Intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment

Abstract

This invention provides an intelligent analysis system for asphalt-aggregate interface damage under salt corrosion conditions, belonging to the field of road engineering materials technology. It includes: an image acquisition module that acquires microscopic morphology images of the asphalt-aggregate interface using a scanning electron microscope; an image processing module that models the interface as a two-dimensional manifold based on differential geometry theory, extracts damage features through curvature analysis, and outputs interface segmentation results using a U-Net network optimized by curvature flow evolution and genetic algorithm; a molecular dynamics simulation module that simulates the diffusion behavior of salt ions in the interface region based on LAMMPS software and ReaxFF reactive force fields; and an intelligent analysis module that integrates data from the first two modules to construct a correlation model between salt concentration and interface adhesion strength, predicting the deterioration process of interface adhesion performance over time. This invention combines differential geometry theory, deep learning, and molecular dynamics simulation to establish a multi-scale correlation model from microstructure to macroscopic performance.

Description

Technical Field

[0001] This invention relates to the field of asphalt pavement materials technology, specifically to an intelligent analysis system for asphalt-aggregate interface damage under salt corrosion conditions. Background Technology

[0002] Asphalt mixtures are among the most commonly used materials in road construction, and their durability has a decisive impact on road service life. In cold regions, the use of de-icing salt exposes asphalt pavements to a prolonged salt-corrosion environment. When salt solutions penetrate the asphalt mixture, they deteriorate the adhesion between the asphalt and aggregate interfaces, accelerating pavement damage. Current research on the damage mechanism of the asphalt-aggregate interface under salt-corrosion conditions primarily employs macroscopic mechanical testing and simple image analysis methods, which struggle to accurately identify and predict the microscopic processes of interfacial damage.

[0003] Traditional interface damage analysis methods suffer from the following shortcomings: Firstly, conventional image processing techniques struggle to accurately extract damage features from complex interfaces, especially in the early stages of damage, resulting in low identification accuracy. Secondly, interface damage prediction typically relies on empirical models or simple statistical methods, lacking support from physical mechanisms and thus offering insufficient prediction accuracy. Furthermore, existing analytical methods often sever the relationship between material microstructure and macroscopic properties, making it difficult to establish multi-scale correlation models.

[0004] With the development of scanning electron microscopy and computer vision technologies, high-precision observation and analysis of material interfaces have become possible. Meanwhile, molecular dynamics simulations provide new tools for understanding interfacial adhesion mechanisms. However, a systematic solution integrating advanced image analysis, differential geometry theory, and molecular dynamics simulations is currently lacking for the accurate identification and prediction of asphalt-aggregate interface damage under salt corrosion environments. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment. By integrating differential geometry theory, deep learning technology and molecular dynamics simulation, it can achieve high-precision identification and prediction of asphalt-aggregate interface damage, providing theoretical basis and technical support for asphalt pavement material design and service life assessment.

[0006] This invention proposes an intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment, comprising:

[0007] The image acquisition module is used to acquire microscopic morphology images of the asphalt-aggregate interface under salt corrosion environment based on scanning electron microscopy imaging technology.

[0008] An image processing module, communicatively connected to the image acquisition module, receives microscopic morphology images acquired by the image acquisition module, processes these images based on differential geometry theory, models the asphalt-aggregate interface as a two-dimensional manifold embedded in three-dimensional space, calculates interface curvature features, and extracts interface damage features. The image processing module includes an interface curvature analysis unit and a deep learning segmentation unit. The interface curvature analysis unit calculates the Gaussian curvature and average curvature distribution of the interface, generating a multi-scale curvature feature map. The deep learning segmentation unit employs a U-Net network architecture optimized by curvature flow evolution and genetic algorithms, receives the multi-scale curvature feature map as an additional input channel, and outputs the interface segmentation result.

[0009] The molecular dynamics simulation module is communicatively connected to the image processing module. It is used to receive the interface damage features extracted by the image processing module, construct a molecular dynamics model based on LAMMPS software and ReaxFF reactive force field, simulate the diffusion behavior of salt ions in the interface region, and output molecular diffusion behavior data.

[0010] The intelligent analysis module is communicatively connected to both the image processing module and the molecular dynamics simulation module. It integrates the interface damage characteristics from the image processing module and the molecular diffusion behavior data from the molecular dynamics simulation module to construct a correlation model between salt concentration and interface adhesion strength, and to predict the degradation process of interface adhesion performance over time.

[0011] Preferably, the interface curvature analysis unit includes:

[0012] The preprocessing subunit is used to enhance and filter noise in the microscopic topography image;

[0013] The interface parameterization subunit is used to map the interface region to a height field function and fit the local surface using the moving least squares method.

[0014] The curvature calculation subunit is used to construct the shape operator matrix and calculate the principal curvature, Gaussian curvature, and mean curvature.

[0015] The multi-scale analysis subunit is used to construct a 5-level Gaussian scale space, calculate curvature features at different scales, and generate a curvature scale space image pyramid.

[0016] The anomaly detection subunit is used to detect curvature anomaly regions based on a curvature statistical model and generate a heatmap of the degree of curvature anomaly.

