Freeze-thaw damage prediction method and system for high-ductility cement-based composite material fused with surface multi-scale structure recognition

Abstract

The invention provides a high-ductility cement-based composite material freeze-thaw damage prediction method and system fused with surface multi-scale structure recognition. The method comprises the following steps: S1, obtaining ECC surface damage information after a freeze-thaw cycle effect; s2, constructing an optimized ECC freeze-thaw damage strength prediction model based on a machine learning method; s3, according to the parameters obtained in the step S1, the surface damage coefficient of the ECC is determined through the prediction model in the step S2; and S4, predicting the damage strength of the ECC after freezing and thawing according to the surface damage coefficient. The invention provides an ECC freeze-thaw damage prediction model based on surface recognition, and the damage strength of a material can be speculated by analyzing surface damage picture information. According to the model, image information such as fiber exposure and surface peeling is directly associated with damage strength by using an image feature extraction and machine learning algorithm, and efficient and accurate prediction of the performance of the ECC material after freeze-thaw damage is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of ECC material damage prediction, and in particular to a freeze-thaw damage prediction method and system for high-ductility cement-based composite materials integrating surface multi-scale structure recognition. Background Art

[0002] High ductility cementitious composites (ECC) are widely used in fields with extremely high durability requirements such as bridges, roads and marine structures due to their excellent crack resistance. However, in cold environments, due to the high amount of auxiliary cementitious materials, porous structure and hydrophilic nature of ECC, mechanical properties will still degrade, but its manifestation is quite different from that of ordinary concrete. Specifically: 1. Ordinary concrete usually shows single cracks or large-area surface spalling during freeze-thaw process, and the crack width is large, while ECC shows micro-crack expansion after freeze-thaw cycle, but the cracks remain small and dispersed (usually less than 100 microns in width) and will not develop into large-scale fractures; 2. Since ECC contains fine aggregates (such as quartz sand) and appropriate amounts of fibers (such as PVA fibers), without coarse aggregates with larger particle sizes, its surface damage is uniform, some fibers are exposed, and the matrix material spalling is small and less, and relatively intact.

[0003] The reasons why the current methods based on concrete surface damage identification cannot be applied to ECC are as follows: 1. The damage of ECC is mainly a multi-level, nonlinear network of small cracks, which is often accompanied by fiber exposure or pullout, forming complex three-dimensional interface damage. It is difficult to fully capture these features by shooting from a single angle, and it is necessary to reconstruct the interactive behavior of cracks and fibers through multi-angle shooting and 3D structure identification; 2. Due to the superior performance of ECC materials, the width and depth of cracks are small, and the surface damage depth and performance of microcracks are not directly linearly related to the degree of damage like conventional concrete. In addition, the continuous exposure of fibers has limited impact on performance, but its complex manifestation increases the difficulty of evaluation, and traditional macro-geometric evaluation methods are difficult to apply; 3. ECC damage identification needs to adapt to the needs of 3D structure reconstruction, be able to finely capture the complex details of microcracks and fiber interfaces, and combine crack evolution with mechanical simulation to achieve multi-stage evaluation. In addition, ECC needs to pay more attention to crack propagation patterns and long-term performance degradation, rather than simple evaluation of single damage states.

[0004] Therefore, for these special freeze-thaw damage characteristics, conventional concrete testing methods (mainly focusing on crack detection and surface spalling) are not suitable for evaluating the damage state of ECC. Understanding and accurately identifying these specific damage characteristics is crucial for the accurate prediction and evaluation of ECC freeze-thaw damage. Summary of the invention

[0005] In view of this, the purpose of the present invention is to provide a freeze-thaw damage prediction method and system for high-ductility cement-based composite materials that integrates surface multi-scale structure recognition, and the damage strength of the material can be inferred by analyzing the surface damage photo information. The model uses image feature extraction and machine learning algorithms to establish a direct correlation between image information such as fiber exposure and surface peeling and damage intensity, thereby achieving efficient and accurate prediction of ECC material performance after freeze-thaw damage.

