Concrete strength prediction method based on deep learning

Abstract

The invention provides a concrete strength prediction method based on deep learning, and the method comprises the steps: obtaining a coarse aggregate surface image through employing a three-dimensional scanning method, recognizing the surface roughness of the coarse aggregate, and extracting the pore distribution characteristics of the surface of the coarse aggregate through combining with an image segmentation algorithm; carrying out feature extraction on the interface transition area by adopting wavelet transform to obtain local features of the interface transition area, and carrying out weighted calculation on the local features in combination with an adaptive weighted fusion algorithm to obtain enhanced local feature representation; fusing the marked microstructure parameters with the macroparameters of the whole concrete, constructing a concrete strength prediction model, and outputting a predicted value of the whole strength of the concrete; and predicting the weights of the regions corresponding to different intensities through an adaptive weighted fusion algorithm to obtain local features contributed to the target intensity.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a method for predicting concrete strength based on deep learning. Background Art

[0002] In the research of concrete strength prediction, the surface roughness and porosity distribution of coarse aggregates have a significant impact on the strength characteristics of the interfacial transition zone. The interfacial transition zone is a weak area between the aggregate and the cement paste in concrete, and the non-uniformity of its strength distribution mainly stems from the complexity of the aggregate surface morphology and the randomness of the porosity distribution. When the surface roughness of the aggregate is high, more local stress concentration points will be formed in the microstructure of the interfacial transition zone, resulting in significant local differences in the strength distribution. At the same time, the non-uniform distribution of porosity will further exacerbate the volatility of this strength distribution. When using a deep learning model to predict concrete strength, the attention mechanism is usually adopted to capture key features. However, when the aggregate surface morphology leads to uneven strength distribution in the interfacial transition zone, the dynamic adjustment mechanism of the attention weight matrix faces challenges. Existing attention mechanisms often have difficulty accurately capturing the subtle differences of local features when dealing with such complex scenarios, resulting in deviations in the prediction results. In addition, the coupling mechanism between the dynamic adjustment strategy of the attention weight matrix and local feature extraction is not clear, which further limits the performance of the model in practical applications. Therefore, how to design an attention mechanism in the deep learning framework that can adaptively adjust the attention area and accurately extract local features has become the core difficulty in solving this technical problem. Summary of the Invention

[0003] The present invention provides a method for predicting concrete strength based on deep learning, mainly including:

[0004] Using a three-dimensional scanning method to obtain the surface image of the coarse aggregate, identifying the surface roughness of the coarse aggregate, and combining an image segmentation algorithm to extract the pore distribution characteristics on the surface of the coarse aggregate;

[0005] According to the surface roughness and pore distribution characteristics, and based on a deep learning network, calculating the mapping relationship between the interfacial strength and the aggregate morphology, analyzing the strength distribution in the interfacial transition zone, outputting the non-uniformity characteristics of the strength distribution, and determining the interfacial transition zone in the region with uneven strength distribution;

[0006] Using a scanning electron microscope or X-ray computed tomography to identify the local strength differences in the interfacial transition zone, analyzing the microstructural parameters and macroscopic parameters inside the interfacial transition zone, and if obvious local strength differences are found, marking the corresponding microstructural parameters;

[0007] Performing feature extraction on the interfacial transition zone using wavelet transform to obtain the local features of the interfacial transition zone, and combining an adaptive weighted fusion algorithm to perform weighted calculation on the local features to obtain an enhanced local feature representation;

[0008] Fuse the marked microstructure parameters with the macroscopic parameters of the whole concrete to construct a concrete strength prediction model and output the predicted value of the overall strength of the concrete;

[0009] Obtain the weights of different strength corresponding regions predicted by the adaptive weighted fusion algorithm to obtain the local features contributing to the target strength;

[0010] If the accuracy rate and recall rate of the local features in the interfacial transition zone are improved to the target values after introducing the adaptive weighted fusion algorithm, generate the internal strength distribution map of the concrete, and analyze the agreement between the strength prediction result and the actual strength test value in combination with the surface roughness and porosity data of the coarse aggregate.

[0011] Furthermore, adopt a three-dimensional scanning method to obtain the surface image of the coarse aggregate, identify the surface roughness of the coarse aggregate, and extract the pore distribution characteristics on the surface of the coarse aggregate in combination with the image segmentation algorithm, including: adjusting the light source intensity and scanning angle of the three-dimensional scanner according to the surface reflection characteristics of the coarse aggregate, using a continuous scanning method to scan the coarse aggregate layer by layer, and generating the first group of three-dimensional point cloud data through point cloud data processing software. Perform noise filtering and data smoothing processing on the first group of three-dimensional point cloud data, use the least squares method to perform surface fitting on the point cloud data to obtain the second group of three-dimensional point cloud data. Perform watershed algorithm segmentation on the second group of three-dimensional point cloud data according to the local curvature value, calculate the standard deviation of the Euclidean distance between adjacent points in each segmentation region to obtain the regional roughness value. Classify the segmentation regions according to the preset roughness threshold. If the regional roughness value is greater than the high threshold, it is marked as a pit region; if the regional roughness value is less than the low threshold, it is marked as a flat region. Perform morphological processing on the marked pit regions, extract the connected regions through the region growing algorithm, and calculate the area, depth, and spacing parameters of each connected region. Use the density clustering algorithm to perform clustering analysis on the connected regions, identify the pore regions according to the area threshold and depth threshold, and calculate the shortest distance matrix between the pore regions. Perform network connectivity analysis on the pore regions according to the shortest distance matrix, and extract the spatial distribution characteristics of the pore clusters, including pore density, connectivity, and directionality parameters.

[0012] Furthermore, based on the surface roughness and pore distribution characteristics, the mapping relationship between the interface strength and aggregate morphology is calculated using a deep learning network, the strength distribution in the interfacial transition zone is analyzed, the non-uniformity characteristics of the strength distribution are output, and the interfacial transition zone with non-uniform strength distribution is determined, including: constructing a first morphology feature vector containing surface height distribution, mean roughness, pore depth, pore density, and pore connectivity based on the surface roughness parameters and pore distribution characteristics of the coarse aggregate, normalizing the feature data to the interval [0, 1] using the maximum-minimum normalization method, and extracting the deep expression of the morphology features through a convolutional neural network. For the normalized deep features, a three-layer fully connected network is used for feature transformation, with the number of nodes in each layer set to 128, 64, and 32 respectively, the ReLU function is used as the activation function, and the interface bonding strength is predicted through a residual connection structure to obtain the interface micro-region bonding strength value. Cubic spline interpolation operation is performed on the predicted interface micro-region bonding strength value, and an opening morphological operator is used for regional smoothing processing to calculate the strength difference between adjacent micro-regions and obtain the interface bonding strength gradient distribution map. The mean strength and standard deviation of the local region are calculated based on the strength gradient distribution map. If the standard deviation is greater than the preset threshold, the region is determined as a region with non-uniform strength distribution, and the spatial position coordinates of the region are obtained. Connected component analysis is performed on the marked region with non-uniform strength distribution, the area and perimeter of each connected component are calculated, and the shape factor is used to characterize the morphology of the non-uniform region to obtain the strength distribution characteristic parameters of the interfacial transition zone. A weight matrix is established based on the strength distribution characteristic parameters of the interfacial transition zone, and the strength values of the non-uniform regions are weighted to obtain the quantitative characterization result of the transition zone strength distribution.

