A method and device for multi-scale direct coupling calculation of hardened cement paste

Abstract

The application provides a hardened cement paste multi-scale direct coupling calculation method and device, the method comprises the following steps: in-situ loading and computer tomography are carried out on hardened cement paste, multi-scale volume data at different stages are generated, and multi-phase segmentation is carried out to obtain multi-valued volume data mask; a damage path prediction deep learning model is used to predict the damage path of the multi-valued volume data mask and divide the multi-scale grid, a finite element software is used to directly simulate and calculate the multi-scale grid structure, a multi-scale damage result is generated, a deep learning model is used to predict the potential damage path based on the high information density data obtained from the image, and modeling is carried out on the basis, the calculation complexity is simplified, the microstructure evolution characteristics and the macro mechanical property response of the cement paste in the hydration process are considered on the premise of ensuring the simulation accuracy, the numerical calculation cost of the multi-scale structure is effectively reduced, and an efficient and expandable technical path is provided for the modeling of the evolution process of complex multi-phase materials.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a hardened cement paste multi-scale direct coupling calculation method and device. BACKGROUND

[0002] Multi-scale performance research of cement-based materials is the key to understanding the mechanical behavior of its heterogeneous composite materials. The macroscopic mechanical response of the material is closely related to its meso and microstructure. It is particularly important to analyze the influence of material composition and structure at different scales on macroscopic performance. Multi-scale performance research has become a key means to understand the mechanical behavior of concrete. The combination of experiments, numerical simulation and theoretical analysis enables researchers to systematically analyze the evolution mechanism of material performance from micro to macro. However, existing research still faces many challenges: microscale testing is limited by sample preparation and testing errors, and there is a large dispersion in the data; numerical simulation, while improving the accuracy of calculation, still relies on empirical formulas and is limited by the cost of calculation, making it difficult to fully cover the complex process of hydration evolution; although the theoretical analysis method can provide effective prediction between scales, there are still applicability problems when considering interface effects and heterogeneity.

[0003] In one type of existing technology, multi-scale performance simulation mainly adopts the traditional hierarchical multi-scale modeling method. This method takes representative volume element (RVE) as the core, establishes constitutive models at micro, meso and macro scales respectively, and transfers parameters through homogenization method, thereby constructing a cross-scale numerical simulation system. Another type of existing technology constructs a cement material microstructure model through an image-driven multi-scale reconstruction method, combines image segmentation, reconstruction and parameter fitting, and improves the modeling accuracy.

[0004] However, although the traditional hierarchical multi-scale modeling method has certain theoretical completeness, the physical quantity mapping between the upper and lower scales is only through numerical averaging, which is difficult to truly reflect the continuous influence of microstructure heterogeneity and evolution on macroscopic performance; the simulation at different scales is based on certain assumptions, which means that errors will accumulate at each scale, causing inaccurate prediction results, and there may be error accumulation in scale transmission; the hydration is simplified as static input or predefined parameters, lacking dynamic simulation, and it is difficult to reflect the evolution characteristics of cement-based materials in the early age, and the hydration process is difficult to couple; the RVE unit is mostly dependent on ideal geometric assumptions, lacking effective coupling with real test data, and unable to accurately reproduce the real microstructure and phase distribution, lacking of structure realism. Although the image-driven modeling technology improves the restoration degree of microstructure, the coupling between image scale and physical quantity is insufficient, and the image cannot be directly mapped to internal process variables such as water state and reaction degree, and the image modeling is fragmented from material evolution; the reconstruction relies on specific sample images, and the model has poor transferability and weak generalization ability; with the increasing complexity of microstructure details, the model freedom degree increases exponentially, resulting in reduced simulation efficiency, making it difficult to quickly apply in large-scale engineering problems, and the calculation cost is high. SUMMARY

[0005] An object of the present application is to provide a hardened cement paste multi-scale direct coupling calculation method, which is based on high information density data obtained by image acquisition, and uses a trained deep learning model to simplify non-important areas and retain important areas, directly coupling image multi-scale and hydration process, realizing the coordinated evolution of microstructure and performance of cement paste during hydration process, improving the structure realism, improving the prediction accuracy and model generalization ability through multi-modal fusion mechanism, and greatly reducing the calculation cost. Another object of the present application is to provide a hardened cement paste multi-scale direct coupling calculation device. Still another object of the present application is to provide a computer readable medium. Still another object of the present application is to provide a computer device.

[0006] To achieve the above objects, one aspect of the present application discloses a hardened cement paste multi-scale direct coupling calculation method, comprising:

[0007] The prepared hardened cement paste is subjected to sample in-situ loading and computer tomography to generate multi-scale volume data at different stages;

[0008] The multi-scale volume data is subjected to multi-phase segmentation to generate multi-valued volume data mask;

[0009] The multi-valued volume data mask is predicted by a pre-constructed damage path prediction deep learning model to generate a predicted damage path;

[0010] The predicted damage path is subjected to multi-scale grid division to generate a divided multi-scale grid structure;

[0011] Direct simulation calculation is performed on the multi-scale grid structure by a finite element software to generate multi-scale damage results.

[0012] Preferably, the prepared hardened cement paste is subjected to sample in-situ loading and computer tomography scanning to generate multi-scale volume data at different stages, including:

[0013] The hardened cement paste is subjected to sample in-situ loading by an online loading device, and the hardened cement paste is scanned by a computer tomography scanning device during the loading process to generate multi-scale in-situ characterization data at different stages;

[0014] The multi-scale in-situ characterization data is subjected to data reconstruction by a reconstruction algorithm to generate multi-scale volume data at different stages.

[0015] Preferably, the multi-scale volume data is subjected to multi-phase segmentation to generate multi-valued volume data masks, including:

[0016] The multi-scale volume data is subjected to super-resolution by a super-resolution model to generate a microstructure;

[0017] The microstructure is subjected to semantic segmentation by a computer vision semantic segmentation model to generate multi-valued volume data masks.

[0018] Preferably, the multi-valued volume data masks are initial-time multi-valued volume data masks;

[0019] The multi-valued volume data masks are subjected to prediction by a pre-constructed damage path prediction deep learning model to generate predicted damage paths, including:

[0020] The initial-time multi-valued volume data masks are subjected to vector representation by a multi-layer perception and a time embedding function to generate input sequences;

[0021] The input sequences are subjected to encoding and decoding processing by an encoding and decoding structure to generate predicted damage paths corresponding to prediction time.

