Method and device for characterizing the water state evolution of cement hydration

Abstract

The application provides a cement hydration water state evolution characterization method and device, and the method comprises the following steps: scanning and imaging a cement slurry by X-CT to generate an original sample volume data set, respectively performing magnetic resonance characterization and modal conversion multi-value segmentation to generate a cement hydration water state evolution sample data set and a multi-value sample volume data set; model training is performed on an encoder-decoder structure to construct a water state evolution deep learning model; water state evolution data is generated by performing water state evolution on original volume data to be characterized, the space-time evolution distribution reconstruction of the water state in the cement material is realized without relying on high-resolution NMR, the four-dimensional distribution of the water state data in the hydration process in time and space is promoted, the spatial resolution is improved, the corresponding relationship between the water distribution and the microstructure is clarified, different water state distributions in small-scale structures can be effectively distinguished, and the occurrence positions of various water in the microstructure can be accurately positioned.

Description

Technical Field

[0001] This invention relates to the field of computer technology, particularly to the field of artificial intelligence technology, and especially to a method and apparatus for characterizing the water-state evolution of cement hydration. Background Technology

[0002] Moisture distribution has a significant impact on the evolution of the properties of cement-based materials, a process that begins in the initial stage of hydration, specifically in the binary system composed of cement and water. The formation and distribution of different water states directly affect key indicators such as the microstructure, mechanical properties, drying shrinkage, and durability of cement-based materials. Therefore, accurately identifying and distinguishing the spatial distribution characteristics of different water states is of great scientific significance for revealing the performance evolution laws during the hydration process of cement materials. However, the monitoring and differentiation of moisture, especially at the spatial scale, still faces significant challenges. This is mainly due to: cement-based materials being opaque, multiphase, and porous media, making moisture distribution difficult to visualize directly; the materials possess significant multi-scale structural characteristics, especially the microstructure of the calcium silicate hydrate (CSH) gel phase, which is not yet fully understood; and the different water states themselves existing in a continuous transitional state without clear boundaries, making them difficult to distinguish using characterization methods.

[0003] Among related technologies, magnetic resonance (NMR) is one of the few methods currently used to observe the moisture state in cement materials. NMR works by detecting hydrogen nuclei (… 1 The relaxation behavior of hydrogen nuclei (H) indirectly yields moisture-related information, but NMR does not directly reflect the spatial structure of the sample; it only inverts the structure through changes in hydrogen nucleus relaxation signals. Therefore, for... 1 The strong dependence of H signals limits NMR's spatial resolution, making it difficult to clearly define the correspondence between water distribution and microstructure. In 3D imaging, due to limitations in NMR signal acquisition speed and signal-to-noise ratio, the obtained image resolution is far lower than that of X-ray computed tomography (XCT) and other imaging methods, making it impossible to effectively distinguish the distribution of different water states in small-scale structures. In real materials, water states exhibit continuous changes and overlap (e.g., the boundary between gel water and interlayer water is unclear), making it impossible for NMR technology to accurately locate the occurrence of various water states in the microstructure. Summary of the Invention

[0004] The present application aims to provide a method for characterizing the water state evolution of cement hydration, which reconstructs the spatial and temporal evolution distribution of the water state in cement materials without relying on high-resolution NMR, promotes the four-dimensional distribution of water state data in time and space during the hydration process, improves the spatial resolution, clarifies the correspondence between water distribution and microstructure, effectively distinguishes different water state distributions in small-scale structures, and accurately locates the occurrence position of various water in the microstructure. Another object of the present application is to provide a device for characterizing the water state evolution of cement hydration. Another object of the present application is to provide a computer readable medium. Another object of the present application is to provide a computer device.

[0005] To achieve the above objects, the present application discloses a method for characterizing the water state evolution of cement hydration, comprising:

[0006] The prepared cement paste is scanned and imaged by computer tomography technology to generate an original sample volume data set;

[0007] The original sample volume data set is characterized by magnetic resonance to generate a water state evolution sample data set of cement hydration;

[0008] The original sample volume data set is subjected to modal conversion and multi-value segmentation to generate a multi-value sample volume data set;

[0009] According to the water state evolution sample data set and the multi-value sample volume data set, a model training is performed on a preset encoder-decoder structure to construct a water state evolution deep learning model;

[0010] The water state evolution of the original volume data to be characterized is performed by the water state evolution deep learning model to generate water state evolution data.

[0011] Preferably, the water state evolution sample data set includes sample global water distribution information, sample Z-axis direction water distribution information, and sample three-dimensional water distribution information;

[0012] The original sample volume data set is characterized by magnetic resonance to generate a water state evolution sample data of cement hydration, comprising:

[0013] The original sample volume data set is sampled by a multi-layer echo sequence to obtain sample global water distribution information at different times;

[0014] The original sample volume data set is sampled by a constant gradient selection sequence to obtain sample Z-axis direction water distribution information at different times;

[0015] The original sample volume data set is subjected to spatial encoding and time encoding to obtain sample three-dimensional water distribution information at different times.

[0016] Preferably, the original sample volume dataset is subjected to mode transformation and multi-valued segmentation to generate a multi-valued sample volume dataset, including:

[0017] The original sample volume data in the original sample volume dataset is converted into electron microscope scan sample images by a pre-trained modality conversion network, thus generating an electron microscope scan sample volume dataset.

[0018] Image segmentation of electron microscope scan sample volume dataset is performed using a pre-trained deep learning segmentation network to generate a multi-valued sample volume dataset.

[0019] Preferably, an electron microscope scanned sample volume dataset is segmented using a pre-trained deep learning segmentation network to generate a multi-valued sample volume dataset, including:

[0020] A deep learning segmentation network is used to coarsely segment electron microscope scanned sample images in the electron microscope scanned sample volume dataset, generating coarsely segmented multi-valued images.

[0021] Extract image regions corresponding to preset key categories from coarsely segmented multi-valued images;

[0022] The image regions corresponding to key categories are refined to generate multi-valued image refinement regions;

[0023] A multi-channel input tensor is constructed based on the refined regions of the multi-valued image and the coarsely segmented multi-valued image to generate a multi-valued sample volume dataset.

