A pavement crack analysis method

Abstract

The present invention discloses a pavement crack analysis method, comprising: preparing a near-surface deformation structure sample of pavement steel bars, performing near-surface feature extraction on the sample to obtain surface deformation feature samples, constructing a near-surface deformation feature matrix comprising N*K samples, performing corrosion feature extraction on the near-surface deformation structure sample, constructing a corrosion deformation feature matrix comprising N*K samples based on the corrosion deformation feature samples, extracting a life label matrix, constructing a training set comprising N*K supervised learning sample pairs, performing model training based on the training set comprising N*K supervised learning sample pairs, obtaining a corrosion life prediction model, predicting the corrosion life of the pavement steel bars, and performing pavement crack analysis. The present invention achieves a comprehensive analysis of pavement cracks caused by steel bar corrosion by comprehensively analyzing the local deformation structure and microstructure characteristics of the pavement steel bars.

Description

Technical Field

[0001] The present invention relates to the field of pavement crack analysis, and in particular to a pavement crack analysis method. Background Art

[0002] In traditional road maintenance, cracks are often identified a posteriori through surface imaging, displacement monitoring, or stress measurement. However, these methods lack a reliable early prediction mechanism for structural damage caused by internal steel corrosion. Current research focuses primarily on the macroscopic level, typically modeling and evaluating cracks through stress corrosion models, empirical formulas, or finite element simulations. However, such methods often fail to capture the true corrosion evolution of steel at the microscopic scale. Specifically, the corrosion behavior of steel in complex service environments is significantly affected by its surface microstructure. Microscopic features such as local pitting initiation, grain evolution, and second-phase particle distribution all play a key role in the initiation and propagation of corrosion cracks. However, existing studies generally focus on the external environment of concrete, electrochemical reactions, or macroscopic structural mechanical behavior, ignoring the local deformation structure formed by the steel itself during the forming process and its pre-induction effect on corrosion behavior.

[0003] For example, during hot rolling, milling, and other forming processes, localized deformation structures such as micro-strain concentration, grain distortion, and second-phase enrichment often appear on or near the surface of steel bars. The heterogeneity and stability differences of these structures directly affect the stability of the surface passivation film and have a substantial impact on the location of pitting, the propagation path, and the rate of corrosion cracks.

[0004] In summary, steel bar corrosion is a complex evolutionary process driven by both internal microstructural characteristics and external deformation structures. However, existing technologies lack methods that effectively couple these two structural factors and conduct corrosion behavior modeling and comprehensive analysis of pavement cracks based on their joint evolution mechanism. Summary of the Invention

[0005] In response to the deficiencies of the prior art, the present invention provides a pavement crack analysis method that solves the technical problems raised in the background technology through comprehensive modeling and analysis of near-surface deformation structure and microscopic characteristics.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] A pavement crack analysis method comprises the following steps:

[0008] S1. Prepare near-surface deformation structure specimens of pavement reinforcement;

[0009] S2. extracting near-surface features of the near-surface deformation structure specimens to obtain near-surface deformation feature samples of each near-surface deformation structure specimen;

[0010] S3. Constructing a near-surface deformation feature matrix containing N*K samples based on the near-surface deformation feature samples;

[0011] S4. Extracting corrosion features of the near-surface deformation structure specimens to obtain corrosion deformation feature samples of each near-surface deformation structure specimen;

[0012] S5. Constructing a corrosion deformation feature matrix containing N*K samples based on the corrosion deformation feature samples;

[0013] S6. Extracting a life label matrix based on the corrosion deformation feature matrix;

[0014] S7, combining the near-surface deformation feature matrix and the lifespan label matrix to construct a training set containing N*K supervised learning sample pairs;

[0015] S8. Perform model training based on a training set containing N*K supervised learning sample pairs to obtain a corrosion life prediction model;

[0016] S9. Use the corrosion life prediction model to predict the corrosion life of pavement reinforcement to perform pavement crack analysis.

[0017] In some specific embodiments, feature extraction is performed on a near-surface deformation structure specimen to obtain a near-surface deformation feature sample of each near-surface deformation structure specimen, including:

[0018] S2-1, obtaining M first microstructural features of the near-surface deformed structure specimen;

[0019] S2-2, performing a performance test on the near-surface deformation structure specimen to obtain a corresponding fracture specimen;

[0020] S2-3, obtaining M second microstructural features at the fracture surface of the fractured specimen;

[0021] S2-4. Construct the near-surface deformation feature sample according to the M first microstructure features and the M second microstructure features.