[0022] Preferably, the method further comprises:

[0023] The prepared hardened cement paste samples at different ages are subjected to sample in-situ loading and computer tomography scanning to generate multi-scale sample volume data at different ages and different stages;

[0024] The multi-scale sample volume data is subjected to multi-phase segmentation to generate multi-valued sample volume data masks;

[0025] The multi-valued sample volume data masks are divided to generate sample damage paths at different ages and different loading stages;

[0026] According to the multi-valued sample body data mask and the sample damage path of different ages and different loading stages, a data set is constructed;

[0027] Through the data set, a deep learning model is trained to construct a damage path prediction deep learning model.

[0028] Preferably, the data set comprises a multi-valued sample body data mask and a sample damage path of different ages and different loading stages;

[0029] Through the data set, a deep learning model is trained to construct a damage path prediction deep learning model, comprising:

[0030] Through the deep learning model, the multi-valued sample body data mask is used for damage path prediction to generate sample predicted damage paths corresponding to different time points;

[0031] The sample predicted damage paths corresponding to different time points and the sample damage paths of different ages and different loading stages are subjected to loss calculation to obtain a loss result;

[0032] According to the loss result, model parameter iterative optimization is performed on the deep learning model to generate a damage path prediction deep learning model.

[0033] Preferably, the predicted damage path is subjected to multi-scale grid division to generate a divided multi-scale grid structure, comprising:

[0034] A high-risk damage area is screened out from the predicted damage path;

[0035] A finite element model is established, the high-risk damage area is subjected to refined grid division, the non-high-risk damage area is subjected to homogenized coarse grid processing and the grid transition area is corrected to generate a multi-scale grid structure.

[0036] The application also discloses a hardening cement paste multi-scale direct coupling calculation device, comprising:

[0037] A multi-scale body data generation unit is used for in-situ loading of the prepared hardening cement paste and computer tomography to generate multi-scale body data at different stages;

[0038] A multi-phase segmentation unit is used for multi-phase segmentation of the multi-scale body data to generate a multi-valued body data mask;

[0039] A damage path prediction unit is used for predicting the multi-valued body data mask through a pre-constructed damage path prediction deep learning model to generate a predicted damage path;

[0040] A grid division unit is used for multi-scale grid division of the predicted damage path to generate a divided multi-scale grid structure;

[0041] A direct coupling calculation unit is configured to perform direct simulation calculation on the multi-scale grid structure by using finite element software to generate multi-scale damage results.

[0042] The application further discloses a computer readable medium, which stores a computer program, and the computer program is executed by a processor to implement the method.

[0043] The application further discloses a computer device, which comprises a memory and a processor, the memory is configured to store information comprising program instructions, and the processor is configured to control execution of the program instructions, and the processor implements the method when executing the program.

[0044] The application further discloses a computer program product, which comprises computer program / instructions, and the computer program / instructions are executed by a processor to implement the method.

[0045] The application performs sample in-situ loading and computer tomography on the prepared hardened cement paste to generate multi-scale volume data at different stages, performs multi-phase segmentation on the multi-scale volume data to generate multi-valued volume data masks, performs prediction on the multi-valued volume data masks by using a pre-constructed damage path prediction deep learning model to generate predicted damage paths, performs multi-scale grid division on the predicted damage paths to generate divided multi-scale grid structures, performs direct simulation calculation on the multi-scale grid structures by using finite element software to generate multi-scale damage results, and based on high information density data obtained from images, the method of training a deep learning model is used to simplify non-important regions and retain important regions, image multi-scale is directly coupled with a hydration process, the collaborative evolution of microstructure and performance of cement paste in the hydration process is realized, the structural realism is improved, the prediction accuracy and model generalization ability are improved through a multi-modal fusion mechanism, and the calculation cost is greatly reduced. BRIEF DESCRIPTION OF DRAWINGS

[0046] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can also obtain other drawings according to these drawings without creative labor.

[0047] Figure 1 A flowchart of a hardened cement paste multi-scale direct coupling calculation method provided by the embodiment of the present application is shown in the figure.

[0048] Figure 2 A flowchart of another hardened cement paste multi-scale direct coupling calculation method provided by the embodiment of the present application is shown in the figure.

[0049] Figure 3A structural schematic diagram of a hardened cement paste multi-scale direct coupling calculation device provided by an embodiment of the present application is provided.

[0050] Figure 4 A structural schematic diagram of a computer device provided by an embodiment of the present application is provided. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0052] It should be noted that the hardened cement paste multi-scale direct coupling calculation method and device disclosed in the present application can be used in the field of artificial intelligence technology, and can also be used in any field other than the field of artificial intelligence technology. The application field of the hardened cement paste multi-scale direct coupling calculation method and device disclosed in the present application is not limited.

[0053] In order to facilitate understanding of the technical solutions provided by the present application, the related contents of the technical solutions of the present application will be described first. The present application optimizes the multi-scale model construction method, directly constructs a calculation model covering macro, meso and micro across six orders of magnitude, so that the hydration evolution, pore structure and mechanical response evolve self-consistently in a unified modeling framework. By combining high-resolution test data for model correction, the prediction accuracy is improved, and the stability and applicability of the model in the full scale are ensured. The present application aims to obtain high information density data based on images, and uses a trained deep learning model to simplify non-important areas and retain important areas, so as to directly simulate and calculate through finite element software, and solve the problems of inaccurate generation of hydration dynamics microstructure model and large amount of calculation.

[0054] The implementation process of the hardened cement paste multi-scale direct coupling calculation method provided by the embodiments of the present application will be described below with the hardened cement paste multi-scale direct coupling calculation device as an example. It can be understood that the execution subject of the hardened cement paste multi-scale direct coupling calculation method provided by the embodiments of the present application includes but is not limited to the hardened cement paste multi-scale direct coupling calculation device.

[0055] Figure 1 A flowchart of the hardened cement paste multi-scale direct coupling calculation method provided by an embodiment of the present application is shown in Figure 1 The method comprises the following steps.

[0056] Step 101, sample in-situ loading and computer tomography are performed on the prepared hardened cement paste to generate multi-scale volume data at different stages.

[0057] In the embodiment of the present application, the computer tomography is X-ray computer tomography (X-CT). Specifically, in-situ loading tests are carried out on the prepared hardened cement paste by the online loading device respectively, the same loading mode is loaded, the X-CT device is used to carry out rapid scanning in an equal interval manner, X-CT multi-scale in-situ characterization data at different loading stages are obtained, and multi-scale volume data at different stages are formed through a reconstruction algorithm.

[0058] Step 102, multi-phase segmentation is carried out on the multi-scale volume data to generate multi-valued volume data masks.

[0059] In the embodiment of the present application, the X-CT multi-scale volume data of rapid scanning is subjected to semantic segmentation through the super-resolution model and the computer vision semantic segmentation model, and the multi-phase structure of the hydration product composition and the crack composition is obtained.