[0024] Preferably, based on the water state evolution sample dataset and the multi-valued sample volume dataset, a pre-defined encoder-decoder structure is trained to construct a deep learning model for water state evolution, including:

[0025] Data registration was performed on the water state evolution sample dataset and the multi-valued sample volume dataset to construct the model dataset;

[0026] Using the model dataset, the encoder-decoder structure is trained to construct a deep learning model for water state evolution.

[0027] Preferably, a deep learning model of water state evolution is constructed by training the encoder-decoder structure using a model dataset, including:

[0028] The encoder performs multi-layer convolutional encoding on the multi-valued sample volume dataset to generate sample feature maps.

[0029] By using a physical constraint layer and a decoder, the sample feature map is mapped and decoded to generate preliminary predictions of water distribution.

[0030] According to different regions and directions, the preliminary water distribution prediction results are accumulated and pooled to generate global water distribution prediction information, Z-axis water distribution prediction information, and three-dimensional water distribution prediction information.

[0031] By constructing a comprehensive loss function, loss results are generated based on the predicted global moisture distribution information, the predicted Z-axis moisture distribution information, the predicted three-dimensional moisture distribution information, the global moisture distribution information of the sample, the moisture distribution information of the sample Z-axis, and the three-dimensional moisture distribution information of the sample.

[0032] By using the gradient descent algorithm, backpropagation is performed based on the loss results to iteratively update the model parameters, thereby constructing a deep learning model for water state evolution.

[0033] Preferably, the method further includes:

[0034] Based on the preliminary predicted moisture distribution results, uncertainty indicators are generated, including the distribution mean and the distribution standard deviation.

[0035] Uncertainty images are generated based on the distribution mean and distribution standard deviation indicators for different regions.

[0036] This invention also discloses a device for characterizing the aqueous evolution of cement hydration, comprising:

[0037] The scanning unit is used to scan and image the prepared cement slurry using computed tomography (CT) technology to generate a raw sample volume dataset.

[0038] The magnetic resonance characterization unit is used to perform magnetic resonance characterization on the original sample volume dataset to generate a water state evolution sample dataset of cement hydration.

[0039] The modality transformation and multi-valued unit is used to perform modality transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset.

[0040] The model training unit is used to train the model on the preset encoder-decoder structure based on the water evolution sample dataset and the multi-valued sample volume dataset, and to build a deep learning model for water evolution.

[0041] The water state evolution unit is used to perform water state evolution on the original volume data to be represented through a water state evolution deep learning model, thereby generating water state evolution data.

[0042] The present invention also discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0043] The present invention also discloses a computer device, including a memory and a processor, wherein the memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions, wherein the processor executes the program to implement the method described above.

[0044] The present invention also discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method described above.

[0045] This invention utilizes computed tomography (CT) to scan and image prepared cement slurry, generating an original sample volume dataset. Magnetic resonance (MR) characterization of the original sample volume dataset generates a water-state evolution sample dataset for cement hydration. Mode transformation and multi-valued segmentation are then performed on the original sample volume dataset to generate a multi-valued sample volume dataset. Based on the water-state evolution sample dataset and the multi-valued sample volume dataset, a pre-defined encoder-decoder structure is trained to construct a deep learning model for water-state evolution. This deep learning model is used to perform water-state evolution on the original volume data to be characterized, generating water-state evolution data. Without relying on high-resolution NMR, this invention reconstructs the spatiotemporal evolution distribution of water in cement materials, effectively promoting the four-dimensional distribution of water-state data in time and space during the hydration process. This improves spatial resolution, clarifies the correspondence between water distribution and microstructure, effectively distinguishes different water-state distributions in small-scale structures, and accurately locates the position of various water types within the microstructure. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating a method for characterizing the water-state evolution of cement hydration, as provided in an embodiment of the present invention;

[0048] Figure 2 A flowchart illustrating another method for characterizing the water-state evolution of cement hydration provided in this embodiment of the invention;

[0049] Figure 3 This is a schematic diagram of the structure of a water-state evolution characterization device for cement hydration provided in an embodiment of the present invention;

[0050] Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] It should be noted that the water state evolution characterization method and apparatus for cement hydration disclosed in this application can be used in the field of artificial intelligence technology, or in any field other than artificial intelligence technology. The application field of the water state evolution characterization method and apparatus for cement hydration disclosed in this application is not limited.

[0053] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution will be explained below. Water reacts chemically with cement and gradually transforms into "bound water" in different states, i.e., different water states. When water is fixed in the crystal structure, it forms so-called "bound water," such as calcium hydroxide (CH) and ettringite (Aft); when water enters the interior of the CSH gel phase, it can form interlayer water and gel water; if water is retained in capillary pores, it manifests as capillary water. With the improvement of computing power and the development of deep learning technology, artificial intelligence methods have shown significant advantages in complex mapping modeling. Based on this, the multimodal problem of NMR and X-CT characterization is integrated, and the results of X-CT characterization are used as a basis to deduce the results of NMR, which can provide technical support for the synergistic use of the two.

[0054] This invention fully integrates the advantages of X-CT and NMR characterization. By constructing a multimodal dataset and introducing a physically constrained neural network model, it achieves the mapping from microstructural images to water state distribution, thereby obtaining four-dimensional water state distribution information (i.e., three-dimensional spatial + temporal evolution) during cement hydration. Based on the water storage mechanism in various cement phases (e.g., CSH, CH, Aft, pores, etc.), and combining NMR experiments and X-CT image analysis, a numerical mapping relationship between structure and water state is established, forming a physical constraint prior. The microstructural images obtained by X-CT (after segmentation and multi-channel processing) are temporally registered with the global, one-dimensional, and three-dimensional water state data obtained by NMR. A deep learning model is constructed, including multi-scale convolutional extraction, physical constraint mapping, and forward inference modules. Inputting structural image data, it outputs predicted values ​​of water state distribution for each pixel or region, achieving the reconstruction of the spatiotemporal evolution distribution of water state in cement materials without relying on high-resolution NMR. This invention can effectively facilitate the four-dimensional (temporal + spatial) distribution of water state data during the hydration process.

[0055] The following uses a water-state evolution characterization device for cement hydration as an example to illustrate the implementation process of the water-state evolution characterization method for cement hydration provided in this embodiment of the invention. It is understood that the executing entity of the water-state evolution characterization method for cement hydration provided in this embodiment of the invention includes, but is not limited to, a water-state evolution characterization device for cement hydration.