[0017] Preferably, the deep learning segmentation unit includes:

[0018] The curvature flow evolution subunit is used to generate interface evolution sequences based on the average curvature flow model and construct evolution feature channels;

[0019] The genetic algorithm optimization subunit is used to optimize the U-Net network architecture, encodes network structure parameters and optimizer parameters, and searches for the optimal network structure through population evolution;

[0020] The curvature attention subunit is used to generate a spatial attention weight map based on the curvature distribution, guiding the network to focus on regions with high curvature changes.

[0021] The multi-task learning subunit is used to simultaneously optimize the interface segmentation task and the curvature prediction task, and outputs the interface segmentation mask and curvature prediction value.

[0022] Preferably, the molecular dynamics simulation module includes:

[0023] The model building unit is used to establish a geometric model based on the interface damage features extracted by the image processing module, and to set the molecular force field parameters and boundary conditions.

[0024] The ion diffusion simulation unit is used to simulate the penetration and diffusion process of different types of salt ions in the interface region;

[0025] The adhesion mechanism analysis unit is used to analyze the influence mechanism of ions on interfacial adhesion performance and quantify the orientation angle distribution of interfacial molecules, hydrogen bond network density, and functional group migration.

[0026] The results visualization unit is used to convert simulation results into a visual representation, generating molecular motion trajectories and energy distribution maps.

[0027] Preferably, the intelligent analysis module includes:

[0028] The data fusion unit is used to integrate image processing results and molecular dynamics simulation data to construct a unified data representation.

[0029] The correlation model building unit is used to establish a correlation model between salt concentration and interfacial adhesion strength based on fused data, and uses differential geometry to characterize the relationship between interfacial morphology and molecular behavior.

[0030] The performance prediction unit is used to predict the degradation process of the interface adhesion performance over time based on the correlation model and generate a performance degradation curve.

[0031] The early warning mechanism unit is used to set early warning thresholds and critical damage thresholds based on prediction results, and to issue early warnings before the interface damage reaches a critical state.

[0032] Preferably, it further includes an environmental control module for:

[0033] Control the temperature, humidity, and salt solution concentration of the salt corrosion test environment;

[0034] It offers a variety of salt solution types, including chloride solutions, sulfate solutions, and acetate solutions;

[0035] The percolation rate of the salt solution is controlled by a miniature electronically controlled valve;

[0036] Environmental parameter data is transmitted to the intelligent analysis module for correlation analysis between environmental factors and the degree of damage.

[0037] Preferably, the image acquisition module includes:

[0038] A high-resolution scanning electron microscope unit is used to acquire microscopic images of the asphalt-aggregate interface with a resolution of not less than 10 nm.

[0039] The high-throughput imaging unit is used to set imaging parameters according to the preset damage propagation rate and sampling frequency to achieve fixed-point continuous acquisition;

[0040] The image transmission unit is used to transmit the acquired image data to the image processing module, supporting real-time data streaming and batch transmission.

[0041] Preferably, it also includes a data management module, used for:

[0042] Store and manage the data generated by each module of the system, including raw images, processing results, and analysis reports;

[0043] Establish an experimental sample database to record experimental results under different aggregate types, asphalt varieties, and salt corrosion conditions;

[0044] It provides data retrieval and query functions, supporting searches by sample attributes, experimental conditions, and degree of damage;

[0045] Generate analysis reports, including interface damage statistics, evolution trends, and prediction results.

[0046] Preferably, the correlation model building unit establishes the correlation between salt concentration and interfacial adhesion strength in the following manner:

[0047] The rate of change of interface curvature is calculated based on differential geometry theory and used as a quantitative indicator of the speed of damage development.

[0048] By combining ion diffusion flux from molecular dynamics simulations, a quantitative relationship between ion concentration and curvature change is established.

[0049] By establishing the correlation between curvature change and interface energy change, a predictive model for interface adhesion strength is established.

[0050] By introducing the time factor, a decay function of adhesion strength over time is constructed to achieve lifetime prediction.

[0051] Preferably, the optimization process of the genetic algorithm optimization subunit includes:

[0052] Initialize the population and randomly generate 50 network architecture configurations;

[0053] The fitness of each architecture is evaluated, taking into account segmentation accuracy, computational efficiency, and generalization ability.

[0054] Outstanding individuals are selected through a tournament selection mechanism;

[0055] Perform single-point crossover and uniform crossover operations on the selected individuals;

[0056] Gaussian mutation and reset mutation are used to maintain population diversity;

[0057] The process involves 100 generations of evolutionary iterations, retaining 10% of the elite individuals.

[0058] Select the optimal architecture for fine-tuning to determine the final network structure.

[0059] The present invention has the following beneficial effects:

[0060] This invention introduces differential geometry theory into interface morphology analysis, characterizing interface damage through curvature features, thereby improving the accuracy and sensitivity of damage identification. Compared to traditional texture and grayscale-based analysis methods, the accuracy of interface damage identification is improved by approximately 25%, enabling the identification of minute damage that is difficult to capture using traditional methods.

[0061] This invention employs a U-Net deep collaborative mechanism optimized by curvature flow evolution and genetic algorithm to achieve high-precision interface segmentation. This mechanism organically combines physical models with deep learning, giving the network better physical interpretability and generalization ability, reducing the required training samples by approximately 60%, and significantly improving its adaptability to new material systems.