[0013] Furthermore, a scanning electron microscope or X-ray computed tomography is used to identify the local strength differences in the interfacial transition zone, and the microstructural parameters and macroscopic parameters inside the interfacial transition zone are analyzed. If obvious local strength differences are found, the corresponding microstructural parameters are marked, including: performing layered scanning imaging of the interfacial transition zone with a scanning electron microscope, obtaining the first gray-scale image sequence of the interfacial transition zone through backscattered electron signals, enhancing the image contrast by using the histogram equalization method to obtain the second gray-scale image sequence. Performing image segmentation on the second gray-scale image sequence by using the maximum entropy threshold method to obtain the binary result of each layer of the image, and performing three-dimensional reconstruction on the segmented image by using the iterative backprojection algorithm to obtain the structural reconstruction image of the interfacial transition zone. Performing voxel labeling on the reconstructed image, setting the voxel size to the cubic micron level, calculating the porosity through three-dimensional connected domain analysis, and identifying the regional distribution of hydration products by using the density clustering method. For the hydration product region, calculating the contrast, entropy value, and angular second moment of the gray-level co-occurrence matrix, setting the direction angles to 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and extracting the morphological feature vector of the hydration products. Classifying the hydration products by using a support vector machine according to the morphological feature vector to obtain the distribution map of the hydration product types, and calculating the volume fraction of each type of hydration product. Using the local adaptive threshold method to mark the strength difference region. If the intensity variance within the region is more than twice the global variance, the porosity, hydration product type distribution, and volume fraction of this region are extracted as characteristic parameters. Establishing a feature mapping table for the characteristic parameters within the marked region, and recording the position coordinates of the strength difference region and the corresponding microstructural characteristic parameter values.

[0014] Furthermore, wavelet transform is used to extract features from the interface transition zone to obtain the local features of the interface transition zone. Combining with the adaptive weighted fusion algorithm, the local features are weighted and calculated to obtain the enhanced local feature representation, including: using Haar wavelet to decompose the structure image of the interface transition zone into four layers to obtain the high-frequency sub-band images and low-frequency sub-band images from the first layer to the fourth layer. Detail feature maps in the horizontal, vertical, and diagonal directions are obtained from the high-frequency sub-bands, and an approximate feature map is obtained from the low-frequency sub-band. The local variance and energy distribution of each sub-band image are calculated according to the wavelet coefficient amplitude. The Laplacian operator edge feature and gray-level co-occurrence matrix texture feature are extracted from the high-frequency sub-bands, and the gray-level histogram and shape moment feature are extracted from the low-frequency sub-bands. The information entropy values of each sub-band feature map are calculated, including gray entropy, texture entropy, and edge entropy. The weight coefficients are set for the sub-band features according to the entropy values. The higher the entropy value, the higher the weight coefficient. A bilateral filter is used to perform joint spatial and range filtering on the weighted sub-band feature maps, and the spatial standard deviation and range standard deviation are set to obtain the filtered feature maps. The filtered feature maps are enhanced using an adaptive threshold method based on the local mean, calculating the gray-level mean of the local region of the image and determining the threshold parameter according to the mean. The feature weights are adjusted according to the local region variance, calculating the variance contribution rate of each scale feature. The larger the variance contribution rate, the larger the corresponding weight. A weighted average algorithm is used to fuse the multi-scale features, and the features are combined through weight normalization to generate a local feature description vector containing edge, texture, and shape features.

[0015] Furthermore, the marked microstructure parameters are fused with the macroscopic parameters of the concrete as a whole to construct a concrete strength prediction model and output the predicted value of the concrete strength as a whole, including: performing maximum-minimum normalization on the marked microstructure parameters of the interface transition zone, converting the porosity, hydration product distribution, and interface bonding degree into the range of [0, 1], and extracting the top three feature components with the highest contribution rate through principal component transformation to construct the first eigenvector. According to the concrete mix parameters, including water-cement ratio, binder dosage, and admixture dosage, and the physical properties parameters of the aggregate, including apparent density, water absorption rate, and crushing index, the second eigenvector is established. The hierarchical clustering algorithm is used to calculate the Euclidean distance between the first eigenvector and the second eigenvector to generate a clustering tree structure, and the parameter correlation degree is determined according to the distance threshold between the nodes of the clustering tree. A five-layer neural network structure is constructed, with the number of input layer nodes being the same as the feature dimension, the hidden layer adopting a three-layer structure, the number of output layer nodes being 1, and the activation function adopting the ReLU function. The feature data are input into the corresponding layers of the network according to the microscale and macroscale respectively, and initial weight coefficients are set for each scale feature, with the sum of the weight coefficients being 1. The random gradient descent method is used to optimize the network weights, the loss function adopts the mean square error, and the learning rate is set in a dynamically decaying manner, updating the weight values in each iteration. According to the optimized network weights, a concrete strength prediction model is constructed, with the input being the multi-scale feature vector and the output being the predicted value of the concrete strength.

[0016] Furthermore, the weights of different intensity corresponding regions predicted by the adaptive weighted fusion algorithm are used to obtain the local features contributing to the target intensity, including: dividing the interfacial transition zone according to the predicted concrete strength value, calculating the weight of each region using an adaptive Gaussian kernel function, where the kernel bandwidth parameter is adaptively adjusted according to the region size, and iteratively optimizing the weight by the gradient descent method to obtain the first region contribution value. Performing local response normalization on the first region contribution value, and the calculation method is to divide the difference between the region contribution and the local mean by the local standard deviation to obtain the second region contribution value. Setting two thresholds, a high threshold and a low threshold, for the second region contribution value. The high threshold is taken as 80% of the maximum contribution of the local region, and the low threshold is taken as 20% of the minimum contribution of the local region, and extracting the weight change region located between the two thresholds. Performing boundary smoothing on the weight change region using morphological closing operation, and the structuring element is a disk-shaped with a radius of 5 pixels. Extracting the contour curve according to the smoothed region boundary, calculating the curvature value and the curvature change rate of the curve to obtain the regional morphology characteristic value. Calculating the surface roughness of the region using the moving average method, with the window size set to 5% of the region area and the sliding step set to 25% of the window size. Establishing a mapping relationship table between the morphology characteristic value and the weight change amount, and recording the corresponding relationship between the curvature value, the curvature change rate, the surface roughness and the weight adjustment amplitude. Calculating the correlation coefficient between the morphology parameter and the weight change through correlation analysis, and judging the morphology influence direction and degree according to the positive or negative sign and the numerical value of the correlation coefficient.