[0056] Figure 1 A flowchart illustrating a method for characterizing the water-state evolution of cement hydration, as provided in an embodiment of the present invention, is shown below. Figure 1 As shown, the method includes:

[0057] Step 101: Use X-CT to scan and image the prepared cement slurry to generate the original sample volume dataset.

[0058] In this embodiment of the invention, water and cement are thoroughly mixed to prepare cement slurry; as the cement slurry ages, it is scanned multiple times using an X-CT device, and a three-dimensional reconstruction algorithm is used to reconstruct the volume data during the cement hydration process, thereby constructing an original sample volume dataset.

[0059] Step 102: Perform NMR characterization on the original sample volume dataset to generate a water state evolution sample dataset of cement hydration.

[0060] In this embodiment of the invention, the water state evolution sample dataset includes global water distribution information of the sample, water distribution information along the Z-axis of the sample, and three-dimensional water distribution information of the sample.

[0061] Specifically, NMR characterization was performed on the original sample volume dataset. Multilayer echo-CPMG (MSE-CPMG) sequence was used to obtain global water distribution information (including crystal water, interlayer water, gel water, capillary water, and volume water) of the original sample volume dataset at different times. Constant gradient layer selection sequence was used to obtain the water distribution of the sample in the Z-axis direction at different times of the original sample volume dataset. Spatial encoding and temporal encoding were used to obtain the three-dimensional water distribution of the sample at different times of the original sample volume dataset.

[0062] Step 103: Perform modality transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset.

[0063] In this embodiment of the invention, X-CT data in the original sample volume dataset is converted into electron microscopy (BSE) sample images through a pre-trained modality conversion network to construct a BSE sample volume dataset; a pre-trained deep learning segmentation network is used to coarsely segment the BSE rock volume images and refine the key categories to generate multi-channel input tensors.

[0064] Step 104: Based on the water state evolution sample dataset and the multi-valued sample volume dataset, train the model on the preset encoder-decoder structure to construct a deep learning model for water state evolution.

[0065] In this embodiment of the invention, the encoder-decoder structure includes an encoder, a physical constraint layer, a dropout layer, and a decoder.

[0066] Step 105: Using a deep learning model for water state evolution, perform water state evolution on the original volume data to be represented, and generate water state evolution data.

[0067] In this embodiment of the invention, the original volume data is X-CT data.

[0068] Specifically, the original volume data to be represented is input into the water state evolution deep learning model to perform water state evolution, and the output is water state evolution data and uncertainty images of each category of region.

[0069] In the technical solution provided by this invention, the prepared cement slurry is scanned and imaged using computed tomography (CT) technology to generate an original sample volume dataset. The original sample volume dataset is then characterized using magnetic resonance imaging (MRI) to generate a water state evolution sample dataset for cement hydration. The original sample volume dataset is further subjected to mode transformation and multi-valued segmentation to generate a multi-valued sample volume dataset. Based on the water state evolution sample dataset and the multi-valued sample volume dataset, a pre-defined encoder-decoder structure is trained to construct a deep learning model for water state evolution. This deep learning model is used to perform water state evolution on the original volume data to be characterized, generating water state evolution data. Without relying on high-resolution NMR, the spatiotemporal evolution distribution of water in cement materials is reconstructed, effectively promoting the four-dimensional distribution of water state data in the hydration process in time and space, improving spatial resolution, clarifying the correspondence between water distribution and microstructure, effectively distinguishing different water state distributions in small-scale structures, and accurately locating the location of various types of water in the microstructure.

[0070] Figure 2 A flowchart of another method for characterizing the water-state evolution of cement hydration provided in this embodiment of the invention is shown below. Figure 2 As shown, the method includes:

[0071] Step 201: Use X-CT to scan and image the prepared cement slurry to generate the original sample volume dataset.

[0072] In this embodiment of the invention, each step is performed by a water-state evolution characterization device for cement hydration.

[0073] In this embodiment of the invention, water and cement are thoroughly mixed to prepare cement slurry. As an optional method, ordinary silicate cement is selected, and different silicate cement slurries with a water-cement ratio of 0.5 are prepared by mixing with water. After stirring for 30 seconds, the mixture is vibrated for 30 seconds and stirred again to ensure uniform mixing. The mixture is then placed into a polyvinyl alcohol (PVA) thin-walled sample tube with a diameter of 10 mm and a height of 100 mm.

[0074] In this embodiment of the invention, as the cement slurry ages, multiple scans are performed using an X-CT device, and a three-dimensional reconstruction algorithm is used to reconstruct the volume data during the cement hydration process, thus constructing an original sample volume dataset. As an optional approach, the X-CT device uses a voltage of 60 kV, a current of 50 μA, a voxel resolution of r = 5 μm / voxel, and a frame rate of 1080 FPS. Data for each sample is captured for 1 hour (h), 4 hours, 8 hours, 12 hours, 24 hours, 1 day (d), 3 days, 7 days, 28 days, and 56 days to obtain the original X-CT volume dataset. The data size is 2500 pixels × 2500 pixels × 25000 pixels, with a total of 25,000 slice images along the z-axis, each image being 2500 pixels × 2500 pixels.

[0075] Step 202: Using the MSE-CPMG sequence, sample the original sample volume dataset to obtain global moisture distribution information of the sample at different times.

[0076] In this embodiment of the invention, the MSE-CPMG sequence is used to obtain the global moisture distribution information of the original sample volume dataset at different times. The global moisture distribution information of the sample includes crystal water, interlayer water, gel water, capillary water and volume water.

[0077] As an alternative, the MSE-CPMG sequence is used with a sampling frequency of 5000 kHz, a sampling time of 0.1 ms, an echo interval of 0.05 ms, and a MSE-τ value of 0.002. Φ The test was conducted using 0.016 MSE-τ as the main test parameter to obtain the attenuation signal at different times, denoted as 0dw(t). Wherein, MSE-τ... Φ MSE-τ and MSE-τ are two parameters of the MSE sequence. Φ is the half-echo interval between π pulses, and its value is set to 0.002; MSE-τ is the delay time after the initial free induction decay (FID) sequence, and its value is set to 0.016.