[0006] To achieve the above object, the present invention adopts the following technical solution: a method for predicting freeze-thaw damage of high-ductility cement-based composite materials integrating surface multi-scale structure recognition, comprising the following steps:

[0007] Step S1, obtaining surface damage image information of the ECC material under the action of freeze-thaw cycles, including surface peeling, fiber exposure, microcracks and pores, and photographing the material surface with a high-resolution camera or microscope to capture the microscopic characteristics of the surface damage after freeze-thaw;

[0008] Step S2, based on the surface damage image information obtained in step S1, preprocessing is performed on it, and the multi-level and multi-scale characteristic performance of the surface of ECC under freeze-thaw damage is combined, and the deep convolutional neural network ResNet, U-Net semantic segmentation and XGBoost classifier algorithm are used to accurately extract key damage features; then, these extracted features are combined with the known information of the number of freeze-thaw cycles and material ratio for annotation, so as to construct a high-quality data set for subsequent model training and evaluation;

[0009] Step S3, training the damage strength prediction model and defining the damage coefficient. Based on the data set annotated in step S2, the ECC freeze-thaw damage strength prediction model is trained using a machine learning algorithm. Through training, the image features are associated with the actual tested strength data, thereby establishing a mapping relationship between the image features and the damage strength. At the same time, the surface damage coefficient is calculated through the model to quantify the degree of surface damage and reflect the degradation of the microstructure after freeze-thaw cycles.

[0010] Step S4, based on the surface damage coefficient, the prediction model is used to predict the strength of the ECC after freeze-thaw damage, and a damage strength index is obtained, thereby evaluating the durability of the ECC material after freeze-thaw cycles.

[0011] In a preferred embodiment, in step S1, after obtaining the surface damage information of the ECC material under the freeze-thaw cycle, a database containing high-quality surface damage images is constructed; the process of establishing this database includes the following steps:

[0012] Step S11: Sample collection and pretreatment: collect surface images of the ECC material after it has been subjected to different freeze-thaw cycles; the images are obtained by scanning electron microscope SEM or high-resolution camera involving the use of optical microscope or camera, high-resolution scanning, electron microscope or computer tomography to ensure that microscopic cracks, spalling, fiber exposure and pore expansion damage characteristics are captured;

[0013] Step S12: Data labeling: classify and label the collected image data, including freeze-thaw cycle number, material ratio, and environmental parameters; ensure that the label information of each image in the database is accurate for subsequent model training;

[0014] Step S13: Image archiving and storage: The images are classified and stored according to the number of freeze-thaw cycles and damage types to facilitate subsequent retrieval.

[0015] In a preferred embodiment, step S2 specifically includes the following steps:

[0016] Step S21: converting the collected color image into a grayscale image using linear weighted RGB;

[0017] Step S22: Image denoising and enhancement: Use algorithms such as Gaussian filtering or median filtering to remove noise in the image; at the same time, use the CLAHE method in OpenCV to perform contrast-limited adaptive histogram equalization to further enhance the contrast of the local damaged area;

[0018] Step S23: Use the deep convolutional neural network ResNet to extract features from the preprocessed grayscale image, provide deep semantic feature extraction capabilities, and capture local and global features of the damage; at the same time, combine the hybrid attention mechanism, i.e., SE module + TransformerAttention, so that the model can automatically focus on areas with significant damage features;

[0019] Step S24: Use U-Net to perform semantic segmentation of the damaged area and extract areas such as microcracks, pores, and fiber exposure; use Open3D tools to reconstruct the three-dimensional point cloud of the damaged area and analyze the multi-scale relationship between microscopic and macroscopic features; predict the damage type through a classifier and evaluate the progression of each type of damage during the freeze-thaw cycle; input the features extracted by U-Net into the XGBoost classifier to classify the damage features in the image.

[0020] In a preferred embodiment, classifying the damage features in the image specifically includes:

[0021] 1) Fiber exposure: The fiber area is extracted by grayscale difference, and the area and height of the fiber exposure area are identified as the first damage parameter;

[0022] 2) Cement-based spalling: The area and depth information of surface spalling are extracted based on grayscale differentiation and defined as the second damage parameter;

[0023] 3) Porosity and microcracks: The feature maps generated by ResNet and U-net can capture the basic locations of cracks.