[0022] In some specific embodiments, constructing the near-surface deformation feature sample according to the M first microstructure features and the M second microstructure features includes:

[0023] S2-4-1. Assign consecutive specimen numbers to each near-surface deformation structure specimen;

[0024] S2-4-2. According to the sample number, obtain a first microstructural feature of the near-surface deformation structure sample corresponding to the sample number;

[0025] S2-4-3, inherit the specimen number of each fracture specimen’s near-surface deformation structure specimen before fracture;

[0026] S2-4-4. According to the inherited sample number, obtain a second microstructural feature of the fractured sample corresponding to the sample number;

[0027] S2-4-5. Pair the sample number, the first microstructure feature, and the second microstructure feature to construct a near-surface deformation feature sample.

[0028] In some specific embodiments, a near-surface deformation feature matrix containing N*K samples is constructed based on the near-surface deformation feature samples, including:

[0029] S3-1. Obtain N near-surface deformation feature samples in each performance test;

[0030] S3-2. Arrange the N near-surface deformation feature samples in each performance test in order according to the sample numbers and construct them into matrix row vectors;

[0031] S3-3. Assign consecutive test round numbers to each matrix row vector according to the test order of the performance test;

[0032] S3-4. Sort the test round numbers up and down according to the test order to obtain K matrix row vectors;

[0033] S3-5. Define the K matrix row vectors as the near-surface deformation feature matrix.

[0034] In some specific embodiments, the corrosion feature extraction is performed on the near-surface deformation structure specimen to obtain the corrosion deformation feature sample of each near-surface deformation structure specimen, including:

[0035] S4-1. performing corrosion treatment on each near-surface deformed structure specimen to obtain a corrosion specimen;

[0036] S4-2, obtaining M corrosion evolution characteristics of the corrosion sample during the corrosion treatment process;

[0037] S4-3, obtaining M corrosion structure characteristics of the corrosion sample after corrosion is completed;

[0038] S4-4. Pairing the sample numbers, corrosion evolution characteristics, and corrosion structure characteristics to construct the corrosion deformation characteristic sample;

[0039] In some specific embodiments, a corrosion deformation feature matrix containing N*K samples is constructed based on the corrosion deformation feature samples, including:

[0040] S5-1. Obtain N corrosion deformation characteristic samples in each corrosion test;

[0041] S5-2. Arrange the N corrosion deformation characteristic samples in each corrosion test in order of sample number and construct them into matrix row vectors;

[0042] S5-3, assigning consecutive corrosion test round numbers to each matrix row vector according to the execution order of the corrosion test;

[0043] S5-4, sorting the corrosion test round numbers up and down according to the test order to obtain K matrix row vectors;

[0044] S5-5. Define the K matrix row vectors as the corrosion deformation feature matrix.

[0045] In some specific embodiments, extracting a life signature matrix from the corrosion deformation feature matrix includes:

[0046] S6-1. Extracting time-based life indicators from corrosion evolution characteristics;

[0047] S6-2. Extract length or quality indicators from corrosion structure characteristics;

[0048] S6-3. According to a preset life evaluation standard, convert the time-related, length-related, or quality-related indicators into a life label with a unified dimension;

[0049] S6-4. Arrange all the life labels in order according to the sample numbers and the test round numbers to construct the life label matrix.

[0050] In some specific embodiments, the near-surface deformation feature matrix and the lifespan label matrix are combined to construct a training set containing N*K supervised learning sample pairs, including:

[0051] S7-1. Obtain the specimen number and test round number of each sample in the near-surface deformation feature matrix;

[0052] S7-2, performing ID splicing on the specimen number and the test round number of each sample to generate a near-surface feature number;

[0053] S7-3, obtaining the sample number and test round number of each tag in the life tag matrix;

[0054] S7-4, concatenate the sample number and test round number of each tag to generate a life tag number;

[0055] S7-5. If the near-surface feature number and the life label number are the same, pair the corresponding near-surface deformation feature sample with the life label to obtain a supervised learning sample pair;

[0056] S7-6, traverse the near-surface deformation feature matrix and the life label matrix until N*K supervised learning sample pairs are obtained;

[0057] S7-7. Combine N*K supervised learning samples to obtain the training set.

[0058] In some specific embodiments, model training is performed based on a training set containing N*K supervised learning sample pairs to obtain a corrosion life prediction model, including:

[0059] S8-1, extracting the first batch of sample pairs of the training set;

[0060] S8-2. Input the first batch of sample pairs into the multi-layer perception model, and after forward propagation, generate the first batch of predicted lifespans at the output layer;

[0061] S8-3, calculating the loss between the predicted lifespan of the first batch and the lifespan labels of the first batch of sample pairs;

[0062] S8-4. If the loss is lower than the threshold, the training ends; otherwise, the loss is calculated to be equivalent to the gradient of the current multi-layer perception model parameters;

[0063] S8-5. Reversely update the multi-layer perception model parameters at a preset learning rate according to the gradient to obtain an updated model;

[0064] S8-6, selecting the next batch of sample pairs from the training set;

[0065] S8-7, input the next batch of sample pairs into the updated model to obtain the loss of the next batch;

[0066] S8-8. Iterate multiple batches of samples until the loss is lower than a threshold, and then export the updated model as a corrosion life prediction model.