[0017] Furthermore, if the accuracy and recall rate of the local features in the interface transition zone are improved to the target values after introducing the adaptive weighted fusion algorithm, a concrete internal strength distribution map is generated. Combining the surface roughness and porosity data of the coarse aggregate, the degree of agreement between the strength prediction result and the actual strength test value is analyzed, including: According to the calculation result of the accuracy of the local features in the interface transition zone, the adaptive weighted fusion algorithm is used to iteratively update the feature weights. The accuracy and recall rate of the updated features are calculated by the cross-validation method. The accuracy threshold is set to 0.85 and the recall rate threshold is set to 0.80. If both indicators exceed the thresholds, the bilinear interpolation method is used to generate the first strength distribution map. The first strength distribution map is gridded, and the grid size is set to a square with a side length of 5 mm. The strength prediction values are extracted at each grid node to generate the second strength distribution map. A strength distribution curve is established by combining the surface roughness data of the coarse aggregate, and the curve is smoothed using a cubic spline function to obtain the third strength distribution map. The test point positions are selected by the stratified sampling method, and the sampling ratio is set to 30% of the total to obtain the actual strength test data and construct a validation dataset. The sum of the squared residuals between the third strength distribution map and the validation dataset is calculated using the least squares method, and 1000 repeated samplings are performed by the Bootstrap method to obtain the residual statistical data. The kernel density estimation is performed on the residual statistical data, and the bandwidth parameter is determined by the Silverman criterion to draw the residual probability density curve. A piecewise linear compensation function is established according to the surface porosity data of the coarse aggregate, and different compensation coefficients are set for different porosity intervals. The third strength distribution map is corrected using the compensation function to obtain the final strength distribution result, and the corrected strength values and prediction errors are recorded.

[0018] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0019] The present invention discloses a method for predicting the strength of concrete based on deep learning. This method obtains the surface image of the coarse aggregate through three-dimensional scanning technology, identifies the surface roughness and pore distribution characteristics, and constructs a strength distribution model for the interface transition zone in combination with a deep learning network. Further, the microstructure of the interface transition zone is analyzed using a scanning electron microscope or X-ray tomography, and wavelet transform and adaptive weighted fusion algorithm are used to extract enhanced local features. Finally, the microstructure parameters and concrete macroscopic parameters are fused to construct a multi-scale concrete strength prediction model. The present invention realizes the accurate prediction of the internal strength distribution of concrete by analyzing the influence of the aggregate morphology on the interface transition zone, providing new ideas and methods for optimizing the performance of concrete. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a method for predicting the strength of concrete based on deep learning according to the present invention.

[0021] Figure 2 Schematic diagram of a concrete strength prediction method based on deep learning according to the present invention. Detailed implementation manners

[0022] To further understand the content of the present invention, the present invention will be described in detail in combination with the accompanying drawings and embodiments. The present application will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the related invention, rather than limiting the invention. In addition, it should be noted that, for the sake of convenience of description, only the parts related to the invention are shown in the drawings.

[0023] As Figure 1 , a concrete strength prediction method based on deep learning in this embodiment may specifically include:

[0024] S101. Use three-dimensional scanning technology to collect image data on the surface of coarse aggregates to identify their roughness characteristics, and at the same time combine image segmentation algorithms to extract pore distribution characteristics, and generate spatial distribution parameters of pore groups.

[0025] S1011. During the process of collecting surface data of coarse aggregates, adjust the light source intensity and incident angle of the three-dimensional scanner according to its light reflection characteristics to optimize the scanning effect. Adopt a continuous scanning method to layer by layer obtain the point cloud data of coarse aggregates, and generate an initial three-dimensional point cloud data set through point cloud processing software. Among them, the light source power of the light-colored surface is set to 350 milliwatts and the incident angle is 45 degrees, while the dark-colored surface is adjusted to 500 milliwatts and 30 degrees. The scanning spacing is kept at 0.5 millimeters to ensure data accuracy. Subsequently, perform Gaussian filtering and noise reduction processing on the initial point cloud data. The filtering window is 5×5 pixels and the standard deviation is set to 1.5 to remove the noise points caused by equipment jitter, and at the same time retain the integrity of edge details to obtain uniform and low-noise point cloud data.

[0026] S1012. Perform surface fitting and feature segmentation on the generated initial point cloud data. Among them, use the least squares method to perform third-order surface fitting on the point cloud data, and control the fitting residual within 0.1 millimeter to generate a smooth second set of three-dimensional point cloud data. Then, based on the local curvature value, use the watershed algorithm to divide the second set of point cloud data into regions. The neighborhood radius is set to 2 millimeters, calculate the principal curvature and Gaussian curvature of each point, and take the segmentation threshold of 0.5 to separate different feature regions. Calculate the standard deviation of the Euclidean distance between adjacent points in each region as the roughness index. If the standard deviation is greater than 0.8 millimeters, it is marked as a pit region, and if it is less than 0.3 millimeters, it is marked as a flat region. Subsequently, perform morphological processing on the pit region. Use a circular structural element with a radius of 3 millimeters to smooth the boundary through dilation and erosion operations, and use the pit region as the seed point to expand the connected region through the region growing algorithm. The stop condition is that the curvature of the boundary point is less than 0.3, and extract the area, depth and spacing parameters of the connected region.

[0027] S1013. Further analyze the marked pit area to extract the pore distribution characteristics. Among them, use the density clustering algorithm to cluster the connected areas, set the clustering radius to 10 mm, the minimum number of points to 3, and the areas with an area greater than 25 square millimeters and a depth greater than 2 mm are determined as pore areas. Calculate the shortest distance matrix between the pore areas, and take the distance threshold as 15 mm to analyze the connectivity of the pore groups. Finally, extract the spatial distribution parameters of the pore groups, including characteristics such as the pore density per unit area, the connectivity between adjacent pores, and the main distribution direction, etc. These parameters provide key inputs for the interface transition zone strength modeling.

[0028] Through the application of 3D scanning technology, high-precision topography data of the coarse aggregate surface can be efficiently obtained. Combining the watershed algorithm and the density clustering algorithm can accurately identify the roughness and pore distribution characteristics, laying a foundation for the training of subsequent deep learning models. The obtained spatial distribution parameters not only reflect the physical characteristics of the coarse aggregate surface but also provide reliable data support for the strength analysis of the interface transition zone.

[0029] It can be understood that the embodiments of the present invention do not overly limit the 3D scanning device model or specific algorithm parameters. Those skilled in the art can adjust the light source settings or clustering conditions according to the actual application scenarios to adapt to the feature extraction requirements of different types of coarse aggregates.

[0030] S102. According to the roughness and pore distribution characteristics of the coarse aggregate surface, use a deep learning network to establish a mapping relationship between the interface strength and the aggregate morphology, construct an interface transition zone strength distribution model to output its non-uniformity characteristics, and determine the areas with non-uniform strength distribution.

[0031] In practical applications, first extract key parameters such as the height distribution, roughness mean, and pore depth from the coarse aggregate surface. Among them, the height distribution reflects the change of the surface undulation, the roughness mean characterizes the overall roughness of the surface, and the pore depth reveals the depth characteristics of the surface microstructure. To ensure data consistency, use the maximum-minimum normalization method to normalize these parameters to the range of 0 to 1, and then extract the deep expression of the morphology characteristics through a convolutional neural network. The convolutional neural network can capture the spatial distribution laws of roughness and pores through multiple convolutional and pooling operations. For example, the convolution kernel size can be set to 3×3, the stride to 1, and the pooling layer uses 2×2 max pooling to gradually abstract the high-order information of the surface characteristics.