[0078] Step 203: Using a constant gradient layer selection sequence, sample the original sample volume dataset to obtain the water distribution information of the sample in the Z-axis direction at different times.

[0079] In this embodiment of the invention, the moisture distribution along the Z-axis of the original sample volume dataset at different times is obtained by using a constant gradient layer selection sequence.

[0080] As an alternative approach, a constant gradient layer selection sequence is used, with a 10mm interval between layers, to obtain the CPMG attenuation signal 1dw(t) of each layer in the original sample volume dataset.

[0081] Step 204: Perform spatial and temporal encoding on the original sample volume dataset to obtain the three-dimensional moisture distribution information of the sample at different times.

[0082] In this embodiment of the invention, spatial and temporal encoding is performed on the original sample volume dataset to obtain the three-dimensional spatial distribution of water content in the original sample volume dataset at different times, denoted as 3dw(t) data.

[0083] Step 205: Convert the original sample volume data in the original sample volume dataset into BSE sample images using a pre-trained modality conversion network to generate the BSE sample volume dataset.

[0084] In this embodiment of the invention, based on a deep learning diffusion model, a BSE image is used as a supervision signal to train the deep learning diffusion model to convert CT images to BSE images, thereby constructing a modality conversion network.

[0085] Specifically, the original sample volume data in the original sample volume dataset is input into the modality transformation network to transform the original sample volume data (CT sample images) into BSE sample images, thus constructing the BSE sample volume dataset.

[0086] Step 206: Perform image segmentation on the BSE sample volume dataset using a pre-trained deep learning segmentation network to generate a multi-valued sample volume dataset.

[0087] In this embodiment of the invention, the deep learning segmentation network is trained based on a deep learning model and is capable of performing multi-valued segmentation of images according to categories.

[0088] In this embodiment of the invention, step 206 specifically includes:

[0089] Step 2061: Use a deep learning segmentation network to coarsely segment the electron microscope scanned sample images in the electron microscope scanned sample volume dataset to generate coarsely segmented multi-valued images.

[0090] Specifically, the BSE sample volume dataset is input into a deep learning segmentation network to perform coarse segmentation on the BSE sample images. The BSE sample images are then segmented according to categories, including but not limited to unhydrated cement particles, low-density CSH, high-density CSH, calcium hydroxide (CH), ettringite (Aft), small capillaries, large capillaries, cracks, and pores.

[0091] In this embodiment of the invention, in the multi-valued image obtained after coarse segmentation, each pixel obtains 0. The integer label 8 clearly identifies each category.

[0092] It is worth noting that the coarse segmentation multi-valued image also includes volume data for various categories, that is: different categories of substances composed of spatial pixels, whose volumes can be directly obtained based on the number of pixels and pixel resolution. Specifically, this includes: the volume of unhydrated cement particles, the volume of low-density CSH, the volume of high-density CSH, the volume of calcium hydroxide (CH), the volume of ettringite (Aft), the volume of small capillaries, the volume of large capillaries, the volume of cracks, and the volume of pores.

[0093] Step 2062: Select the image regions corresponding to the preset key categories from the coarsely segmented multi-valued image.

[0094] In this embodiment of the invention, based on coarse segmentation, image regions corresponding to preset key categories are extracted from the multi-valued image obtained after coarse segmentation, and the image regions corresponding to preset key categories are refined.

[0095] It is worth noting that the key categories can be set according to actual needs, and this embodiment of the invention does not limit this. As an optional solution, the key categories include high-density CSH, assuming the label is 2.

[0096] Taking high-density CSH as the key category as an example, high-density CSH regions (ROIs) are extracted from the multi-valued image obtained after coarse segmentation.

[0097] Step 2063: Refine the image regions corresponding to the key categories to generate multi-valued image refinement regions.

[0098] In this embodiment of the invention, the refinement process specifically includes: using morphological operations to remove isolated noise from the image regions corresponding to key categories, and using a sliding window method to calculate the local gray-scale mean within the image regions. With local standard deviation .

[0099] Taking high-density CSH as the key category as an example, the refinement process is as follows: morphological operations are used to remove isolated noise in the high-density CSH region (ROI); within the ROI, a sliding window method is used to calculate the local gray-level mean. With local standard deviation This is to reflect the potential differences in local grayscale distribution and chemical composition.

[0100] Step 2064: Construct a multi-channel input tensor based on the refined region of the multi-valued image and the coarse segmentation of the multi-valued image to generate a multi-valued sample volume dataset.

[0101] In this embodiment of the invention, discrete labels for each pixel are generated based on the coarsely segmented multi-valued image. The original BSE grayscale of the coarsely segmented multi-valued image Discrete labels represent the coarse segmentation result, that is, they identify the category of the pixel; based on the multi-valued image refinement region, the local gray-level mean of the image region corresponding to each key category is obtained. With local standard deviation Discrete labels Original BSE grayscale Local mean and local standard deviation Multichannel input tensors combined into pixels:

[0102]

[0103] in, For a multi-channel input tensor, H is the length of the image and W is the width of the image.

[0104] The multi-valued sample volume dataset is obtained by using the set of multi-channel input tensors for each pixel of each BSE sample image.

[0105] Step 207: Perform data registration on the water state evolution sample dataset and the multi-valued sample volume dataset to construct the model dataset.

[0106] In this embodiment of the invention, the multi-valued sample volume dataset obtained through BSE and the water state evolution sample dataset obtained through NMR characterization are matched one-to-one according to the corresponding images to construct the model dataset required for the deep learning model. The model dataset includes matching pairs of multi-channel input tensors and corresponding global water distribution information, water distribution information along the Z-axis, and three-dimensional water distribution information of the samples.

[0107] Furthermore, the model dataset is divided into a training set and a validation set in a 3:1 ratio. The training set is used for subsequent model training, and the validation set is used for model validation.

[0108] Step 208: Using the model dataset, train the encoder-decoder structure to construct a deep learning model for water state evolution.

[0109] In this embodiment of the invention, step 208 specifically includes:

[0110] Step 2081: Using an encoder, perform multi-layer convolutional encoding on the multi-valued sample volume dataset to generate sample feature maps.

