A method and system for optimizing long-life pavement structures based on artificial intelligence

Abstract

This invention relates to the field of intelligent road engineering technology, and more particularly to an artificial intelligence-based method and system for optimizing long-life pavement structures. The method includes the following steps: acquiring pavement structure data; constructing a structural mechanics model based on the pavement structure data to obtain the structural mechanics model; simulating fatigue loss based on the structural mechanics model to obtain fatigue loss data; performing rapid physical simulation based on the fatigue loss data to obtain loss simulation data; and predicting the physical prior life based on the loss simulation data to obtain life prediction data. This invention achieves accurate quantification of the spatial distribution and evolution trend of fatigue damage through fracture entropy field modeling and crack density analysis. Combined with AI-driven multi-timeframe evolution simulation, it can significantly improve the accuracy of pavement structure life prediction and the targeted nature of design optimization.

Description

Technical Field

[0001] This invention relates to the field of intelligent road engineering technology, and in particular to a method and system for optimizing long-life pavement structures based on artificial intelligence. Background Technology

[0002] Long-life pavement refers to a highly durable pavement structure whose main structural layers (such as base course, subbase course, and subgrade) will not experience structural failure within a design service life of more than 30 years under normal traffic loads and appropriate maintenance. It requires only periodic surface maintenance to maintain its service performance (such as smoothness and skid resistance). Existing pavement structure design methods largely rely on traditional mechanics theories and empirical formulas to select structural parameters and assess lifespan. However, during long-term service, nonlinear factors at the microscopic level of pavement materials can significantly affect structural performance. Furthermore, factors such as traffic loads and environmental changes in actual operation make it difficult for traditional methods to accurately predict structural lifespan, thus affecting the scientific validity and economic efficiency of pavement maintenance strategies. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention proposes an artificial intelligence-based method and system for optimizing long-life pavement structures, thereby resolving at least one of the aforementioned technical issues.

[0004] This application provides an artificial intelligence-based method for optimizing long-life pavement structures, the method comprising:

[0005] S1. Obtain pavement structure data, construct a structural mechanics model based on the pavement structure data, and obtain the structural mechanics model;

[0006] S2. Perform fatigue loss simulation based on the structural mechanics model to obtain fatigue loss data;

[0007] S3. Perform rapid physical simulation based on fatigue loss data to obtain loss simulation data;

[0008] S4. Based on the loss simulation data, perform physical prior lifetime prediction to obtain lifetime prediction data.

[0009] This invention achieves efficient characterization of the multi-dimensional performance of pavement structures by integrating structural mechanics modeling with AI-driven fatigue loss simulation. Compared to traditional design methods relying on empirical formulas or single mechanical calculations, this method can dynamically reflect the evolution of micro-damage caused by material performance degradation, cumulative load effects, and environmental factors. Rapid physical simulation technology reduces numerical computation costs, improving structural evaluation efficiency while ensuring accuracy. Combined with a physical prior life prediction model, it enables quantitative prediction of the durability performance of different design schemes during their service life, possessing superior engineering adaptability and promotional value, and contributing to the intelligent optimization and green sustainable development of long-life pavement structures.

[0010] Optionally, S1 includes:

[0011] Obtain pavement structure data, divide the pavement structure into structural layers based on the pavement structure data, and obtain structural layer data, which includes surface layer data, intermediate layer data, base layer data, and subgrade data;

[0012] By setting inter-layer conditions for the structural layer data, an inter-layer model is obtained.

[0013] Physical conditions are set for the interlayer model to obtain the structural physical model;

[0014] By setting material property degradation conditions for the structural physical model, a structural mechanical model is obtained.

[0015] This invention achieves high-fidelity construction of a structural mechanics model by layering and setting conditions for pavement structure data, demonstrating significant advantages in physical representation. The hierarchical division of the structure ensures that each functional layer (such as surface layer, intermediate layer, base layer, and subgrade) has clear mechanical boundaries in the model, facilitating the simulation of their respective stress and deformation responses. The introduction of interlayer conditions enhances the model's ability to reflect local failure behaviors such as interlayer shear slip and bond failure. By setting physical boundaries and material degradation functions, it is possible not only to dynamically simulate the impact of environmental and usage factors such as temperature and humidity coupling and load accumulation on material properties, but also to provide a time-series evolution basis for life assessment. This modeling process effectively improves the accuracy and adaptability of pavement structure simulation, providing a solid data and physical foundation for fatigue analysis and life prediction.

[0016] Optionally, S2 includes:

[0017] Fracture entropy field data is obtained by generating the fracture entropy field based on the structural mechanics model.

[0018] Crack density was calculated based on the structural mechanics model to obtain crack density data;

[0019] The structural mechanics model was annotated based on fracture entropy field data and crack density data to obtain fatigue loss data.

[0020] This invention effectively enhances the physical interpretability and spatial resolution of fatigue loss simulation by utilizing the dual characteristics of fracture entropy field and crack density. The fracture entropy field reflects the local energy dissipation and damage accumulation trends of the material, possessing a highly sensitive ability to identify damage precursors; crack density characterizes the actual evolution of microscopic damage from both quantitative and spatial distribution dimensions. Integrating these two features for the annotation of structural mechanics models not only strengthens the model's ability to express the degree of damage in various heterogeneous regions but also achieves an effective mapping from microscale features to macroscopic fatigue performance, significantly improving the accuracy and reliability of fatigue simulation.

[0021] Optionally, the generation of the fracture entropy field includes:

[0022] Based on the structural mechanics model, the stress distribution of the structural layer data under external load disturbance is locally scanned to obtain local scan data.

[0023] Based on the local scan data, the equivalent stress fluctuation amplitude, shear stress gradient, and bond interface instability index were calculated, and the equivalent stress fluctuation amplitude data, shear stress gradient data, and bond interface instability data were obtained respectively.