[0032] S1021. For the measurement data on the surface of coarse aggregates, construct a feature vector including height distribution, mean roughness, pore depth, pore density, and pore connectivity. The height value range is usually from -2.5 mm to 2.5 mm, the mean roughness is about 0.85 mm, the pore depth ranges from 0.5 mm to 3 mm, the pore density is 8 per square centimeter, and the connectivity is 0.65. Subsequently, map the feature data to a unified interval through min-max normalization and input it into a convolutional neural network. Extract deep features through three-layer convolutional operations. The number of convolutional kernels in each layer is 32, 64, and 128 respectively. The ReLU activation function is used to enhance the non-linear expression ability, generating deep expression data of the morphology features.

[0033] S1022. After obtaining the deep features, use a fully connected network for feature transformation and predict the interfacial bonding strength. The fully connected network contains three layers, and the number of nodes is set to 128, 64, and 32 respectively. The ReLU activation function is introduced between each layer to introduce non-linear characteristics. At the same time, a residual connection structure is introduced to alleviate the vanishing gradient problem in the deep network. The residual connection adds the input feature and the transformed feature to ensure the effective transmission of information in the network, and outputs the bonding strength value of the interfacial micro-region, which usually ranges from 3 MPa to 12 MPa, reflecting the strength characteristics of the local area of the interface.

[0034] Subsequently, smooth the predicted interfacial micro-region bonding strength value. Through the cubic spline interpolation algorithm, make the strength values between adjacent sampling points transition more naturally and eliminate the noise impact brought by mutations. Then, use morphological opening operation to smooth the intensity distribution. The operator uses a circular structuring element with a radius of 5 pixels. Through the operations of erosion first and then dilation, remove small-scale noise and retain the main intensity distribution trend. On this basis, calculate the intensity difference between adjacent micro-regions to generate an intensity gradient distribution map. Areas with higher gradient values often correspond to roughness mutations or pore densifications. For example, the maximum gradient value can reach 4 MPa / mm, indicating significant intensity changes in the interface transition region.

[0035] For the intensity gradient distribution map, further calculate the mean intensity and standard deviation of the local area to quantify its distribution characteristics. For example, the mean intensity of a certain area may be 7.5 MPa, and the standard deviation is 2.1 MPa. If the standard deviation exceeds a preset threshold, such as 1.8 MPa, then determine that this area is an area with uneven intensity distribution and record its spatial coordinates. These uneven areas usually appear in the depressions or pore densifications on the surface of coarse aggregates and have a high risk of stress concentration.

[0036] S1023. For the marked uneven intensity distribution region, use the connected component analysis method to calculate its morphological features. The area and perimeter of the connected component are obtained through pixel statistics and boundary tracking respectively. For example, the area of the largest connected component can reach 12 square millimeters and the perimeter is 15 millimeters. Subsequently, calculate the shape factor as a morphological characterization index. The shape factor is defined as 4π times the area divided by the square of the perimeter, and the typical value is 0.72. Based on these characteristic parameters, construct a weight matrix, where the weight value is proportional to the area and shape factor of the connected component. The larger the area or the more irregular the shape of the region, the higher the weight. Through weighted calculation, obtain the intensity distribution characteristic parameters of the interface transition region, showing that the intensity value gradually increases from 3.5 MPa to 9.8 MPa in the thickness direction.

[0037] The key to constructing the intensity distribution model of the interface transition region lies in the deep extraction ability of the deep learning network for morphological features. The convolutional neural network can effectively capture the spatial correlation of roughness and pores, and the residual connection structure enhances the model's prediction ability for complex features. In addition, through the combination of cubic spline interpolation and morphological processing, the continuity and accuracy of the intensity distribution map are ensured. The obtained characteristic parameters of the uneven region not only reveal the intensity change law of the interface transition region but also provide a reliable basis for multi-scale parameter fusion.

[0038] It can be understood that technicians can adjust the network structure or standardization method according to the specific type of coarse aggregate. For example, increase the depth of the convolutional layer or modify the calculation rule of the weight matrix to meet the intensity prediction requirements in different scenarios. These adjustments do not deviate from the core technical solution of the embodiments of the present invention.

[0039] S103. Analyze the microstructure and macroscopic parameters of the interface transition region by scanning electron microscopy or X-ray computed tomography method, and mark the corresponding microscopic characteristic parameters when significant intensity differences are found.

[0040] First, use scanning electron microscopy to perform high-precision imaging on the interface transition region. By adjusting the electron beam voltage to 20 kV and setting the working distance to 15 mm, backscattered electron signals are obtained. The scanning process is carried out layer by layer, with a spacing of 0.5 μm, and the magnification range is between 1000 times and 10000 times. Finally, a sequence of the first grayscale images with a resolution of 2048×2048 pixels is generated. These images reflect the microscopic structural details inside the interface transition region, such as the distribution of pores and hydration products. To improve the image quality, the histogram equalization method is used to enhance the contrast, increasing the number of gray levels from the initial 156 levels to 243 levels, thus making the microscopic features more prominent.

[0041] Such as Figure 2As shown in the figure, after preprocessing the first grayscale image sequence, the enhanced second grayscale image sequence is segmented by the maximum entropy threshold method. The maximum entropy algorithm determines the segmentation threshold by iteratively calculating the maximum value of the image information entropy. For example, the threshold may be 127, which divides the image into two parts: foreground and background, generating a binary image. Subsequently, the iterative back-projection algorithm is used to perform three-dimensional reconstruction on the binary image. During the reconstruction process, the projection angle interval is set to 1 degree and the number of iterations is 50 times to ensure the reconstruction accuracy. The spatial resolution of the obtained reconstructed image reaches 0.2 micrometers, clearly presenting the three-dimensional structural characteristics of the interface transition zone.

[0042] Next, the reconstructed image is voxelized. The voxel size is set to 0.5×0.5×0.5 cubic micrometers. The pore structure is identified through 26-neighborhood connectivity analysis and the porosity is calculated. The result shows that the average porosity of the interface transition zone is 15.3%. On this basis, the density clustering method is used to identify the hydrated product regions. The clustering radius is set to 2 micrometers and the minimum number of points is 5, which can effectively distinguish different types of hydrated products such as ettringite and calcium silicate hydrate gel. This clustering method is based on the spatial density distribution and gradually converges by iterating the clustering center to ensure the robustness of the identification result.

[0043] S1032. For the identified hydrated product regions, calculate the gray-level co-occurrence matrix to extract morphological features. The displacement distance is set to 1 pixel, and the contrast, entropy value, and angular second moment are calculated in four directions: 0 degree, 45 degrees, 90 degrees, and 135 degrees, forming a 12-dimensional feature vector. The contrast reflects the texture coarseness, the entropy value measures the randomness of the distribution, and the angular second moment represents the uniformity. Subsequently, the feature vector is input into the support vector machine for classification. The support vector machine uses the radial basis kernel function, learns the boundary of the hydrated product types through training samples, and the classification accuracy can reach 92.5%. Output the distribution map of the hydrated product types and calculate the volume fractions of various hydrated products. For example, the calcium silicate hydrate gel accounts for 32.5% and the unhydrated cement particles account for 18.6%.

