Multi-scale direct coupling calculation method and device for hardened cement paste

Abstract

The invention provides a hardened cement paste multi-scale direct coupling calculation method and device, and the method comprises the steps: carrying out the in-situ loading and computed tomography of hardened cement paste, generating multi-scale volume data of different stages, and carrying out the multi-phase segmentation to obtain a multi-valued volume data mask; the damage path prediction deep learning model is used to predict the damage path of the multi-valued volume data mask and divide a multi-scale grid, the finite element software is used to carry out direct simulation calculation on the multi-scale grid structure to generate a multi-scale damage result, and the high information density data obtained based on the image can be used to predict the damage path of the multi-valued volume data mask. A potential damage path is predicted through a deep learning model, modeling is performed on the basis, calculation complexity is simplified, on the premise that simulation precision is guaranteed, microstructure evolution characteristics and macroscopic mechanical property response of cement paste in the hydration process are considered, the numerical calculation cost of a multi-scale structure is effectively reduced, and the method is suitable for popularization and application. And an efficient and extensible technical path is provided for evolution process modeling of complex multi-phase materials.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a multi-scale direct coupling calculation method and apparatus for hardened cement slurry. Background Technology

[0002] Multi-scale performance studies of cement-based materials are crucial for understanding their heterogeneous composite mechanical behavior. The macroscopic mechanical response of materials is closely related to their mesoscopic and microscopic structures, making in-depth analysis of the influence of material composition and structure at different scales on macroscopic properties particularly important. Multi-scale performance studies have become a key means of understanding the mechanical behavior of concrete. The combination of experiments, numerical simulations, and theoretical analysis allows researchers to systematically analyze the evolution mechanism of material properties from the microscopic to the macroscopic level. However, existing research still faces many challenges: microscale testing is limited by sample preparation and testing errors, resulting in significant data dispersion; while numerical simulations improve computational accuracy, they still rely on empirical formulas and are limited by computational costs, making it difficult to fully cover the complex process of hydration evolution; although theoretical analysis methods can provide effective predictions across scales, their applicability remains problematic when considering interface effects and heterogeneity.

[0003] In one category of existing technologies, multi-scale performance simulation primarily employs traditional hierarchical multi-scale modeling methods. This method uses the Representative Volume Unit (RVE) as its core, establishing constitutive models at the micro, meso, and macro scales, and transferring parameters through homogenization methods to construct a cross-scale numerical simulation system. Another category of existing technologies utilizes image-driven multi-scale reconstruction methods to construct microstructural models of cement materials, combining image segmentation, reconstruction, and parameter fitting to enhance model realism.

[0004] However, although traditional hierarchical multi-scale modeling methods have a certain degree of theoretical completeness, they only map physical quantities between different scales through numerical averaging, making it difficult to truly reflect the heterogeneity of microstructures and the continuous impact of evolution on macroscopic performance. Simulations at different scales are all based on certain assumptions, which means that errors will accumulate at each scale, resulting in inaccurate predictions, and errors may accumulate during scale transfer. Simplifying hydration into static inputs or predefined parameters lacks dynamic simulation, making it difficult to reflect the evolutionary characteristics of cement-based materials in the early stages, and the hydration process is difficult to couple. RVE units mostly rely on ideal geometric assumptions, lack effective coupling with real experimental data, and cannot accurately replicate the real microporous structure and phase distribution, resulting in insufficient structural realism. While image-driven modeling improves the accuracy of microstructure reconstruction, it suffers from insufficient coupling between image scale and physical quantities. Images cannot be directly mapped to internal process variables such as moisture state and reaction degree, resulting in image modeling being disconnected from material evolution. Reconstruction relies on specific sample images, leading to poor model transferability and weak model generalization ability. As the complexity of microstructure details increases, the model's degrees of freedom grow exponentially, resulting in reduced simulation efficiency, making it difficult to apply quickly to large-scale engineering problems, and incurring high computational costs. Summary of the Invention

[0005] One objective of this invention is to provide a multi-scale direct coupling computational method for hardened cement slurry. Based on high-information-density data acquired from images, and utilizing a deep learning model for training, non-critical regions are simplified while critical regions are preserved. The image is directly coupled across multiple scales with the hydration process, enabling the co-evolution of the microstructure and properties of the cement slurry during hydration, thus improving structural realism. A multi-modal fusion mechanism enhances prediction accuracy and model generalization ability, significantly reducing computational costs. Another objective of this invention is to provide a multi-scale direct coupling computational device for hardened cement slurry. A further objective is to provide a computer-readable medium. A final objective is to provide a computer device.

[0006] To achieve the above objectives, this invention discloses a multi-scale direct coupling calculation method for hardened cement slurry, comprising: The prepared hardened cement slurry was subjected to in-situ loading and computed tomography to generate multi-scale volumetric data at different stages. Multi-phase segmentation is performed on multi-scale volumetric data to generate multi-valued volumetric data masks; By using a pre-built deep learning model for predicting destruction paths, multi-valued volumetric data masks are predicted to generate predicted destruction paths. The predicted damage path is divided into multi-scale grids to generate the resulting multi-scale grid structure. The finite element method (FEM) software is used to directly simulate and calculate multi-scale mesh structures, generating multi-scale failure results.

[0007] Preferably, the prepared hardened cement slurry is subjected to in-situ loading and computed tomography to generate multi-scale volumetric data at different stages, including: The hardened cement slurry was loaded in situ using an online loading device, and the hardened cement slurry was scanned during the loading process using a computed tomography (CT) scanner to generate multi-scale in situ characterization data at different stages. By using a reconstruction algorithm, multi-scale in-situ characterization data is reconstructed to generate multi-scale volume data at different stages.

[0008] Preferably, the multi-scale volumetric data is segmented into multiple phases to generate a multi-valued volumetric data mask, including: By using a super-resolution model, multi-scale volumetric data can be super-resolutiond to generate microstructures. Using a computer vision semantic segmentation model, the microstructure is semantically segmented to generate a multi-valued volume data mask.