[0111] In this embodiment of the invention, the encoder input is a multi-channel input tensor. ,go through L Layer convolution is used for encoding, generating sample feature maps step by step. Its expression is:

[0112]

[0113]

[0114] Where Conv represents the convolution operation, BN represents batch normalization, ReLU represents the ReLU activation function, and X represents the multi-channel input tensor of all pixels in the current image. The set, This represents the result of the (k-1)th convolutional encoding. This represents the result of the first convolutional encoding. This represents the feature map of the sample.

[0115] Step 2082: Map and decode the sample feature map through the physical constraint layer and decoder to generate preliminary prediction results of water distribution.

[0116] In this embodiment of the invention, for each pixel , by discrete channels Determine its category And define a moisture range for this category according to a preset. A fully connected layer is used to map the feature maps of each sample layer to vectors, and then the mapping is activated by the Sigmoid function. The interval, specifically:

[0117]

[0118] MLP stands for fully connected layer. It is the Sigmoid activation function. For sample feature maps, This is the sample mapping vector.

[0119] Physical constraints can be expressed as:

[0120]

[0121] in, The constrained sample vector, For category The minimum value of the moisture range, For category The maximum value of the moisture range, This is the sample mapping vector.

[0122] By fusing physical constraints with the decoder through skip links and upsampling:

[0123]

[0124] in, To predict preliminary moisture distribution results, For sample feature maps, The constrained sample vector, This represents the result of the (k-1)th convolutional encoding.

[0125] It is worth noting that the preliminary prediction of moisture distribution results The moisture distribution is shown at the pixel level.

[0126] In this embodiment of the invention, the moisture range for each category The calculation process is as follows:

[0127] Based on the main areas of water accumulation, the following assumptions about water distribution can be made:

[0128]

[0129] in, and These represent the proportion of interlayer water per unit volume in high-density CSH and low-density CSH, respectively. and These represent the proportion of gel water per unit volume in high-density CSH and low-density CSH, respectively (both determined by the CSH gel model). and These are the volumes of the high-density CSH and the low-density CSH, respectively. The volumes of the high-density CSH and the low-density CSH can be obtained from the coarsely segmented multi-valued image. This refers to the volume of water between layers. The volume of the gel water; and These are the proportions of water of crystallization per unit volume in CH and Aft (determined by their chemical formulas). For the volume of bound water; For capillary water volume; The pore volume can be obtained from the coarsely segmented multi-valued image; The proportion of water that can fill the capillary pores (typically 0.6–0.9, considering saturation and capillary size) can be used to determine the mass / mass range of water based on the density of water at different scales.

[0130]

[0131] in, Let i be the mass of water. For the volume of water i, the coarse segmentation multi-valued image is volume data of different categories, that is: different categories of substances composed of spatial pixels. The volume of each water can be obtained directly based on the number of pixels and the pixel resolution. Let be the density of water i. Wherein, =1.00g / cm³, =2.24 g / cm³, =2.60~2.80g / cm³, =2.1~2.3g / cm³.

[0132] It is worth noting that since the density of water is distributed in intervals, the corresponding mass of water is also distributed in intervals, that is, the mass range of water.

[0133] Based on the principle of LF-NMR, by using NMR to obtain the signal percentage of different water masses, the mass ratio of different water masses can be determined. Therefore, based on the total water content, the following can be obtained: , , , Based on this, density can be used to obtain... , , , Then, based on the aforementioned volume relationships, solve for... , , , , , as well as This involves calculating the water content of different categories within categories such as HD-CSH, and then, based on the law of conservation of water mass, summing the water content of different categories within the same category to obtain the water content range for any pixel. . There may be various types of water in the image. For example, in a multi-valued image, the region at pixel location (214, 212) contains HD-CSH, and the proportions of interlayer water and gel water can be calculated. and Therefore, the moisture mass of this pixel location is actually the sum of the mass of the interlayer water and the mass of the gel water in that pixel location. Due to the law of conservation of mass, this part of the moisture originates from the initial mixing water, so its moisture content is the percentage of the moisture mass relative to the initial mixing water mass. Based on the known initial water mass, the corresponding moisture content range can be calculated.

[0134] Furthermore, based on the preliminary moisture distribution results predicted by the model, uncertainty indices are generated, including the distribution mean and distribution standard deviation. Uncertainty images are then generated according to different region categories, based on the distribution mean and distribution standard deviation indices. Specifically, Dropout layers are embedded in the encoder and physical constraint layer, and Dropout activation is maintained during the optimization phase. K forward propagations are performed on the same input to obtain the prediction set. Calculate the mean and standard deviation As an indicator of output and uncertainty, the uncertainty indicator is plotted according to the pixel distribution of the image to obtain an uncertainty image.

[0135] Step 2083: According to different regions and directions, perform cumulative pooling on the preliminary water distribution prediction results to generate global water distribution prediction information, Z-axis water distribution prediction information, and three-dimensional water distribution prediction information.

[0136] In this embodiment of the invention, since the global signal can be decomposed into moisture information at different scales, the preliminary predicted moisture distribution results are divided into different regions c (e.g., CH, Aft, high-density CSH, low-density CSH, capillaries, etc.), and the moisture in each region is accumulated as follows: ,in, To predict preliminary moisture distribution results, This represents the cumulative water content in region c; based on the relaxation model. Calculate the NMR signal contribution of region c: This allows for the accumulation of information to predict the global water distribution. .

[0137] Since the one-dimensional moisture distribution information is mainly capillary water along the Z-axis, the preliminary moisture distribution prediction results are screened to identify the capillary region, and the distribution information is calculated layer by layer along the Z-axis. The small capillary water, large capillary water, and volumetric water in each region of each layer along the Z-axis are accumulated, and the accumulation method is consistent with the global signal processing method. And through the relaxation model Calculate and obtain the global NMR signal of each layer region along the Z-axis, and stitch them together along the Z-axis to predict the water distribution information along the Z-axis. It is worth noting that region c here refers to the capillary region.

[0138] The three-dimensional moisture distribution signal is the coarsest moisture distribution, and small-scale moisture signals are lost for verification. Therefore, pooling is directly applied to the preliminary predicted moisture distribution results to convert them into predicted three-dimensional moisture distribution information. .