[0024] Discrete meshes are generated based on equivalent stress fluctuation amplitude data, shear stress gradient data, and bond interface instability data to obtain mesh microstate data.

[0025] The fracture entropy field data is obtained by calculating the local fracture entropy based on the grid microstate data.

[0026] This invention achieves high-resolution modeling of the structural micro-damage potential through the extraction of multidimensional stress response characteristics and local energy dissipation analysis. By performing local scanning of stress distribution within structural layers, local stress concentration areas under external load disturbances can be accurately captured, overcoming the deficiency of traditional overall analysis in neglecting weak instability zones. Through equivalent stress fluctuation amplitude, shear stress gradient, and bond interface instability indices, the potential fracture risk of the structure can be revealed from different mechanical perspectives. Combined with a discrete mesh generation strategy, the mechanical response data is transformed into mesh elements with location semantics and physical state, improving the computability and expressive power of fracture entropy.

[0027] Optionally, the crack density calculation includes:

[0028] Crack data were obtained by using a structural mechanics model to determine crack criterion.

[0029] Based on the crack data, the number density and length density are calculated to obtain the number density data and length density data.

[0030] Spatial crack density map data is obtained by mapping spatial crack density map based on quantity density data;

[0031] Local density mutation data are obtained by extracting local density mutation data from the spatial crack density map data.

[0032] The crack density data is obtained by integrating the number density data, length density data, and local density abrupt change data.

[0033] In this invention, the crack criterion identifies key damage locations based on a structural mechanics model, ensuring the physical reliability of the input data. Through dual-index modeling of quantity density and length density, it not only captures the frequency of crack occurrence but also quantifies the scale of crack development, helping to assess the rate and extent of structural integrity degradation. The mapping of the spatial crack density map projects the original crack data onto a distribution map with spatial semantics. Combined with the extraction of local density abrupt changes, it effectively reveals local damage clusters and risk hotspots within the structure.

[0034] Optionally, the model annotation includes:

[0035] Crack direction features are extracted from crack data to obtain crack direction feature data;

[0036] Anisotropic features are extracted based on fracture entropy field data and crack direction feature data to obtain anisotropic feature data.

[0037] The fracture entropy field data, crack density data, and anisotropic characteristic data are used to annotate the structural mechanics model to obtain fatigue loss data.

[0038] In this invention, crack direction feature extraction can accurately characterize the dominant crack propagation path, avoiding the neglect of crack orientation influence in traditional fatigue assessment. Coupled with fracture entropy field data, anisotropic feature extraction effectively reflects the differences in energy dissipation and fracture evolution in different directions, enhancing the model's adaptability to heterogeneous and asymmetric structural responses. By integrating fracture entropy field, crack density, and anisotropic features into the structural mechanics model, a fatigue loss labeling system driven by physical mechanisms is formed, enabling the model to possess clear damage semantic identification and traceability during prediction and optimization.

[0039] Optionally, S3 includes:

[0040] Multi-time frame data is extracted from fatigue loss data to obtain multi-time frame data;

[0041] Crack evolution slope data is obtained by calculating the crack evolution slope based on multi-time frame data.

[0042] Entropy growth rate data is obtained by extracting the entropy growth rate data from the fracture entropy field data corresponding to multiple time frames.

[0043] The evolution of hot spots in the risk area is analyzed based on multi-time frame data to obtain hot spot data of the risk area;

[0044] Based on crack evolution slope data, entropy growth rate data, and hot spot data in the risk area, a reduced-order modeling spatial expansion simulation was performed to obtain loss simulation data.

[0045] This invention significantly improves the analytical accuracy and simulation efficiency of the damage evolution process by combining multi-timeframe sequence analysis and multi-physics field fusion calculation. The crack evolution slope reflects the temporal intensity characteristics of crack growth and can reveal the main trend of fatigue propagation; the entropy growth rate index, combined with the changes in the fracture entropy field, captures the dynamic energy dissipation caused by microstructural instability; and the hot spot evolution in the risk region depicts the local damage accumulation path from a spatial distribution perspective. By integrating the above-mentioned multi-source dynamic features and adopting a reduced-order modeling strategy for spatial expansion simulation, not only is the computational complexity effectively reduced, but also rapid simulation of high-dimensional structural states can be achieved while maintaining physical reliability, enhancing the applicability and response speed for life assessment and structural optimization in practical engineering.

[0046] Optionally, the entropy growth rate extraction includes:

[0047] The global entropy growth rate and the neighborhood entropy growth rate are calculated based on the fracture entropy field data corresponding to the multi-time frame data, and the global entropy growth rate data and the neighborhood entropy growth rate data are distributed.

[0048] The entropy increase time-varying variance is calculated based on the global entropy growth rate data and the neighborhood entropy growth rate data to obtain the entropy increase time-varying variance data;

[0049] Local entropy flow graphs are constructed based on entropy-increasing time-varying variance data to obtain local entropy flow graph data;

[0050] Entropy growth rate data is obtained by performing cross-resolution entropy fusion based on local entropy flow graph data.

[0051] This invention introduces multi-scale, multi-timeframe fracture entropy evolution features, which not only enhances the dynamic perception of local instability trends in materials but also strengthens the physical interpretability of structural fatigue evolution prediction. Global entropy growth rate calculation macroscopically reflects the overall energy dissipation trend, while neighborhood entropy growth rate calculation accurately captures the microscopic disturbance response in local areas. Time-varying variance analysis of entropy growth enables the system to sensitively perceive the fluctuations in instability behavior across different time periods, facilitating the identification of potential critical evolution inflection points. The construction of local entropy flow graphs enables modeling of information flow in spatiotemporal micro-intervals, while the cross-resolution entropy fusion mechanism effectively coordinates feature consistency across high and low resolutions, improving the model's adaptability and accuracy to damage propagation in heterogeneous scenarios.