[0060] Step 103, the multi-valued volume data mask is predicted through the pre-constructed damage path prediction deep learning model to generate a predicted damage path.

[0061] In the embodiment of the present application, the damage path prediction deep learning model is a model constructed by using ViT (Vision Transformer) as a backbone network for training, which can meet the demand of predicting the final damage path from the initial microstructure.

[0062] Specifically, the multi-valued volume data mask image at the initial state is input into the damage path prediction deep learning model, and the predicted damage path at the prediction time is output.

[0063] Step 104, multi-scale grid division is carried out on the predicted damage path to generate a divided multi-scale grid structure.

[0064] In the embodiment of the present application, for the high-risk damage area in the predicted damage path predicted by the damage path prediction deep learning model, the finite element model is used to carry out refined grid division by using a cohesive element; for other areas, the complex microstructure is first equivalent to a mean matrix by using the Mori-Tanaka method, and then a homogenized coarse grid is used for processing, after the establishment, the transition area of the grid is corrected to obtain the divided multi-scale grid structure, so that the calculation can be carried out smoothly.

[0065] Step 105, direct simulation calculation is carried out on the multi-scale grid structure by using the finite element software to generate a multi-scale damage result.

[0066] In the embodiment of the present application, the multi-scale grid structure is introduced into the finite element software, the cohesive element is set and the finite element direct calculation coupling model is carried out, the multi-scale grid structure is simulated and simulated, and the multi-scale damage result is generated. Specifically, the cohesive or cohesive element is embedded in the high-risk area in the finite element software, and the failure criterion such as critical stress or energy release rate is defined. Specifically, the global stress field is solved, the stress concentration area is identified, the multi-scale damage result is output, and the element failure condition of the refined area is dynamically updated. The multi-scale damage result is the crack evolution morphology.

[0067] Further, the crack evolution morphology simulated by the multi-scale finite element direct calculation coupling model is compared with the actual observation data of X-CT in step 101, and the damage path prediction deep learning model and the multi-scale grid division are optimized according to the comparison result.

[0068] The technical scheme provided by the embodiment of the present application generates multi-scale volume data at different stages by in-situ loading and computer tomography of the prepared hardened cement paste sample. The multi-scale volume data is multi-phase segmented to generate multi-valued volume data mask. The multi-valued volume data mask is predicted by the pre-constructed damage path prediction deep learning model to generate a predicted damage path. The predicted damage path is divided into a multi-scale grid structure. The multi-scale grid structure is directly simulated by the finite element software to generate a multi-scale damage result. The high information density data obtained by the image is simplified in the important area by training the deep learning model. The image multi-scale and the hydration process are directly coupled to realize the collaborative evolution of the microstructure and performance of the cement paste in the hydration process, improve the structure realism, and improve the prediction accuracy and model generalization ability through the multi-modal fusion mechanism, and greatly reduce the calculation cost.

[0069] Figure 2 The flowchart of another multi-scale direct coupling calculation method for hardened cement paste provided by the embodiment of the present application is shown in Figure 2 The method comprises the following steps.

[0070] In step 201, the different age hardened cement paste samples are in-situ loaded and computer tomography is performed to generate multi-scale sample volume data at different stages.

[0071] In the embodiment of the present application, each step is executed by a hardened cement paste multi-scale direct coupling calculation device.

[0072] In the embodiment of the present application, a hardened cement paste is prepared in a binary system of cement and moisture, and the hydration is terminated by isopropanol solution to obtain cement paste samples of different hydration termination ages. Specifically, ordinary Portland cement and deionized water are used as raw materials, and a cement paste sample is prepared with a preset water-cement ratio. The hydration is terminated by soaking in isopropanol solution at different ages to obtain hardened cement paste samples of different ages.

[0073] As an option, ordinary Portland cement and deionized water are used as raw materials, and a cement paste sample is prepared with a water-cement ratio of 0.5. The hydration is terminated by soaking in isopropanol solution at 1 day (d), 3d, 7d, and 28d to obtain hardened cement paste samples of different ages.

[0074] In the embodiment of the present application, computer tomography is X-ray computer tomography (X-CT). In-situ loading tests are carried out on hardened cement paste samples of different ages by an online loading device, and the same loading mode is used. Fast scanning is carried out by the X-CT device at equal intervals to obtain X-CT multi-scale in-situ characterization data at different loading stages. The multi-scale sample volume data at different ages and different loading stages are formed by a reconstruction algorithm.

[0075] As an option, in-situ loading tests are carried out on samples of different ages by an online loading device. Uniaxial compression tests are carried out for loading, and the loading is carried out in increments of 10 micrometers (um) displacement steps. The loading is stopped after the displacement, and the fast scanning is carried out by the X-CT device after waiting for 30 seconds. The X-CT multi-scale in-situ characterization data at different loading stages are obtained, and the original multi-scale sample volume data gray distribution at different ages and different loading states is formed by the reconstruction algorithm of the X-CT device.

[0076] Step 202, multi-phase segmentation is carried out on the multi-scale sample volume data to generate multi-valued sample volume data masks.

[0077] In the embodiment of the present application, the X-CT multi-scale sample volume data of fast scanning is segmented by a super-resolution model and a computer vision semantic segmentation model to obtain a multi-phase structure of hydration product components and cracks.

[0078] Specifically, the multi-scale sample volume data is super-resolved by the super-resolution model to obtain fine microstructures. The microstructures are segmented by the computer vision semantic segmentation model to generate a multi-phase structure 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, and form multi-valued sample volume data masks.

[0079] Step 203, according to the multi-value sample body data mask, the different age and different loading stage of the sample damage path is generated.

[0080] In the embodiment of the application, the multi-value sample body 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, and the specific extension rules are as follows:

[0081] The multi-value sample body data includes a plurality of voxels, and the multi-value sample body data mask includes: ; wherein, represents that the voxel x belongs to the identified crack region; respectively represent unhydrated particles, high-density hydrated calcium silicate (HD-CSH), low-density hydrated calcium silicate (LD-CSH), calcium hydroxide (CH), ettringite (Aft) and pores; represents an irrelevant background region. The voxel set of the crack region is denoted as F, and the boundary thereof is denoted as , so as to obtain the distribution of the initial crack in space.

[0082] The distance function of the crack region F is defined as:

[0083]

[0084] wherein, is the distance between the voxel x and the crack region F; y is a voxel in the crack region F; is the three-dimensional Euclidean distance between the voxel x and the voxel y.

[0085] The inflation threshold δ is set, and all points satisfying: are regarded as the expanded inflation region , that is, a fixed distance δ shell outside the crack region. The inflation region F δ is taken as the intersection of each phase region (F ) in the multi-value sample body data mask except the crack region and the background region, and the overlapping region is obtained one by one:

[0086]

[0087] wherein, is the overlapping region between the i-th phase region and the inflation region, F δ is the inflation region, is the i-th phase region composed of the voxel x.