[0067] The present invention provides a pavement crack analysis method, which has the following beneficial effects:

[0068] This invention introduces the local deformation structure formed during steel bar processing as a key factor affecting corrosion behavior, combined with microstructural characteristics, breaking through the single mode of corrosion modeling in existing technologies that relies solely on electrochemical or environmental factors.

[0069] Furthermore, by performing multi-dimensional characterization of the microstructure of the specimen before and after fracture, a feature set including indicators such as pore size, grain evolution, and second phase distribution is constructed, so that the structural driving mechanism of the corrosion crack formation process can be quantitatively expressed, thereby enhancing the model's sensitivity to differences in corrosion paths; furthermore, the corrosion life label in the present invention is not generated through empirical fitting, but is derived from real indicators in the corrosion evolution process (such as pitting time, crack length, product quality, etc.), thereby improving the interpretability of the prediction model.

[0070] In summary, the present invention achieves a comprehensive analysis of pavement cracks caused by steel bar corrosion by comprehensively analyzing the local deformation structure and microstructure characteristics of the pavement steel bars. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 This is a flow chart of a pavement crack analysis method according to the present invention;

[0072] Figure 2 This is a flow chart for constructing the near-surface deformation feature sample of the present invention;

[0073] Figure 3 This is a flow chart for constructing the corrosion deformation feature sample described in the present invention. DETAILED DESCRIPTION

[0074] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0075] See also Figures 1 to 3 The present invention provides a pavement crack analysis method, comprising the following steps:

[0076] S1. Prepare near-surface deformation structure specimens of pavement reinforcement;

[0077] Among them, the near-surface deformation structure specimen refers to: the microscopic deformation structure of the surface or near-surface area of the pavement steel bar, which indicates the deformation of the pavement steel bar caused by thermal-mechanical coupling, mechanical processing and other factors during processing (such as hot rolling, milling, cooling, etc.).

[0078] Specifically, this may include:

[0079] Original geometric features such as thread ridges, which are inherent geometric features on the surface of the steel bar, such as threads or raised parts formed during the hot rolling process;

[0080] Deformation characteristics during processing. This characteristic is that after the steel bar undergoes processing such as hot rolling and milling, additional plastic deformation may occur in the near-surface area of the steel bar surface, forming microstructural characteristics that are different from the original geometric shape.

[0081] For example, grain refinement means that during milling or cooling, grain refinement or changes may occur on the surface of the steel bar, affecting the mechanical properties of the steel bar. Grain boundaries and second-phase particles mean that stress during hot working may cause changes in grain boundaries and even form second-phase particles. The presence of grain boundaries and second-phase particles affects the corrosion resistance and strength of the steel bar. Microcracks or pores mean that during processing, microcracks or pores may form on the surface of the steel bar. Microcracks or pores affect the overall performance of the steel bar, especially in corrosive environments, and can easily become a corrosion source.

[0082] In this embodiment, the following steps are used to prepare pavement steel bar samples with different milling parameters:

[0083] ①During the steel bar manufacturing process:

[0084] The steel bars are treated using furnace jet refining and electromagnetic stirring technology to ensure uniform chemical composition of the steel bars.

[0085] Adopt out-of-furnace online degassing and grain refinement technology to improve the microstructure of steel bars and enhance their mechanical properties.

[0086] Combined with hot top oil-lubricated semi-continuous casting technology, the steel bars are initially cast to ensure that the steel bars have suitable surface quality and internal structure.

[0087] ②During the milling process:

[0088] An orthogonal experimental scheme was used for the milling of the steel bar surface, in which different surface characteristics were obtained by adjusting the milling parameters such as cutting speed and cutting depth.

[0089] The cutting speed setting range is 2000m / min to 5000m / min, while the proposed values of axial and radial depth of cut are 0.1mm to 0.4mm respectively.

[0090] During the milling process, since surface roughness has a significant impact on the final structure, in order to ensure the consistency of the steel bar surface geometry and reduce crack initiation, the parameters during the milling process are kept stable to avoid any fluctuations.

[0091] ③ During the temperature and force monitoring process:

[0092] During the milling process, the temperature testing system and cutting force testing system are used to monitor the steel bars in real time to ensure the stability of the processing.