[0009] Preferably, the multivalued volume data mask is the multivalued volume data mask at the initial time. By using a pre-built deep learning model for predicting destruction paths, multi-valued volumetric data masks are predicted to generate predicted destruction paths, including: By using a multilayer perceptron and a temporal embedding function, the multivalued volume data mask at the initial time step is represented by a vector to generate the input sequence; The input sequence is encoded and decoded using an encoding / decoding structure to generate the predicted destruction path corresponding to the predicted time.

[0010] Preferably, the method further includes: In-situ loading and computed tomography were performed on hardened cement slurry samples at different ages to generate multi-scale sample volume data at different ages and stages. Multi-phase segmentation is performed on multi-scale sample volume data to generate multi-valued sample volume data masks; Based on the multi-valued sample volume data mask, the destruction paths of samples at different ages and loading stages are generated; A dataset is constructed based on the multi-valued sample volume data mask and the destruction paths of samples at different ages and loading stages; By training a deep learning model using a dataset, a deep learning model for predicting destructive paths is constructed.

[0011] Preferably, the dataset includes a multi-valued sample volume data mask and sample destruction paths at different ages and loading stages; By training a deep learning model using a dataset, a destructive path prediction deep learning model is constructed, including: Using a deep learning model, destruction path prediction is performed based on multi-valued sample volume data mask, generating sample prediction destruction paths corresponding to different times. Loss calculations were performed on the predicted destruction paths of samples at different times and the destruction paths of samples at different ages and loading stages to obtain the loss results; Based on the loss results, the model parameters of the deep learning model are iteratively optimized to generate a deep learning model for predicting damaged paths.

[0012] Preferably, the predicted damage path is divided into multi-scale grids to generate a multi-scale grid structure, including: High-risk damage areas are identified from predicted damage paths; A finite element model is established, and the high-risk damage area is divided into a fine mesh. The non-high-risk damage area is processed into a homogenized coarse mesh and the mesh transition area is corrected to generate a multi-scale mesh structure.

[0013] This invention also discloses a multi-scale direct coupling computing device for hardened cement slurry, comprising: The multi-scale volumetric data generation unit is used to perform in-situ loading and computed tomography on the prepared hardened cement slurry to generate multi-scale volumetric data at different stages. The multiphase segmentation unit is used to perform multiphase segmentation on multiscale volume data and generate a multi-valued volume data mask. The destruction path prediction unit is used to predict multi-valued volume data masks and generate predicted destruction paths by using a pre-built destruction path prediction deep learning model. Mesh generation unit is used to perform multi-scale mesh generation on the predicted damage path, generating the resulting multi-scale mesh structure. Directly coupled computational units are used to directly simulate and calculate multi-scale mesh structures using finite element software, generating multi-scale failure results.

[0014] 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.

[0015] 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.

[0016] 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.

[0017] This invention involves in-situ loading and computed tomography (CT) scans of prepared hardened cement slurry to generate multi-scale volumetric data at different stages. The multi-scale volumetric data is then segmented into multiple phases to generate a multi-valued volumetric data mask. A pre-built deep learning model for predicting failure paths is used to predict the multi-valued volumetric data mask and generate predicted failure paths. The predicted failure paths are then divided into multi-scale meshes to generate a multi-scale mesh structure. The multi-scale mesh structure is directly simulated using finite element method software to generate multi-scale failure results. Based on high-information-density data acquired from images, and utilizing a deep learning model for training, non-critical regions are simplified while critical regions are retained. The multi-scale images are directly coupled with the hydration process, enabling the co-evolution of microstructure and properties of the cement slurry during hydration, improving structural realism. A multi-modal fusion mechanism enhances prediction accuracy and model generalization ability, significantly reducing computational costs. Attached Figure Description

[0018] 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.

[0019] Figure 1 A flowchart of a multi-scale direct coupling calculation method for hardened cement slurry provided in an embodiment of the present invention; Figure 2 A flowchart of another multi-scale direct coupling calculation method for hardened cement slurry provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a multi-scale direct coupling computing device for hardened cement slurry provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0020] 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.

[0021] It should be noted that the multi-scale direct coupling calculation method and apparatus for hardened cement slurry 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 multi-scale direct coupling calculation method and apparatus for hardened cement slurry disclosed in this application is not limited.

[0022] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution is described below. This invention optimizes the multi-scale model construction method by directly constructing a computational model covering six orders of magnitude across macroscopic, mesoscopic, and microscopic scales, enabling hydration evolution, pore structure, and mechanical response to evolve self-consistently within a unified modeling framework. By combining high-resolution experimental data for model correction, prediction accuracy is improved, and the stability and applicability of the model across all scales are ensured. The objective of this invention is to solve the problems of inaccurate generation of hydration dynamics microstructure models and high computational cost by using high-information-density data acquired from images and by training deep learning models to simplify non-critical regions and retain critical regions, thereby enabling direct simulation calculations using finite element software.

[0023] The following uses a hardened cement slurry multi-scale direct coupling calculation device as an example to illustrate the implementation process of the hardened cement slurry multi-scale direct coupling calculation method provided in this embodiment of the invention. It is understood that the execution subject of the hardened cement slurry multi-scale direct coupling calculation method provided in this embodiment of the invention includes, but is not limited to, a hardened cement slurry multi-scale direct coupling calculation device.

[0024] Figure 1 A flowchart of a multi-scale direct coupling calculation method for hardened cement slurry provided in an embodiment of the present invention is shown below. Figure 1 As shown, the method includes: Step 101: Perform in-situ loading and computed tomography on the prepared hardened cement slurry to generate multi-scale volume data at different stages.

[0025] In this embodiment of the invention, the computed tomography scan is an X-ray computed tomography scan (X-CT). Specifically, in-situ loading tests are conducted on the prepared hardened cement slurry using an online loading device, with the same loading mode. Rapid scanning is performed at equal intervals using an X-CT device to obtain X-CT multi-scale in-situ characterization data at different loading stages. Multi-scale volume data at different stages are then formed using a reconstruction algorithm.

[0026] Step 102: Perform multi-phase segmentation on the multi-scale volume data to generate a multi-valued volume data mask.