[0088] It is worth noting that the expansion threshold δ can be set according to actual needs, and the embodiments of the present application do not limit this. As an optional solution, the expansion threshold δ is 2.

[0089] In the embodiments of the present application, this step can exclude background or invalid regions unrelated to the crack, and at the same time, the microstructure possibly associated with the crack periphery is included in the candidate range of the 'potential damage area'. For the voxels in the 'potential damage area', a 26-neighborhood three-dimensional connected search is used to further identify the connected components thereof, so as to eliminate discrete noise or small isolated voxels, and thus a more stable and continuous periphery microstructure region is obtained, denoted as R The union of all periphery microstructures is obtained as the total periphery microstructure region R δ :

[0090]

[0091] The volume of the initial crack region F is denoted as V F , and if the volume vol(R δ ) of R δ exceeds β times of the volume V F of the original crack, that is, vol(R δ )>βV F , then the periphery of R δ is morphologically contracted or preferentially removed until vol(R δ )≤βV F . Thus, the sample damage path of different ages and different loading stages is obtained.

[0092] It is worth noting that the volume multiple β can be set according to actual needs, and the embodiments of the present application do not limit this. As an optional solution, the volume multiple β is 2.

[0093] Step 204, constructing a data set according to the multi-valued sample body data mask and the sample damage path of different ages and different loading stages.

[0094] In the embodiments of the present application, the multi-valued sample body data mask at the initial time and the sample damage path at the final time are taken as input and supervision, and the intermediate process is taken as key time supervision to establish a data set. Specifically, the multi-valued sample body data mask at the initial time is paired with the sample damage path at the final time, and the sample damage path in the loading process is taken as key time supervision to construct a data set.

[0095] Step 205, training a deep learning model through the data set to construct a damage path prediction deep learning model.

[0096] ​​In the embodiment of the present application, the data set includes a multi-valued sample body data mask and a sample damage path at different ages and different loading stages.

[0097] In the embodiment of the present application, the data set is divided into a training set and a verification set according to a preset division ratio, the multi-valued sample body data mask and the sample damage path at different ages and different loading stages in the training set are used for model training, and the multi-valued sample body data mask and the sample damage path at different ages and different loading stages in the verification set are used for model verification.

[0098] It is worth noting that the division ratio can be set according to actual needs, and the embodiment of the present application does not limit this. As an optional solution, the division ratio of the training set and the verification set is 3:1.

[0099] In the embodiment of the present application, step 205 specifically includes:

[0100] Step 2051, predicting the damage path according to the multi-valued sample body data mask through the deep learning model to generate the sample predicted damage path corresponding to different time points.

[0101] In the embodiment of the present application, in order to meet the demand of predicting the final damage path from the initial microstructure, a ViT (Vision Transformer) is used as a backbone network to construct a deep learning model. The model input is the multi-valued sample body data mask at the initial time, and the output is the damage path at the final time.

[0102] It is worth noting that the data form of the multi-valued sample body data mask is a multi-valued sample image.

[0103] Specifically, the multi-valued sample body data mask is trained by expanding it according to the horizontal cross section, and the multi-valued sample body data mask at the initial time is denoted as: , wherein X(t1) is the multi-valued sample image at the initial time; H is the length of the image; W is the width of the image; and C is the number of channels, including 1-8 categories (such as different phases formed in the hydration process). At different time points t i (i=2,…,n,) during loading, the corresponding potential damage area binary image is divided:

[0104]

[0105] , wherein i=n represents the final time; Y(t i ) represents the sample crack area or the sample potential damage area under the loading state at the i-th time point; H is the length of the image; and W is the width of the image.

[0106] The multi-valued sample image X(t1) at the initial moment is divided into M sub-blocks (patches) on the plane with a size of P×P, and each sub-block is subjected to a multi-layer perception (MLP) to obtain a sample image sub-block vector :

[0107]

[0108] wherein, represents a d-dimensional real number space, that is, a vector space containing d real numbers. Specifically, the original sub-block is compressed into a d-dimensional vector after MLP.

[0109] A learnable time embedding function is introduced to represent the features at the loading moment:

[0110]

[0111] wherein, is the time vector at the loading moment i.

[0112] As an optional solution, the time embedding encoding adopts a lookup table form:

[0113]

[0114] wherein, is the time vector at the loading moment i; is a learnable matrix storing the d-dimensional vectors corresponding to the indexes of 0~n moments, which is updated in the back propagation during training, and according to the time index i directly look up the table during inference.

[0115] The time vector is concatenated with the sample image sub-block vector to obtain the initial sequence of the model:

[0116]

[0117] On this basis, in order to distinguish the cross-sectional positions of the sample image sub-block and the body data in the Z-axis, a corresponding spatial position encoding and a cross-sectional perpendicular direction encoding are added to each sub-block to form the input sequence:

[0118]

[0119] wherein, is the sample input sequence, is the sample initial sequence, is the sample image sub-block vector corresponding to the spatial position encoding, is the sample initial sequence Corresponding cross-section vertical direction coding.

[0120] Suppose the entire ViT network contains L layers of encoding-decoding structure, each layer is passed through an encoder Encode the input sequence:

[0121]

[0122] Wherein, is the sample coding sequence, is the sample input sequence.

[0123] Let the decoder Map the sub-vector of the sample coding sequence Back to the original image resolution, output the same size segmentation probability map:

[0124]

[0125] Wherein, is the sample prediction damage path corresponding to time i, is the sample coding sequence, H is the length of the image, and W is the width of the image.

[0126] In the embodiment of the application, the network is allowed to predict the damage path corresponding to the time through the time index i.

[0127] Step 2052, loss calculation is performed on the sample prediction damage path corresponding to different times and the sample damage path of different ages and different loading stages, to obtain a loss result.

[0128] In the embodiment of the application, the loss calculation is performed on the output image at time i and the binarization result at different times through weighted sum loss,

[0129]

[0130] Wherein, is the loss result; is the sample prediction damage path corresponding to time i; Y(t i ) is the sample damage path at different times; is the training weight, used to balance the importance of intermediate time and final time in training.

[0131] Step 2053, model parameter iterative optimization is performed on the deep learning model according to the loss result, to generate a damage path prediction deep learning model.

[0132] In the embodiment of the application, the network parameters θ in the deep learning model are updated through back propagation at the same time, so that the model can better capture the evolution law of the crack from initiation to final in a multi-time supervision manner.