[0093] S2. extracting near-surface features of the near-surface deformation structure specimens to obtain near-surface deformation feature samples of each near-surface deformation structure specimen;

[0094] S3. Constructing a near-surface deformation feature matrix containing N*K samples based on the near-surface deformation feature samples;

[0095] S4. Extracting corrosion features of the near-surface deformation structure specimens to obtain corrosion deformation feature samples of each near-surface deformation structure specimen;

[0096] S5. Constructing a corrosion deformation feature matrix containing N*K samples based on the corrosion deformation feature samples;

[0097] S6. Extracting a life label matrix based on the corrosion deformation feature matrix;

[0098] S7, combining the near-surface deformation feature matrix and the lifespan label matrix to construct a training set containing N*K supervised learning sample pairs;

[0099] S8. Perform model training based on a training set containing N*K supervised learning sample pairs to obtain a corrosion life prediction model;

[0100] S9. Use the corrosion life prediction model to predict the corrosion life of pavement reinforcement to perform pavement crack analysis.

[0101] In this example, a crack analysis method based on the near-surface deformation structure of pavement reinforcement was developed, forming a complete technical process from specimen preparation, multidimensional microstructure extraction, corrosion behavior observation, to lifespan label construction and prediction model training. This implementation establishes a data-driven structure-performance mapping mechanism by linking actual processing parameters with microstructural responses. Combined with a supervised learning model, this method quantitatively predicts the lifespan of reinforcement corrosion crack propagation, providing a solution for pavement structure durability analysis based on material microevolution mechanisms.

[0102] In this embodiment, step S2 specifically includes:

[0103] S2-1, obtaining M first microstructural features of the near-surface deformed structure specimen;

[0104] In step S2-1, the first microstructural feature is used to characterize the microscopic deformation structure of the near-surface of the road steel bar before fracture. In this embodiment, the process of obtaining the first microstructural feature is as follows:

[0105] Measuring pore size: The bubble method is used to measure the pore size of the deformed structure near the steel bar surface, evaluate the surface porosity, and obtain the pore size characteristics of the steel bar near the surface area.

[0106] Obtaining element distribution: Scanning electron microscopy (SEM) combined with energy dispersive spectroscopy (EDS) was used to observe the surface morphology, element distribution, and grain orientation of the near-surface deformed structure of the steel bar to obtain the element distribution characteristics near the surface of the steel bar.

[0107] Characterization of three-dimensional structure: X-ray tomography technology is used to characterize the three-dimensional characteristics of the near-surface deformation structure of the steel bar and obtain the three-dimensional structural characteristics of the near-surface steel bar.

[0108] Obtaining particle evolution characteristics: Through metallographic microscopy (OM), transmission electron microscopy (TEM) and X-ray diffraction (XRD) and other technical means, the dynamic evolution characteristics of grains, grain boundaries and second-phase particles in the near-surface area of the steel bar are studied to obtain the particle evolution characteristics near the surface of the steel bar.

[0109] S2-2, performing a performance test on the near-surface deformation structure specimen to obtain a corresponding fracture specimen;

[0110] Among them, performance tests include mechanical performance tests (such as yield strength, elastic modulus), hardness tests and conductivity tests, and the mechanical performance test ends with the fracture of the near-surface deformation structure specimen to form a fracture specimen, that is, a fracture specimen can be obtained from the test.

[0111] S2-3, obtaining M second microstructural features at the fracture surface of the fractured specimen;

[0112] Specifically, step S2-3 is to perform microstructural analysis on the fracture surface of each fractured sample to obtain M second microstructural features. The second microstructural features mainly characterize the microstructural structure of the fracture surface of the fractured sample, such as grains, grain boundaries, and second phase particles.

[0113] S2-4. Construct the near-surface deformation feature sample according to the M first microstructure features and the M second microstructure features.

[0114] In this example, two types of microstructural features were collected by observing the microstructure of pavement reinforcement before and after fracture. The first microstructural feature describes the surface state of the processed but undamaged reinforcement, encompassing multidimensional information such as pore size, element distribution, three-dimensional structure, and second-phase particle evolution. The second microstructural feature characterizes the structural state at the fracture site, reflecting the accumulation of microscopic damage. This implementation achieves a complete characterization of the microscopic features before and after the structural state.

[0115] Furthermore, the step S2-4 specifically includes:

[0116] S2-4-1. Assign consecutive specimen numbers to each near-surface deformation structure specimen;

[0117] S2-4-2. According to the sample number, obtain a first microstructural feature of the near-surface deformation structure sample corresponding to the sample number;

[0118] S2-4-3, inherit the specimen number of each fracture specimen’s near-surface deformation structure specimen before fracture;

[0119] S2-4-4. According to the inherited sample number, obtain a second microstructural feature of the fractured sample corresponding to the sample number;

[0120] S2-4-5. Pair the sample number, the first microstructure feature, and the second microstructure feature to construct a near-surface deformation feature sample.