[0052] Optionally, the evolution of hotspots in the risk area includes:

[0053] Fatigue concentration areas are selected based on multi-time frame data to obtain hot spot data;

[0054] Contour extraction is performed on the hot spot data to obtain hot spot contour data;

[0055] The centroid trajectory is extracted based on the hot spot contour data to obtain the hot spot centroid data;

[0056] Dynamic evolution was performed based on hotspot centroid data to obtain hotspot data for the risk area.

[0057] This invention effectively improves the identification accuracy and temporal evolution analysis capability of local high-risk areas by progressively extracting and tracking the spatial features and dynamic response behavior of fatigue concentration areas across multiple time frames. Preliminary screening of fatigue concentration areas can focus on potential failure core regions, significantly reducing redundant analysis overhead; contour extraction clearly defines the boundary morphology of hot spots, providing accurate geometric basis for subsequent structural change modeling; center of gravity trajectory extraction introduces spatial-temporal bidirectional information fusion, enhancing the ability to continuously model regional displacement and evolution trends; dynamic evolution analysis based on center of gravity data not only captures regional stress concentration, expansion, or migration behavior but also uncovers the dominant mechanism of hot spot development.

[0058] Optionally, this application also provides an artificial intelligence-based long-life pavement structure optimization system for executing the artificial intelligence-based long-life pavement structure optimization method described above, wherein the artificial intelligence-based long-life pavement structure optimization system includes:

[0059] The structural mechanics model building module is used to acquire pavement structure data and build a structural mechanics model based on the pavement structure data to obtain the structural mechanics model.

[0060] The fatigue loss simulation module is used to simulate fatigue loss based on the structural mechanics model and obtain fatigue loss data.

[0061] The rapid physical simulation module is used to perform rapid physical simulations based on fatigue loss data to obtain loss simulation data.

[0062] The physical prior lifetime prediction module is used to predict physical prior lifetime based on loss simulation data, and obtain lifetime prediction data.

[0063] The purpose of this invention is as follows: Step S1 involves introducing real pavement structure data to construct a structural mechanics model that includes interlayer contact stress, bonding conditions, and environmental coupling boundaries, thereby improving the physical realism and adaptability of the modeling; Step S2 integrates fatigue indices such as crack density and fracture entropy on this model to simulate the microscopic damage evolution process under real working conditions, ensuring that fatigue loss assessment has material degradation and anisotropic characteristics; Step S3, based on fatigue loss data, combines multi-time frame sequences and hot spot evolution paths to achieve rapid physical-level reduced-order modeling and extended simulation, significantly reducing the computational cost of traditional simulation; Step S4 integrates loss simulation results to construct a priori life prediction model, enabling a forward-looking assessment of pavement service life and suggestions for structural life extension strategies. Attached Figure Description

[0064] Other features, objects, and advantages of this application will become more apparent from the following detailed description of the non-limiting embodiments, taken with reference to the accompanying drawings:

[0065] Figure 1 A flowchart illustrating the steps of an artificial intelligence-based long-life pavement structure optimization method according to one embodiment is shown.

[0066] Figure 2 A flowchart illustrating the steps of a structural mechanics model construction method according to one embodiment is shown.

[0067] Figure 3 A flowchart illustrating the steps of a fatigue loss simulation method according to one embodiment is shown.

[0068] Figure 4 A flowchart illustrating the steps of a fast physical simulation method according to one embodiment is shown.

[0069] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0070] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0071] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. Functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0072] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0073] Please see Figures 1 to 4 This application provides an artificial intelligence-based method for optimizing long-life pavement structures, the method comprising:

[0074] S1. Obtain pavement structure data, construct a structural mechanics model based on the pavement structure data, and obtain the structural mechanics model;

[0075] Specifically, key material parameters of each pavement structure layer (including surface layer, base layer, subbase layer, and subgrade) are collected. These parameters include the thickness, elastic modulus, Poisson's ratio, and other basic mechanical properties of each layer, used to characterize the structural stiffness and deformation capacity. Corresponding traffic load data are acquired, including typical axle load values, tire contact pressure distribution, vehicle speed, and load repetition frequency. This is combined with external environmental conditions, including regional temperature and humidity ranges and diurnal temperature variation cycles. Based on the above structural layer data and load information, a multi-layer structural mechanical analysis model is constructed. The model employs a layered modeling approach, treating each structural layer as a sub-structural unit with independent mechanical properties, and establishing contact surfaces between layers to simulate stress transfer and deformation coordination. In setting interlayer boundary conditions, an interface bonding model is considered, preferably a coupled combination of a Coulomb friction model and a shear slip model, to simulate the frictional bonding and relative slip behavior between structural layers. A temperature-humidity-load coupling factor is set in the model to combine parameters such as temperature gradient and humidity penetration depth with the stress-strain relationship of structural elements, simulating the physical behavior caused by thermal expansion and contraction and changes in moisture content. A time decay function is set in the model. The time decay function is represented in an exponential form, that is, let the th... The modulus degradation behavior of layered materials is determined by the function The description, in which Indicates the first Layer at time stiffness attenuation factor This is the material degradation rate coefficient. Evolution time.

[0076] S2. Perform fatigue loss simulation based on the structural mechanics model to obtain fatigue loss data;

[0077] Specifically, time-varying traffic loads are applied to the constructed structural mechanics model. The traffic load simulation considers typical traffic flow patterns and is constructed based on parameters such as axle load spectrum, tire contact area, repetitive traffic frequency, and vehicle speed. The loads are applied to the surface nodes of the surface layer and distributed according to time sequence to form a dynamic load input field.

[0078] After the load is applied, the system solves the stress-strain response of each structural element in the model and extracts mechanical variables such as axial stress, shear stress, and principal strain at each time point, providing basic data for subsequent damage calculation.