[0133] At the model deployment, only the multi-valued sample image X(t1) at the initial time is given and the final time The final damage path is directly obtained by forward propagation of the network:

[0134]

[0135] wherein, is the sample input sequence; is the time vector at the final time n; is the sample image sub-region vector; is the encoder; is the decoder; is the sample encoding sequence; is the sample final damage path corresponding to the final time n, that is, the sample final damage path.

[0136] In the embodiment of the application, only one forward inference is required in the whole process, and there is no need to explicitly input or generate the image at the intermediate time, thereby realizing efficient prediction of the final damage path.

[0137] As an optional environment, the model training environment of the damage prediction deep learning model is:

[0138] Operating system: Ubuntu 20.04LTS;

[0139] Model running environment: Pytorch+CUDA;

[0140] Graphics card: NVIDIA 4090x4;

[0141] Graphics card driver version: 520.56.06;

[0142] CUDA version: 11.8;

[0143] Pytorch version: Stable (1.13.1).

[0144] Step 206, model verification is performed on the damage prediction deep learning model by constructing the test set.

[0145] In the embodiment of the application, the test set can be constructed in the manner of steps 201 to 204, which will not be repeated here. As an optional solution, samples with a new sample age of 28 days are used as tests, the test set is constructed in the manner of steps 201 to 204, and the damage prediction deep learning model is verified.

[0146] Step 207, sample in-situ loading is performed on the cement paste by an online loading device, and the cement paste is scanned by a computer tomography device during the loading process to generate multi-scale in-situ characterization data at different stages.

[0147] In the embodiment of the application, the computer tomography is X-ray computer tomography (X-CT). The sample in-situ loading is performed on the cement paste by the online loading device, and the cement paste is scanned by the X-CT device in an equal interval manner during the loading process to generate multi-scale in-situ characterization data at different stages.

[0148] As an optional solution, the loading can be performed by a uniaxial compression test, the loading is performed in increments of a displacement step of 10 um, the loading is stopped after the displacement, and then 30 s is waited, the X-CT device is used to perform rapid scanning, and after the end, the loading is continued to obtain X-CT multi-scale in-situ characterization data at different stages.

[0149] Step 208, data reconstruction is performed on the multi-scale in-situ characterization data by a reconstruction algorithm to generate multi-scale volume data at different stages.

[0150] Specifically, the data reconstruction is performed on the multi-scale in-situ characterization data by a reconstruction algorithm of the X-CT to generate multi-scale volume data at different stages. The gray scale distribution of the multi-scale volume data is [0, 65535].

[0151] Step 209, super-resolution is performed on the multi-scale volume data by a super-resolution model to generate a microstructure.

[0152] Specifically, the multi-scale volume data is input into the super-resolution model to perform super-resolution, and the microstructure is output.

[0153] Step 210, semantic segmentation is performed on the microstructure by a computer vision semantic segmentation model to generate multi-valued volume data masks.

[0154] In the embodiment of the application, the multi-valued volume data mask is a multi-valued volume data mask at an initial moment.

[0155] Specifically, the microstructure is input into the computer vision semantic segmentation model to perform semantic segmentation, and the multi-valued volume data mask is generated. 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.

[0156] Step 211, vector representation is performed on the multi-valued volume data mask at the initial moment by an MLP and a time embedding function to generate an input sequence.

[0157] In the embodiment of the present application, the initial time multi-value body data mask is denoted as: , wherein, is the initial time multi-value image; H is the length of the image; W is the width of the image; and C is the number of channels, including 1-8 categories (such as different phases formed in the hydration process). At different time t i (i=2,…,n,) are divided into corresponding potential damage area binary images:

[0158]

[0159] wherein i=n represents the final time; represents the crack area or potential damage area at the loading state of the i-th time; H is the length of the image; and W is the width of the image.

[0160] The initial time multi-value image is divided into M sub-patches (patches) on the plane with a size of P*P, and each sub-patch is subjected to a multi-layer perception (MLP) to obtain an image sub-patch vector .

[0161]

[0162] A learnable time embedding function is introduced, and the time embedding coding adopts a lookup table form:

[0163]

[0164] wherein, is the time vector at the loading time i; is a learnable matrix, storing the d-dimensional vector corresponding to the index of each time from 0 to n. During inference, according to the time index i , the time vector is directly looked up in the table.

[0165] The time vector is concatenated with the image sub-patch vector to obtain the initial sequence of the model:

[0166]

[0167] On this basis, in order to distinguish the cross-sectional position of the image sub-patch and the body data in the Z-axis direction, a corresponding spatial position code and a cross-sectional vertical direction code are added to each sub-patch to form the input sequence:

[0168]

[0169] wherein, is the input sequence, for initial sequence, for image sub-region vector corresponding spatial position encoding, for initial sequence corresponding cross-section vertical direction encoding.

[0170] In the embodiment of the present application, the position encoding adopts the form of sine-cosine encoding:

[0171] The spatial position encoding of the sub-region adopts two-dimensional sine-cosine encoding, for row:

[0172]

[0173] wherein, represents the one-dimensional encoding of the sub-region on the two-dimensional image; is the row encoding of the 2kth sub-region, r is the row number, d row is the dimension in the row direction; is the row encoding of the 2k+1th sub-region.

[0174] For column:

[0175]

[0176] wherein, represents the one-dimensional encoding of the sub-region on the two-dimensional image; is the column encoding of the 2kth sub-region, c is the column number, d col is the dimension in the column direction; is the column encoding of the 2k+1th sub-region.

[0177] It is worth noting that the sine-cosine and the encoding mode can be distinguished by the value of k.

[0178] The row encoding and the column encoding of the sub-region are spliced to obtain 2D encoding:

[0179]

[0180] wherein, is the spatial position encoding of the sub-region, is the row encoding of the sub-region, is the column encoding of the sub-region.

[0181] In the z-axis direction, that is, the stacking direction of each picture in the multi-scale body data, through 1D sine-cosine encoding, that is:

[0182]

[0183] wherein, is the cross-section vertical direction encoding of the 2kth sub-region; The cross-section vertical direction coding of the 2k+1th sub-region, z is the index position of the sub-region in the z direction, that is, the slice number in the picture stacking direction in the three-dimensional body data; d is the dimension of the sub-region.

[0184] Step 212, the input sequence is coded and decoded by the coding and decoding structure to generate a prediction destruction path corresponding to the prediction moment.

[0185] In the embodiment of the application, the entire ViT network comprises an L-layer coding-decoding structure, each layer is coded and decoded by an encoder The input sequence is coded:

[0186]

[0187] Among them, is the coded sequence, is the input sequence.