[0121] In this embodiment, a unique specimen number is assigned to each near-surface deformed structure specimen, and the number inheritance relationship is maintained during the feature acquisition process before and after fracture. This achieves a precise correspondence between the first and second microstructural features. This numbering mechanism ensures the consistency of the data structure, allowing the microstructural changes between the original processing state and the fracture state to be compared at the sample level.

[0122] In this embodiment, step S3 specifically includes:

[0123] S3-1. Obtain N near-surface deformation feature samples in each performance test;

[0124] S3-2. Arrange the N near-surface deformation feature samples in each performance test in order according to the sample numbers and construct them into matrix row vectors;

[0125] S3-3. Assign consecutive test round numbers to each matrix row vector according to the test order of the performance test;

[0126] S3-4. Sort the test round numbers up and down according to the test order to obtain K matrix row vectors;

[0127] S3-5. Define the K matrix row vectors as the near-surface deformation feature matrix.

[0128] In this embodiment, a structured representation of multi-round test data is achieved by sorting the N near-surface deformation feature samples obtained from each performance test by specimen number and constructing matrix row vectors based on the test round number. This step not only ensures the consistency of sample arrangement but also provides a clear temporal relationship for the feature data, facilitating the capture of changing trends in near-surface microstructure under different processing parameters or service conditions. The resulting near-surface deformation feature matrix is a two-dimensional data structure with a clear dimension (N*K) and numbering system. It can be used as a standard input format during model training and supports subsequent fusion with corrosion feature labels at the sample level.

[0129] The step S4 specifically includes:

[0130] S4-1. performing corrosion treatment on each near-surface deformed structure specimen to obtain a corrosion specimen;

[0131] Among them, the corrosion treatment is to conduct accelerated corrosion tests on the samples in a set controlled environment (including the type of corrosive medium, hot and humid conditions and corrosive ion concentration) to simulate the corrosion behavior of steel bars in actual service environment.

[0132] S4-2, obtaining M corrosion evolution characteristics of the corrosion sample during the corrosion treatment process;

[0133] In step S4-2, the corrosion evolution characteristics are used to characterize the dynamic evolution behavior of the near-surface deformation structure of the steel bar during the corrosion process. In this embodiment, the corrosion evolution characteristics are obtained as follows:

[0134] A single CCD camera system is used to take in-situ photos of the steel bar surface morphology during the corrosion process;

[0135] Extract the initiation time, expansion path, and expansion rate of corrosion cracks;

[0136] Extract the volume change and accumulation rate of corrosion products;

[0137] Analyze the geometric changes during pitting initiation and crack connection.

[0138] Thus, M time-series corrosion evolution characteristics are obtained during the corrosion process.

[0139] S4-3, obtaining M corrosion structure characteristics of the corrosion sample after corrosion is completed;

[0140] In step S4-3, the corrosion structure characteristics are used to characterize the microstructural changes in the near-surface area of the steel bar after the corrosion is completed. In this embodiment, the process of obtaining the corrosion structure characteristics is as follows:

[0141] Using metallographic microscope (OM), scanning electron microscope (SEM), transmission electron microscope (TEM), atomic force microscope (AFM) and Kelvin probe scanning (SKPFM) and other means;

[0142] Extract the number, size, and distribution density of pits on the surface of corroded steel bars;

[0143] Obtain the evolution of second phase particles and their spatial correlation with corrosion pits;

[0144] The potential difference distribution between phases in the corrosion area, the morphology of the corrosion crack endpoint, and other information are extracted to construct M corrosion microstructure features after the corrosion is completed.

[0145] S4-4. Pairing the sample numbers, corrosion evolution characteristics, and corrosion structure characteristics to construct the corrosion deformation characteristic sample;

[0146] In step S4-4, a set of corrosion deformation feature samples is established for each near-surface deformation structure specimen. The samples combine the dynamic behavior of the corrosion process and the microstructure state after corrosion to characterize the corrosion response characteristics of the specimen in a typical corrosion environment.

[0147] In this embodiment, step S5 specifically includes:

[0148] S5-1. Obtain N corrosion deformation characteristic samples in each corrosion test;

[0149] That is, in each round of corrosion test, corresponding corrosion deformation feature samples are extracted for N near-surface deformation structure specimens.