[0079] Subsequently, based on the disturbance response results, a fracture entropy field generation model was used to analyze the local energy fluctuations of each element. This model combines information such as stress fluctuation amplitude, shear stress gradient, and interface stability indices to measure the degree of energy imbalance in each structural element under external load disturbances, outputting fracture entropy field data representing the local thermal imbalance trend of the system, used to identify potential damage-active regions. The system identifies potential crack locations based on the maximum principal stress criterion. That is, for a given structural element, when its principal stress exceeds the material's set fracture threshold, it is considered a region with potential crack risk. Within the identified region, the number of cracks per unit area (i.e., crack number density) and the total crack length (i.e., crack length density) are calculated. Based on the acquired fracture entropy field data and crack density data, combined with information on the local principal stress direction and crack propagation direction, anisotropic damage characteristics of each structural element are extracted. The fracture entropy data, crack density data, and anisotropic characteristics are fused / annotated, and the spatial annotation of damage factors is completed in the structural mechanics model, outputting fatigue loss data for each structural element. The fatigue loss data includes, but is not limited to, fatigue damage index (representing the cumulative degree of damage), energy dissipation rate (reflecting the energy dissipation capacity per unit time), and critical approach rate (used to represent the relative distance between the current state and the material failure threshold).

[0080] S3. Perform rapid physical simulation based on fatigue loss data to obtain loss simulation data;

[0081] Specifically, based on the fatigue loss data obtained above, a multi-time-frame sequence is constructed. Each time frame corresponds to the fatigue state of the pavement structure after continuous loading under a certain working condition, including time-varying indices such as crack density, fracture entropy value, and hot spot characteristics. Based on this multi-time-frame sequence, the system calculates the crack evolution slope, representing the rate of change of the total crack length or crack density per unit area per unit time. The local fracture entropy growth rate is calculated by the change in fracture entropy values ​​of adjacent structural units in consecutive time frames, i.e., the energy imbalance growth per unit time. The hot spot trajectory evolution characteristics are obtained by extracting the spatial contour of the hot spot region based on the high-value clustered areas in the aforementioned entropy field, and by fitting the trajectory of the hot spot centroid coordinates in consecutive time frames to obtain the spatial migration path and evolution direction of the hot spot. The system uses principal component analysis or an autoencoder neural network model to perform feature compression processing on the above high-dimensional time-series fields, generating a low-dimensional potential expression space. Based on the compressed representation space, a physical simulation network for spatial expansion simulation is constructed. This network combines time frame indexes, loading conditions (such as wheel and axle loads, temperature and humidity environment), structural location, and low-dimensional state variables to predict the evolution of structural fatigue state in the spatial domain at future time steps. This simulation network can provide rapid simulation support for different loading conditions (such as high temperature and high frequency, humid erosion, etc.).

[0082] S4. Based on the loss simulation data, perform physical prior lifetime prediction to obtain lifetime prediction data.

[0083] Specifically, based on the aforementioned rapid physical simulation results, the system extracts high-risk regions within the structure and tracks the cumulative damage data for these regions. The cumulative damage data is a continuous sequence of damage factors across multiple time frames, denoted as... ,in Indicates the first The system inputs local damage intensity indices (such as fracture entropy and crack density) at each frame. This time-series data is then fed into a deep learning model built on a Long Short-Term Memory (LSTM) network for lifetime trend prediction. The model structure is as follows: Input is a multidimensional damage sequence. ,in This represents the damage value. This is the spatial location information of the corresponding structural unit (which can be embedded as a vector). The system uses a feature vector representing operating conditions (such as temperature, humidity, and load frequency) as input to an intermediate layer. This intermediate layer includes an LSTM encoding layer containing one or two hidden units, with each layer having 64 to 128 nodes. A fully connected output layer outputs a damage prediction sequence for the next N frames. Mean squared error (MSE) is used as the training metric. The system pre-sets physical prior data based on empirical data or expert knowledge, setting fatigue limit conditions as life assessment thresholds. These include, but are not limited to, a fatigue life threshold (representing the maximum allowable number of cycles for a material under a specific load condition), a critical crack density threshold (representing the maximum allowable crack density before structural failure), and a maximum fracture entropy threshold (characterizing the energy dissipation limit). The system intersects these thresholds with the predicted curves output by the LSTM, identifying the position in the prediction sequence where the threshold is first exceeded as the remaining life of the structural unit. This life is defined as the time difference between the current moment and the intersection point, expressed in days or load cycles. Based on the unit life prediction, the system performs weighted fusion processing according to structural levels (surface layer, intermediate layer, base layer, subgrade). Specifically, the lifetime of each layer is calculated separately, and weights are assigned based on layer thickness, modulus, and structural importance factors to obtain the overall remaining lifetime index: ,in This refers to the remaining lifespan data of the overall structure. For the index of the structure layer, The number of structural layers (e.g., surface layer, intermediate layer, base layer, and subgrade, a total of 4 layers). For the first The weighting coefficients for each layer are set based on empirical data or expert knowledge. For the first The system predicts the lifetime of the layered structure and outputs a lifetime contribution analysis map of each structural layer to identify the layer with the shortest lifespan. Based on the lifetime prediction results, the system generates structural reinforcement suggestions and material alternatives through a preset suggestion parameter library, including but not limited to setting additional thickness suggestions for critical life layers; recommending new structural materials with higher fatigue strength or bonding performance; and outputting a visualized lifetime extension simulation map for decision-making.

[0084] Optionally, S1 includes:

[0085] S11. Obtain pavement structure data, divide the pavement structure into structural layers based on the pavement structure data, and obtain structural layer data, which includes surface layer data, intermediate layer data, base layer data and subgrade data.

