voxel data is traversed to find the voxel boundaries belonging to the slurry. Boundary voxels can be determined by judging the changes in the attribute values ​​of adjacent voxels. For example, when a voxel has the attribute value of (slurry) and its adjacent voxel has the attribute value of (sand layer), this voxel may be a boundary voxel. For these boundary voxels, a contour extraction algorithm (such as the boundary tracking algorithm) is used to obtain the contour of the slurry region. The basic idea of ​​the boundary tracking algorithm is to start from a boundary point, sequentially search for adjacent boundary points along the boundary until returning to the starting point, thus forming a closed contour.

[0127] (3) Calculate the coverage area: Based on the extracted contour, calculate the coverage area using polygon area calculation methods. For simple polygons (such as contours that are typically simple polygons), Green's formula can be used to calculate the area. Assume the vertex coordinates of the polygon are (x... i ,y i (i = 1, 2, ..., n, where n is the number of vertices), then the area (where x) n+1 =x1,y n+1=y1). For models with different time series, repeat the above process, calculate the coverage area at each time point, and observe its changing trend over time to evaluate the lateral diffusion range of the slurry in the sand layer.

[0128] Diffusion rate calculation is performed based on the changes in grout diffusion volume or coverage area at different time points, combined with the time interval between two adjacent scans. Diffusion rate can be calculated from both volume and area changes, reflecting the dynamic characteristics of grout diffusion in water-rich sand layers from different perspectives. Analysis of diffusion rates at different stages helps understand the rate of grout diffusion, providing a basis for grouting engineering design and optimization. Specifically, this includes:

[0129] (1) Calculation of diffusion rate based on volume change: Select two adjacent time series t n-1 and t n Calculate the change in slurry diffusion volume ΔV = V between these two moments. n -V n-1 V n and V n-1 They are t n-1 and t n The volume of slurry diffusion at time t. Given the time interval Δt between two adjacent scans, calculate the diffusion rate based on the volume change.

[0130] (2) Calculation of diffusion rate based on area change: Similarly, two adjacent time series t are selected. n-1 and t n Calculate the change in slurry coverage area ΔA = A between these two moments. n -A n-1 A n and A n-1 They are t n and t n-1 The area covered by the slurry at any given time. Then the diffusion rate is based on the change in area.

[0131] Analysis of diffusion rate changes:

[0132] By calculating the diffusion rate at different stages (whether based on volume or area), we can understand the changing pace of grout diffusion. For example, if the diffusion rate is fast in the initial stage of grouting and gradually slows down over time, it may mean that the grout diffusion in the sand layer is encountering increasing resistance, such as the gradual filling of sand layer pores. This information can provide a basis for the design and optimization of grouting projects, such as adjusting grouting pressure, grout volume, or the proportion of grouting materials to achieve better grouting results. Furthermore, comparing the diffusion rates calculated based on volume and area allows for a comprehensive understanding of the grout diffusion characteristics from different perspectives. For instance, when the volume diffusion rate does not change significantly but the area diffusion rate decreases markedly, it may indicate that the grout is diffusing deeper into the sand layer while lateral diffusion is relatively slower.

[0133] This embodiment employs a transparent experimental device constructed from plexiglass and transparent glass sand to simulate a water-rich sand layer environment and conduct grouting tests. An X-ray scanner is used to capture subtle changes in the grout and generate time-series voxel data. The voxel data is then processed using a convolutional neural network to identify the grout penetration path, determine the diffusion boundary range, and establish a three-dimensional model for quantitative analysis. This reconstructs the entire grouting process within the experimental device, providing a powerful tool for the research of grouting technology in water-rich sand layers.

[0134] In a more specific implementation example, the method for modeling and analyzing the permeability grouting model test of water-rich sand layers is as follows:

[0135] (i) In a water-rich sand layer reinforcement project of an underground engineering project, the method of the present invention was used to conduct permeation grouting model test modeling and analysis.

[0136] First, a visual simulation device for permeation grouting in water-rich sand layers was constructed. A cubic transparent permeation grouting test chamber was made of high-strength transparent acrylic material, and filled with a sand layer simulating the characteristics of water-rich sand layers in actual engineering projects. The grouting system adopted a two-liquid grouting mode, and a grout storage tank, a high-performance grouting pump, and grouting pipes were prepared, ensuring a tight connection between the grouting system and the test chamber. The water pressure control system consisted of a large-capacity high-head water storage tank and a water inlet pipe. After being adjusted according to the pore water pressure requirements of the actual project, it was connected to the test chamber to provide stable pore water pressure for the simulated grouting device in water-rich sand layers. The data acquisition system used an advanced X-ray scanner, and data storage equipment and processing software were prepared.