[0027] In this embodiment of the invention, a super-resolution model and a computer vision semantic segmentation model are used to perform semantic segmentation on rapidly scanned X-CT multi-scale volume data to obtain the multiphase structure of hydration product components and crack composition.

[0028] Step 103: Using a pre-built deep learning model for predicting destruction paths, predict the multi-valued volume data mask to generate predicted destruction paths.

[0029] In this embodiment of the invention, the deep learning model for predicting the destruction path is a model trained using ViT (Vision Transformer) as the backbone network, which can meet the requirement of predicting the final destruction path from the initial microstructure.

[0030] Specifically, the multi-valued volumetric data mask image of the initial state is input into the deep learning model for predicting the destruction path, and the predicted destruction path at the prediction time is output.

[0031] Step 104: Divide the predicted damage path into a multi-scale grid and generate the resulting multi-scale grid structure.

[0032] In this embodiment of the invention, for high-risk damage areas in the predicted damage path predicted by the deep learning model for damage path prediction, a finite element model is used with cohesive elements to refine the mesh. For other areas, the complex microstructure is first equivalent to a mean matrix using the Mori-Tanaka method, and then processed using a homogenized coarse mesh. After the mesh is established, the transition region of the mesh is corrected to obtain the multi-scale mesh structure after division, so that the calculation can proceed smoothly.

[0033] Step 105: Perform direct simulation calculations on the multi-scale mesh structure using finite element software to generate multi-scale failure results.

[0034] In this embodiment of the invention, a multi-scale mesh structure is imported into finite element software, cohesive elements are set, and a direct finite element calculation coupled model is performed to simulate the multi-scale mesh structure and generate multi-scale failure results. Specifically, in the finite element software, bonded or cohesive elements are embedded in high-risk areas, and failure criteria such as critical stress or energy release rate are defined. Specifically, the global stress field is solved, stress concentration regions are identified, multi-scale failure results are output, and the element failure conditions of the refined regions are dynamically updated. The multi-scale failure results represent crack evolution morphology.

[0035] Furthermore, the crack evolution morphology obtained by the multi-scale finite element direct calculation coupled model simulation is compared with the actual X-CT observation data in step 101. Based on the comparison results, the deep learning model for predicting the failure path and the multi-scale mesh generation are backtracked and optimized.

[0036] In the technical solution provided by this invention, the prepared hardened cement slurry is subjected to in-situ loading and computed tomography to generate multi-scale volume data at different stages; the multi-scale volume data is segmented into multiple phases to generate a multi-valued volume data mask; a pre-constructed deep learning model for predicting failure paths is used to predict the multi-valued volume data mask to generate predicted failure paths; the predicted failure paths are divided into multi-scale meshes to generate a multi-scale mesh structure; the multi-scale mesh structure is directly simulated and calculated using finite element software to generate multi-scale failure results. Based on high-information-density data acquired from images, and by using a deep learning model for training, non-critical regions are simplified while critical regions are retained. The multi-scale images are directly coupled with the hydration process, realizing the co-evolution of microstructure and properties of cement slurry during hydration, improving structural realism, and enhancing prediction accuracy and model generalization ability through a multi-modal fusion mechanism, thereby greatly reducing computational costs.

[0037] Figure 2 A flowchart of another multi-scale direct coupling calculation method for hardened cement slurry provided in an embodiment of the present invention is shown below. Figure 2 As shown, the method includes: Step 201: Perform in-situ loading and computed tomography on the prepared hardened cement slurry samples at different ages to generate multi-scale sample volume data at different ages and stages.

[0038] In this embodiment of the invention, each step is executed by a multi-scale direct-coupled computing device for hardened cement slurry.

[0039] In this embodiment of the invention, a hardened cement paste is prepared using a binary system of cement and water. Hydration is terminated by isopropanol solution, and cement paste samples with different ages at which hydration has been terminated are obtained. Specifically, ordinary silicate cement and deionized water are used as raw materials, and cement paste samples are prepared at a preset water-cement ratio. Hydration is terminated by soaking in isopropanol solution at different ages, and hardened cement paste samples with different ages are obtained.

[0040] As an alternative, cement paste samples were prepared using ordinary silicate cement and deionized water as raw materials with a water-cement ratio of 0.5. The hydration was terminated by soaking in isopropanol solution at curing times of 1 day (d), 3 days, 7 days, and 28 days, and hardened cement paste samples at different curing times were obtained.

[0041] In this embodiment of the invention, the computed tomography scan is an X-ray computed tomography scan (X-CT). In-situ loading tests are carried out on hardened cement slurry samples of different ages using an online loading device, with the same loading mode. Rapid scanning is performed at equal intervals using an X-CT device to obtain X-CT multi-scale in-situ characterization data at different loading stages. Multi-scale sample volume data at different ages and loading stages are formed by reconstruction algorithm.

[0042] As an alternative approach, in-situ loading tests are conducted on samples of different ages using an online loading device. Loading is performed through a uniaxial compression test with a displacement increment of 10 micrometers (µm). After loading reaches this displacement, loading is stopped and waited for 30 seconds before a rapid scan is performed using an X-CT device. Loading is then resumed to obtain X-CT multi-scale in-situ characterization data at different loading stages. The original multi-scale sample volume data of different ages and loading states are generated using an X-CT reconstruction algorithm, with the grayscale distribution in the range of [0, 65535].

[0043] Step 202: Perform multi-phase segmentation on the multi-scale sample volume data to generate a multi-valued sample volume data mask.

[0044] In this embodiment of the invention, a super-resolution model and a computer vision semantic segmentation model are used to perform semantic segmentation on rapidly scanned X-CT multi-scale sample data to obtain the multiphase structure of hydration product components and crack composition.

[0045] Specifically, a super-resolution model is used to perform super-resolution on multi-scale sample volume data to obtain fine microstructures; a computer vision semantic segmentation model is used to perform semantic segmentation on the microstructures to generate a multiphase structure composed of cracks, unhydrated particles, high-density hydrated calcium silicate (HD-CSH), low-density hydrated calcium silicate (LD-CSH), calcium hydroxide (CH), ettringite (Aft), and pores, forming a multi-valued sample volume data mask.