[0139] In this embodiment of the invention, preliminary moisture distribution results are predicted. Three types of NMR signals were simulated in the forward direction (predicting global moisture distribution information, predicting moisture distribution information in the Z-axis direction, and predicting three-dimensional moisture distribution information) to correspond to the data (sample global moisture distribution information, sample Z-axis direction moisture distribution information, and sample three-dimensional moisture distribution information) obtained in the corresponding experiment through MSE-CPMG, constant gradient layer selection, and spatial encoding.

[0140] Step 2084: Using the constructed comprehensive loss function, generate loss results based on the predicted global moisture distribution information, the predicted Z-axis moisture distribution information, the predicted three-dimensional moisture distribution information, the global moisture distribution information of the sample, the moisture distribution information of the sample Z-axis, and the three-dimensional moisture distribution information of the sample.

[0141] In this embodiment of the invention, the predicted global moisture distribution information, the predicted Z-axis moisture distribution information, and the predicted three-dimensional moisture distribution information are spliced ​​together to generate a predicted NMR signal; the sample's global moisture distribution information, the sample's Z-axis moisture distribution information, and the sample's three-dimensional moisture distribution information are spliced ​​together to obtain the actual NMR signal.

[0142] Specifically, the loss is calculated for the predicted NMR signal and the actual NMR signal using the following comprehensive loss functions to generate the loss result. The comprehensive loss functions include the mean squared error loss function, the difference penalty loss function, and the total variation regularization loss function.

[0143] The mean squared error loss function is:

[0144]

[0145] in, The mean squared error loss is N, where N is the total number of losses. This is the i-th predicted NMR signal; Let be the i-th actual NMR signal.

[0146] A difference penalty is applied to the total moisture content (e.g., CH, Aft, high-density CSH, low-density CSH, capillary moisture, etc.) of each region compared to the preset range. For each region c, if the predicted regional moisture content... If the value exceeds the preset range, a penalty is applied. The difference penalty loss function is:

[0147]

[0148] in, The difference penalty loss is denoted by c, where c is the region. The maximum value of the predicted moisture range in region c; This represents the minimum value of the moisture range in region c; This represents the cumulative moisture content in region c.

[0149] It is worth noting that, and The calculation method and and The calculation method is the same, and you can refer to step 2082 for details. This embodiment of the invention does not limit this.

[0150] Total variation regularization is used to ensure smooth water distribution within the same phase, while allowing appropriate abrupt changes at phase boundaries. The total variation regularization loss function is:

[0151]

[0152] in, For total variation regularization loss, Let represent the moisture distribution of the pixel located at (i+1, j). Let (i, j) represent the moisture distribution of the pixel located at (i, j). Let represent the water distribution of the pixel located at (i, j+1).

[0153] In summary, the final total loss is as follows:

[0154]

[0155] in, and For hyperparameters; This is the total variation regularization loss; This is the mean square error loss; L represents the difference penalty loss; L represents the final total loss.

[0156] Step 2085: Using the gradient descent algorithm, backpropagation is performed based on the loss results to iteratively update the model parameters and construct a deep learning model for water state evolution.

[0157] In this embodiment of the invention, the initial water distribution is used as input. A simulated NMR signal is calculated using a forward model to calculate the final total loss. The parameters related to the water mapping in the network are then updated using backpropagation (or the water distribution map is directly adjusted). The update process employs a gradient descent algorithm (such as Adam), iterating multiple times until the simulated NMR signal matches the experimental data, thus completing the construction of the deep learning model for water state evolution.

[0158] Furthermore, the deep learning model for water state evolution was validated through internal and external validation. Specifically, internal validation involved using the initial water distribution obtained through inversion optimization. A forward model was used to generate predicted NMR signals (predicting global moisture distribution, Z-axis moisture distribution, and 3D moisture distribution), which were then compared with the actual NMR signals obtained experimentally (global moisture distribution, Z-axis moisture distribution, and 3D moisture distribution) to verify the rationality of the overall moisture proportion and local distribution. For external validation, thermogravimetric analysis (TGA) or other chemical characterization methods were used to examine the moisture proportion in each major region. Finally, the NMR-corrected spatiotemporal distribution map of moisture and uncertainty maps for each region were output as important bases for the quantitative analysis of the hydration mechanism of hardened cement slurry.

[0159] Step 209: Using a water state evolution deep learning model, perform water state evolution on the original volume data to be represented, and generate water state evolution data.

[0160] Specifically, the raw volume data to be characterized is input into the water state evolution deep learning model to perform water state evolution, and the output is water state evolution data and uncertainty maps for each region. The water state evolution data is the predicted water distribution result.

[0161] In this embodiment of the invention, uncertainty indices are generated based on water state evolution data. These uncertainty indices include a distribution mean index and a distribution standard deviation index. Uncertainty images are then generated according to different regional categories, based on the distribution mean index and the distribution standard deviation index. Specifically, the mean and standard deviation of the water state evolution data are calculated; the calculated mean and standard deviation are determined as uncertainty indices; and the uncertainty indices are plotted according to the pixel distribution of the image to obtain the uncertainty image.

[0162] It is worth noting that the water state evolution data can be presented as a spatiotemporal distribution map of water for users to view.

[0163] This invention, based on multimodal data fusion and physical constraint deep learning, achieves accurate mapping of microstructure to moisture distribution. It can quantitatively predict the spatiotemporal evolution characteristics of different types of water during the hydration process of cement-based materials, improving the accuracy and efficiency of material structure analysis and performance prediction. It has significant engineering application value and scientific research significance.

[0164] It is worth noting that the acquisition, storage, use, and processing of data in the technical solution of this application all comply with relevant laws and regulations. The user information in the embodiments of this application was obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the client.

[0165] It is worth noting that the information collected in this application is information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data all comply with the relevant laws, regulations and standards of the relevant countries and regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation portals are provided for users to choose to authorize or refuse.