[0044] When analyzing the local intensity differences, the local adaptive threshold method is used to mark the significant regions. Calculate the local variance through a sliding window. The window size is set to 21×21 pixels. If the intensity variance of a certain region exceeds twice the global variance, it is marked as a region with significant intensity differences. The porosity of these regions is usually relatively high, for example, reaching 23.7%, and the distribution of hydrated products is uneven, showing a lower content of calcium silicate hydrate gel and a higher content of unhydrated particles. After marking, establish a feature mapping table to record the position coordinates and microstructure parameters of these regions, such as porosity and the volume fraction of hydrated products. It is found that the intensity difference regions are mostly distributed in the range of 15 micrometers to 45 micrometers from the aggregate surface.

[0045] The combination of scanning electron microscopy and X-ray tomography can efficiently obtain the microscopic information of the interfacial transition zone. The application of the maximum entropy threshold method and the iterative back-projection algorithm ensures the accuracy of 3D reconstruction, while the combined use of the gray-level co-occurrence matrix and the support vector machine realizes the accurate classification of the types of hydration products. These steps not only reveal the causes of local strength differences but also provide key input data for strength prediction.

[0046] It is understandable that technicians can adjust the scanning parameters or clustering conditions according to the sample characteristics, such as increasing the voxel resolution or modifying the type of kernel function, to further improve the accuracy of feature extraction. These adjustments are all within the technical scope of the embodiments of the present invention.

[0047] S104. Perform multi-scale decomposition on the structural image of the interfacial transition zone using wavelet transform technology to obtain detail and approximation features, and combine the adaptive weighted fusion algorithm to weight these features to generate an enhanced local feature representation.

[0048] First, perform four-layer decomposition on the structural image of the interfacial transition zone using Haar wavelets. The decomposition process gradually splits the image into high-frequency sub-bands and low-frequency sub-bands. Among them, the high-frequency sub-bands capture the detail information in the horizontal, vertical, and diagonal directions, and the low-frequency sub-bands retain the main contour and approximation information of the image. Through this multi-scale analysis, the microscopic structural characteristics of the interfacial transition zone, such as edge mutations, texture distributions, and overall shapes, can be revealed at different resolutions. After decomposition, detail feature maps are extracted from the high-frequency sub-bands. For example, the horizontal high-frequency sub-band highlights the vertical edges, the vertical high-frequency sub-band reflects the horizontal edges, the diagonal high-frequency sub-band presents diagonal details, and the low-frequency sub-band provides global structural information.

[0049] For the sub-band images obtained by decomposition, further calculate the local variance and energy distribution of the wavelet coefficients to quantify the feature intensity. The local window size is set to 8×8 pixels, and by statistically analyzing the fluctuations of the coefficients within the window, the richness of details in the sub-band can be reflected. In the high-frequency sub-bands, edge features are extracted using the Laplacian operator with a 3×3 pixel operator template, which can effectively detect regions of intensity mutation; texture features are calculated using the gray-level co-occurrence matrix, considering four directions of 0 degrees, 45 degrees, 90 degrees, and 135 degrees, and the displacement distance is set to 2 pixels to capture the texture directionality and coarseness of the microscopic structure. In the low-frequency sub-band, the gray-level histogram is used to extract the brightness distribution feature with 256-level quantization, and the shape moments calculate 7 invariant moments to describe the overall geometric characteristics of the interfacial transition zone. These features together constitute the basis of the multi-scale feature expression.

[0050] S1041. After feature extraction is completed, calculate the information entropy of each sub-band feature map to measure its information content, including gray entropy, texture entropy, and edge entropy. Among them, gray entropy reflects the uniformity of brightness distribution, texture entropy represents the complexity of texture, and edge entropy quantifies the significance of edges. In actual calculation, the gray entropy value ranges from 2.5 to 4.2, the texture entropy value ranges from 1.8 to 3.5, and the edge entropy value ranges from 1.2 to 2.8. Subsequently, dynamically set the weight coefficient according to the entropy value. The larger the entropy value, the more effective information the sub-band contains, and the higher the corresponding weight coefficient. For example, the weight of the first-layer high-frequency sub-band can reach 0.35, while the weight of the fourth-layer low-frequency sub-band may be 0.15. This adaptive weight allocation method ensures the pertinence of feature fusion.

[0051] Subsequently, use a bilateral filter to process the weighted sub-band feature map to smooth the noise while retaining the edge information. Bilateral filtering performs filtering in two dimensions, the spatial domain and the range domain. The spatial standard deviation is set to 3 pixels to control the influence range of the spatial neighborhood, and the range standard deviation is set to 0.15 to adjust the weight of gray similarity. After filtering, the feature map reaches a balance between detail clarity and noise suppression. Immediately afterwards, further enhance the features through an adaptive threshold method based on local mean. The local region size is set to 16×16 pixels, and the gray mean value within the region is calculated as the threshold basis. The threshold parameter usually fluctuates between 0.4 and 0.7, thereby highlighting the significant features and weakening the background interference.

[0052] S1042. To optimize the feature weight allocation, calculate the local variance contribution rate of each scale feature to further adjust the weight. Among them, the variance contribution rate reflects the contribution degree of the feature to the overall information change. For example, the weight of the region where the variance contribution rate is greater than 0.25 increases by 0.15, and the weight of the region less than 0.1 decreases by 0.12. Dynamically optimize the weight distribution in this way. Subsequently, use a weighted average algorithm to fuse multi-scale features, integrate edge, texture, and shape features into a unified feature description vector, and finally ensure that the numerical distribution of each component of the vector is between -1 and 1 through weight normalization processing, generating a local feature representation with a dimension of 128. Among them, the edge feature accounts for 32 dimensions, the texture feature accounts for 64 dimensions, and the shape feature accounts for 32 dimensions, comprehensively characterizing the structural characteristics of the interface transition zone.

[0053] The multi-scale decomposition ability of wavelet transform significantly improves the resolution and diversity of feature extraction. Especially, the high-frequency sub-band can accurately capture the local mutation features of the interface transition zone, while the low-frequency sub-band ensures the stability of the overall structure. Combining with the adaptive weighted fusion algorithm not only enhances the robustness of feature representation but also realizes the key area focus through the dynamic adjustment of entropy value and variance. The finally generated feature vector provides high-quality input data for subsequent strength prediction and can effectively reflect the correlation between the microscopic and macroscopic characteristics of the interface transition zone.

[0054] It can be foreseen that technicians can adjust the decomposition layer number or filtering parameters according to actual needs. For example, increasing the decomposition level to five layers can obtain more detailed features, or modifying the standard deviation of bilateral filtering to adapt to different noise levels. These optimizations are all within the technical framework of the embodiments of the present invention.

[0055] S105. Integrate the marked microstructure parameters of the interfacial transition zone with the macroscopic parameters of concrete, and output the strength prediction value of concrete by constructing a multi-scale parameter prediction model.