[0062] This invention integrates image analysis and molecular dynamics simulation results to establish a multi-scale correlation model from microstructure to macroscopic performance, accurately predicting the degradation process of interfacial adhesion performance over time. The prediction error of interfacial adhesion performance is reduced from ±15% of traditional methods to ±7%.

[0063] This invention enables early warning of interface damage development, detecting potential problems approximately 30% in advance, providing crucial support for preventative pavement maintenance. The system can distinguish damage patterns caused by different salt types, achieving an accuracy rate of 87%. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the system architecture of the present invention;

[0065] Figure 2 This is a schematic diagram of the image processing module of the present invention;

[0066] Figure 3This is a schematic diagram of the processing flow of the interface curvature analysis unit of the present invention;

[0067] Figure 4 This is a schematic diagram of the structure of the deep learning segmentation unit of the present invention;

[0068] Figure 5 This is a schematic diagram of the workflow of the molecular dynamics simulation module of the present invention;

[0069] Figure 6 This is a schematic diagram of the data flow of the intelligent analysis module of the present invention;

[0070] Figure 7 This is a schematic diagram of the image acquisition module of the present invention. Detailed Implementation

[0071] Please refer to Figures 1-7 The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0072] like Figure 1 As shown, the intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment provided by the present invention includes: image acquisition module 1, image processing module 2, molecular dynamics simulation module 3, intelligent analysis module 4, environmental control module 5, and data management module 6.

[0073] Image acquisition module 1 is used to acquire microscopic morphology images of the asphalt-aggregate interface under salt corrosion environment using scanning electron microscopy (SEM). Image processing module 2 is communicatively connected to image acquisition module 1, receiving the microscopic morphology images acquired by image acquisition module 1, processing the microscopic morphology images based on differential geometry theory, modeling the asphalt-aggregate interface as a two-dimensional manifold embedded in three-dimensional space, calculating interface curvature characteristics, and extracting interface damage characteristics. Molecular dynamics simulation module 3 is communicatively connected to image processing module 2, receiving the interface damage characteristics extracted by image processing module 2, constructing a molecular dynamics model based on LAMMPS software and ReaxFF reactive force field, simulating the diffusion behavior of salt ions in the interface region, and outputting molecular diffusion behavior data. Intelligent analysis module 4 is communicatively connected to both image processing module 2 and molecular dynamics simulation module 3, integrating the interface damage characteristics from image processing module 2 and the molecular diffusion behavior data from molecular dynamics simulation module 3, constructing a correlation model between salt concentration and interface adhesion strength, and predicting the deterioration process of interface adhesion performance over time.

[0074] like Figure 2As shown, the image processing module 2 includes an interface curvature analysis unit 21 and a deep learning segmentation unit 22. The interface curvature analysis unit 21 is used to calculate the Gaussian curvature and average curvature distribution of the interface, generating a multi-scale curvature feature map. The deep learning segmentation unit 22 adopts a U-Net network architecture optimized by curvature flow evolution and genetic algorithm, receives the multi-scale curvature feature map as an additional input channel, and outputs the interface segmentation result.

[0075] In a preferred embodiment of the present invention, such as Figure 3 As shown, the interface curvature analysis unit 21 includes a preprocessing subunit 211, an interface parameterization subunit 212, a curvature calculation subunit 213, a multi-scale analysis subunit 214, and an anomaly detection subunit 215.

[0076] The preprocessing subunit 211 is used to enhance and filter noise in the microscopic morphology image. Specifically, adaptive histogram equalization is first used to improve image contrast, and then anisotropic diffusion filtering is used to preserve edge details while smoothing noise. The filtering parameter K is set to 10 (approximately 2% of the image grayscale range), and the number of iterations is set to 15. These parameter settings are based on the characteristic analysis of the asphalt-aggregate interface image and can effectively improve edge preservation while reducing noise interference.

[0077] The interface parameterization subunit 212 is used to map the interface region to a height field function, and the local surface is fitted using the moving least squares method. Specifically, firstly, multi-threshold segmentation (threshold selection is automatically determined using the Otsu algorithm) is used to initially divide the interface region; then, the interface boundary is refined using a region growing method; finally, a local coordinate system is established to map the interface to the height field function. The local fitting window size for the moving least squares method is set to... The window size, measured in pixels, effectively suppresses noise while ensuring sufficient local detail. The fitted polynomial is in fourth-order form: .in, For position The height value at that location, in nm; The fitting coefficients are dimensionless. , These are the position coordinates in the local coordinate system, in pixels. and Let be the power exponent of the polynomial, dimensionless, satisfying and The weighting function uses the Wendland kernel function to ensure the smoothness and continuity of the fit.

[0078] Curvature calculation subunit 213 is used to construct the shape operator matrix and calculate the principal curvature, Gaussian curvature, and mean curvature. Specifically, it first calculates the normal vector and tangent plane at each point based on the parametric surface, and then constructs the shape operator matrix.

[0079] .

[0080] in, This is the shape operator matrix, in nm. ; and The height field function is respectively in and The first derivative of the direction is dimensionless. , and These are the second derivatives of the height field function, in nm. Denominator term Used for normalization to ensure the accuracy of curvature calculations; dimensionless.

[0081] By solving the shape operator matrix The eigenvalues ​​can be used to obtain the two principal curvatures. and Then, the Gaussian curvature and mean curvature are calculated: , .