[0024] In a preferred embodiment, the pores and microcracks are further processed and verified by:

[0025] (1) Morphological operation: Morphological operation is applied to highlight the crack edges and remove falsely detected noise points;

[0026] (2) Edge detection algorithm: Use Canny edge detection or Sobel operator on the feature map to further extract the clear edges of the cracks and identify the exact location and shape of pores and cracks;

[0027] (3) Generative Adversarial Network (GAN) correction: GAN is used to generate the real crack distribution and compare it with the pore and crack features extracted from the feature map to correct noise or artifacts and ensure the accuracy of damaged area identification.

[0028] In a preferred embodiment, step S3 specifically includes the following steps:

[0029] Step S31: Model training: The feature data extracted in step S2 are trained with the actual strength test data, and a mapping relationship between damage features and actual strength is established through a machine learning algorithm CNN to construct a strength prediction model for ECC after freeze-thaw damage;

[0030] Step S32: Calculation of surface damage coefficient: Based on the model output, determine a surface damage coefficient D s , which is used to quantify the degree of surface damage of ECC materials; the surface damage coefficient D s It reflects the degradation of the microstructure of the material after freeze-thaw cycles, which serves as an important parameter for subsequent strength prediction. A dynamic weight allocation algorithm is introduced for damage characteristics, and the contribution weight of each feature to damage strength prediction is adjusted according to the number of freeze-thaw cycles and environmental factors. The damage strength coefficient is expanded from a single indicator to a multidimensional parameter matrix, and a comprehensive index describing the degree of material degradation is established:

[0031]

[0032] Where: i represents the weight of the i-th microscopic damage parameter, which is set according to its influence on the damage degree; P i represents the i-th microscopic damage parameter obtained in step S24.

[0033] In a preferred embodiment, in step S4, the strength of the ECC after freeze-thaw damage is predicted based on the surface damage coefficient to obtain a damage strength index, thereby evaluating the durability of the ECC material after freeze-thaw cycles;

[0034] f after =f before ·(1-k·D s )

[0035] Where: f after Indicates the strength of the material after freeze-thaw cycles; f before represents the strength of the material before freeze-thaw cycles; k represents an empirical constant used to adjust the influence of surface damage coefficient on strength prediction.

[0036] In a preferred embodiment, step S4 specifically includes:

[0037] Step S41: Strength prediction verification and error analysis: Use the model to predict new data, use cross-validation to compare the error between the predicted value and the actual test value, and verify the model accuracy;

[0038] Step S42: damage strength trend analysis: based on the predicted damage strength value, analyze the strength variation trend with the number of freeze-thaw cycles to quantify the durability attenuation; construct time series data by sequentially collecting surface damage images after multiple freeze-thaw cycles, and introduce recurrent neural network RNN ​​to analyze the dynamic evolution of damage characteristics with the number of freeze-thaw cycles;

[0039] Step S43: Durability and remaining life assessment: Determine the durability level of the material by analyzing the strength drop, and use a regression model to predict the remaining service life; combine historical inspection data of buildings or bridges, and provide intelligent decision support for material replacement or maintenance through a cloud-based big data analysis platform.

[0040] The present invention also provides a freeze-thaw damage prediction system for high-ductility cement-based composite materials that integrates surface multi-scale structure recognition, and runs the freeze-thaw damage prediction method for high-ductility cement-based composite materials that integrates surface multi-scale structure recognition, including an ECC surface damage information acquisition module, an ECC image feature extraction module, a damage strength prediction model training and surface damage coefficient determination module, and a strength prediction module.

[0041] The present invention also provides a computer-readable storage medium, comprising: a computer program is stored on the computer-readable storage medium; when the computer program is run by a processor, the method for predicting freeze-thaw damage of high-ductility cement-based composite materials integrating surface multi-scale structure identification is executed.

[0042] Compared with the prior art, the present invention has the following beneficial effects: the ECC freeze-thaw damage prediction model based on surface recognition provided by the present invention can detect materials by non-destructive means, avoiding the limitations of destructive experiments, and can achieve efficient prediction of the strength of materials after freeze-thaw damage. At the same time, by combining machine learning algorithms with surface image feature extraction, the prediction accuracy is improved, providing effective technical support for the long-term service of ECC materials in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 Schematic diagram of the process of step S1 of the preferred embodiment of the present invention;

[0044] Figure 2 Schematic diagram of the process of step S2 of the preferred embodiment of the present invention;

[0045] Figure 3 Schematic diagram of the process of step S3 of the preferred embodiment of the present invention;