[0046] Step 203: Divide the data according to the multi-valued sample volume mask to generate sample destruction paths for different ages and loading stages.

[0047] In this embodiment of the invention, the multi-valued sample volume data mask includes cracks, unhydrated particles, high-density hydrated calcium silicate (HD-CSH), low-density hydrated calcium silicate (LD-CSH), calcium hydroxide (CH), ettringite (Aft), and pores. Taking the crack region as the core and the microstructure region at the crack boundary as the direction, the crack region is extended outward to form a damage path. The specific extension rules are as follows: Multivalued sample volume data includes multiple voxels; multivalued sample volume data mask: ;in, This indicates that voxel x belongs to the identified crack region; These represent unhydrated particles, high-density hydrated calcium silicate (HD-CSH), low-density hydrated calcium silicate (LD-CSH), calcium hydroxide (CH), ettringite (Aft), and pores, respectively. The irrelevant background region is represented. The voxel set of the crack region is denoted as F, and its boundary is denoted as... This allows us to obtain the initial distribution of cracks in space.

[0048] Define the distance function for the crack region F:

[0049] in, y is the distance between voxel x and the crack region F; y is the voxel within the crack region F. Let x be the three-dimensional Euclidean distance between voxels x and y.

[0050] Setting an expansion threshold δ will satisfy: All points are considered as the expanded region. That is, outside the crack region, at a fixed distance δ of the outer shell. The expansion region F... δ The regions of each phase in the multi-valued sample volume data mask, excluding crack and background regions ( Take the intersection of each region and obtain the overlapping regions one by one:

[0051] in, F represents the overlapping region between the region of the i-th phase and the expansion region. δ For expansion region, Let be the region of the i-th phase composed of voxels x.

[0052] It is worth noting that the expansion threshold δ can be set according to actual needs, and this embodiment of the invention does not limit it. As an optional solution, the expansion threshold δ is 2.

[0053] In this embodiment of the invention, this step can exclude background or invalid regions unrelated to the crack, while including microstructures that may be associated with the crack in the candidate range of "potential damage zones". For For voxels in the dataset, a 26-neighborhood three-dimensional connectivity search is used to further identify their connected components, thereby eliminating discrete noise or tiny isolated voxels and obtaining a more stable and continuous surrounding microstructure region, denoted as . The total region R of the surrounding microstructures is obtained by taking the union of all surrounding microstructures. δ :

[0054] Let the volume of the initial crack region F be V. F If R δ volume vol(R) δ (Exceeding the original crack volume V) F β times (β is a user-defined multiple), i.e.: vol(R) δ )>βV F Then for R δMorphological shrinkage or preferential elimination is performed on the periphery until vol(R) is satisfied. δ )≤βV F This allowed us to obtain the destruction pathways of samples at different ages and loading stages. .

[0055] It is worth noting that the volume factor β can be set according to actual needs, and this embodiment of the invention does not limit it. As an optional solution, the volume factor β is 2.

[0056] Step 204: Construct a dataset based on the multi-valued sample body data mask and the sample destruction paths at different ages and loading stages.

[0057] In this embodiment of the invention, the dataset is built using the multi-valued sample volume data mask at the initial moment and the sample destruction path at the final moment as input and supervision, and the intermediate process as key moment supervision. Specifically, the multi-valued sample volume data mask at the initial moment is paired with the sample destruction path at the final moment, and the sample destruction path during the loading process is used as key moment supervision to construct the dataset.

[0058] Step 205: Train the deep learning model using the dataset to build a deep learning model for predicting disrupted paths.

[0059] In this embodiment of the invention, the dataset includes a multi-valued sample body data mask and sample destruction paths at different ages and loading stages.

[0060] In this embodiment of the invention, the dataset is divided into a training set and a validation set according to a preset division ratio. The multi-valued sample volume data mask and the sample destruction paths of different ages and loading stages in the training set are used for model training, and the multi-valued sample volume data mask and the sample destruction paths of different ages and loading stages in the validation set are used for model validation.

[0061] It is worth noting that the partitioning ratio can be set according to actual needs, and this embodiment of the invention does not limit it. As an optional scheme, the partitioning ratio of the training set to the validation set is 3:1.

[0062] In this embodiment of the invention, step 205 specifically includes: Step 2051: Using a deep learning model, predict the destruction path based on the multi-valued sample volume data mask, and generate the sample prediction destruction path corresponding to different times.

[0063] In this embodiment of the invention, to meet the requirement of predicting the final destruction path from the initial microstructure, a deep learning model is constructed using ViT (Vision Transformer) as the backbone network. The model input is a multi-valued sample volume data mask at the initial time step, and the output is the destruction path at the final time step.

[0064] It is worth noting that the data format of the multi-valued sample volume data mask is a multi-valued sample image.

[0065] Specifically, the multi-valued sample volume data mask is trained by expanding it along the horizontal cross section, and the initial multi-valued sample volume data mask is denoted as: Where X(t1) is the multi-valued sample image at the initial time; H is the image length; W is the image width; and C is the number of channels, containing 1 to 8 categories (such as different phases formed during hydration). At different times t during loading... i On the range (i=2,…,n), corresponding binarized images of potential damaged regions were defined:

[0066] Where i=n represents the final time; Y(t) i ) represents the sample crack region or potential damage region under the loading state at the i-th time step; H is the length of the image; W is the width of the image.

[0067] The multi-valued sample image X(t1) at the initial time is divided into M sub-patterns of size P×P on the plane. Each sub-pattern is then processed by a multilayer perceptron (MLP) to obtain the sample image sub-region vector. :

[0068] in, This represents a d-dimensional real space, that is, a vector space containing d real numbers. Specifically, the original sub-blocks are compressed into a d-dimensional vector after passing through the MLP.

[0069] Introduce a learnable temporal embedding function to characterize the features at the loading moment:

[0070] in, Let i be the time vector at loading time i.