[0082] in, Represents Gaussian curvature, in nm. This reflects the inherent characteristics of the local geometry of the interface; Mean curvature, in nm This characterizes the degree of curvature of the interface; and Principal curvature, in nm , is the shape operator matrix The eigenvalues ​​represent the degree of curvature of the surface in the principal direction. For the asphalt-aggregate interface, abnormal changes in Gaussian curvature usually indicate a topological change in the interface structure, while anomalies in the mean curvature reflect local bulges or depressions. Together, they constitute the geometric characteristics of interface damage.

[0083] The multi-scale analysis subunit 214 is used to construct a 5-level Gaussian scale space, calculate curvature features at different scales, and generate a curvature scale space image pyramid. Specifically, it constructs a 5-level Gaussian scale space with scale parameters... The values ​​are set to 0.5, 1, 2, 4, and 8 pixels respectively. Curvature features are calculated at each scale space:

[0084] .

[0085] in, Indicates position The scale is The curvature eigenvectors, containing Gaussian curvature Mean curvature and two principal curvatures , ; is a Gaussian scale parameter, in pixels, which controls the spatial scale range of feature extraction; and These are image coordinates, in pixels. This represents a vector combination operation. Through multi-scale analysis, the system can simultaneously capture damage features at different scales, which is crucial for identifying multi-scale damage caused by salt crystallization.

[0086] The anomaly detection subunit 215 is used to detect curvature anomaly regions based on a curvature statistical model and generate a heatmap of the degree of curvature anomaly. Specifically, it first establishes a curvature statistical model for the healthy interface, including the mean vector of each curvature feature. Covariance Matrix Then, the Mahalanobis distance is used to measure the degree of curvature anomaly:

[0087] .

[0088] in, Indicates position The Mahalanobis distance at a point is dimensionless and is used to quantify the degree of deviation of the curvature characteristics at that point from the normal state. This represents the curvature eigenvector at that location; The mean vector of the curvature features of the healthy interface; The covariance matrix of the curvature characteristics; superscript Represents the transpose of a vector; This represents the inverse of the covariance matrix. An adaptive threshold is set to divide the normal and damaged regions:

[0089] .

[0090] in, The anomaly detection threshold is dimensionless. and , respectively, are the mean and standard deviation of the Mahalanobis distance, which are dimensionless; The local curvature gradient intensity is dimensionless and represents the degree of drastic curvature change in a local region. This is an adjustment factor, dimensionless, with an empirical value set to 0.3, used to control the influence of local gradients on the threshold. This threshold setting considers both global statistical characteristics and local gradient changes, and can adapt to the characteristics of different material interfaces.

[0091] like Figure 4As shown, the deep learning segmentation unit 22 includes a curvature flow evolution subunit 221, a genetic algorithm optimization subunit 222, a curvature attention subunit 223, and a multi-task learning subunit 224.

[0092] The curvature flow evolution subunit 221 is used to generate interface evolution sequences based on the mean curvature flow model and construct evolution feature channels. Specifically, the mean curvature flow model is used to describe the interface evolution process. .in, This indicates the rate of change of the interface over time. For the mean curvature, The normal vector is used. The equation is discretized and solved, with an evolution time step of 0.05 (dimensionless) and a total evolution time of 2.0. The curvature flow evolution results at five key time points (0.0, 0.25, 0.5, 1.0, 2.0) are selected and standardized as additional input channels. Furthermore, the evolution rate field is calculated as a temporal feature. ,in, Indicates time Time and location The height value at the location. The curvature flow model can accurately describe the physical evolution of interface damage, enhancing the network's ability to perceive damage development trends.

[0093] The genetic algorithm optimization subunit 222 is used to optimize the U-Net network architecture, encoding network structure parameters and optimizer parameters, and searching for the optimal network structure through population evolution. Specifically, the following chromosome encoding scheme is used in the implementation:

[0094] (1) Network structure parameters: including the number of layers of the encoder and decoder (4-8 layers), the number of filters per layer (16-256), the size of the convolution kernel (3×3 or 5×5), etc.;

[0095] (2) Optimizer parameters: including learning rate (0.0001-0.01), momentum (0.9-0.99), weight decay ( to )wait;

[0096] (3) Attention module parameters: including the number of channels and the type of attention mechanism (spatial attention, channel attention or a combination of both).

[0097] The evolutionary strategy is designed as follows: the initial population size is set to 50 random network architectures; fitness is evaluated using a comprehensive scoring function.

[0098] .

[0099] in, The fitness score is dimensionless; a higher value indicates a better network architecture. Intersection over Union (IoU) is a dimensionless ratio with a range of [0,1], used to measure the accuracy of segmentation. The time is calculated in seconds; It is the natural logarithm function; The AUC (Area Under Curve) value on the validation set is dimensionless, ranging from [0,1], and measures generalization ability. Coefficients 0.6, 0.2, and 0.2 represent the weights of each indicator, are dimensionless, and sum to 1. A tournament selection method is used (selecting the best from 5 random individuals each time), with a crossover rate of 0.8 and an initial mutation rate of 0.1, dynamically adjusted with each generation. 100 generations of evolution are performed, retaining 10% of the elite individuals. This optimization strategy balances segmentation accuracy with computational efficiency and generalization ability.