[0046] Figure 4 Schematic diagram of the process of step S4 of the preferred embodiment of the present invention;

[0047] Figure 5 This is a surface damage image of a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0049] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0050] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0051] A freeze-thaw damage prediction method for high-ductility cementitious composites integrating surface multi-scale structure identification, reference Figure 1-5 , specifically including the following steps:

[0052] Step S1, obtaining surface damage information of the ECC material under the action of freeze-thaw cycles, including surface peeling, fiber exposure, microcracks, pores, etc., and photographing the material surface with a high-resolution camera or microscope to capture the microscopic characteristics of the surface damage after freeze-thaw;

[0053] Step S2, based on the surface damage image information obtained in step S1, pre-process it, combine the multi-level and multi-scale characteristic performance of the ECC surface under freeze-thaw damage, use deep convolutional neural network (ResNet), U-Net semantic segmentation and XGBoost classifier algorithm to accurately extract key damage features (such as surface peeling, fiber exposure, pores and microcracks, etc.). Subsequently, these extracted features are combined with known information such as the number of freeze-thaw cycles and material ratio for annotation to build a high-quality data set for subsequent model training and evaluation.

[0054] Step S3, training the damage intensity prediction model and defining the damage coefficient, based on the data set annotated in step S2, using a machine learning algorithm to train the ECC freeze-thaw damage intensity prediction model. Through training, the image features are associated with the actual tested strength data, thereby establishing a mapping relationship between the image features and the damage intensity. At the same time, the surface damage coefficient is calculated through the model to quantify the degree of surface damage and reflect the degradation of the microstructure after freeze-thaw cycles.

[0055] Step S4, based on the surface damage coefficient, the prediction model is used to predict the strength of the ECC after freeze-thaw damage, and a damage strength index is obtained, thereby evaluating the durability of the ECC material after freeze-thaw cycles.

[0056] In step S1, after obtaining the surface damage information of the ECC material under the action of freeze-thaw cycles, a database containing high-quality surface damage images is constructed. The process of establishing this database includes the following steps:

[0057] Step S11: Sample collection and pretreatment: The surface images of the ECC material are collected after being subjected to different freeze-thaw cycles. The images can be obtained by scanning electron microscopy (SEM) or high-resolution camera, involving the use of optical microscope or camera, high-resolution scanning, electron microscopy or computer tomography, to ensure that damage features such as micro cracks, spalling, fiber exposure and pore expansion are captured.

[0058] Step S12: Data labeling: Classify and label the collected image data, including freeze-thaw cycle number, material ratio (such as fiber content, auxiliary cementitious material ratio), and environmental parameters (such as temperature, salt solution concentration). Ensure that the label information of each image in the database is accurate for subsequent model training.

[0059] Step S13: Image archiving and storage: The images are classified and stored according to the number of freeze-thaw cycles and damage types to facilitate subsequent retrieval.

[0060] In step S2, in order to extract the characteristic information of the ECC material after freeze-thaw damage, the image must first be grayed to make it more suitable for the processing flow of the computer vision algorithm.

[0061] Step S21, image grayscale: The collected color image is converted into a grayscale image using linear weighted RGB to highlight the texture details of the surface damage area. Grayscale processing can remove color information, making the microcracks and pore structures in the image more obvious in the grayscale difference, which is convenient for subsequent algorithms to detect damage features.

[0062] Step 22, image denoising and enhancement: Use algorithms such as Gaussian filtering or median filtering to remove noise in the image and reduce errors caused by equipment or environment. At the same time, the CLAHE method in OpenCV (such as cv2.createCLAHE()) can be used for contrast-limited adaptive histogram equalization to further enhance the contrast of local damaged areas and make details more obvious, so as to more accurately identify and extract features.

[0063] In step S2, in order to effectively identify and quantify the characteristics of ECC materials after freeze-thaw damage, a deep convolutional neural network (ResNet), U-Net semantic segmentation and XGBoost classifier were combined.

[0064] Step S23, use a deep convolutional neural network (ResNet) to extract features from the preprocessed grayscale image, provide deep semantic feature extraction capabilities, and capture local and global features of damage. At the same time, combined with a hybrid attention mechanism (SE module + TransformerAttention), the model can automatically focus on areas with significant damage features (such as fiber exposure areas, crack extension areas), while reducing interference with non-damaged areas.