[0137] After the grouting test began, the X-ray scanner was first calibrated. Based on the test chamber size and sand layer characteristics, a suitable resolution of 0.5 mm / pixel, a scanning range covering the entire test chamber, and an exposure time of 2 seconds were selected to ensure the scanned image closely matched the actual size and density distribution of the standard object. When grout was injected into the test chamber, the X-ray scanner was used for continuous scanning, and the scanned data was stored in the computer as two-dimensional image data. Subsequently, the two-dimensional image data underwent denoising and enhancement processing, employing median filtering and histogram equalization enhancement methods to improve image quality. Based on the differences in X-ray absorption by the water-rich sand layer, grout, and test chamber, a threshold segmentation method was used for image segmentation. The three-dimensional space of the scanned area was divided into small cubic voxels, each with a side length of 0.5 mm. Each voxel was assigned corresponding attribute values ​​based on the image segmentation results, such as sand layer, grout, or test chamber wall.

[0138] Next, a convolutional neural network architecture suitable for voxel data was constructed. In terms of network layer design, multiple convolutional layers were designed, each with a kernel size of 3×3, max pooling layers were configured, and fully connected layers were added at the end of the network. The ReLU activation function was chosen to introduce nonlinearity into the convolutional and fully connected layers. The mean squared error loss function was chosen, and the Adam algorithm was used to optimize the network parameters during training. In the feature extraction and 3D model reconstruction stages, the generated voxel data was linearly normalized to a range between 0 and 1. Based on the data volume, the voxel data was divided into a 70% training set, a 15% validation set, and a 15% test set. The training set was input into the constructed convolutional neural network for 500 iterations of training. After training, the test set was input into the trained convolutional neural network to extract features from the voxel data. A voxel-based modeling algorithm was then used to reconstruct a 3D model for each time series of voxel data, generating a 3D model that visually demonstrates the diffusion process of slurry in a water-rich sand layer.

[0139] Finally, quantitative analysis was performed based on the 3D model for each time series. The number of voxels within the grout diffusion area was counted, and multiplied by the volume of each voxel to obtain the grout diffusion volume. For the calculated cross-section of the sand layer surface, the boundaries and contours of the grout elements on the cross-section were identified and extracted, and the coverage area was calculated using Green's formula. Based on the changes in grout diffusion volume and coverage area under different time series, and combined with the time interval between two adjacent scans, the grout diffusion rate was calculated from both volume and area change perspectives, providing an accurate basis for grouting engineering design and optimization.

[0140] (ii) In a foundation reinforcement project of a certain water conservancy hub, the method of the present invention was used to conduct a permeable grouting model test of a water-rich sand layer.

[0141] When constructing the visualization simulation device, a cube-shaped transparent permeable grouting test chamber was made of specially made transparent tempered glass to ensure it could withstand high water pressure and grouting pressure. The grouting system adopted a single-liquid grouting mode, with a large-capacity grout storage tank, a high-pressure grouting pump, and durable grouting pipes prepared and connected to the test chamber. The water pressure control system consisted of a large high-head water storage tank and robust water inlet pipes, adjusted according to the actual pore water pressure parameters of the project before being connected to the test chamber. The data acquisition system used a high-precision X-ray scanner and was equipped with a professional data processing workstation.

[0142] During the grouting test, the X-ray scanner was calibrated with a resolution of 0.4 mm / pixel, a scanning range that completely covered the test chamber, and an exposure time of 1.5 seconds. After the grout injection began, the X-ray scanner continuously scanned, storing the data as two-dimensional images. Gaussian filtering and adaptive histogram equalization were applied to the two-dimensional image data for noise reduction. Based on the differences in X-ray absorption by the water-rich sand layer, the grout, and the test chamber, a region growing method was used for image segmentation. The three-dimensional space was divided into voxels with a side length of 0.4 mm, and corresponding attribute values ​​were assigned to each voxel.

[0143] When constructing the convolutional neural network architecture, deep convolutional layers with a kernel size of 5×5 were designed, along with average pooling layers and multiple fully connected layers. The LeakyReLU activation function was chosen. The cross-entropy loss function was selected, and the Adagrad algorithm was used for network parameter optimization training. In the feature extraction and 3D model reconstruction stages, the voxel data was normalized to a range between -1 and 1. The voxel data was divided into a 60% training set, a 20% validation set, and a 20% test set. The training set was input into the network for 800 iterations of training. After training, the test set was input into the network, features were extracted, and the Marching Cubes algorithm was used to reconstruct the 3D model from the voxel data at each time step.

[0144] Quantitative analysis is performed based on the reconstructed 3D model. The number of voxels in the slurry diffusion area is counted, and the volume of each voxel is multiplied to calculate the slurry diffusion volume. For a specific vertical section, the slurry element boundaries are identified and their contours are extracted using an edge detection algorithm, and the coverage area is calculated using an integral method. Based on the changes in diffusion volume and coverage area at different time intervals, combined with the time interval between two adjacent scans, the slurry diffusion rate is calculated, providing a scientific basis for decision-making regarding the foundation reinforcement of water conservancy projects.