[0188] The coding adopts a combination of multi-head attention mechanism (MHA) and feedforward neural network (FFN), and for the input sequence of the lth layer The l+1th layer can be expressed as:

[0189]

[0190] Among them, LN is a layer normalization operation; is the input sequence of the l+1th layer; is the output sequence of the lth layer; is the input sequence of the lth layer.

[0191] MHA is specifically:

[0192]

[0193] Among them, are all coding matrix parameters, which can be obtained by training; x is a voxel.

[0194] FFN adopts an MLP structure, that is,

[0195]

[0196] Among them, sigma is a ReLU activation function; are all coding matrix parameters, which can be obtained by training; x is a voxel; and are all bias parameters.

[0197] Let the decoder inversely map the sub-region-vector of to return the original picture resolution, and output a segmentation probability map of the same size:

[0198]

[0199] wherein, is the predicted damage path corresponding to time i, is the encoded sequence, H is the length of the image, and W is the width of the image.

[0200] For the encoder, the sub-zone vectors are first taken out, that is: In the above formula, the first time vector is removed, and only the subsequent sub-zone vectors are retained and denoted as Linear mapping is then performed to map the dimension d back to a single-channel result, corresponding to the binarized potential damage zone:

[0201]

[0202] wherein, is the kth reserved sub-zone vector; is the kth binarized potential damage zone; and are learnable parameters in the decoder. Each is reconstructed into a P*P small block probability map and spliced to obtain a predicted binarized potential damage area, that is, a predicted damage path corresponding to time i:

[0203]

[0204] wherein, is the predicted damage path corresponding to time i; is the Mth small block probability map, H is the length of the image, and W is the width of the image.

[0205] Step 213, screening a high-risk damage area from the predicted damage path.

[0206] In the embodiment of the application, the predicted damage path includes a high-risk damage area ( ) and an area other than the high-risk damage area ( ).

[0207] Step 214, establishing a finite element model, performing refined grid division on the high-risk damage area, performing homogenized coarse grid processing on the non-high-risk damage area, and correcting the grid transition area to generate a multi-scale grid structure.

[0208] ​Specifically, using meshing software such as Hypermesh, for the high-risk potential damage area identified in the prediction model, a refined meshing strategy is adopted to improve the simulation accuracy of the local stress field and crack propagation behavior; for the non-high-risk area predicted in the model, the mechanical properties of these areas are homogenized by the Mori-Tanaka equivalent method, and they are equivalent to the "matrix region" with average performance. Based on this equivalent matrix, a relatively coarse grid element is used for division, thereby reducing the consumption of computing resources. After completing the fine grid and coarse grid division, the transition area between them needs to be handled. In order to avoid grid discontinuity, calculation anomaly or stress concentration phenomenon, the following transition correction strategy is adopted: (1) one or more layers of transition grid elements are inserted at the junction of fine grid and coarse grid, and the size of the elements in this area gradually changes from fine to coarse; (2) the shared node method is used for the joint boundary to ensure that the grids on both sides can be correctly coupled; (3) finally, the mesh quality checking tool in HyperMesh is used to optimize the transition area, ensuring that all elements meet the calculation accuracy and stability requirements. Through the above combination of refined and equivalent coarse partition modeling strategies, both the calculation efficiency of the model and the simulation accuracy of the high-risk area are ensured.

[0209] Step 215, import the multi-scale grid structure into the finite element software, set the cohesive element and perform finite element direct calculation coupling model, simulate the multi-scale grid structure, and generate multi-scale damage results.

[0210] In the embodiment of the application, the finite element software is Abaqus. Specifically, in Abaqus, the contact-based cohesive modeling strategy is adopted, the contact surface between the entity grids in the damage path area is defined, and the contact relationship with cohesive behavior is applied. The cohesive contact model uses the Traction-Separation rule to represent the interface mechanical response, and combines the damage maximum stress criterion to control the damage process. This strategy can automatically generate crack propagation behavior when the contact surface meets the fracture condition, avoiding the modeling limitations brought by pre-fixed path, and is suitable for multi-scale damage analysis problems with dynamic evolution of path. After definition, direct calculation is performed to obtain multi-scale damage results.

[0211] Further, the damage path obtained by finite element calculation is compared with the actual observed damage path, the intersection and union ratio (IoU) of the binary data (damage area is 1, non-damage area is 0) of the two is calculated, if the result is less than 0.7, the damage prediction deep learning model and the meshing strategy are traced back, and the model optimization and strategy optimization are performed.

[0212] It should be noted that the acquisition, storage, use, processing, etc. of data in the technical solutions in the present application comply with relevant provisions of laws and regulations. The user information in the embodiments of the present application is obtained through legal and compliant channels, and the acquisition, storage, use, processing, etc. of the user information is authorized and agreed by the client.

[0213] It should be noted that the information collected in the present 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, etc. of the relevant data comply with relevant laws, regulations and standards of countries and regions, necessary security measures are taken, do not violate public order and good customs, and provide corresponding operation portals for users to choose to authorize or refuse.

[0214] It should be noted that the technical solutions provided by the present application provide corresponding operation portals for users to choose to agree or refuse the automatic decision result; if the user chooses to refuse, the expert decision process is entered.

[0215] The technical scheme of the hardened cement paste multi-scale direct coupling calculation method provided by the embodiments of the present application comprises: sample in-situ loading and computer tomography are performed on the prepared hardened cement paste to generate multi-scale volume data at different stages; multi-phase segmentation is performed on the multi-scale volume data to generate multi-valued volume data masks; a pre-constructed damage path prediction deep learning model is used to predict the multi-valued volume data masks to generate predicted damage paths; multi-scale grid division is performed on the predicted damage paths to generate divided multi-scale grid structures; direct simulation calculation is performed on the multi-scale grid structures by using a finite element software to generate multi-scale damage results, high information density data obtained based on images are used, and a trained deep learning model is used to simplify unimportant regions and retain important regions, image multi-scale and hydration process are directly coupled, the collaborative evolution of microstructure and performance of cement paste in the hydration process is realized, the structure realism is improved, the prediction accuracy and model generalization ability are improved through the multi-modal fusion mechanism, and the calculation cost is greatly reduced.

[0216] Figure 3 A structure diagram of a hardened cement paste multi-scale direct coupling calculation device provided by the embodiments of the present application, which is used to execute the above-mentioned hardened cement paste multi-scale direct coupling calculation method, as shown in Figure 3 The device comprises a multi-scale volume data generation unit 11, a multi-phase segmentation unit 12, a damage path prediction unit 13, a grid division unit 14 and a direct coupling calculation unit 15.

[0217] The multi-scale volume data generation unit 11 is used to perform sample in-situ loading and computer tomography on the prepared hardened cement paste to generate multi-scale volume data at different stages.