[0166] It is worth noting that the technical solution provided in this application provides users with a corresponding operation entry point, allowing users to choose to agree to or reject the automated decision-making result; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0167] The technical solution of the water state evolution characterization method for cement hydration provided in this invention involves using computed tomography (CT) to scan and image the prepared cement slurry, generating an original sample volume dataset; performing magnetic resonance characterization on the original sample volume dataset to generate a water state evolution sample dataset for cement hydration; performing mode transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset; training a model on a preset encoder-decoder structure based on the water state evolution sample dataset and the multi-valued sample volume dataset to construct a deep learning model for water state evolution; and using the deep learning model for water state evolution to perform water state evolution on the original volume data to be characterized, generating water state evolution data. This achieves the reconstruction of the spatiotemporal evolution distribution of water in cement materials without relying on high-resolution NMR, effectively promoting the four-dimensional distribution of water state data in time and space during the hydration process, improving spatial resolution, clarifying the correspondence between water distribution and microstructure, effectively distinguishing different water state distributions in small-scale structures, and accurately locating the location of various types of water in the microstructure.

[0168] Figure 3 This is a schematic diagram of a device for characterizing the water-state evolution of cement hydration according to an embodiment of the present invention. This device is used to perform the aforementioned method for characterizing the water-state evolution of cement hydration. Figure 3 As shown, the device includes: a scanning unit 11, a magnetic resonance characterization unit 12, a mode conversion and multi-valued unit 13, a model training unit 14, and a water state evolution unit 15.

[0169] The scanning unit 11 is used to scan and image the prepared cement slurry using computed tomography (CT) technology to generate an original sample volume dataset.

[0170] The magnetic resonance characterization unit 12 is used to perform magnetic resonance characterization on the original sample volume dataset to generate a water state evolution sample dataset of cement hydration.

[0171] The modality transformation and multi-valued unit 13 is used to perform modality transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset.

[0172] The model training unit 14 is used to train the model on the preset encoder-decoder structure based on the water evolution sample dataset and the multi-valued sample volume dataset, and to build a deep learning model for water evolution.

[0173] The water state evolution unit 15 is used to perform water state evolution on the original volume data to be represented through a water state evolution deep learning model, and generate water state evolution data.

[0174] In this embodiment of the invention, the water state evolution sample dataset includes global water distribution information, Z-axis water distribution information, and three-dimensional water distribution information. The magnetic resonance characterization unit 12 is specifically used to sample the original sample volume dataset through a multi-layer echo sequence to obtain global water distribution information at different times; to sample the original sample volume dataset through a constant gradient layer selection sequence to obtain Z-axis water distribution information at different times; and to perform spatial and temporal encoding on the original sample volume dataset to obtain three-dimensional water distribution information at different times.

[0175] In this embodiment of the invention, the modality conversion and multi-valued unit 13 is specifically used to convert the original sample volume data in the original sample volume dataset into electron microscope scan sample images through a pre-trained modality conversion network, thereby generating an electron microscope scan sample volume dataset; and to perform image segmentation on the electron microscope scan sample volume dataset through a pre-trained deep learning segmentation network, thereby generating a multi-valued sample volume dataset.

[0176] In this embodiment of the invention, the mode conversion and multi-valuedness unit 13 is specifically used to perform coarse segmentation on electron microscope scanned sample images in the electron microscope scanned sample volume dataset through a deep learning segmentation network to generate a coarse segmented multi-valued image; to select image regions corresponding to preset key categories from the coarse segmented multi-valued image; to refine the image regions corresponding to the key categories to generate a multi-valued image refinement region; and to construct a multi-channel input tensor based on the multi-valued image refinement region and the coarse segmented multi-valued image to generate a multi-valued sample volume dataset.

[0177] In this embodiment of the invention, the model training unit 14 is specifically used to perform data registration on the water state evolution sample dataset and the multi-valued sample volume dataset to construct a model dataset; and to train the encoder-decoder structure using the model dataset to construct a deep learning model for water state evolution.

[0178] In this embodiment of the invention, the model training unit 14 is specifically used to perform multi-layer convolutional encoding on the multi-valued sample volume dataset through an encoder to generate sample feature maps; to map and decode the sample feature maps through a physical constraint layer and a decoder to generate preliminary predicted water distribution results; to perform cumulative pooling on the preliminary predicted water distribution results according to different regions and directions to generate predicted global water distribution information, predicted Z-axis water distribution information, and predicted three-dimensional water distribution information; to generate a loss result based on the predicted global water distribution information, predicted Z-axis water distribution information, predicted three-dimensional water distribution information, sample global water distribution information, sample Z-axis water distribution information, and sample three-dimensional water distribution information through a constructed comprehensive loss function; and to perform backpropagation iteratively update the model parameters based on the loss result through a gradient descent algorithm to construct a deep learning model for water state evolution.

[0179] In this embodiment of the invention, the device further includes: an uncertainty index generation unit 16 and an uncertainty image generation unit 17.

[0180] The uncertainty index generation unit 16 is used to generate uncertainty indices based on the preliminary predicted water distribution results. The uncertainty indices include the distribution mean index and the distribution standard deviation index.

[0181] The uncertainty image generation unit 17 is used to generate uncertainty images according to different categories of regions, based on the distribution mean index and the distribution standard deviation index.

[0182] In this embodiment of the invention, the prepared cement slurry is scanned and imaged using computed tomography (CT) technology to generate an original sample volume dataset. The original sample volume dataset is then characterized using magnetic resonance imaging (MRI) to generate a water state evolution sample dataset for cement hydration. The original sample volume dataset is further processed by mode transformation and multi-valued segmentation to generate a multi-valued sample volume dataset. Based on the water state evolution sample dataset and the multi-valued sample volume dataset, a pre-defined encoder-decoder structure is trained to construct a deep learning model for water state evolution. This deep learning model is used to perform water state evolution on the original volume data to be characterized, generating water state evolution data. Without relying on high-resolution NMR, the spatiotemporal evolution distribution of water in cement materials is reconstructed, effectively promoting the four-dimensional distribution of water state data in the hydration process in time and space, improving spatial resolution, clarifying the correspondence between water distribution and microstructure, effectively distinguishing different water state distributions in small-scale structures, and accurately locating the location of various types of water in the microstructure.

[0183] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer device, specifically, a computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0184] This invention provides a computer device, including a memory and a processor. The memory is used to store information including program instructions, and the processor is used to control the execution of the program instructions. When the program instructions are loaded and executed by the processor, they implement the steps of the above-described embodiment of the water state evolution characterization method for cement hydration. For a detailed description, please refer to the above-described embodiment of the water state evolution characterization method for cement hydration.