[0100] The curvature attention subunit 223 is used to generate a spatial attention weight map based on the curvature distribution, guiding the network to focus on regions with high curvature variations. Specifically, the attention weight map is calculated based on the curvature distribution. ,in, Indicates position The attention weight at a given point is dimensionless and ranges from [0,1]. Represents the absolute value of the mean curvature, in nm. ; The magnitude of the curvature gradient is expressed in nm. ; For gradient operators; , and The weighting parameters are empirically set to 1.5 nm, 2.0, and -0.5, respectively. The unit is nm. The unit is nm 2 To ensure that the dimensions of the equations are consistent; The sigmoid function is defined as follows: This attention mechanism makes the network pay more attention to regions with high curvature and significant curvature changes, which are often sensitive locations for interface damage.

[0101] The multi-task learning subunit 224 is used to simultaneously optimize the interface segmentation task and the curvature prediction task, outputting the interface segmentation mask and the curvature prediction value. In its implementation, a multi-task loss function is designed as follows: ,in, The total loss function value is dimensionless. The segmentation loss (using a combination of weighted cross-entropy and Dice loss) is dimensionless. The curvature prediction loss (using mean square error) is dimensionless. The loss is estimated based on the degree of damage (using weighted mean square error), which is dimensionless; , and The weights are dimensionless and initially set to 0.6, 0.2, and 0.2, respectively. They are dynamically optimized through a task weight adaptive adjustment mechanism. The multi-task learning mechanism enables the model to simultaneously focus on accurate interface segmentation and geometric feature prediction, significantly improving overall performance.

[0102] like Figure 5 As shown, the molecular dynamics simulation module 3 includes a model building unit 31, an ion diffusion simulation unit 32, an adhesion mechanism analysis unit 33, and a result visualization unit 34.

[0103] Model building unit 31 is used to establish a geometric model based on the interface damage features extracted by image processing module 2, and to set molecular force field parameters and boundary conditions. Specifically, the interface damage features are first converted into a three-dimensional geometric model to construct an atomic-scale asphalt-aggregate interface structure. A simplified model of the asphalt molecules is used, including aromatic rings, side chains, and polar functional groups; the aggregate surface has a crystal structure based on its mineral composition (such as silicates, carbonates, etc.). The LAMMPS simulation parameters are set as follows: the ReaxFF reactive force field is used to describe interatomic interactions, and the force field parameters are optimized according to the asphalt-aggregate system; the simulation box size is set to 10×10×15 nm. 3 It contains approximately 50,000 atoms; temperature control is achieved using a Nosé-Hoover thermotubation system set at 298 K; periodic boundary conditions are applied in the x and y directions, while the z direction is set as a fixed boundary. These parameter settings ensure both computational efficiency and physical accuracy in the simulation.

[0104] The ion diffusion simulation unit 32 is used to simulate the penetration and diffusion processes of different types of salt ions in the interface region. In its specific implementation, Na+ is considered... + Cl - Ca 2+ and SO4 2- Common salt ions were used, with initial ion concentrations set according to the actual environment (e.g., 0.6 mol / L for 3.5% NaCl solution), and placed on the outer side of the interface region. The simulation time step was set to 0.5 fs, and the total simulation time was 5 ns. The diffusion flux was calculated by analyzing the diffusion trajectory and distribution of the ions. ,in, This refers to ion diffusion flux, expressed in mol / (m²). 2 ·s); This is the ion diffusion coefficient, in units of m. 2 / s, calculated using mean square displacement; This represents the ion concentration gradient, in mol / m³. 4 ; This is the gradient operator. The diffusion behavior of different ions reflects their penetration ability in the interfacial region and the strength of their interaction with the interface.

[0105] The adhesion mechanism analysis unit 33 is used to analyze the influence mechanism of ions on interfacial adhesion properties, quantifying the orientation angle distribution of interfacial molecules, hydrogen bond network density, and functional group migration. Specifically, it calculates the interfacial binding energy: ,in, , which is the interfacial binding energy, expressed in kJ / mol, and is a quantitative indicator of interfacial adhesion strength; The total energy of the system is expressed in kJ / mol. This represents the molecular energy of asphalt, expressed in kJ / mol. This represents the surface energy of the aggregate, expressed in kJ / mol. The value represents the ion energy, expressed in kJ / mol. Furthermore, the analysis of interfacial molecular configuration changes, including orientation angle distribution, hydrogen bond network, and functional group migration, directly relates to interfacial adhesion properties.

[0106] The results visualization unit 34 is used to convert simulation results into a visual representation, generating molecular motion trajectories and energy distribution maps. In practice, conventional molecular dynamics visualization techniques, such as VMD or OVITO software, are used to generate atomic trajectory animations, density distribution maps, and energy evolution curves, intuitively displaying the molecular behavior and interactions in the interface region.

[0107] like Figure 6 As shown, the intelligent analysis module 4 includes a data fusion unit 41, an association model construction unit 42, a performance prediction unit 43, and an early warning mechanism unit 44.