[0086] Specifically, the system collects and summarizes the original structural parameter data related to the target road section. These parameters include the thickness (in centimeters) of each structural layer, material type identification, construction batch information, and compaction index. This data can be derived from design drawings, on-site inspection records, and historical construction archives. Based on the standard hierarchical classification logic commonly used in road engineering, the system deconstructs the original data into hierarchical structural layer datasets. Surface layer data: This layer uses asphalt mixtures, such as AC-13 or AC-20 particle size distributions, with a typical paving thickness ranging from 4 to 8 centimeters. The system determines the surface layer type based on the material identification field and extracts its thickness and construction density as surface layer structural elements. Intermediate layer data: This layer is mainly used for fatigue buffering and stress diffusion, and its construction forms include stress-absorbing layers (SAMI) or chip seal layers. The system detects the presence of the intermediate layer based on the interlayer material identification field and marks its fatigue-resistant material type and structural layer location. Subbase data: The subbase is the main load-bearing structural layer, typically consisting of a cement-stabilized crushed stone layer or a lime-fly ash crushed stone layer, with a thickness between 20 and 35 cm. The system automatically extracts the thickness, material composition, mix proportion, and construction strength grade of this layer. Subbase data: This layer is either natural foundation or artificial fill. The system obtains its California bearing capacity index (CBR value), compaction degree, and moisture content, etc., based on in-situ testing or geological survey data, as input for the mechanical properties of the structure's base. Through the above standardized structural layer division process, the system can output a structural layer dataset with hierarchical identification, physical parameters, and construction attributes.

[0087] S12. Set inter-layer conditions for the structural layer data to obtain the inter-layer model;

[0088] Specifically, the system identifies the contact interfaces between adjacent layers in the structural layer data and sets corresponding interlayer boundary conditions for each layer pair (such as surface layer and intermediate layer, intermediate layer and base layer, base layer and subgrade). Based on road construction process records, design parameters, or sensor feedback information, the system determines the contact type between each pair of structural layers and classifies it into one of the following three contact states: fully bonded state (ideal bond), indicating that a completely firm connection is formed between the upper and lower structural layers without relative slippage. The system sets the friction coefficient of this contact interface to be close to infinity and does not consider the influence of sliding or interface separation; partially bonded state (friction bond), indicating that there is a certain degree of friction bond effect at the contact interface, allowing limited slippage but still having a certain shear resistance. The system sets the friction coefficient μ to a range of 0.3 to 0.6 based on the on-site material conditions; unbonded state (slipping contact), indicating that no effective bond is formed between the upper and lower structural layers, allowing complete slippage. The system sets the friction coefficient to zero, and the interface contact only bears positive compressive stress and does not have shear transfer capability. The system performs mechanical modeling on the aforementioned contact states, employing a nonlinear contact element modeling method to construct the interlayer interaction mechanism. Specifically, the Coulomb friction model can be used, where the shear slip force is jointly determined by the friction coefficient μ and the normal contact force, and it possesses a shear limit determination function to achieve dynamic simulation of physical behaviors such as interlayer bond failure and slip propagation. The system outputs the number of each structural layer pair, the contact boundary type (including contact state and friction parameters), and the contact model form as a standard interlayer model data structure.

[0089] S13. Set the physical conditions for the interlayer model to obtain the structural physical model;

[0090] Specifically, the system applies stress loading conditions to the road structure. Based on preset parameters, a tire-structure coupled contact model is set on the top working surface of the structural model, considering the contact pressure distribution and dynamic transmission characteristics between the tire and the road surface. The load type can be selected as a moving load (simulating vehicle movement) or a cyclic load (simulating repetitive traffic). The peak axle load P is set according to different vehicle types, ranging from 40 to 80 kN. The frequency parameter of the traffic load is set based on historical experience or preset parameters, between 1 and 5 Hz. This frequency parameter, along with the Poisson's ratio of the structural material, affects the lateral stress distribution and dynamic response of the structure; the system performs coupled term modeling in the calculation. The system sets the boundary conditions of the structural model, specifically including symmetrical boundary settings. For the geometric centerline or mirror region of the structural model, offset symmetrical boundary conditions are set, i.e., zero displacement constraints are applied in the horizontal and vertical directions at this boundary to avoid non-physical deformation caused by artificial boundaries. The bottom boundary is set according to the foundation treatment method, including setting it as a completely fixed boundary, i.e., the bottom nodes have no displacement in three dimensions; or setting it as an elastic support boundary, introducing the foundation elastic coefficient k (unit: kN). This is to reflect the elastic response characteristics of the soil foundation.

[0091] S14. Set material property degradation conditions for the structural physical model to obtain the structural mechanical model.

[0092] Specifically, a time-dependent material degradation function, such as modulus degradation, is set for each structural layer. ,in for The elastic modulus of the material at any given time (unit: MPa). The initial elastic modulus, This is the degradation factor, which is the rate at which the stiffness of a material decreases with time or the number of load cycles during the loading process. Its value is determined based on material fatigue test results or empirical estimation, for example, 0.005-0.05 / year. It is an exponential function. To load the duration or usage time; or to use a degradation model based on cumulative strain, such as... ,in The modulus after N cycles. To accumulate traffic load cycles (e.g., equivalent standard axle load cycles). This is the logarithmic degradation coefficient of the modulus, such as 0.02-0.1. The system is configured with load cycle count; fatigue performance parameters are set, including the fatigue limit of the surface layer material. ,in This is the fatigue strength proportionality coefficient (usually an empirical value, depending on the material type). The current elastic modulus; the critical strain for base layer cracking. The system embeds the aforementioned degradation model and fatigue performance parameters into the finite element material definition module, enabling the material parameters to be dynamically updated over time or load cycles, forming a spatiotemporally coupled structural mechanics model. This model can adjust material stiffness, load-bearing capacity, and fatigue tolerance in real time during physical simulation.