[0185] The following is for reference. Figure 4 It shows a schematic diagram of the structure of a computer device 600 suitable for implementing the embodiments of this application.

[0186] like Figure 4 As shown, the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate tasks and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer device 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0187] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal feedback (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed in storage section 608 as needed.

[0188] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program tangibly embodied on a machine-readable medium, the computer program including program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611.

[0189] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0190] For ease of description, the above devices are described separately by function as various units. Of course, in implementing this application, the functions of each unit can be implemented in one or more software and / or hardware.

[0191] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0192] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0193] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0194] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0195] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0196] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, they do not mean that the applicant has used or necessarily used the solution.

[0197] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0198] This application can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0199] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0200] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for characterizing the water-state evolution of cement hydration, characterized in that, The method includes: The prepared cement slurry was scanned and imaged using computed tomography (CT) technology to generate a dataset of the original sample volume. Magnetic resonance characterization was performed on the original sample volume dataset to generate a water-state evolution sample dataset of cement hydration; The original sample volume dataset is subjected to mode transformation and multi-valued segmentation to generate a multi-valued sample volume dataset; Based on the water state evolution sample dataset and the multi-valued sample volume dataset, a model is trained on the preset encoder-decoder structure to construct a deep learning model for water state evolution. The water state evolution deep learning model is used to perform water state evolution on the original volume data to be represented, generating water state evolution data.

2. The method for characterizing the aqueous evolution of cement hydration according to claim 1, characterized in that, The water state evolution sample dataset includes global water distribution information, Z-axis water distribution information, and three-dimensional water distribution information. The process of performing magnetic resonance characterization on the original sample volume dataset to generate water-state evolution sample data of cement hydration includes: By sampling the original sample volume dataset using multi-layer echo sequences, global moisture distribution information of the sample at different times is obtained. By using a constant gradient layer selection sequence, the original sample volume dataset is sampled to obtain the water distribution information of the sample along the Z-axis at different times; Spatial and temporal encodings were performed on the original sample volume dataset to obtain three-dimensional moisture distribution information of the samples at different times.

3. The method for characterizing the aqueous evolution of cement hydration according to claim 1, characterized in that, The process of performing mode transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset includes: The original sample volume data in the original sample volume dataset is converted into electron microscope scan sample images by a pre-trained modality conversion network, thereby generating an electron microscope scan sample volume dataset. The electron microscope scan sample volume dataset is segmented using a pre-trained deep learning segmentation network to generate a multi-valued sample volume dataset.

4. The method for characterizing the aqueous evolution of cement hydration according to claim 3, characterized in that, The process of segmenting the electron microscope scanned sample volume dataset using a pre-trained deep learning segmentation network to generate a multi-valued sample volume dataset includes: The deep learning segmentation network is used to coarsely segment the electron microscope scanned sample images in the electron microscope scanned sample volume dataset to generate coarsely segmented multi-valued images. Image regions corresponding to preset key categories are selected from the coarsely segmented multi-valued image; The image regions corresponding to key categories are refined to generate multi-valued image refinement regions; Based on the refined regions of the multivalued image and the coarsely segmented multivalued image, a multi-channel input tensor is constructed to generate a multivalued sample volume dataset.

5. The method for characterizing the aqueous evolution of cement hydration according to claim 1, characterized in that, The step of training a model on a preset encoder-decoder structure based on the water state evolution sample dataset and the multi-valued sample volume dataset to construct a deep learning model for water state evolution includes: Data registration was performed on the water state evolution sample dataset and the multi-valued sample volume dataset to construct the model dataset; Using the model dataset, the encoder-decoder structure is trained to construct a deep learning model for water state evolution.

6. The method for characterizing the aqueous evolution of cement hydration according to claim 5, characterized in that, The step of training the encoder-decoder structure using the model dataset to construct a deep learning model for water state evolution includes: The multi-level convolutional encoding of the multi-valued sample volume dataset is performed by an encoder to generate sample feature maps. The sample feature map is mapped and decoded through a physical constraint layer and a decoder to generate a preliminary prediction of water distribution. According to different regions and directions, the preliminary predicted water distribution results are accumulated and pooled to generate predicted global water distribution information, predicted water distribution information in the Z-axis direction, and predicted three-dimensional water distribution information. By constructing a comprehensive loss function, a loss result is generated based on the predicted global moisture distribution information, the predicted Z-axis moisture distribution information, the predicted three-dimensional moisture distribution information, the sample global moisture distribution information, the sample Z-axis moisture distribution information, and the sample three-dimensional moisture distribution information. By using the gradient descent algorithm, backpropagation is performed based on the loss results to iteratively update the model parameters and construct a deep learning model for water state evolution.

7. The method for characterizing the aqueous evolution of cement hydration according to claim 6, characterized in that, The method further includes: Based on the preliminary predicted moisture distribution results, uncertainty indicators are generated, including the distribution mean index and the distribution standard deviation index. Uncertainty images are generated based on the distribution mean index and distribution standard deviation index according to different category regions.

8. A device for characterizing the water-state evolution of cement hydration, characterized in that, The device includes: The scanning unit is used to scan and image the prepared cement slurry using computed tomography (CT) technology to generate a raw sample volume dataset. A magnetic resonance characterization unit is used to perform magnetic resonance characterization on the original sample volume dataset to generate a water state evolution sample dataset of cement hydration. The modality transformation and multi-valued unit is used to perform modality transformation and multi-valued segmentation on the original sample volume dataset to generate a multi-valued sample volume dataset. The model training unit is used to train the model on the preset encoder-decoder structure based on the water state evolution sample dataset and the multi-valued sample volume dataset, and to build a deep learning model for water state evolution. The water state evolution unit is used to perform water state evolution on the original volume data to be represented through the water state evolution deep learning model to generate water state evolution data.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the water-state evolution characterization method for cement hydration as described in any one of claims 1 to 7.

10. A computer device comprising a memory and a processor, the memory for storing information including program instructions, and the processor for controlling the execution of the program instructions, characterized in that, When the program instructions are loaded and executed by the processor, the water state evolution characterization method for cement hydration as described in any one of claims 1 to 7 is implemented.

11. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the water-state evolution characterization method for cement hydration as described in any one of claims 1 to 7.

Images