[0108] The data fusion unit 41 is used to integrate image processing results and molecular dynamics simulation data to construct a unified data representation. Specifically, it aligns curvature feature data and molecular dynamics simulation results in time and space to construct a fused data structure. ,in, To merge datasets; For the first Curvature feature vector of each data point; This represents the corresponding ion diffusion flux, in mol / (m²). 2 ·s); The interfacial binding energy is expressed in kJ / mol. The time point is in seconds (s). This represents the number of data points. This indicates a set construction operation. Data fusion employs a spatiotemporal registration strategy to ensure spatiotemporal consistency of data from different sources. Simultaneously, all types of data undergo normalization processing to unify the scale of different data sources, preventing any single type of data from dominating the analysis results.

[0109] The correlation model building unit 42 is used to establish a correlation model between salt concentration and interfacial adhesion strength based on fused data, and uses differential geometry to characterize the relationship between interfacial morphology and molecular behavior. Specifically, it calculates the rate of change of interfacial curvature based on differential geometry theory. and As a quantitative indicator of the rate of damage development: , ,in, For position The rate of change of the Gaussian curvature at a given point over time, in nm. / s; For position The rate of change of the mean curvature at a given point over time, in nm. / s; This represents the partial derivative with respect to time. It is combined with the ion diffusion flux from molecular dynamics simulations. Establish a quantitative relationship between ion concentration and curvature change: , ,in, The correlation coefficient between the average curvature change and ion flux is given in nm·s / mol. The coefficient representing the correlation between the average curvature change and the ion concentration is given in nm. ·L / mol; The rate of change of basic curvature, in nm. / s; The correlation coefficient between Gaussian curvature change and ion flux is given in nm. ·m 2 ·s / mol; The coefficient representing the correlation between Gaussian curvature change and ion concentration is given in nm. ·L / mol; The rate of change of the basic Gaussian curvature, in nm. / s; The local ion concentration is expressed in mol / L. A predictive model for interfacial adhesion strength is established based on the correlation between curvature changes and interfacial energy changes. ,in, This refers to the interfacial adhesion strength, expressed in MPa. Initial adhesion strength, in MPa; The coefficient representing the influence of average curvature variation is expressed in MPa·s·nm. The Gaussian curvature variation influence coefficient is expressed in MPa·s·nm. 2 ; Indicates taking the absolute value; Indicates from time 0 to The integral; Let be the integral variable, representing time in seconds. For limestone aggregate, The empirical value is approximately 0.15 MPa·s·nm. Approximately 0.08 MPa·s·nm 2 For granite aggregates, Approximately 0.12 MPa·s·nm, Approximately 0.10 MPa·s·nm 2 This correlation model, based on differential geometry and molecular dynamics, has clear physical meaning and can accurately predict changes in interfacial adhesion properties.

[0110] The performance prediction unit 43 is used to predict the degradation process of the interface adhesion performance over time based on the correlation model, and generate a performance degradation curve. Specifically, a time factor is introduced to construct a function for the degradation of adhesion strength over time. ,in, For time The interfacial adhesion strength at that time, expressed in MPa; Initial adhesion strength, in MPa; This is the degradation rate constant, in days. It is related to the type and concentration of salt; Time, in days; It is a time exponent, dimensionless, and reflects the nonlinear characteristics of the degradation process. For a NaCl solution (3.5% concentration)... The typical value is 0.015 days. , Approximately 0.7; for CaCl2 solution (3.5% concentration). Approximately 0.022 days , It is approximately 0.75. By predicting the change in interfacial adhesion strength over time, the service life of materials in salt-corrosion environments can be assessed.

[0111] The early warning mechanism unit 44 is used to set an early warning threshold and a critical damage threshold based on the prediction results, and to issue an early warning before the interface damage reaches a critical state. Specifically, the early warning threshold for the relative loss rate of adhesion strength is set as follows: ,in, The value represents the relative loss rate of adhesion strength, expressed in % (%). Initial adhesion strength, in MPa; For time The adhesion strength at a given time, expressed in MPa. When the early warning threshold is reached (e.g., 20%), the system issues an early warning. When a critical damage threshold (e.g., 50%) is reached, a severe damage warning is issued. This multi-level warning mechanism can effectively support preventative maintenance decisions.

[0112] The environmental control module 5 is used to control the temperature, humidity and salt solution concentration of the salt corrosion test environment; it provides a variety of salt solution types, including chloride solution, sulfate solution and acetate solution; it controls the seepage rate of the salt solution through a micro-electric valve; and it transmits environmental parameter data to the intelligent analysis module 4 for correlation analysis between environmental factors and damage degree.

[0113] In practice, the temperature control range is -20℃ to 60℃ with an accuracy of ±0.5℃; the humidity control range is 10% to 95% with an accuracy of ±2%; and the salt solution concentration can be adjusted from 0.1% to 10% with an accuracy of ±0.1%. The salt solution percolation rate is controlled from 0.1 to 10 ml / min, precisely controlled by a micro-electric valve. Environmental parameters are collected every 30 seconds to form time-series data, which is transmitted to the intelligent analysis module 4 for correlation analysis. This precise environmental control ensures the repeatability of experimental conditions and the reliability of data.

[0114] like Figure 7 As shown, the image acquisition module 1 includes a high-resolution scanning electron microscope unit 11, a high-throughput imaging unit 12, and an image transmission unit 13.