[0065] Step S24, use U-Net to perform semantic segmentation of the damaged area, and extract areas such as microcracks, pores, and fiber exposure. Use Open3D tools to reconstruct the three-dimensional point cloud of the damaged area and analyze the multi-scale relationship between microscopic features (such as crack density, pore distribution) and macroscopic features (fiber exposure area, peeling area). Use a classifier to predict the type of damage and evaluate the progression of each type of damage during the freeze-thaw cycle. This is a process included in the classification and recognition process. The features extracted by U-Net are input into the XGBoost classifier to classify the damage features in the image. Specific recognition classification includes:

[0066] Fiber exposure: The fiber area is extracted by grayscale difference, and the area and height of the fiber exposure area are identified as damage parameters1.

[0067] Cement-based spalling: The area and depth information of surface spalling are extracted based on grayscale differentiation and defined as damage parameter 2.

[0068] Porosity and microcracks: The feature maps generated by ResNet and U-net can capture the basic location of cracks, but since microcracks are small and irregular, there may still be noise and misjudgment. Further processing and verification are carried out in the following ways:

[0069] Morphological operations: Apply morphological operations (such as dilation and erosion operations) to highlight crack edges and remove false positives.

[0070] Especially for intermittent cracks, morphological operations can connect them into more complete lines.

[0071] Edge detection algorithm: Use Canny edge detection or Sobel operator on the feature map to further extract cracks

[0072] Clear edges can be used to identify the exact location and shape of pores and cracks.

[0073] Generative Adversarial Network (GAN) Correction: Use GAN to generate the real crack distribution, compare it with the pore and crack (including area and penetration) features extracted from the feature map, correct noise or artifacts, and ensure the accuracy of damaged area identification.

[0074] In step S3, the damage intensity prediction model is trained and the surface damage coefficient is determined. Based on the damage features extracted in S2 and the annotated damage parameters, a damage intensity prediction model is constructed and trained, and the surface damage coefficient is calculated, which specifically includes the following steps:

[0075] Step S31: Model training: The feature data extracted in S2 are trained with the actual strength test data, and a mapping relationship between damage features and actual strength is established through a machine learning algorithm (CNN) to construct a strength prediction model for ECC after freeze-thaw damage.

[0076] Step S32: Calculation of surface damage coefficient: Based on the model output, determine a surface damage coefficient D s, which is used to quantify the degree of surface damage of ECC materials. This coefficient reflects the degradation of the microstructure of the material after freeze-thaw cycles and can be used as an important parameter for subsequent strength prediction. A dynamic weight distribution algorithm is introduced for damage characteristics (such as fiber exposure area, crack density, pore size, etc.), and the contribution weight of each feature to damage strength prediction is adjusted according to the number of freeze-thaw cycles and environmental factors (such as temperature fluctuation amplitude, salt solution concentration, etc.). In addition, the damage intensity coefficient is expanded from a single indicator to a multidimensional parameter matrix to establish a comprehensive index that describes the degree of material degradation.

[0077]

[0078] Where: i --The weight of the i-th microscopic parameter is set according to its influence on the damage degree;

[0079] P i —The i-th damage parameter obtained in S24. i is used to represent the index of the damage parameter, corresponding to the weight ω i and damage parameter P i The i-th item of .

[0080] The index i refers to the i-th microscopic damage parameter.

[0081] It is also used to represent the association between the corresponding weight and the damage parameter, that is, ω i and P i There is a one-to-one correspondence.

[0082] In step S4, the strength of the ECC after freeze-thaw damage is predicted based on the surface damage coefficient to obtain a damage strength index, thereby evaluating the durability of the ECC material after freeze-thaw cycles.

[0083] f after =f before ·(1-k·D s )

[0084] Where: f after --The strength of the material after freeze-thaw cycles;

[0085] f before -- Strength of materials before freeze-thaw cycles;

[0086] k - an empirical constant used to adjust the effect of the surface damage factor on strength prediction.

[0087] Step S41: Strength prediction verification and error analysis: Use the model to predict new data, use cross-validation to compare the error between the predicted value and the actual test value, and verify the model accuracy.