[0056] First, obtain the microstructure parameters of the interfacial transition zone, including porosity, hydration product distribution, and interfacial bonding degree. These parameters reflect the microscopic characteristics of the interfacial transition zone. For example, the porosity usually fluctuates between 5% and 35%, the characteristic value range of the hydration product distribution is between 0.2 and 0.8, and the interfacial bonding degree is between 0.3 and 0.9. To eliminate the dimension difference and facilitate subsequent processing, the maximum-minimum normalization method is used to map these parameters to the interval of 0 to 1. Subsequently, the dominant components of the features are extracted by principal component analysis. The cumulative contribution rate of the first three principal components can reach 87.5%. Among them, the first principal component is mainly related to porosity, the second principal component captures the characteristics of hydration product distribution, and the third principal component is associated with the change of interfacial bonding degree. Based on this, the first feature vector is constructed to characterize the microscopic scale characteristics.

[0057] S1051. For the overall characteristics of concrete, collect macroscopic parameters to construct the second feature vector. The macroscopic parameters include mix proportion parameters and aggregate physical property parameters. The mix proportion parameters cover water-cement ratio, binder dosage, and admixture dosage. For example, the water-cement ratio ranges from 0.35 to 0.65, the binder dosage is from 350 to 550 kg / m³, and the admixture dosage accounts for 0.8% to 2.5% of the binder mass. The aggregate physical property parameters include apparent density, water absorption rate, and crushing index. The apparent density is usually between 2.6 and 2.8, the water absorption rate is between 0.5% and 2.8%, and the crushing index is between 8% and 16%. These parameters are also normalized to form a unified macroscopic feature vector, reflecting the potential influence of material composition and aggregate quality on strength.

[0058] Next, use the hierarchical clustering algorithm to analyze the correlation between the microscopic and macroscopic feature vectors. Hierarchical clustering gradually merges similar features by calculating the Euclidean distance to generate a clustering tree structure. In actual calculation, the maximum hierarchical distance may be 0.85. If the distance threshold is set to 0.6, reasonable parameter correlation groups can be divided. The correlation coefficient between the water-cement ratio and porosity is as high as 0.82, indicating that the two have a strong synergistic effect in strength formation; the correlation coefficient between the aggregate crushing index and interfacial bonding degree is 0.75, showing the direct influence of aggregate quality on interfacial strength. This correlation analysis provides a theoretical basis for feature fusion, ensuring that the model can capture the internal relationship between multi-scale parameters.

[0059] S1052. On the basis of feature fusion, a five-layer neural network model is constructed to achieve strength prediction. The number of nodes in the input layer is the same as the feature dimension, for example, set to 12, which includes the total dimension of micro and macro features. The hidden layer is designed with three layers, and the number of nodes is 24, 16, and 8 in sequence. The ReLU activation function is used to introduce the non-linear mapping ability. The number of nodes in the output layer is 1, and the strength prediction value is directly output. The micro features and macro features are respectively assigned initial weights. For example, the micro weight is 0.6 and the macro weight is 0.4, and the sum is 1. Subsequently, the network weights are optimized by the stochastic gradient descent method. The mean square error is selected as the loss function, and the initial learning rate is set to 0.01, and it is dynamically adjusted in the way of decaying to 0.8 times every 50 iterations. When training to 500 iterations, the loss value drops from 0.45 to 0.08. The optimized weights show that the proportion of micro features rises to 0.65 and the macro features are 0.35, highlighting the dominant role of the micro structure in strength.

[0060] When constructing the prediction model, the fusion of micro and macro features significantly improves the prediction accuracy. For example, for concrete samples with the same water-cement ratio but different porosity in the interfacial transition zone, the predicted strength difference can reach 5 MPa to 8 MPa, indicating the key role of microstructural features in strength change. The finally output strength prediction value can effectively reflect the coupling relationship between the internal micro characteristics and macro performance of concrete through the comprehensive analysis of multi-scale features.

[0061] The combination of hierarchical clustering and neural network provides an efficient way for multi-scale parameter fusion. By standardizing the microstructural parameters and extracting the principal components, the effectiveness of the features is ensured, and the incorporation of macro parameters enhances the adaptability of the model to the overall characteristics of the material. Technicians can adjust the number of network layers or the learning rate strategy according to actual needs, such as increasing the hidden layer to four layers or adopting an adaptive learning rate optimization algorithm to further improve the prediction accuracy. These improvements are all in line with the technical framework of the embodiments of the present invention.

[0062] S106. Use the adaptive weighted fusion algorithm to calculate the weights of each region in the interfacial transition zone to obtain the contribution degree of local features, and judge the influence of aggregate morphology on the characteristics of the interfacial transition zone through the mapping relationship between the morphology features and the weight change.

[0063] First, divide the interfacial transition zone according to the predicted concrete strength value, and calculate the weight of each region using an adaptive Gaussian kernel function. The core of the Gaussian kernel function lies in controlling the distribution range of the weight through the bandwidth parameter. The bandwidth size is adaptively adjusted according to the area of the region. For example, for a region with an area of 100 square micrometers, the initial bandwidth is set to 10 micrometers. Subsequently, the weight is iteratively optimized by the gradient descent method. The initial value of the learning rate is 0.01, and the adjustment amplitude for each iteration is 10% of the current value. When the weight change rate is less than 0.1%, the iteration stops. Usually, about 50 rounds are required, and finally the bandwidth may converge to 7.5 micrometers to obtain the contribution value of the first region, reflecting the preliminary influence degree of each region on the strength.

[0064] S1061. Perform local response normalization on the contribution value of the first region. Standardize the data by calculating the mean and standard deviation within the window. The window size is set to 15×15 pixels. The specific operation is to subtract the window mean from the contribution value of each region and then divide by the standard deviation. The result values are distributed between -2.5 and 2.5. Subsequently, set high and low thresholds for the contribution value of the second region to extract the weight change region. The high threshold is taken as 80% of the local maximum contribution, for example, about 2.0, and the low threshold is taken as 20% of the minimum contribution, about -0.5. The extracted change region is concentrated in the middle part of the interfacial transition zone. This normalization and threshold setting method effectively highlights the relative differences in the contribution degrees between regions, providing a clear basis for analysis.

[0065] Next, perform boundary smoothing on the extracted weight change region. Use morphological closing operation, and the structure element is selected as a disk shape with a radius of 5 pixels. Eliminate boundary noise through the operations of dilation first and then erosion to make the region contour more continuous and smooth. On this basis, extract the smoothed boundary contour curve, and calculate the curvature value by the three-point method. The sampling point spacing is set to 2 micrometers. The curvature value ranges from 0.05 to 0.8, and the curvature change rate ranges from 0.02 to 0.15. Further, use the moving average method to calculate the surface roughness of the region. The window size is dynamically adjusted according to the area of the region. For example, for a 200-square-micrometer region, the window is set to 10 square micrometers, and the step size is 2.5 square micrometers. The roughness value is distributed between 0.2 and 1.5 micrometers. These morphological features provide multi-dimensional data support for the mapping relationship.