[0115] The high-resolution scanning electron microscope (SEM) unit 11 is used to acquire microscopic images of the asphalt-aggregate interface with a resolution of not less than 10 nm. In specific implementation, it uses SEM parameters of 15 kV voltage and 10 mm working distance, with magnification adjustable from 500 to 50,000 times, which can clearly observe the interface microstructure and early damage characteristics.

[0116] The high-throughput imaging unit 12 is used to set imaging parameters according to the preset damage propagation rate and sampling frequency to achieve continuous acquisition at fixed points. In specific implementation, an appropriate sampling frequency (once every 4 to 24 hours) is set according to the estimated damage propagation rate (approximately 0.1-10 μm / day), and marker tracking technology is used to ensure the consistency of position during long-term observation.

[0117] The image transmission unit 13 is used to transmit the acquired image data to the image processing module 2, supporting real-time data streaming and batch transmission. Specifically, a high-speed data transmission interface (such as USB 3.0 or Gigabit Ethernet) is used to achieve real-time image transmission, while also supporting local caching and batch transmission modes to improve the system's fault tolerance and flexibility.

[0118] Data management module 6 is used to store and manage the data generated by each module of the system, including raw images, processing results and analysis reports; establish an experimental sample database to record experimental results of different aggregate types, asphalt varieties and salt corrosion conditions; provide data retrieval and query functions, supporting retrieval by sample attributes, experimental conditions and damage degree; and generate analysis reports, including interface damage statistics, evolution trends and prediction results.

[0119] In practice, a hybrid architecture combining relational databases (such as PostgreSQL) and non-relational databases (such as MongoDB) is adopted. The relational database stores experimental metadata and relational information, while the non-relational database stores large volumes of images and computational results. Data retrieval supports multi-condition combined queries, with response time optimized to within 100 ms. Analysis reports utilize an interactive web interface, supporting multi-dimensional data visualization and export functions.

[0120] The working process of this invention is as follows: First, the image acquisition module 1 acquires the microscopic morphology image of the asphalt-aggregate interface under salt corrosion environment; then, the image processing module 2 processes the acquired image, extracts the interface curvature features based on differential geometry theory, and outputs the interface segmentation results through a deep learning network; next, the molecular dynamics simulation module 3 constructs a molecular model based on the interface damage features, simulates the diffusion behavior of salt ions in the interface region, and outputs molecular diffusion behavior data; finally, the intelligent analysis module 4 integrates the image processing and molecular dynamics simulation results, constructs a correlation model between salt concentration and interface adhesion strength, predicts the deterioration process of interface adhesion performance over time, and realizes damage early warning.

[0121] The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion conditions provided by this invention can achieve high-precision identification and prediction of interface damage, providing a scientific basis for asphalt pavement material design and service life assessment. The system organically combines differential geometry theory, deep learning technology, and molecular dynamics simulation to construct a multi-scale correlation model from microstructure to macroscopic performance, showing broad application prospects.

[0122] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment, characterized in that, include: The image acquisition module is used to acquire microscopic morphology images of the asphalt-aggregate interface under salt corrosion environment based on scanning electron microscopy imaging technology. An image processing module, communicatively connected to the image acquisition module, receives microscopic morphology images acquired by the image acquisition module, processes these images based on differential geometry theory, models the asphalt-aggregate interface as a two-dimensional manifold embedded in three-dimensional space, calculates interface curvature features, and extracts interface damage features. The image processing module includes an interface curvature analysis unit and a deep learning segmentation unit. The interface curvature analysis unit calculates the Gaussian curvature and average curvature distribution of the interface, generating a multi-scale curvature feature map. The deep learning segmentation unit employs a U-Net network architecture optimized by curvature flow evolution and genetic algorithms, receives the multi-scale curvature feature map as an additional input channel, and outputs the interface segmentation result. The molecular dynamics simulation module is communicatively connected to the image processing module. It is used to receive the interface damage features extracted by the image processing module, construct a molecular dynamics model based on LAMMPS software and ReaxFF reactive force field, simulate the diffusion behavior of salt ions in the interface region, and output molecular diffusion behavior data. The intelligent analysis module is communicatively connected to the image processing module and the molecular dynamics simulation module, respectively. It is used to integrate the interface damage characteristics of the image processing module and the molecular diffusion behavior data of the molecular dynamics simulation module, construct a correlation model between salt concentration and interface adhesion strength, and predict the degradation process of interface adhesion performance over time. Calculation of the rate of change of interface curvature based on differential geometry theory and As a quantitative indicator of the rate of damage development: , ,in, For position The rate of change of the Gaussian curvature at a given point over time, in nm. / s; For position The rate of change of the mean curvature over time, in nm / s; This represents the partial derivative with respect to time, combined with the ion diffusion flux from molecular dynamics simulations. Establish a quantitative relationship between ion concentration and curvature change: , ,in, The correlation coefficient between the average curvature change and ion flux is given in nm·s / mol. The coefficient representing the correlation between the average curvature change and the ion concentration is given in nm. ·L / mol; The rate of change of basic curvature, in nm. / s; The correlation coefficient between Gaussian curvature change and ion flux is given in nm. ·m 2 ·s / mol; The coefficient representing the correlation between Gaussian curvature change and ion concentration is given in nm. ·L / mol; The rate of change of the basic Gaussian curvature, in nm. / s; The local ion concentration is expressed in mol / L. A predictive model for interfacial adhesion strength is established based on the correlation between curvature changes and interfacial energy changes. ,in, The interfacial adhesion strength is expressed in MPa. Initial adhesion strength, in MPa; The coefficient representing the influence of average curvature variation is expressed in MPa·s·nm. The Gaussian curvature variation influence coefficient is expressed in MPa·s·nm. 2 ; Indicates taking the absolute value; Indicates from time 0 to The integral; Let be the integral variable, representing time, in seconds (s).

2. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, The interface curvature analysis unit includes: The preprocessing subunit is used to enhance and filter noise in the microscopic topography image; The interface parameterization subunit is used to map the interface region to a height field function and fit the local surface using the moving least squares method. The curvature calculation subunit is used to construct the shape operator matrix and calculate the principal curvature, Gaussian curvature, and mean curvature. The multi-scale analysis subunit is used to construct a 5-level Gaussian scale space, calculate curvature features at different scales, and generate a curvature scale space image pyramid. The anomaly detection subunit is used to detect curvature anomaly regions based on a curvature statistical model and generate a heatmap of the degree of curvature anomaly.

3. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, The deep learning segmentation unit includes: The curvature flow evolution subunit is used to generate interface evolution sequences based on the average curvature flow model and construct evolution feature channels; The genetic algorithm optimization subunit is used to optimize the U-Net network architecture, encodes network structure parameters and optimizer parameters, and searches for the optimal network structure through population evolution; The curvature attention subunit is used to generate a spatial attention weight map based on the curvature distribution, guiding the network to focus on regions with high curvature changes. The multi-task learning subunit is used to simultaneously optimize the interface segmentation task and the curvature prediction task, and outputs the interface segmentation mask and curvature prediction value.

4. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, The molecular dynamics simulation module includes: The model building unit is used to establish a geometric model based on the interface damage features extracted by the image processing module, and to set the molecular force field parameters and boundary conditions. The ion diffusion simulation unit is used to simulate the penetration and diffusion process of different types of salt ions in the interface region; The adhesion mechanism analysis unit is used to analyze the influence mechanism of ions on interfacial adhesion performance and quantify the orientation angle distribution of interfacial molecules, hydrogen bond network density, and functional group migration. The results visualization unit is used to convert simulation results into a visual representation, generating molecular motion trajectories and energy distribution maps.

5. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, The intelligent analysis module includes: The data fusion unit is used to integrate image processing results and molecular dynamics simulation data to construct a unified data representation. The correlation model building unit is used to establish a correlation model between salt concentration and interfacial adhesion strength based on fused data, and uses differential geometry to characterize the relationship between interfacial morphology and molecular behavior. The performance prediction unit is used to predict the degradation process of the interface adhesion performance over time based on the correlation model and generate a performance degradation curve. The early warning mechanism unit is used to set early warning thresholds and critical damage thresholds based on prediction results, and to issue early warnings before the interface damage reaches a critical state.

6. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, It also includes an environmental control module, used for: Control the temperature, humidity, and salt solution concentration of the salt corrosion test environment; Offers a variety of salt solution types; The percolation rate of the salt solution is controlled by a miniature electronically controlled valve; Environmental parameter data is transmitted to the intelligent analysis module for correlation analysis between environmental factors and the degree of damage.

7. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, The image acquisition module includes: A high-resolution scanning electron microscope unit is used to acquire microscopic images of the asphalt-aggregate interface with a resolution of not less than 10 nm. The high-throughput imaging unit is used to set imaging parameters according to the preset damage propagation rate and sampling frequency to achieve fixed-point continuous acquisition; The image transmission unit is used to transmit the acquired image data to the image processing module, supporting real-time data streaming and batch transmission.

8. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 1, characterized in that, It also includes a data management module, used for: Store and manage the data generated by each module of the system, including raw images, processing results, and analysis reports; Establish an experimental sample database to record experimental results under different aggregate types, asphalt varieties, and salt corrosion conditions; It provides data retrieval and query functions, supporting searches by sample attributes, experimental conditions, and degree of damage; Generate analysis reports, including interface damage statistics, evolution trends, and prediction results.

9. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 5, characterized in that, The correlation model building unit establishes the correlation between salt concentration and interfacial adhesion strength in the following manner: The rate of change of interface curvature is calculated based on differential geometry theory and used as a quantitative indicator of the speed of damage development. By combining ion diffusion flux from molecular dynamics simulations, a quantitative relationship between ion concentration and curvature change is established. By establishing the correlation between curvature change and interface energy change, a predictive model for interface adhesion strength is established. By introducing the time factor, a decay function of adhesion strength over time is constructed to achieve lifetime prediction.

10. The intelligent analysis system for asphalt-aggregate interface damage under salt corrosion environment according to claim 3, characterized in that, The optimization process of the genetic algorithm optimization subunit includes: Initialize the population and randomly generate 50 network architecture configurations; The fitness of each architecture is evaluated, taking into account segmentation accuracy, computational efficiency, and generalization ability. Outstanding individuals are selected through a tournament selection mechanism; Perform single-point crossover and uniform crossover operations on the selected individuals; Gaussian mutation and reset mutation are used to maintain population diversity; The process involves 100 generations of evolutionary iterations, retaining 10% of the elite individuals. Select the optimal architecture for fine-tuning to determine the final network structure.

Images