[0088] Step S42: Damage strength trend analysis: Based on the predicted damage strength value, analyze the strength change trend with the number of freeze-thaw cycles to quantify the durability attenuation. By sequentially collecting surface damage images after multiple freeze-thaw cycles, constructing time series data, and introducing a recurrent neural network (RNN) to analyze the dynamic evolution of damage characteristics with the number of freeze-thaw cycles.

[0089] Step S43: Durability and remaining life assessment: Determine the durability level of the material by analyzing the strength drop, and use a regression model to predict the remaining service life. Combined with the historical inspection data of the building or bridge, the cloud big data analysis platform provides intelligent decision support for material replacement or maintenance.

[0090] Based on the above-mentioned freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure identification, the present invention also provides a freeze-thaw damage prediction system for high-ductility cement-based composite materials integrating surface multi-scale structure identification, which can directly obtain the surface damage characteristics of ECC materials on site and evaluate their strength and durability in real time.

[0091] Table 1: Extraction of post-damage feature information

[0092]

Claims

1. A freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure identification, characterized in that: The following steps are involved: Step S1, obtaining surface damage image information of the ECC material under the action of freeze-thaw cycles, including surface peeling, fiber exposure, microcracks and pores, and photographing the material surface with a high-resolution camera or microscope to capture the microscopic characteristics of the surface damage after freeze-thaw; Step S2, based on the surface damage image information obtained in step S1, preprocessing is performed on it, and the multi-level and multi-scale characteristic performance of the surface of ECC under freeze-thaw damage is combined, and the deep convolutional neural network ResNet, U-Net semantic segmentation and XGBoost classifier algorithm are used to accurately extract key damage features; then, these extracted features are combined with the known information of the number of freeze-thaw cycles and material ratio for annotation, so as to construct a high-quality data set for subsequent model training and evaluation; Step S3, training the damage strength prediction model and defining the damage coefficient. Based on the data set annotated in step S2, the ECC freeze-thaw damage strength prediction model is trained using a machine learning algorithm. Through training, the image features are associated with the actual tested strength data, thereby establishing a mapping relationship between the image features and the damage strength. At the same time, the surface damage coefficient is calculated through the model to quantify the degree of surface damage and reflect the degradation of the microstructure after freeze-thaw cycles. Step S4, based on the surface damage coefficient, the prediction model is used to predict the strength of the ECC after freeze-thaw damage, and a damage strength index is obtained, thereby evaluating the durability of the ECC material after freeze-thaw cycles.

2. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 1 is characterized in that: In step S1, after obtaining the surface damage information of the ECC material under the freeze-thaw cycle, a database containing high-quality surface damage images is constructed; the process of establishing this database includes the following steps: Step S11: Sample collection and pretreatment: collect surface images of the ECC material after it has been subjected to different freeze-thaw cycles; the images are obtained by scanning electron microscope SEM or high-resolution camera involving the use of optical microscope or camera, high-resolution scanning, electron microscope or computer tomography to ensure that microscopic cracks, spalling, fiber exposure and pore expansion damage characteristics are captured; Step S12: Data labeling: classify and label the collected image data, including freeze-thaw cycle number, material ratio, and environmental parameters; ensure that the label information of each image in the database is accurate for subsequent model training; Step S13: Image archiving and storage: The images are classified and stored according to the number of freeze-thaw cycles and damage types to facilitate subsequent retrieval.

3. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 1, characterized in that: Step S2 specifically includes the following steps: Step S21: converting the collected color image into a grayscale image using linear weighted RGB; Step S22: Image denoising and enhancement: Use algorithms such as Gaussian filtering or median filtering to remove noise in the image; at the same time, use the CLAHE method in OpenCV to perform contrast-limited adaptive histogram equalization to further enhance the contrast of the local damaged area; Step S23: Use the deep convolutional neural network ResNet to extract features from the preprocessed grayscale image, provide deep semantic feature extraction capabilities, and capture local and global features of the damage; at the same time, combine the hybrid attention mechanism, i.e., SE module + TransformerAttention, so that the model can automatically focus on areas with significant damage features; Step S24: Use U-Net to perform semantic segmentation of the damaged area and extract areas such as microcracks, pores, and fiber exposure; use Open3D tools to reconstruct the three-dimensional point cloud of the damaged area and analyze the multi-scale relationship between microscopic and macroscopic features; predict the damage type through a classifier and evaluate the progression of each type of damage during the freeze-thaw cycle; input the features extracted by U-Net into the XGBoost classifier to classify the damage features in the image.

4. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 3 is characterized in that: Classifying damage features in images specifically includes: 1) Fiber exposure: The fiber area is extracted by grayscale difference, and the area and height of the fiber exposure area are identified as the first damage parameter; 2) Cement-based spalling: The area and depth information of surface spalling are extracted based on grayscale differentiation and defined as the second damage parameter; 3) Porosity and microcracks: The feature maps generated by ResNet and U-net can capture the basic locations of cracks.

5. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 4, characterized in that: The pores and microcracks were further processed and verified by; (1) Morphological operation: Morphological operation is applied to highlight the crack edges and remove falsely detected noise points; (2) Edge detection algorithm: Use Canny edge detection or Sobel operator on the feature map to further extract the clear edges of the cracks and identify the exact location and shape of pores and cracks; (3) Generative Adversarial Network (GAN) correction: GAN is used to generate the real crack distribution and compare it with the pore and crack features extracted from the feature map to correct noise or artifacts and ensure the accuracy of damaged area identification.

6. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 1, characterized in that: Step S3 specifically includes the following steps: Step S31: Model training: The feature data extracted in step S2 are trained with the actual strength test data, and a mapping relationship between damage features and actual strength is established through a machine learning algorithm CNN to construct a strength prediction model for ECC after freeze-thaw damage; Step S32: Calculation of surface damage coefficient: Based on the model output, determine a surface damage coefficient D s , which is used to quantify the degree of surface damage of ECC materials; the surface damage coefficient D s It reflects the degradation of the microstructure of the material after freeze-thaw cycles, which serves as an important parameter for subsequent strength prediction. A dynamic weight allocation algorithm is introduced for damage characteristics, and the contribution weight of each feature to damage strength prediction is adjusted according to the number of freeze-thaw cycles and environmental factors. The damage strength coefficient is expanded from a single indicator to a multidimensional parameter matrix, and a comprehensive index describing the degree of material degradation is established: Where: i represents the weight of the i-th microscopic damage parameter, which is set according to its influence on the damage degree; P i represents the i-th microscopic damage parameter obtained in step S24.

7. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 1, characterized in that: In step S4, the strength of the ECC after freeze-thaw damage is predicted based on the surface damage coefficient to obtain a damage strength index, thereby evaluating the durability of the ECC material after freeze-thaw cycles; f after =f before ·(1-k·D s ) Where: f after Indicates the strength of the material after freeze-thaw cycles; f before represents the strength of the material before freeze-thaw cycles; k represents an empirical constant used to adjust the influence of surface damage coefficient on strength prediction.

8. The freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition according to claim 7, characterized in that: Step 4 specifically includes: Step 41: Strength prediction verification and error analysis: Use the model to predict new data, use cross-validation to compare the error between the predicted value and the actual test value, and verify the accuracy of the model; Step 42: Damage strength trend analysis: Based on the predicted damage strength value, analyze the change trend of strength with the number of freeze-thaw cycles to quantify durability attenuation; construct time series data by sequentially collecting surface damage images after multiple freeze-thaw cycles, and introduce recurrent neural network RNN ​​to analyze the dynamic evolution of damage characteristics with the number of freeze-thaw cycles; Step 43: Durability and remaining life assessment: Determine the durability level of the material by analyzing the extent of strength decline, and use a regression model to predict the remaining service life; combine historical inspection data of buildings or bridges, and provide intelligent decision support for material replacement or maintenance through a cloud-based big data analysis platform.

9. A freeze-thaw damage prediction system for high-ductility cement-based composite materials that integrates multi-scale surface structure recognition, characterized in that A freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure recognition as described in any one of claims 1 to 8 is run, including an ECC surface damage information acquisition module, an ECC image feature extraction module, a damage strength prediction model training and surface damage coefficient determination module, and a strength prediction module.

10. A computer-readable storage medium, characterized in that: include: The computer-readable storage medium stores a computer program; when the computer program is executed by a processor, a freeze-thaw damage prediction method for high-ductility cement-based composite materials integrating surface multi-scale structure identification as described in any one of claims 1 to 8 is executed.

Images