[0066] S1062. After obtaining the morphological features, establish a mapping relationship table between the curvature value, the curvature change rate, the surface roughness, and the weight adjustment amplitude. When the curvature value exceeds 0.5 and the curvature change rate is greater than 0.1, the weight adjustment amplitude can reach more than 50%, indicating that the region with drastic morphological changes contributes significantly to the strength. Subsequently, calculate the correlation coefficient between the morphological parameters and the weight change through correlation analysis. The results show that the curvature value is positively correlated with the weight change, with a correlation coefficient of 0.75, and the surface roughness is negatively correlated with the weight change, with a correlation coefficient of -0.62, revealing the dual influence mechanism of the morphological features on the strength of the interfacial transition zone, that is, the high-curvature region may cause stress concentration, while the surface with higher roughness improves the interfacial stability by increasing the bonding area.

[0067] The combination of the adaptive Gaussian kernel function and the gradient descent method ensures the flexibility and accuracy of the weight calculation, and the local response normalization enhances the comparability of the regional contribution degree. Through morphological processing and morphological feature extraction, the complexity of the local structure of the interfacial transition zone is clearly characterized. The mapping relationship table and the correlation analysis provide quantitative evidence for the mechanism of the aggregate morphology affecting the strength, indicating that the morphological features significantly affect the overall strength distribution by adjusting the interfacial properties.

[0068] It can be foreseen that technicians can adjust the Gaussian kernel bandwidth or the window size according to actual needs. For example, set the initial bandwidth value to 15 microns or increase the window to 20×20 pixels to adapt to the feature analysis at different scales. These optimizations are all within the technical scope of the embodiments of the present invention.

[0069] S107. If the adaptive weighted fusion algorithm improves the accuracy and recall rate of the local features of the interfacial transition zone to the target level, generate the internal strength distribution map of the concrete, and combine the surface roughness and porosity data analysis of the coarse aggregate to verify the consistency between the prediction result and the measured value to verify the effectiveness of the model.

[0070] First, use the adaptive weighted fusion algorithm to iteratively update the weights of the local features of the interfacial transition zone. The initial weight is set to 0.5, and it is dynamically adjusted according to the change in accuracy in each round of iteration. For example, when the accuracy improvement exceeds 2%, increase it by 0.05, and when it is lower than 1%, decrease it by 0.03. After about 25 rounds of iteration, the accuracy increases from 0.78 to 0.87, and the recall rate increases from 0.75 to 0.83. Subsequently, evaluate the performance of the updated features through the cross-validation method. Set the accuracy threshold to 0.85 and the recall rate threshold to 0.80. If both meet the standards, use the bilinear interpolation method to generate the first strength distribution map, with a resolution of up to 1 square millimeter, ensuring that the spatial details of the distribution map are retained.

[0071] S1071. Perform grid processing on the first strength distribution map. The grid size is set as a square with a side length of 5 mm. Each grid contains multiple strength prediction points. The strength difference between adjacent nodes is usually between 0.5 MPa and 2 MPa. Especially in the interface transition zone, it shows a significant gradient change, and the maximum gradient can reach 4 MPa / mm. Subsequently, combine the surface roughness data of coarse aggregates to construct a strength distribution curve. When the roughness is between 0.8 mm and 1.2 mm, the interface bonding strength is the highest, reaching 12 MPa. When it is lower than 0.5 mm or higher than 1.5 mm, the strength decreases significantly, down to 6 MPa at the lowest. Then, smooth the curve through a cubic spline function to eliminate local mutations and maintain continuity, generating the third strength distribution map.

[0072] Next, use the stratified sampling method to select test points to obtain actual strength data. The sampling ratio is set at 30%. Approximately 90 test points are collected in a 100-square-millimeter area, covering different depths in the interface transition zone. Based on these data, calculate the sum of squared residuals between the third strength distribution map and the measured values through the least squares method, and use the Bootstrap method for 1000 repeated samplings to obtain the residual statistics. The average deviation is about 1.2 MPa, and the standard deviation is 0.8 MPa. Further, use kernel density estimation to analyze the residual distribution. The bandwidth parameter is set at 0.35 according to the Silverman criterion. The drawn probability density curve is approximately normal, and the peak residuals are concentrated between -0.5 MPa and 0.5 MPa, indicating that the predicted values and the measured values are generally close, but there are still local deviations.

[0073] S1072. To further improve the accuracy, design a piecewise linear compensation function according to the surface porosity data of coarse aggregates. When the porosity is less than 10%, the compensation coefficient is 1.05. When it is between 10% and 20%, it increases to 1.15. When it exceeds 20%, it is 1.25. Use this function to correct the third strength distribution map to generate the final strength distribution result. After correction, the average deviation between the predicted value and the measured value drops to 0.6 MPa, and the maximum error decreases from 3.5 MPa to 1.8 MPa. Especially in the area with a higher porosity, the prediction accuracy is significantly improved, reflecting the precise correction effect of the compensation function on the influence of the microstructure.

[0074] The adaptive weighted fusion algorithm improves the feature representation ability through iterative optimization, and the grid processing and cubic spline smoothing ensure the continuity and readability of the strength distribution map. Combining the analysis of roughness and porosity not only verifies the prediction ability of the model for the strength in the interface transition zone, but also achieves a high degree of agreement with the measured data through piecewise compensation.

[0075] It can be foreseen that technicians can adjust the grid size or compensation coefficient according to specific requirements. For example, reduce the grid to 3 mm or refine the porosity interval to further optimize the accuracy of the distribution map. These adjustments are all within the technical scope of the embodiments of the present invention.

[0076] Only some preferred embodiments of the present invention are listed above, but the present invention is not limited thereto, and many improvements and transformations can be made. As long as the improvements and transformations are made based on the basic principles of the present invention, they shall be regarded as falling within the protection scope of the present invention.

Claims

1. A concrete strength prediction method based on deep learning, characterized in that, The method includes: Obtaining the surface image of coarse aggregate by using a three-dimensional scanning method, identifying the surface roughness of the coarse aggregate, and extracting the pore distribution characteristics on the surface of the coarse aggregate by combining an image segmentation algorithm; According to the surface roughness and pore distribution characteristics, and based on a deep learning network, calculating the mapping relationship between the interface strength and the aggregate morphology, analyzing the strength distribution in the interfacial transition zone, outputting the non-uniformity characteristics of the strength distribution, and determining the interfacial transition zone in the region with non-uniform strength distribution; Using a scanning electron microscope or X-ray computed tomography to identify the local strength differences in the interfacial transition zone, analyzing the microscopic and macroscopic parameters inside the interfacial transition zone, and if obvious local strength differences are found, marking the corresponding microscopic structure parameters; Performing feature extraction on the interfacial transition zone by using wavelet transform to obtain the local features of the interfacial transition zone, and combining an adaptive weighted fusion algorithm to perform weighted calculation on the local features to obtain an enhanced local feature representation; Fusing the marked microscopic structure parameters with the macroscopic parameters of the whole concrete to construct a concrete strength prediction model and output the predicted value of the overall strength of the concrete; Obtaining the local features contributing to the target strength through the weights of different strength corresponding regions predicted by the adaptive weighted fusion algorithm; If the accuracy and recall rate of the local features in the interfacial transition zone are improved to the target values after introducing the adaptive weighted fusion algorithm, generating a strength distribution map inside the concrete, and combining the surface roughness and porosity data of the coarse aggregate to analyze the coincidence degree between the strength prediction result and the actual strength test value.