[0150] S5-2. Arrange the N corrosion deformation characteristic samples in each corrosion test in order of sample number and construct them into matrix row vectors;

[0151] The consistency of the characteristic arrangement of each sample in different rounds of testing is maintained through consistency numbering.

[0152] S5-3, assigning consecutive corrosion test round numbers to each matrix row vector according to the execution order of the corrosion test;

[0153] Each round of corrosion test generates a set of row vectors, which are numbered to maintain their upper and lower order in the entire matrix.

[0154] S5-4, sorting the corrosion test round numbers up and down according to the test order to obtain K matrix row vectors;

[0155] Where K is the total number of corrosion test rounds, and finally a corrosion deformation feature matrix with K rows and N columns is formed.

[0156] S5-5. Define the K matrix row vectors as the corrosion deformation feature matrix.

[0157] In this example, the N corrosion deformation feature samples extracted from each corrosion test round are numbered and sorted, and a matrix row vector is constructed with the corrosion test round as the dimension. This ultimately forms a corrosion deformation feature matrix with strict arrangement rules. This matrix maintains the same sample numbering sequence as the near-surface deformation feature matrix, thus achieving accurate correspondence between cross-dimensional features.

[0158] In this embodiment, step S6 specifically includes:

[0159] S6-1. Extracting time-based life indicators from corrosion evolution characteristics;

[0160] For example, the extraction process of time-based life indicators is to collect the pitting initiation time and crack start time during the corrosion process; and record the time required for the corrosion crack to reach a specified length (such as 10μm, 100μm).

[0161] S6-2. Extract length or quality indicators from corrosion structure characteristics;

[0162] Exemplarily, the process of extracting length-type or mass-type indicators is to extract the final crack length and maximum pitting depth through image processing and microscopic measurement; and calculate the total weight (or mass increment) of corrosion products.

[0163] S6-3. According to a preset life evaluation standard, convert the time-related, length-related, or quality-related indicators into a life label with a unified dimension;

[0164] For example, by setting a critical corrosion threshold, it can be quantified into a lifespan indicator. The measures can be:

[0165] The “failure” standard is set when the corrosion crack length ≥ Xμm or the product weight gain ≥ Ymg;

[0166] The actual test data is compared with the threshold to obtain a unified numerical label, such as: crack growth life (unit: hours) or mass gain life (unit: mg) or equivalent corrosion level (classification labels: mild, moderate, severe).

[0167] S6-4. Arrange all the life labels in order according to the sample numbers and the test round numbers to construct the life label matrix.

[0168] In this example, dynamic evolution data during the corrosion process (e.g., pitting initiation time and crack propagation time) and structural observation data after corrosion completion (e.g., final crack length, pitting depth, and product quality) are structured to extract physically meaningful temporal and spatial life indicators. Furthermore, based on a set failure threshold, the corrosion characterization data from different dimensions is standardized into a unified life label value. All label data are arranged according to specimen number and test round sequence to construct a life label matrix consistent with the structure of the corrosion deformation feature matrix. This label matrix retains a quantitative representation of key behaviors during the corrosion process, providing output targets with clear physical meaning for the supervised learning model.

[0169] In this embodiment, step S7 specifically includes:

[0170] S7-1. Obtain the specimen number and test round number of each sample in the near-surface deformation feature matrix;

[0171] S7-2, performing ID splicing on the specimen number and the test round number of each sample to generate a near-surface feature number;

[0172] S7-3, obtaining the sample number and test round number of each tag in the life tag matrix;

[0173] S7-4, concatenate the sample number and test round number of each tag to generate a life tag number;

[0174] S7-5. If the near-surface feature number and the life label number are the same, pair the corresponding near-surface deformation feature sample with the life label to obtain a supervised learning sample pair;

[0175] S7-6, traverse the near-surface deformation feature matrix and the life label matrix until N*K supervised learning sample pairs are obtained;

[0176] S7-7. Combine N*K supervised learning samples to obtain the training set.

[0177] In this example, a one-to-one mapping between feature samples and target labels is established by concatenating the specimen number and test run number for each sample in the near-surface deformation feature matrix and the lifetime label matrix to generate a unique ID. This mapping ensures perfect alignment of input features and output labels at the data structure level, avoiding the misalignment and confusion common in cross-dimensional data pairing. After all number matching is completed, eligible features and labels are systematically paired to generate structurally standardized supervised learning sample pairs, which are ultimately combined into a complete training set.