[0145] Example 2

[0146] The purpose of this embodiment is to provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the above-described method.

[0147] Example 3

[0148] The purpose of this embodiment is to provide a computer-readable storage medium.

[0149] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the above method.

[0150] Example 4

[0151] The purpose of this embodiment is to provide a modeling and analysis system for permeability grouting model tests in water-rich sand layers, including:

[0152] The time-series voxel data generation module is configured to: acquire the three-dimensional structural information inside the simulation device during the water-rich sand layer permeation grouting test and generate time-series voxel data;

[0153] The 3D model reconstruction module is configured to: extract features from voxel data using a convolutional neural network, and reconstruct a 3D model for each time-series voxel data based on the extracted features;

[0154] The quantitative analysis module is configured to perform quantitative analysis of the slurry diffusion process based on the three-dimensional model of each time series and calculate the experimental parameters.

[0155] Example 5

[0156] The purpose of this embodiment is to provide a computer program product containing instructions that, when run on a computer, cause the computer to perform the methods and functions involved in any of the above embodiments.

[0157] Example 6

[0158] The purpose of this embodiment is to provide a modeling and analysis device for permeable grouting model test in water-rich sand layers, including: a grouting simulation test device for water-rich sand layers and a computer;

[0159] The grouting simulation test device for water-rich sand layers includes: a grouting system, a water pressure control system, a visualization simulation system for water-rich sand layers, and a data acquisition system;

[0160] The grouting system is directly connected to the water-rich sand layer simulation grouting device to meet the needs of different grouting tests.

[0161] The water pressure control system is directly connected to the water-rich sand layer simulated grouting device to provide it with pore water pressure.

[0162] The data acquisition system is used to collect structural information of the water-rich sand layer during the grouting process;

[0163] The data acquisition system communicates with a computer, which is configured to:

[0164] The system receives three-dimensional structural information of the water-rich sand layer inside the grouting process from the data acquisition system and generates time-series voxel data.

[0165] Features are extracted from voxel data using convolutional neural networks, and a 3D model is reconstructed from the voxel data of each time series based on the extracted features.

[0166] Based on the three-dimensional model of each time series, the slurry diffusion process is quantitatively analyzed and experimental parameters are calculated.

[0167] The computer can be configured with high-end computing capabilities, featuring an Intel Core i9 or AMD Ryzen 9 series multi-core processor, such as the i9-13900K, for powerful computing performance. At least 32GB of RAM is required to handle large datasets and complex tasks, ensuring smooth system operation. A 1TB or larger NVMe SSD is recommended for faster data access thanks to its high read / write speeds. A high-performance NVIDIA GeForce RTX 40 series or AMD Radeon RX 7000 series graphics card is used to accelerate specific algorithms. Windows 11 or a Linux distribution is chosen for stable compatibility. A high-resolution large-screen monitor (2560×1440 or higher) is included for precise analysis. An efficient cooling system maintains stable operation under high loads, and a Gigabit Ethernet or higher network interface ensures fast data transfer. A high-power, stable power supply is also included, balancing compatibility, expandability, and chassis space, allowing for flexible configuration adjustments and optimization based on actual needs.

[0168] The steps and methods involved in the apparatus of the above embodiments correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0169] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0170] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A modeling analysis method for a water-rich sand layer permeation grouting model test, characterized in that, The application relates to a visualization simulation device for water-rich sand layer permeation grouting. Three-dimensional structure information of the inside of a simulation device in a water-rich sand layer permeation grouting test process is acquired, and time-series voxel data is generated. The acquisition of the three-dimensional structure information of the inside of the simulation device in the water-rich sand layer permeation grouting test process comprises the following steps: initial scanning is performed before the sand layer is filled and before grout is injected; a suitable scanning mode is selected, scanning parameters are set, and the parameters are consistent with the parameters determined in the debugging stage, and the initial state of the water-rich sand layer is recorded; grouting is started, and continuous scanning is performed; a scanning time interval is set, and the sand layer and the grouting condition in the test box are scanned once at a fixed time interval; the process of generating the time-series voxel data comprises the following steps: two-dimensional image data obtained through scanning is subjected to denoising treatment; two-dimensional image data after denoising is subjected to image enhancement operation; according to the difference in X-ray absorption degree of the water-rich sand layer, grout and the test box, a threshold segmentation method is used to segment the enhanced image; according to the size of the test box and the scanning resolution, the three-dimensional space of the scanning area is divided into small cubic voxels, and for each voxel, a corresponding attribute value is given according to the position of the voxel in the image and the image segmentation result; all voxel data are stored in the form of a three-dimensional array, and the value of each element is the attribute value of the corresponding voxel; features in the voxel data are extracted by using a convolutional neural network, and a three-dimensional model is reconstructed for each time-series voxel data according to the extracted features; based on the three-dimensional model of each time series, the grout diffusion process is quantitatively analyzed, and test parameters are calculated. The features in the voxel data are extracted by using a convolutional neural network, wherein the convolutional neural network comprises:

2. The modeling analysis method of water-rich sand layer permeation grouting model test according to claim 1, characterized in that, a first convolutional layer, a second convolutional layer, a third convolutional layer, a maximum pooling layer and a fully connected layer; the first convolutional layer is used for preliminary feature extraction of the voxel data; the second convolutional layer is used for extracting more complex features from the voxel data; the third convolutional layer can capture more subtle and higher-level features in the voxel data; a maximum pooling layer is connected after each convolutional layer, and the maximum pooling layer is used for downsampling of a feature map output by the connected convolutional layer; a fully connected layer is connected after the maximum pooling layer, and the fully connected layer is used for integrating the extracted features and is used for final classification or regression tasks. The three-dimensional model is reconstructed for each time-series voxel data according to the extracted features, and the specific process comprises the following steps:

3. The modeling analysis method of water-rich sand layer permeation grouting model test according to claim 1, characterized in that, a three-dimensional space is divided into small cubes, i.e. voxels, and for each small cube, the intersection of the small cube with an isosurface is determined according to the attribute value of the vertex of the small cube; then, according to a pre-defined lookup table, the topological structure of the isosurface on the small cube is determined; finally, the coordinates of points on the isosurface are calculated through interpolation, and the isosurfaces on all small cubes are spliced, so that the three-dimensional model is obtained; the model can directly and intuitively show the diffusion process of the grout in the water-rich sand layer, so that subsequent visualization and quantitative analysis can be performed. The application relates to a visualization simulation device for water-rich sand layer permeation grouting.

4. A modeling analysis system for model test of water-rich sand layer permeation grouting, characterized in that, A time-series voxel data generation module is configured to acquire three-dimensional structure information of the inside of a simulation device in a water-rich sand layer permeation grouting test process, and generate time-series voxel data. ​ The three-dimensional structure information inside the simulation device in the process of the water-rich sand layer permeation grouting test is acquired, including: Before the sand layer filling is completed and the slurry is not injected, an initial scanning is performed; A suitable scanning mode is selected, the scanning parameters are set to be consistent with the parameters determined in the debugging stage, and the initial state of the water-rich sand layer is recorded; Grouting is started, and continuous scanning is performed; A scanning time interval is set, and the sand layer and grouting in the test box are scanned once at the fixed time interval; The process of generating time-series voxel data is as follows: The two-dimensional image data obtained by scanning is subjected to denoising processing; The two-dimensional image data after denoising is subjected to image enhancement operation; According to the difference in X-ray absorption degree of the water-rich sand layer, the slurry and the test box, a threshold segmentation method is used to perform image segmentation on the enhanced image; According to the size of the test box and the scanning resolution, the three-dimensional space of the scanning region is divided into small cubic voxels, and for each voxel, a corresponding attribute value is assigned according to its position in the image and the image segmentation result; All voxel data are stored in the form of a three-dimensional array, and the value of each element is the attribute value of the corresponding voxel; The three-dimensional model reconstruction module is configured to extract features in the voxel data using a convolutional neural network, and reconstruct a three-dimensional model for each time-series voxel data according to the extracted features; The quantitative analysis module is configured to quantitatively analyze the slurry diffusion process based on the three-dimensional model of each time series, and calculate the test parameters.

5. The modeling analysis device for the model test of water-rich sand layer permeation grouting, characterized in that, It comprises: A water-rich sand layer grouting simulation test device and a computer; The water-rich sand layer grouting simulation test device comprises a grouting system, a water pressure control system, a visual water-rich sand layer simulation system and a data acquisition system; The grouting system is directly connected with the water-rich sand layer simulation grouting device to meet different grouting test requirements; The water pressure control system is directly connected with the water-rich sand layer simulation grouting device to provide pore water pressure for it; The data acquisition system is used to acquire the internal structure information of the water-rich sand layer during grouting; The data acquisition system communicates with the computer, and the computer is configured to: Receive the three-dimensional structure information of the water-rich sand layer inside during grouting acquired by the data acquisition system, and generate time-series voxel data; Extract features in the voxel data using a convolutional neural network, and reconstruct a three-dimensional model for each time-series voxel data according to the extracted features; Quantitatively analyze the slurry diffusion process based on the three-dimensional model of each time series, and calculate the test parameters.

6. A computer program product comprising a computer program, characterized in that, The computer program is executed by the processor to implement the method of any one of claims 1 to 3.

7. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps of the method of any one of claims 1 to 3.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The program is executed by the processor to perform the steps of the method of any one of claims 1 to 3.

Images