[0093] Optionally, S2 includes:

[0094] S21. Generate the fracture entropy field based on the structural mechanics model to obtain fracture entropy field data;

[0095] Specifically, for each element mesh or voxel in the structural mechanics model, the following mechanical responses are extracted based on external disturbances (such as vehicle loads and temperature stress): equivalent stress fluctuation amplitude (equivalent stress fluctuation amplitude = maximum principal stress − minimum principal stress), shear stress gradient (shear stress gradient = square root of the sum of the squares of the rates of change of shear stress in the horizontal and vertical directions), and bond interface instability indices (such as shear slip or normal crack opening). The system performs local fracture entropy calculation, using the following formula: ,in For the first The fracture entropy value of each unit. This is the Boltzmann constant, which can be set as a normalization constant or an empirical weight. For micro-response category index, This represents the total number of micro-response categories counted, including equivalent stress fluctuation amplitude, shear stress gradient, and bond interface instability indices. For the first The unit exhibits the [missing information] under multiple disturbance loading conditions. The probability of a microscopic response (which can be obtained based on finite element substep statistics). The system outputs the corresponding fracture entropy value for each mesh element; forming a three-dimensional tensor as fracture entropy field data.

[0096] S22. Calculate the crack density based on the structural mechanics model to obtain crack density data;

[0097] Specifically, the system determines which element regions have a risk of crack initiation based on preset crack criteria. When the maximum principal strain of a certain element exceeds the material's critical principal strain threshold, it is considered to have a risk of crack initiation. Alternatively, if the stress intensity factor within a fixed spatial window of the crack tip is greater than the material's fracture toughness threshold (i.e., fracture strength factor), then that location is marked as a potential crack initiation region. The crack number density is calculated within each preset spatial window (e.g., 10 cm). 10 centimeters A 2 cm three-dimensional region is defined, and the number of crack center points that satisfy the above crack criteria is counted. The crack number density of the region is obtained by dividing the total number of cracks by the volume of the spatial window. The total length of all cracks in the region is calculated and normalized to a unit volume to obtain the crack length density. The two types of density values ​​(crack number density and length density) are mapped back to the mesh elements in the original structural model according to their corresponding positions to construct a three-dimensional spatial distribution map, denoted as the crack density map.

[0098] S23. Based on the fracture entropy field data and crack density data, the structural mechanics model is labeled to obtain fatigue loss data.

[0099] Specifically, based on the crack density map, if crack geometry information can be obtained, the spatial orientation of each crack can be modeled: the system fits the crack morphology to line segments and extracts the principal direction vector of the crack; it statistically analyzes the direction distribution of cracks within the region and constructs a direction histogram; the system constructs a principal direction tensor (i.e., a structural tensor) based on a preset region range (e.g., 3x3x3). The fracture entropy field data and crack direction information are jointly analyzed; the corresponding local fracture entropy value is extracted for each directional dimension (e.g., along 0°, 45°, and 90°); the statistical variance of the fracture entropy value in different directions is calculated to obtain the degree of fatigue anisotropy in that region. Based on existing fracture entropy field data, crack density map, and anisotropy index, the system performs fatigue annotation operations on each element in the structural mechanics model. The following three types of data are spatially mapped and indexed, unified into the structural model mesh element: the fracture entropy value of each element (representing its energy imbalance under multiple perturbation conditions); the crack density value (number density or length density) of each element; and the fatigue anisotropy index of each element.

[0100] Optionally, the generation of the fracture entropy field includes:

[0101] Based on the structural mechanics model, the stress distribution of the structural layer data under external load disturbance is locally scanned to obtain local scan data.

[0102] Specifically, the system applies representative external disturbance boundary conditions to the target structure, including but not limited to simulating typical traffic load conditions, such as periodic axle loads (modeled as sinusoidal loads). ,in For a moment Traffic disturbance load. This represents the peak value of the traffic disturbance load. It is a sine function. The load disturbance angular frequency, (Time-based); localized impact load or short-term overload simulation. Based on historical celestial data, the temperature difference distribution range is set to simulate structural thermal stress caused by diurnal or seasonal temperature variations, specifically using a periodic sinusoidal temperature function. Establish a thermo-mechanical coupling field, in which For a moment Temperature value, The average temperature reference value, The amplitude of the temperature disturbance. It is a sine function. The angular frequency of the temperature difference change. The time variable is used. Uneven settlement displacement boundaries are applied to the bottom or boundaries of the structural model to simulate the dynamic changes of the foundation soil layer under environmental conditions such as rainfall and compaction. After setting the disturbance boundaries, the system scans each finite element in the structural layer time-by-time within the time domain, collecting its stress indices, obtaining the stress tensor data of each element under disturbance, recording its complete stress change trajectory at different disturbance times, and representing this stress response data as a four-dimensional tensor sequence, i.e., constructing a space-time stress response surface to obtain local scan data.

[0103] Based on the local scan data, the equivalent stress fluctuation amplitude, shear stress gradient, and bond interface instability index were calculated, and the equivalent stress fluctuation amplitude data, shear stress gradient data, and bond interface instability data were obtained respectively.

[0104] Specifically, for any point in time, calculate the stress:

[0105] ,

[0106] in This is the equivalent stress value. for Directional normal stress, for Directional normal stress, for Directional normal stress, for Plane shear stress, for Plane shear stress, for Planar shear stress. For all moments within the scanning period, the maximum and minimum values ​​are recorded, and the stress fluctuation amplitude is calculated. ,in This represents the amplitude of equivalent stress fluctuation. For the maximum time point stress calculation, This is for minimum time-point stress calculation. The system calculates the three-dimensional shear stress gradient field: ,in This represents the spatial gradient modulus of shear stress. for Plane shear stress, for Plane shear stress, for Plane shear stress, for Spatial coordinate axes for Spatial coordinate axes for Spatial coordinate axes. For interlayer structures (such as the base-surface interface), extract the normal cracking displacement and tangential slip displacement, and calculate the interface failure index: ,in This is an index of bonding interface instability. This is a tangential slip displacement. For normal crack displacement, This is the critical tangential slip displacement. This is the critical normal cracking displacement. If If so, it is considered that there is a clear trend of instability.

[0107] Discrete meshes are generated based on equivalent stress fluctuation amplitude data, shear stress gradient data, and bond interface instability data to obtain mesh microstate data.