[0178] In this embodiment, step S8 specifically includes:

[0179] S8-1, extracting the first batch of sample pairs of the training set;

[0180] S8-2. Input the first batch of sample pairs into the multi-layer perception model, and after forward propagation, generate the first batch of predicted lifespans at the output layer;

[0181] S8-3, calculating the loss between the predicted lifespan of the first batch and the lifespan labels of the first batch of sample pairs;

[0182] S8-4. If the loss is lower than the threshold, the training ends; otherwise, the loss is calculated to be equivalent to the gradient of the current multi-layer perception model parameters;

[0183] S8-5. Reversely update the multi-layer perception model parameters at a preset learning rate according to the gradient to obtain an updated model;

[0184] S8-6, selecting the next batch of sample pairs from the training set;

[0185] S8-7, input the next batch of sample pairs into the updated model to obtain the loss of the next batch;

[0186] S8-8. Iterate multiple batches of samples until the loss is lower than a threshold, and then export the updated model as a corrosion life prediction model.

[0187] In this embodiment, the corrosion life prediction model employs a multi-layer perceptron (MLP) architecture, consisting of an input layer, two hidden layers, and an output layer. The input layer receives the normalized near-surface deformation feature vector. The hidden layer uses the ReLU activation function to enhance nonlinear fitting capabilities. The output layer uses a linear activation function for regression prediction of corrosion life values. The number of neurons in each layer is determined based on the sample dimensionality and empirical rules. A typical configuration is a d-dimensional input layer with 64 and 32 neurons in the hidden layers, respectively.

[0188] During the training process, the model uses mean square error (MSE) as the loss function to reflect the error range between the predicted life and the actual life. The optimizer uses Adam, and updates the parameters with an adaptive learning rate. The initial learning rate is set to 0.001, and the batch size is 32. In each round of training, a batch of sample pairs is extracted from the training set, and forward propagation is performed to generate the predicted life value. The loss function is calculated based on the label value. If the current loss value is lower than the set threshold (such as 0.001), the training is terminated early; otherwise, backpropagation is performed, and the model parameters are iteratively updated based on the loss gradient. After all iterations, the derived model has the ability to predict the corrosion life of unseen specimens. Therefore, it can be used as an analytical tool in the pavement crack analysis method. By finely modeling the internal corrosion process of road steel bars, it can achieve quantitative deduction from microscopic corrosion behavior to macroscopic crack initiation trends, providing data support for the early identification and reliability prediction of pavement cracks.

[0189] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer program are loaded or executed on a computer, the process or function described in the embodiments of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired method (e.g., infrared, wireless, microwave, etc.).

[0190] The computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., DVD ), or semiconductor media. The semiconductor media may be a solid-state drive.

[0191] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative, and for example, multiple units or components can be combined or integrated into another system, or some features can be omitted or not implemented. In addition, the coupling or direct coupling or communication connection shown or discussed can be through some interface, indirect coupling or communication connection of devices or units, and can be electrical, mechanical, or other forms.

[0192] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A pavement crack analysis method, characterized in that: include: S1. Prepare near-surface deformation structure specimens of pavement reinforcement; S2. extracting near-surface features of the near-surface deformation structure specimens to obtain near-surface deformation feature samples of each near-surface deformation structure specimen; The feature extraction of the near-surface deformation structure specimen to obtain the near-surface deformation feature sample of each near-surface deformation structure specimen includes: S2-1, obtaining M first microstructural features of the near-surface deformed structure specimen; S2-2, performing a performance test on the near-surface deformation structure specimen to obtain a corresponding fracture specimen; S2-3, obtaining M second microstructural features at the fracture surface of the fractured specimen; S2-4, constructing the near-surface deformation feature sample according to the M first microstructure features and the M second microstructure features; The step of constructing the near-surface deformation feature sample according to the M first microstructure features and the M second microstructure features includes: S2-4-1. Assign consecutive specimen numbers to each near-surface deformation structure specimen; S2-4-2. According to the sample number, obtain a first microstructural feature of the near-surface deformation structure sample corresponding to the sample number; S2-4-3, inherit the specimen number of each fracture specimen’s near-surface deformation structure specimen before fracture; S2-4-4. According to the inherited sample number, obtain a second microstructural feature of the fractured sample corresponding to the sample number; S2-4-5, pairing the sample number, the first microstructure feature, and the second microstructure feature to construct a near-surface deformation feature sample; S3. Constructing a near-surface deformation feature matrix containing N*K samples based on the near-surface deformation feature samples; Where N represents the number of samples in the matrix row vector, and K represents the number of rows in the matrix row vector; S4. Extracting corrosion features of the near-surface deformation structure specimens to obtain corrosion deformation feature samples of each near-surface deformation structure specimen; S5. Constructing a corrosion deformation feature matrix containing N*K samples based on the corrosion deformation feature samples; S6. Extracting a life label matrix based on the corrosion deformation feature matrix; S7, combining the near-surface deformation feature matrix and the lifespan label matrix to construct a training set containing N*K supervised learning sample pairs; S8. Perform model training based on a training set containing N*K supervised learning sample pairs to obtain a corrosion life prediction model; S9. Use the corrosion life prediction model to predict the corrosion life of pavement reinforcement to perform pavement crack analysis.