[0218] The multi-phase segmentation unit 12 is configured to perform multi-phase segmentation on the multi-scale body data to generate a multi-valued body data mask.

[0219] The damage path prediction unit 13 is configured to predict the multi-valued body data mask by using a pre-constructed damage path prediction deep learning model to generate a predicted damage path.

[0220] The mesh division unit 14 is configured to perform multi-scale mesh division on the predicted damage path to generate a divided multi-scale mesh structure.

[0221] The direct coupling calculation unit 15 is configured to perform direct simulation calculation on the multi-scale mesh structure by using a finite element software to generate a multi-scale damage result.

[0222] In the embodiment of the present application, the multi-scale body data generation unit 11 is specifically configured to perform sample in-situ loading on the hardened cement paste by using an online loading device, and perform scanning on the hardened cement paste in the loading process by using a computer tomography device to generate multi-scale in-situ characterization data at different stages; and perform data reconstruction on the multi-scale in-situ characterization data by using a reconstruction algorithm to generate multi-scale body data at different stages.

[0223] In the embodiment of the present application, the multi-phase segmentation unit 12 is specifically configured to perform super-resolution on the multi-scale body data by using a super-resolution model to generate a microstructure; and perform semantic segmentation on the microstructure by using a computer vision semantic segmentation model to generate a multi-valued body data mask.

[0224] In the embodiment of the present application, the multi-valued body data mask is a multi-valued body data mask at an initial moment; and the damage path prediction unit 13 is specifically configured to perform vector representation on the multi-valued body data mask at the initial moment by using a multi-layer perception and a time embedding function to generate an input sequence; and perform encoding and decoding processing on the input sequence by using an encoding and decoding structure to generate a predicted damage path corresponding to a prediction moment.

[0225] In the embodiment of the present application, the device further comprises a sample body data generation unit 16, a sample segmentation unit 17, a damage path division unit 18, a data set construction unit 19 and a model training unit 20.

[0226] The sample body data generation unit 16 is configured to perform sample in-situ loading and computer tomography scanning on the prepared hardened cement paste samples at different ages to generate multi-scale sample body data at different ages and different stages.

[0227] The sample segmentation unit 17 is configured to perform multi-phase segmentation on the multi-scale sample body data to generate a multi-valued sample body data mask.

[0228] The damage path division unit 18 is configured to divide the multi-valued sample body data mask to generate sample damage paths at different ages and different loading stages.

[0229] The data set construction unit 19 is configured to construct a data set according to the multi-valued sample body data mask and the sample damage paths in different loading stages of different ages.

[0230] The model training unit 20 is configured to train a deep learning model through the data set, and construct a damage path prediction deep learning model.

[0231] In the embodiment of the present application, the data set includes the multi-valued sample body data mask and the sample damage paths in different loading stages of different ages; the model training unit 20 is specifically configured to generate the sample predicted damage paths corresponding to different time points by performing damage path prediction on the multi-valued sample body data mask through the deep learning model; perform loss calculation on the sample predicted damage paths corresponding to different time points and the sample damage paths in different loading stages of different ages to obtain a loss result; and perform model parameter iterative optimization on the deep learning model according to the loss result to generate the damage path prediction deep learning model.

[0232] In the embodiment of the present application, the grid division unit 14 is specifically configured to screen a high-risk damage area from the predicted damage path; establish a finite element model, perform refined grid division on the high-risk damage area, perform homogenized coarse grid processing on a non-high-risk damage area, and correct a grid transition area to generate a multi-scale grid structure.

[0233] In the scheme of the embodiment of the present application, the prepared hardened cement paste is subjected to sample in-situ loading and computer tomography to generate multi-scale body data in different stages; the multi-scale body data is subjected to multi-phase segmentation to generate a multi-valued body data mask; the multi-valued body data mask is predicted through a pre-constructed damage path prediction deep learning model to generate a predicted damage path; the predicted damage path is subjected to multi-scale grid division to generate a divided multi-scale grid structure; the multi-scale grid structure is simulated through a multi-scale finite element direct calculation coupling model to generate a multi-scale damage result; based on high information density data obtained through images, and by using a training deep learning model, non-important areas are simplified and important areas are reserved, image multi-scale is directly coupled with a hydration process, the collaborative evolution of microstructure and performance of cement paste in the hydration process is realized, the structural realism is improved, the prediction accuracy and model generalization ability are improved through a multi-modal fusion mechanism, and the calculation cost is greatly reduced.

[0234] The systems, apparatuses, modules or units disclosed 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, and specifically, the computer device can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0235] The computer device provided by the embodiment of the present application includes a memory and a processor. The memory is configured to store information including program instructions. The processor is configured to control execution of the program instructions. The program instructions are loaded and executed by the processor to implement each step of the above-described embodiment of the method for multi-scale direct coupling calculation of hardened cement paste. For specific description, refer to the above-described embodiment of the method for multi-scale direct coupling calculation of hardened cement paste.

[0236] Reference is made below to Figure 4 which shows a structural schematic diagram of a computer device 600 suitable for implementing the embodiments of the present application.

[0237] As shown in Figure 4 , the computer device 600 includes a central processing unit (CPU) 601, which can perform various appropriate operations and processes according to programs stored in a read-only memory (ROM) 602 or programs loaded from a storage portion 608 to a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the computer device 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0238] The following components are connected to the I / O interface 605: an input portion 606 including a keyboard, a mouse, and the like; an output portion 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), and the like, and a speaker, and the like; a storage portion 608 including a hard disk, and the like; and a communication portion 609 including a network interface card such as a LAN card, a modem, and the like. The communication portion 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as necessary. A removable medium 611 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is attached to the drive 610 as necessary, so that a computer program read therefrom is installed in the storage portion 608 as necessary.

[0239] In particular, the processes described above with reference to the flow charts can be implemented as computer software programs in accordance with embodiments of the application. For example, embodiments of the application include a computer program product which includes a computer program tangibly embodied on a machine readable medium, the computer program including program code for executing the methods illustrated by the flow charts. In such embodiments, the computer program can be downloaded and installed from a network via the communications portion 609 and / or installed from the removable media 611.

[0240] Computer readable media includes permanent and non-permanent, removable and non-removable media which can be implemented by any method or technology for storage of information. The information can be computer readable instructions, data structures, program modules 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, compact disc read only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer readable media does not include transitory media such as modulated data signals and carrier waves.

[0241] For the sake of description, the above apparatus is described in various units with functions respectively. Of course, the functions of each unit can be implemented in the same or more software and / or hardware when implementing the present application.

[0242] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts 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, a special purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks Figure 1 The device that implements the functions specified in one flow or multiple flows and / or blocks

[0243] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0244] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the Figure 1 function specified in the flow or flows and / or blocks Figure 1 of the block or blocks.