[0108] Specifically, the structural model space is discretized into small grid cells (also called voxels) with a fixed voxel size (e.g., 5 cm³). For each small grid cell, if the cell contains multiple finite element cells or scan point data, the mean or maximum value is used for dimensionality reduction. Alternatively, if the cell contains multi-layer material data, it can be synthesized in a weighted manner (e.g., based on volume percentage, layer position, etc.).

[0109] The fracture entropy field data is obtained by calculating the local fracture entropy based on the grid microstate data.

[0110] Specifically, within the perturbation period (such as time or load cycle), the frequency of occurrence of each state type in the voxel is counted and converted into a probability value. For the th Individual element (i.e., small grid unit), denoted as its first... The probability of a class state occurring is It satisfies the normalization condition, that is, the sum of the probabilities of all states equals 1. Calculate the entropy: ,in For the first The fracture entropy value of individual elements, For state category index, The total number of state categories. For the first The first individual element The static probability of a class state; or by calculation: ,in For the first The dynamic fracture entropy value of individual elements, For time frame indexing, For time frame weighting function, For the first Individual elements in time At the moment The probability of a class state. The fracture entropy value of each voxel is used to form the overall fracture entropy field.

[0111] Optionally, the crack density calculation includes:

[0112] Crack data were obtained by using a structural mechanics model to determine crack criterion.

[0113] Specifically, the three-dimensional stress tensor output from the previous structural mechanics model is used as the input criterion. For each discrete node or mesh element in the structure, its first principal stress value (i.e., the component with the largest stress along the principal axis) is extracted. If this value exceeds the preset critical fracture stress threshold, it is determined that there is a tendency for crack initiation at that location. Alternatively, if the shear stress on the structure exceeds the shear limit of the material, it should be determined that there is a possibility of shear crack formation. Alternatively, the system uses the J-integral method to construct a local path integration region for each suspected crack tip location, calculates the energy release rate of the structural crack tip, and if it is greater than the fracture toughness of the material, it is considered that the crack has a tendency to propagate. For spatial elements that meet any crack criterion, the following structural crack information is extracted: for each mesh element that meets the crack formation / propagation criteria, its coordinates, crack direction (principal stress direction), crack length (if there is extension in the simulation), and time frame are extracted.

[0114] Based on the crack data, the number density and length density are calculated to obtain the number density data and length density data.

[0115] Specifically, within each voxel cell, the number of crack entries located in that space from the crack data set is counted. The crack number density of that voxel region is then defined as the crack number density equal to the number of cracks in that region divided by the voxel volume. For the crack data within each voxel, the length of each crack is extracted and summed within that region. The crack length density of that voxel region is then defined as the crack length density equal to the sum of all crack lengths in that region divided by the voxel volume.

[0116] Spatial crack density map data is obtained by mapping spatial crack density map based on quantity density data;

[0117] Specifically, using the crack number density or length density values ​​at discrete voxel locations as input, spatial interpolation is performed on the entire structural region to generate a continuous density distribution field. Cross-sections are cut at preset heights, depths, or critical stress areas to display the crack density distribution on the corresponding cross-sections. Rendering layers of different colors or grayscale levels are generated based on density value ranges to highlight densely cracked areas or potential fatigue hotspots. The output spatial crack density map data is a continuous distribution mapping of crack density within the structural region, i.e., spatial crack density map data.

[0118] Local density mutation data are obtained by extracting local density mutation data from the spatial crack density map data.

[0119] Specifically, spatial gradient analysis is performed on the crack density map generated in the previous stage. In three-dimensional coordinate space, the first-order partial derivatives of the crack density function in the X, Y, and Z directions are calculated respectively. Based on the above first-order partial derivatives, a density gradient vector field is constructed. The density gradient value reflects the rate and direction of crack density change in the neighborhood of that point. The gradient calculation can be implemented using the finite difference method, the central difference method, or a convolution filter operator (such as the Sobel operator). After obtaining the density gradient field, local abrupt change regions are identified by setting a sudden change judgment threshold. Specifically, this includes setting a sudden change judgment threshold, for example, using the average value of the overall density change rate plus twice the standard deviation as the critical value. For any spatial point, when its density gradient magnitude exceeds the threshold, its corresponding region is determined to be a local density abrupt change region. Optionally, an edge enhancement algorithm (such as a high-pass filter based on the Laplacian operator) is used to enhance the density gradient change. The spatial contour boundaries of local mutation regions are defined based on Sobel, Canny, or region growing methods. The output local density mutation data is a set of mutation points or mutation regions, including the position of each mutation point or region center in three-dimensional structural coordinates; the corresponding density gradient magnitude value, i.e. density change rate intensity; the output results can be organized as a point set, region layer, or density mutation vector tensor, and uniformly represented as a local density mutation dataset.

[0120] The crack density data is obtained by integrating the number density data, length density data, and local density abrupt change data.

[0121] Specifically, for each grid cell, a density description vector is constructed to vectorize the number density data, length density data, and local density abrupt change data to obtain crack density data.

[0122] Optionally, the model annotation includes:

[0123] Crack direction features are extracted from crack data to obtain crack direction feature data;

[0124] Specifically, for each valid crack, the system subtracts the starting point coordinates from the crack endpoint coordinates to obtain the components of the three-dimensional direction vector. This direction vector is then normalized to its magnitude, converting it into a unit direction vector representing the crack propagation direction. Using methods such as K-means clustering, spherical Gaussian mixture model, or principal component analysis, the crack direction distribution within a local area is clustered and summarized to extract the dominant direction or direction distribution pattern. The crack direction vectors within each grid cell are averaged or modeled for distribution to generate a regional crack principal direction density map. The aforementioned data are then integrated to obtain crack direction feature data.

[0125] Anisotropic features are extracted based on fracture entropy field data and crack direction feature data to obtain anisotropic feature data.