[0071] As an alternative, time embedding encoding uses a lookup table:

[0072] in, Let i be the time vector at loading time i; It is a learnable matrix that stores the d-dimensional vectors corresponding to the indices from 0 to n. It is updated during backpropagation during training and during inference based on the time indices. i Look up the table directly.

[0073] time vector With sample image sub-region vector The initial sequence of the model is obtained by concatenation:

[0074] Based on this, to distinguish between sample image sub-regions and volume data cross-sectional positions on the Z-axis, a corresponding spatial location code can be added to each sub-region. Encoding perpendicular to the cross section This forms the input sequence:

[0075] in, For the sample input sequence, The initial sequence of the sample. For sample image sub-region vectors The corresponding spatial location code, Initial sequence of the sample The corresponding cross-section is encoded in the vertical direction.

[0076] Suppose the entire ViT network contains an L-layer encoder-decoder structure, with each layer connected by an encoder. Encode the input sequence:

[0077] in, For the sample encoding sequence, Input sequence for the sample.

[0078] Decoder Sample encoding sequence The sub-region vector is inversely mapped to return the original image resolution, outputting a segmentation probability map of the same size:

[0079] in, Let i be the sample predicted destruction path. Here, H is the image encoding sequence, H is the image length, and W is the image width.

[0080] In this embodiment of the invention, the network is able to predict the destruction path at the corresponding time by using time index i.

[0081] Step 2052: Calculate the loss for the predicted destruction paths of samples at different times and the destruction paths of samples at different ages and loading stages to obtain the loss results.

[0082] In this embodiment of the invention, a weighted summation loss is used to calculate the loss between the output image at time i and the binarization results at different times.

[0083] in, As a result of loss; Y(t) represents the predicted destruction path for the sample at time i. i () represents the sample destruction path at different times; These are the training weights, used to balance the importance of intermediate and final moments during training.

[0084] Step 2053: Iteratively optimize the model parameters of the deep learning model based on the loss results to generate a deep learning model for predicting the destroyed path.

[0085] In this embodiment of the invention, the network parameters θ in the deep learning model are updated simultaneously through backpropagation, allowing the model to better capture the evolution of cracks from their inception to their final stage through multi-time supervision.

[0086] When deploying the model, simply provide the multi-valued sample image X(t1) at the initial time and specify the final time. The final destruction path is obtained directly through network forward propagation:

[0087] in, Input sequence for sample; This is the time vector at the final moment n; The vector of a sub-region of the sample image; For encoder; For decoders; The sample encoding sequence; The predicted destruction path for the sample at the final time n is: the final destruction path of the sample.

[0088] In this embodiment of the invention, the entire process requires only one forward inference, without the need to explicitly input or generate images at intermediate moments, thereby achieving efficient prediction of the final destruction path.

[0089] As an optional environment, the model training environment that disrupts the prediction deep learning model is: Operating system: Ubuntu 20.04LTS; Model runtime environment: PyTorch + CUDA; Graphics card: NVIDIA 4090×4; Graphics card driver version: 520.56.06; CUDA version: 11.8; PyTorch version: Stable (1.13.1).

[0090] Step 206: Validate the deep learning model for destruction prediction using the constructed test set.

[0091] In this embodiment of the invention, the test set can be constructed according to steps 201 to 204, which will not be repeated here. As an optional approach, samples with a new sample age of 28 days are used as tests, and the test set is constructed according to steps 201 to 204 to verify the destruction prediction deep learning model.

[0092] Step 207: Load the cement paste sample in situ using an online loading device, and scan the cement paste during the loading process using a computed tomography (CT) scanner to generate multi-scale in situ characterization data at different stages.

[0093] In this embodiment of the invention, the computed tomography scan is an X-ray computed tomography (X-CT) scan. Cement paste samples are loaded in situ using an online loading device, and the X-CT device scans the cement paste at equal intervals during the loading process to generate multi-scale in-situ characterization data at different stages.

[0094] As an alternative, loading can be performed through a uniaxial compression test, with a displacement increment of 10 μm. After loading reaches this displacement, loading is stopped and a 30-second wait is followed by a rapid X-CT scan. Loading is then resumed to obtain X-CT multi-scale in-situ characterization data at different loading stages.

[0095] Step 208: Reconstruct the multi-scale in-situ representation data using a reconstruction algorithm to generate multi-scale volume data at different stages.

[0096] Specifically, the multi-scale in-situ characterization data is reconstructed using X-CT reconstruction algorithms to generate multi-scale volumetric data at different stages. The grayscale distribution of the multi-scale volumetric data is in the range [0, 65535].

[0097] Step 209: Use a super-resolution model to perform super-resolution on multi-scale volume data to generate microstructures.

[0098] Specifically, multi-scale volumetric data is input into a super-resolution model for super-resolution, and the microstructure is output.

[0099] Step 210: Using a computer vision semantic segmentation model, perform semantic segmentation on the microstructure to generate a multi-valued volume data mask.

[0100] In this embodiment of the invention, the multivalued volume data mask is the multivalued volume data mask at the initial moment.

[0101] Specifically, the microstructure is input into a computer vision semantic segmentation model for semantic segmentation, generating a multi-valued volume data mask. The multi-valued volume data mask includes cracks, unhydrated particles, high-density hydrated calcium silicate (HD-CSH), low-density hydrated calcium silicate (LD-CSH), calcium hydroxide (CH), ettringite (Aft), and pores.

[0102] Step 211: Using MLP and temporal embedding function, the multi-valued volume data mask at the initial time is vectorized to generate the input sequence.

[0103] In this embodiment of the invention, the multi-valued volume data mask at the initial time is denoted as: ,in, The initial multi-valued image; H is the image length; W is the image width; C is the number of channels, containing 1 to 8 categories (e.g., different phases formed during hydration). At different times t during loading... i On the range (i=2,…,n), corresponding binarized images of potential damaged regions were defined:

[0104] Where i=n represents the final time; H represents the crack region or potential damage region under the loading state at the i-th moment; H is the length of the image; W is the width of the image.