[0245] It should also be noted that the term "comprising" or "including" or any other variation thereof is intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element preceded by "comprises... a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that includes the recited element.

[0246] The acquisition, storage, use, processing, etc. of data in the technical solutions of the present application comply with relevant provisions of national laws and regulations.

[0247] It should be noted that in the embodiments of the present application, some software, components, models, etc. of the prior art can be mentioned, which should be considered as exemplary, and the purpose is only to illustrate the feasibility of the implementation of the technical solutions of the present application, but it does not mean that the applicant has or will necessarily use the scheme.

[0248] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented 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.

[0249] The application can be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, and the like, that perform particular tasks or implement particular abstract data types. The application can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in local and remote computer storage media including memory storage devices.

[0250] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, the system embodiments are described simply because they are basically similar to the method embodiments, and the relevant parts can be referred to the description of the method embodiments.

[0251] The above only describes the embodiments of the application and is not intended to limit the application. The application can have various changes and modifications for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the application shall be included in the scope of the claims of the application.

Claims

1. A method for multi-scale direct coupling calculation of hardened cement paste, characterized in that, The method comprises: The prepared hardened cement paste is subjected to in-situ loading and computer tomography scanning to generate multiscale volume data at different stages, specifically comprising: The hardened cement paste is subjected to in-situ loading by an online loading device, and the hardened cement paste is scanned by a computer tomography scanning device during the loading process to generate multiscale in-situ characterization data at different stages; The multiscale in-situ characterization data is subjected to data reconstruction by a reconstruction algorithm to generate multiscale volume data at different stages; The multiscale volume data is subjected to multi-phase segmentation to generate a multi-valued volume data mask, specifically comprising: The multiscale volume data is subjected to super-resolution by a super-resolution model to generate a microstructure; The microstructure is subjected to semantic segmentation by a computer vision semantic segmentation model to generate a multi-valued volume data mask, which is an initial-time multi-valued volume data mask; The multi-valued volume data mask is subjected to prediction by a pre-constructed damage path prediction deep learning model to generate a predicted damage path, specifically comprising: The initial-time multi-valued volume data mask is subjected to vector representation by a multi-layer perception and a time embedding function to generate an input sequence; The input sequence is subjected to encoding and decoding processing by an encoding and decoding structure to generate a predicted damage path corresponding to a prediction time; The predicted damage path is subjected to multiscale grid division to generate a divided multiscale grid structure, specifically comprising: High-risk damage areas are screened from the predicted damage path; A finite element model is established, the high-risk damage areas are subjected to refined grid division, the non-high-risk damage areas are subjected to homogenized coarse grid processing and grid transition area correction to generate a multiscale grid structure; The multiscale grid structure is subjected to direct simulation calculation by a finite element software to generate a multiscale damage result.

2. The method of claim 1, wherein, The method further comprises: The prepared hardened cement paste samples at different ages are subjected to in-situ loading and computer tomography scanning to generate multiscale sample volume data at different ages and different stages; The multiscale sample volume data is subjected to multi-phase segmentation to generate a multi-valued sample volume data mask; The multi-valued sample volume data mask is divided to generate sample damage paths at different ages and different loading stages; A data set is constructed according to the multi-valued sample volume data mask and the sample damage paths at different ages and different loading stages; The deep learning model is trained by the data set to construct the damage path prediction deep learning model.

3. The method for multi-scale direct coupling calculation of hardened cement paste according to claim 2, characterized in that, The data set comprises the multi-valued sample volume data mask and the sample damage paths at different ages and different loading stages; The deep learning model is trained by the data set to construct the damage path prediction deep learning model, comprising: The multi-valued sample volume data mask is subjected to damage path prediction by the deep learning model to generate sample predicted damage paths corresponding to different times; The sample predicted damage paths corresponding to different times and the sample damage paths at different ages and different loading stages are subjected to loss calculation to obtain a loss result; According to the loss result, model parameter iterative optimization is performed on the deep learning model, and the damage path prediction deep learning model is generated.

4. A hardened cement paste multiscale direct coupling computing device, characterized by, The device comprises: A multi-scale body data generation unit is configured to perform sample in-situ loading on the prepared hardened cement paste and computer tomography, and generate multi-scale body data at different stages; A multi-phase segmentation unit is configured to perform multi-phase segmentation on the multi-scale body data, and generate multi-valued body data masks; A damage path prediction unit is configured to perform prediction on the multi-valued body data masks by using a pre-constructed damage path prediction deep learning model, and generate predicted damage paths; A grid division unit is configured to perform multi-scale grid division on the predicted damage paths, and generate divided multi-scale grid structures; A direct coupling calculation unit is configured to perform direct simulation calculation on the multi-scale grid structures by using finite element software, and generate multi-scale damage results. The multi-scale body data generation unit is specifically configured to perform sample in-situ loading on the hardened cement paste by using an online loading device, perform scanning on the hardened cement paste in the loading process by using a computer tomography device, and generate multi-scale in-situ characterization data at different stages; and perform data reconstruction on the multi-scale in-situ characterization data by using a reconstruction algorithm, and generate multi-scale body data at different stages. The multi-phase segmentation unit is specifically configured to perform super-resolution on the multi-scale body data by using a super-resolution model, generate microstructures, perform semantic segmentation on the microstructures by using a computer vision semantic segmentation model, and generate multi-valued body data masks, which are multi-valued body data masks at an initial moment. The damage path prediction unit is specifically configured to perform vector representation on the multi-valued body data masks at the initial moment by using a multi-layer perception and a time embedding function, generate input sequences, perform encoding and decoding processing on the input sequences by using an encoding and decoding structure, and generate predicted damage paths corresponding to a prediction moment. The grid division unit is specifically configured to filter out high-risk damage areas from the predicted damage paths, establish a finite element model, perform refined grid division on the high-risk damage areas, perform homogenized coarse grid processing on non-high-risk damage areas, and correct grid transition areas, and generate multi-scale grid structures.

5. A computer readable medium having stored thereon a computer program, characterized in that, The program is executed by the processor to implement the hardened cement paste multi-scale direct coupling calculation method of any one of claims 1-3.

6. A computer device comprising a memory for storing information including program instructions, and a processor for controlling execution of the program instructions, characterized in that, The program instructions are loaded and executed by the processor to implement the hardened cement paste multi-scale direct coupling calculation method of any one of claims 1-3.

7. A computer program product comprising computer programs / instructions, characterized in that, The computer program / instructions are executed by the processor to implement the hardened cement paste multi-scale direct coupling calculation method of any one of claims 1-3.

Images