2. The method according to claim 1, wherein The obtaining the surface image of coarse aggregate by using a three-dimensional scanning method, identifying the surface roughness of the coarse aggregate, and extracting the pore distribution characteristics on the surface of the coarse aggregate by combining an image segmentation algorithm includes: Obtaining the layer-by-layer scanning data of the coarse aggregate by using a continuous scanning method according to the surface reflection characteristics of the coarse aggregate, and obtaining the first group of three-dimensional point cloud data through point cloud data processing software; Performing surface fitting processing on the first group of three-dimensional point cloud data by using the least squares method to obtain the second group of three-dimensional point cloud data; Calculating the regional roughness value for the second group of three-dimensional point cloud data by using the watershed algorithm, and if the regional roughness value is greater than a preset high threshold, marking it as a pit area; Identifying the pore areas in the pit area by using a density clustering algorithm, and obtaining the spatial distribution characteristics of the pore groups through the shortest distance matrix between the pore areas.

3. The method according to claim 1, characterized in that, The according to the surface roughness and pore distribution characteristics, and based on a deep learning network, calculating the mapping relationship between the interface strength and the aggregate morphology, analyzing the strength distribution in the interfacial transition zone, outputting the non-uniformity characteristics of the strength distribution, and determining the interfacial transition zone in the region with non-uniform strength distribution includes: Obtaining the surface height distribution parameter, roughness mean parameter, and pore depth parameter from the surface of the coarse aggregate, performing normalization processing by using the maximum-minimum normalization method, and obtaining the deep expression data of the morphology characteristics through a convolutional neural network; Performing feature transformation on the deep expression data of the morphology characteristics by using a fully connected network, and predicting the interface bonding strength through a residual connection structure to obtain the interface micro-region bonding strength value; Perform cubic spline interpolation on the interface micro-region bonding strength values, use morphological operators for regional smoothing processing, and obtain the interface bonding strength gradient distribution map by calculating the strength difference between adjacent micro-regions; Calculate the intensity mean and standard deviation of the local region according to the interface bonding strength gradient distribution map. If the standard deviation is greater than the preset threshold, determine that the region is a region with uneven intensity distribution.

4. The method according to claim 1, wherein Use scanning electron microscopy or X-ray computed tomography to identify the local intensity differences in the interface transition zone, analyze the microscopic and macroscopic parameters inside the interface transition zone. If obvious local intensity differences are found, mark the corresponding microscopic structure parameters, including: Use scanning electron microscopy to perform layer-by-layer scanning on the interface transition zone, obtain the first gray-scale image sequence of the interface transition zone through backscattered electron signals, and obtain the second gray-scale image sequence according to the first gray-scale image sequence using the histogram equalization method; Obtain a binary image according to the second gray-scale image sequence using the maximum entropy threshold method, and perform three-dimensional reconstruction on the binary image through the iterative backprojection algorithm to obtain a reconstructed image; Perform voxel labeling on the reconstructed image to obtain a labeled image, and identify the hydrated product region from the labeled image by density clustering method; Calculate the contrast, entropy value, and angular second moment of the gray-level co-occurrence matrix for the hydrated product region to obtain a feature vector, and judge whether obvious local intensity differences occur according to the feature vector. If so, mark the corresponding microscopic structure parameters.

5. The method according to claim 1, characterized in that, Perform feature extraction on the interface transition zone using wavelet transform to obtain the local features of the interface transition zone, and combine the adaptive weighted fusion algorithm to perform weighted calculation on the local features to obtain an enhanced local feature representation, including: Perform four-layer decomposition on the interface transition zone structure image using Haar wavelet to obtain high-frequency sub-band images and low-frequency sub-band images, and obtain detail feature maps in the horizontal, vertical, and diagonal directions from the high-frequency sub-band images; Calculate the local variance and energy distribution for the sub-band images according to the wavelet coefficient amplitude, obtain the Laplacian operator edge features and gray-level co-occurrence matrix texture features from the high-frequency sub-band images, and obtain the gray-level histogram and shape moment features from the low-frequency sub-band images; Calculate the gray entropy, texture entropy, and edge entropy values for the sub-band feature maps, and set weight coefficients for the sub-band features; Use the weighted average algorithm to fuse multi-scale features, and combine the features through weight normalization to obtain an enhanced local feature representation.

6. The method according to claim 1, characterized in that, Fuse the marked microscopic structure parameters with the macroscopic parameters of the concrete as a whole, construct a concrete strength prediction model, and output the predicted value of the concrete strength as a whole, including: Obtain the microscopic structure parameters of the interface transition zone, and obtain the first feature vector by normalizing the microscopic structure parameters using the maximum and minimum normalization method. The microscopic structure parameters include porosity, hydrated product distribution, and interface bonding degree; Construct a second feature vector according to the concrete mix ratio parameters and aggregate physical property parameters. The concrete mix ratio parameters include water-cement ratio, binder dosage, and admixture dosage, and the aggregate physical property parameters include apparent density, water absorption rate, and crushing index; Calculate the Euclidean distance between the first feature vector and the second feature vector using the hierarchical clustering algorithm, and determine the parameter correlation degree according to the distance threshold between the nodes of the clustering tree; Use the first feature vector and the second feature vector as the input data of the concrete strength prediction model, and optimize the neural network weights by the stochastic gradient descent method to obtain the predicted value of the overall concrete strength.

7. The method according to claim 1, characterized in that The weights of different strength corresponding regions predicted by the adaptive weighted fusion algorithm to obtain the local features contributing to the target strength include: Calculate the weights of each region in the interfacial transition zone using the adaptive Gaussian kernel function, and iteratively optimize the weights by the gradient descent method to obtain the contribution degree value of the first region; Obtain the contribution degree value of the second region by subtracting the local mean from the contribution degree value of the first region and then dividing by the local standard deviation; Set high and low thresholds for the contribution degree value of the second region to extract the weight change region, and perform morphological closing operation on the weight change region to obtain a smooth boundary; Extract the contour curve according to the smooth boundary and calculate the curvature value and the curvature change rate; Obtain the local features contributing to the target strength.

8. The method according to claim 1, wherein If the accuracy and recall rate of the local features in the interfacial transition zone are improved to the target values after introducing the adaptive weighted fusion algorithm, generate the internal strength distribution map of the concrete, and analyze the coincidence degree between the strength prediction result and the actual strength test value in combination with the surface roughness and porosity data of the coarse aggregate, including: Iteratively update the weights of the local features in the interfacial transition zone using the adaptive weighted fusion algorithm, calculate the accuracy and recall rate of the updated features by the cross-validation method according to the feature weights. If the accuracy exceeds the accuracy threshold and the recall rate exceeds the recall rate threshold, generate the first strength distribution map; Perform grid processing on the first strength distribution map, extract the strength prediction values according to the grid nodes to obtain the second strength distribution map; Establish a strength distribution curve according to the second strength distribution map in combination with the surface roughness data of the coarse aggregate, and smooth the strength distribution curve by the cubic spline function to obtain the third strength distribution map; Establish a piecewise linear compensation function according to the porosity data on the surface of the coarse aggregate, and correct the third strength distribution map by the piecewise linear compensation function to obtain the final strength distribution result.

Images