[0105] Multivalued image at the initial moment The image is divided into M sub-patterns of size P×P on the plane. Each sub-pattern is processed by a multilayer perceptron (MLP) to obtain the image sub-region vector. :

[0106] We introduce a learnable temporal embedding function, and the temporal embedding encoding uses a lookup table format:

[0107] in, Let i be the time vector at loading time i; It is a learnable matrix that stores d-dimensional vectors corresponding to time indices from 0 to n. During inference, it is used based on the time indices. i Look up the table directly.

[0108] time vector Image sub-region vector The initial sequence of the model is obtained by concatenation:

[0109] Based on this, to distinguish between image sub-regions and the cross-sectional positions of volume data on the Z-axis, a corresponding spatial location code can be added to each sub-region. Encoding perpendicular to the cross section This forms the input sequence:

[0110] in, Given the input sequence, For the initial sequence, Image sub-region vector The corresponding spatial location code, For the initial sequence The corresponding cross-section is encoded in the vertical direction.

[0111] In this embodiment of the invention, the position encoding adopts a sine / cosine encoding form: The spatial location encoding of sub-regions uses two-dimensional sine and cosine coding. For rows:

[0112] in, The one-dimensional encoding of a sub-region on a two-dimensional image; This is the row encoding for the 2kth sub-region, where r is the row number and d is the line number. row The dimension representing the row direction; This is the row code for the 2k+1th sub-region.

[0113] For column:

[0114] in, The one-dimensional encoding of a sub-region on a two-dimensional image; The column encoding for the 2kth sub-region, where c is the column number and d is the column number. col The dimension along the column direction; The column encoding for the 2k+1th sub-region.

[0115] It is worth noting that the sine and cosine values, as well as the encoding method, can be distinguished by the k value.

[0116] The row and column codes of the sub-region are concatenated to obtain the 2D code:

[0117] in, Encoding the spatial location of the sub-region For sub-area row encoding, Encode the sub-area column.

[0118] In the z-axis direction, i.e., in multi-scale volumetric data, the stacking direction of each image is encoded using 1D sine and cosine coding, i.e.:

[0119] in, Encode the vertical direction of the cross section of the 2kth sub-region; The vertical direction encoding of the cross section of the 2k+1th sub-region is given, where z is the index position of the sub-region in the z-direction, i.e., the slice number in the image stacking direction in the 3D volume data; d is the dimension of the sub-region.

[0120] Step 212: Using an encoding and decoding structure, the input sequence is encoded and decoded to generate the predicted destruction path corresponding to the predicted time.

[0121] In this embodiment of the invention, the entire ViT network comprises an L-layer encoder-decoder structure, with each layer consisting of an encoder. Encode the input sequence:

[0122] in, For encoded sequences, The input sequence is .

[0123] The encoding employs a combination of multi-head attention (MHA) and feedforward neural network (FFN) for the input sequence of the l-th layer. Its (l+1)th layer can be represented as:

[0124] Where LN is the layer normalization operation; This is the input sequence for the (l+1)th layer; This is the output sequence of the l-th layer; This is the input sequence for the l-th layer.

[0125] MHA specifically refers to:

[0126] in, All of these are encoding matrix parameters, which can be obtained through training; x is a voxel.

[0127] FFN adopts an MLP structure, that is

[0128] Where σ is the ReLU activation function; All of these are encoding matrix parameters, which can be obtained through training; x represents a voxel. and These are all bias parameters.

[0129] Let the decoder be The sub-region vector is inversely mapped to return the original image resolution, outputting a segmentation probability map of the same size:

[0130] in, Let i be the predicted destruction path at time i. Here, H is the encoding sequence, H is the length of the image, and W is the width of the image.

[0131] for The decoder first extracts the sub-region vector, that is: In the middle, remove the first time vector and keep only the subsequent sub-region vectors, denoted as... Then, a linear mapping is performed to map the dimension d back to the single-channel result, corresponding to the potential corrupted region of binarization:

[0132] in, The vector of the k-th retained sub-region; This represents the k-th potential corrupted region in binarization; and These are the learnable parameters in the decoder. Each Reconstructing a P×P block probability graph The data is then concatenated to obtain the predicted binarized potential damage region, i.e., the predicted damage path corresponding to prediction time i.

[0133] in, The predicted destruction path corresponds to the predicted time i; Let H be the probability map of the Mth small block, where H is the length of the image and W is the width of the image.

[0134] Step 213: Select high-risk damage areas from the predicted damage paths.

[0135] In this embodiment of the invention, the predicted damage path includes high-risk damage areas ( ) and areas other than high-risk damage areas ( ).

[0136] Step 214: Establish a finite element model, refine the mesh for high-risk damage areas, homogenize the coarse mesh for non-high-risk damage areas, and correct the mesh transition area to generate a multi-scale mesh structure.

[0137] Specifically, using mesh generation software such as Hypermesh, a fine mesh generation strategy is adopted for high-risk potential failure areas identified in the prediction model to improve the simulation accuracy of local stress field and crack propagation behavior. For areas predicted as non-high-risk in the model, high-precision modeling is not performed directly. Instead, the Mori-Tanaka equivalent method is used to homogenize the mechanical properties of these areas, making them equivalent to a "matrix region" with average performance. Based on this equivalent matrix, coarser mesh elements are used for generation, thereby reducing computational resource consumption. After completing the generation of fine and coarse meshes, the transition region between them needs to be processed. To avoid mesh discontinuities, computational anomalies, or stress concentration, the following transition correction strategy is adopted: (1) Insert one or more layers of transition mesh elements at the junction of fine and coarse meshes, with the element size gradually changing from fine to coarse in this region; (2) Use shared nodes at the joint boundary to ensure that the meshes on both sides can be correctly coupled; (3) Finally, use the mesh quality check tool in HyperMesh to optimize the transition region to ensure that all elements meet the requirements of computational accuracy and stability. By combining the above-mentioned refinement and equivalent coarsening partitioning modeling strategy, both the computational efficiency of the model and the simulation accuracy of high-risk areas are ensured.

[0138] Step 215: Import the multi-scale mesh structure into the finite element software, set the cohesive elements and perform direct finite element calculation of the coupled model to simulate the multi-scale mesh structure and generate multi-scale failure results.