2. A pavement crack analysis method according to claim 1, characterized in that: According to the near-surface deformation feature samples, a near-surface deformation feature matrix containing N*K samples is constructed, including: S3-1. Obtain N near-surface deformation feature samples in each performance test; S3-2. Arrange the N near-surface deformation feature samples in each performance test in order according to the sample numbers and construct them into matrix row vectors; S3-3. Assign consecutive test round numbers to each matrix row vector according to the test order of the performance test; S3-4. Sort the test round numbers up and down according to the test order to obtain K matrix row vectors; S3-5. Define the K matrix row vectors as the near-surface deformation feature matrix.

3. A pavement crack analysis method according to claim 1, characterized in that: The corrosion characteristics of the near-surface deformation structure specimen are extracted to obtain a corrosion deformation characteristic sample of each near-surface deformation structure specimen, including: S4-1. performing corrosion treatment on each near-surface deformed structure specimen to obtain a corrosion specimen; S4-2, obtaining M corrosion evolution characteristics of the corrosion sample during the corrosion treatment process; S4-3, obtaining M corrosion structure characteristics of the corrosion sample after corrosion is completed; S4-4. Pair the sample number, corrosion evolution characteristics, and corrosion structure characteristics to construct the corrosion deformation characteristic sample.

4. A pavement crack analysis method according to claim 1, characterized in that: According to the corrosion deformation feature samples, a corrosion deformation feature matrix containing N*K samples is constructed, including: S5-1. Obtain N corrosion deformation characteristic samples in each corrosion test; S5-2. Arrange the N corrosion deformation characteristic samples in each corrosion test in order of sample number and construct them into matrix row vectors; S5-3, assigning consecutive corrosion test round numbers to each matrix row vector according to the execution order of the corrosion test; S5-4, sorting the corrosion test round numbers up and down according to the test order to obtain K matrix row vectors; S5-5. Define the K matrix row vectors as the corrosion deformation feature matrix.

5. A pavement crack analysis method according to claim 1, characterized in that: Extracting a life label matrix from the corrosion deformation feature matrix includes: S6-1. Extracting time-based life indicators from corrosion evolution characteristics; S6-2. Extract length or quality indicators from corrosion structure characteristics; S6-3. According to a preset life evaluation standard, convert the time-related, length-related, or quality-related indicators into a life label with a unified dimension; S6-4. Arrange all the life labels in order according to the sample numbers and the test round numbers to construct the life label matrix.

6. A pavement crack analysis method according to claim 1, characterized in that: The near-surface deformation feature matrix and the lifespan label matrix are combined to construct a training set containing N*K supervised learning sample pairs, including: S7-1. Obtain the specimen number and test round number of each sample in the near-surface deformation feature matrix; S7-2, performing ID splicing on the specimen number and the test round number of each sample to generate a near-surface feature number; S7-3, obtaining the sample number and test round number of each tag in the life tag matrix; S7-4, concatenate the sample number and test round number of each tag to generate a life tag number; S7-5. If the near-surface feature number and the life label number are the same, pair the corresponding near-surface deformation feature sample with the life label to obtain a supervised learning sample pair; S7-6, traverse the near-surface deformation feature matrix and the life label matrix until N*K supervised learning sample pairs are obtained; S7-7. Combine N*K supervised learning samples to obtain the training set.

7. A pavement crack analysis method according to claim 1, characterized in that: The model is trained based on a training set containing N*K supervised learning sample pairs to obtain a corrosion life prediction model, including: S8-1, extracting the first batch of sample pairs of the training set; S8-2. Input the first batch of sample pairs into the multi-layer perception model, and after forward propagation, generate the first batch of predicted lifespans at the output layer; S8-3, calculating the loss between the predicted lifespan of the first batch and the lifespan labels of the first batch of sample pairs; S8-4. If the loss is lower than the threshold, the training ends; otherwise, the loss is calculated to be equivalent to the gradient of the current multi-layer perception model parameters; S8-5. Reversely update the multi-layer perception model parameters at a preset learning rate according to the gradient to obtain an updated model; S8-6, selecting the next batch of sample pairs from the training set; S8-7, input the next batch of sample pairs into the updated model to obtain the loss of the next batch; S8-8. Iterate multiple batches of samples until the loss is lower than a threshold, and then export the updated model as a corrosion life prediction model.

Images