[0139] In this embodiment of the invention, the finite element software is Abaqus. Specifically, a contact-based cohesive modeling strategy is adopted in Abaqus. Contact surfaces are defined between the solid meshes in the failure path region, and contact relationships with cohesive behavior are applied. This cohesive contact model uses the traction-separation rule to characterize the interface mechanical response and combines it with the maximum damage stress criterion to control the failure process. This strategy can automatically generate crack propagation behavior when the contact surface meets the fracture conditions, avoiding the modeling limitations caused by pre-fixed paths, and is suitable for multi-scale damage analysis problems with dynamic path evolution. After definition, calculations are performed directly to obtain multi-scale failure results.

[0140] Furthermore, the failure path calculated by the finite element method is compared with the failure path observed in reality. The intersection-union ratio (IoU) of the two binary data (failure area is 1, non-failure area is 0) is calculated. If the result is less than 0.7, the deep learning model for failure prediction and the mesh partitioning strategy are backtracked to optimize the model and the strategy.

[0141] 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.

[0142] 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.

[0143] 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.

[0144] The technical solution of the multi-scale direct coupling calculation method for hardened cement slurry provided in this invention involves in-situ loading and computed tomography (CT) scanning of the prepared hardened cement slurry to generate multi-scale volume data at different stages; multi-phase segmentation of the multi-scale volume data to generate a multi-valued volume data mask; prediction of the multi-valued volume data mask using a pre-built deep learning model for failure path prediction to generate predicted failure paths; multi-scale meshing of the predicted failure paths to generate a multi-scale mesh structure; and direct simulation calculation of the multi-scale mesh structure using finite element software to generate multi-scale failure results. Based on high-information-density data acquired from images, and by using a deep learning model for training, non-critical regions are simplified while critical regions are retained. The multi-scale image is directly coupled with the hydration process, realizing the co-evolution of microstructure and properties of cement slurry during hydration, improving structural realism, and enhancing prediction accuracy and model generalization ability through a multi-modal fusion mechanism, thereby significantly reducing computational costs.

[0145] Figure 3 This is a schematic diagram of a multi-scale direct coupling calculation device for hardened cement slurry provided in an embodiment of the present invention. This device is used to execute the aforementioned multi-scale direct coupling calculation method for hardened cement slurry, such as... Figure 3 As shown, the device includes: a multi-scale volume data generation unit 11, a multi-phase segmentation unit 12, a failure path prediction unit 13, a mesh generation unit 14, and a direct coupling calculation unit 15.

[0146] The multi-scale volume data generation unit 11 is used to perform in-situ loading and computed tomography on the prepared hardened cement slurry to generate multi-scale volume data at different stages.

[0147] The multiphase segmentation unit 12 is used to perform multiphase segmentation on multiscale volume data and generate multi-valued volume data masks.

[0148] The destruction path prediction unit 13 is used to predict multi-valued volume data masks by using a pre-built destruction path prediction deep learning model to generate predicted destruction paths.

[0149] The meshing unit 14 is used to perform multi-scale meshing on the predicted damage path and generate the resulting multi-scale mesh structure.

[0150] The direct-coupled computational unit 15 is used to perform direct simulation calculations on multi-scale mesh structures using finite element software, generating multi-scale failure results.

[0151] In this embodiment of the invention, the multi-scale volume data generation unit 11 is specifically used to load hardened cement slurry samples in situ using an online loading device, and to scan the hardened cement slurry during the loading process using a computed tomography (CT) scanner to generate multi-scale in situ characterization data at different stages; and to reconstruct the multi-scale in situ characterization data using a reconstruction algorithm to generate multi-scale volume data at different stages.

[0152] In this embodiment of the invention, the multiphase segmentation unit 12 is specifically used to perform super-resolution on multi-scale volume data through a super-resolution model to generate microstructures; and to perform semantic segmentation on the microstructures through a computer vision semantic segmentation model to generate multi-valued volume data masks.

[0153] In this embodiment of the invention, the multi-valued volume data mask is the multi-valued volume data mask at the initial time; the destruction path prediction unit 13 is specifically used to perform vector representation of the multi-valued volume data mask at the initial time through a multilayer perceptron and a time embedding function to generate an input sequence; and to perform encoding and decoding processing on the input sequence through an encoding and decoding structure to generate the predicted destruction path corresponding to the prediction time.

[0154] In this embodiment of the invention, the device further includes: a sample data generation unit 16, a sample segmentation unit 17, a destruction path division unit 18, a dataset construction unit 19, and a model training unit 20.

[0155] The sample volume data generation unit 16 is used to perform in-situ loading and computed tomography on hardened cement slurry samples of different ages to generate multi-scale sample volume data of different ages and stages.

[0156] The sample segmentation unit 17 is used to perform multi-phase segmentation on multi-scale sample volume data and generate multi-valued sample volume data masks.

[0157] The destruction path segmentation unit 18 is used to segment the samples based on the multi-valued sample volume data mask, generating destruction paths for samples of different ages and loading stages.

[0158] Dataset construction unit 19 is used to construct a dataset based on multi-valued sample body data mask and sample destruction paths at different ages and loading stages.

[0159] The model training unit 20 is used to train the deep learning model using a dataset to build a deep learning model for predicting destructive paths.

[0160] In this embodiment of the invention, the dataset includes a multi-valued sample volume data mask and sample destruction paths at different ages and loading stages. The model training unit 20 is specifically used to predict destruction paths based on the multi-valued sample volume data mask using a deep learning model, generating predicted destruction paths for samples at different times. Loss calculations are performed on the predicted destruction paths for samples at different times and the destruction paths for samples at different ages and loading stages to obtain loss results. Based on the loss results, the model parameters of the deep learning model are iteratively optimized to generate a deep learning model for destruction path prediction.

[0161] In this embodiment of the invention, the mesh division unit 14 is specifically used to select high-risk damage areas from the predicted damage paths; establish a finite element model, refine the mesh division of the high-risk damage areas, homogenize the non-high-risk damage areas into coarse meshes and correct the mesh transition areas to generate a multi-scale mesh structure.

[0162] In this embodiment of the invention, the prepared hardened cement slurry undergoes in-situ loading and computed tomography to generate multi-scale volumetric data at different stages. The multi-scale volumetric data is then segmented into multiple phases to generate a multi-valued volumetric data mask. A pre-built deep learning model for predicting failure paths is used to predict the multi-valued volumetric data mask, generating predicted failure paths. The predicted failure paths are then divided into multi-scale meshes to generate a multi-scale mesh structure. A multi-scale finite element direct computation coupling model is used to simulate the multi-scale mesh structure, generating multi-scale failure results. Based on high-information-density data acquired from images, and by training a deep learning model, non-critical regions are simplified while critical regions are retained. The multi-scale images are directly coupled with the hydration process, enabling the co-evolution of the microstructure and properties of the cement slurry during hydration, improving structural realism. A multi-modal fusion mechanism enhances prediction accuracy and model generalization ability, significantly reducing computational costs.

[0163] 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.

[0164] 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 multi-scale direct coupling calculation method for hardened cement slurry. For a detailed description, please refer to the above-described embodiment of the multi-scale direct coupling calculation method for hardened cement slurry.

[0165] 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.

[0166] 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.

[0167] 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.

[0168] 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.

[0169] 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.

[0170] 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.

[0171] 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.

[0172] 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.

[0173] 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.

[0174] 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.

[0175] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0176] 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, it does not mean that the applicant has used or necessarily used the solution.

[0177] 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.

[0178] 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.

[0179] 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.

[0180] 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 multi-scale direct coupling calculation method for hardened cement slurry, characterized in that, The method includes: The prepared hardened cement slurry was subjected to in-situ loading and computed tomography to generate multi-scale volumetric data at different stages. The multi-scale volume data is segmented into multiple phases to generate a multi-valued volume data mask; The multi-valued volume data mask is predicted using a pre-built deep learning model for predicting destruction paths, thereby generating predicted destruction paths. The predicted damage path is divided into multi-scale grids to generate a multi-scale grid structure. The finite element method (FEM) software is used to directly simulate and calculate multi-scale mesh structures, generating multi-scale failure results.

2. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 1, characterized in that, The prepared hardened cement slurry is subjected to in-situ loading and computed tomography (CT) scans to generate multi-scale volumetric data at different stages, including: The hardened cement slurry was loaded in situ using an online loading device, and the hardened cement slurry was scanned during the loading process using a computed tomography (CT) scanner to generate multi-scale in situ characterization data at different stages. The multi-scale in-situ characterization data is reconstructed using a reconstruction algorithm to generate multi-scale volume data at different stages.

3. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 1, characterized in that, The step of performing multi-phase segmentation on the multi-scale volume data to generate a multi-valued volume data mask includes: The multi-scale volume data is super-resolution processed using a super-resolution model to generate microstructures. The microstructure is semantically segmented using a computer vision semantic segmentation model to generate a multi-valued volume data mask.

4. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 1, characterized in that, The multi-valued volume data mask is the multi-valued volume data mask at the initial moment; The step of predicting the multi-valued volumetric data mask and generating predicted destruction paths using a pre-built deep learning model for predicting destruction paths includes: The multi-valued volume data mask at the initial moment is represented by a vector using a multilayer perceptron and a temporal embedding function to generate an input sequence. The input sequence is encoded and decoded using an encoding / decoding structure to generate the predicted destruction path corresponding to the predicted time.

5. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 4, characterized in that, The method further includes: In-situ loading and computed tomography were performed on hardened cement slurry samples at different ages to generate multi-scale sample volume data at different ages and stages. The multi-scale sample volume data is segmented into multiple phases to generate a multi-valued sample volume data mask; Based on the multi-valued sample body data mask, the sample destruction paths of different ages and loading stages are generated; A dataset is constructed based on the multi-valued sample volume data mask and the sample destruction paths at different ages and loading stages. The deep learning model is trained using the dataset to construct the deep learning model for predicting the destruction path.

6. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 5, characterized in that, The dataset includes a multi-valued sample volume data mask and sample destruction paths at different ages and loading stages; The step of training a deep learning model using the dataset to construct the disruptive path prediction deep learning model includes: Using the deep learning model, destruction path prediction is performed based on the multi-valued sample volume data mask to generate sample prediction destruction paths corresponding to different times. Loss calculations are performed on the predicted destruction paths of samples at different times and the destruction paths of samples at different ages and loading stages to obtain the loss results; Based on the loss results, the deep learning model is iteratively optimized to generate the destruction path prediction deep learning model.

7. The multi-scale direct coupling calculation method for hardened cement slurry according to claim 1, characterized in that, The step of performing multi-scale grid partitioning on the predicted damage path to generate a partitioned multi-scale grid structure includes: High-risk destruction areas are selected from the predicted destruction paths; A finite element model is established, and the high-risk damage area is divided into a fine mesh. The non-high-risk damage area is processed into a homogenized coarse mesh and the mesh transition area is corrected to generate a multi-scale mesh structure.

8. A multi-scale direct-coupled computing device for hardened cement slurry, characterized in that, The device includes: The multi-scale volumetric data generation unit is used to perform in-situ loading and computed tomography on the prepared hardened cement slurry to generate multi-scale volumetric data at different stages. A multiphase segmentation unit is used to perform multiphase segmentation on the multiscale volume data and generate a multi-valued volume data mask. The destruction path prediction unit is used to predict the multi-valued volume data mask by using a pre-built destruction path prediction deep learning model to generate a predicted destruction path. A grid partitioning unit is used to perform multi-scale grid partitioning on the predicted destruction path, generating a partitioned multi-scale grid structure. Directly coupled computational units are used to directly simulate and calculate multi-scale mesh structures using finite element software, generating multi-scale failure results.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the multi-scale direct coupling calculation method for hardened cement slurry 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 multi-scale direct coupling calculation method for hardened cement slurry 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 multi-scale direct coupling calculation method for hardened cement slurry as described in any one of claims 1 to 7.

Images