[0126] Specifically, the system performs spatial gradient calculations on the fracture entropy field throughout the entire 3D structural mesh, obtaining the entropy gradient vector at each voxel unit. This vector consists of three directional components, representing the rate of change of entropy in the x, y, and z spatial dimensions, respectively, and overall indicating the direction of the most drastic change in fracture entropy. For each mesh unit with a valid crack direction vector, the cosine similarity between the crack's unit direction vector and the entropy gradient vector of the unit is calculated. When the cosine value is close to 1, it indicates that the crack direction is highly consistent with the entropy change direction, suggesting a clear dominant directionality and significant anisotropy in this region. When the cosine value is close to 0 or negative, it indicates that the crack direction is random or inconsistent with the entropy change direction, i.e., the structure tends to be isotropic or there is no dominant energy diffusion path. Within each structural mesh unit or local analysis window, the system summarizes all similarity values ​​between crack directions and entropy gradient directions and averages them to obtain the anisotropy index of the region, which is used to quantitatively represent the degree of directionality. The output anisotropic feature data includes anisotropic feature fields, which assign a scalar value at each spatial location (i.e., grid cell) to represent the directional intensity at that location; and a three-dimensional anisotropic map, organized in the form of a tensor or voxel map.

[0127] The fracture entropy field data, crack density data, and anisotropic characteristic data are used to annotate the structural mechanics model to obtain fatigue loss data.

[0128] Specifically, the system performs 3D mesh registration on fracture entropy field data, crack density data, and anisotropic characteristic data, ensuring they share a unified coordinate domain with the structural finite element mechanical model. If the structural model uses hexahedral elements, nearest neighbor interpolation or trilinear interpolation algorithms are performed based on the node coordinates. After registration, the system constructs a joint feature vector for fatigue characterization for each structural element or node. This vector consists of four components: fracture entropy value (reflecting the degree of disorder in the local evolution of the material), number density and length density (corresponding to the number of cracks and total length per unit volume, respectively), and anisotropic characteristic value (representing the degree of consistency between crack and energy dissipation direction). The system constructs a fatigue loss exponential function to map the multidimensional characterization vector into a single fatigue scoring index. The scoring function can be linearly weighted, assigning adjustable weight coefficients to various characteristic values ​​to form the following expression: the fatigue loss score equals the weighted sum of the fracture entropy value and the number density, length density, and anisotropic characteristics. For composite materials or nonlinear behavior scenarios, lightweight neural networks or regression models can be used alternatively to establish and optimize the mapping function through training. The system maps and assigns the calculated fatigue loss score to the corresponding unit position in the structural mechanics model, forming a three-dimensional fatigue loss field for the structure.

[0129] Optionally, S3 includes:

[0130] S31. Extract multi-time frame data based on fatigue loss data to obtain multi-time frame data;

[0131] Specifically, the system selects fatigue loss data of the structure under different loading cycle stages, and each specific fatigue loss data point is represented at a specific time point. The system measures the fatigue sensitivity of each element at different locations. The total simulation duration is set to T, divided into N equally spaced time frame sequences, for example, every 100,000 loading cycles as the time frame interval. For each time frame, the system records the distribution of fatigue loss in the three-dimensional space of the structure, forming a tensor field. The system combines the fatigue loss tensors from all time frames in a time-series format to form a complete multi-time-frame fatigue loss dataset.

[0132] S32. Calculate the crack evolution slope based on multi-time frame data to obtain crack evolution slope data;

[0133] Specifically, the system first selects index fields to describe the crack state, including crack length density and crack number density, which represent the crack state over time. The system calculates the volume-normalized total crack length and number of cracks at a given time and spatial location. Based on this, the system uses a temporal difference method to estimate the crack evolution rate across adjacent time frames. Taking crack length density as an example, the crack growth slope is defined as follows: ,in This represents the rate of increase in crack length density at that point per unit time. For the first +1 frame time Below, spatial point Crack length density, For the first +1 time frame corresponding to the loading time for Spatial coordinates for Spatial coordinates for Spatial coordinates For the first Frame time Below, spatial point Crack length density, For the first The loading time corresponding to each time frame The system assigns a time frame index number. The system performs spatial mean filtering on the obtained slope field. The system calculates the standard deviation of the slope variation. The output is a crack evolution slope map.

[0134] S33. Extract the entropy growth rate from the fracture entropy field data corresponding to the multi-time frame data to obtain the entropy growth rate data;

[0135] Specifically, the system uses fracture entropy field data corresponding to multiple loading time frames as input, including loading cycle times and spatial coordinates. For two consecutive time frames, the system calculates the global and local entropy growth rates respectively. On the global scale, the system integrates the difference in the entropy field across the entire spatial domain and divides it by the time interval to obtain the average entropy growth rate index; the global entropy growth rate... ,in For the global entropy growth rate, For the first +1 time frame position The fracture entropy value, In three-dimensional space Spatial coordinates In three-dimensional space Spatial coordinates In three-dimensional space Spatial coordinates For the first Location in each time frame The fracture entropy value, For the first +1 time frame corresponding to the loading time For the first Each time frame corresponds to a loading moment. At a local scale, the system calculates the rate of entropy change per unit time point by point, forming a local entropy acceleration field. ,in The rate of increase of local entropy indicates the position. The local unit-time entropy growth is calculated, and other parameters can be referenced from previous data. The system calculates the temporal variance of the entropy growth rate at each point in multiple time frames to obtain a local entropy growth fluctuation index. Integrating the above, the entropy growth rate data is obtained.

[0136] S34. Based on the multi-time frame data, the hot spot evolution in the risk area is analyzed to obtain the hot spot data of the risk area;

[0137] Specifically, the system identifies regions where fatigue loss values ​​are significantly higher than the overall average level at each time frame. A threshold for the fatigue loss field is set (e.g., the 90th percentile is selected as the global quantile threshold). The set of spatial locations where the local fatigue loss exceeds the threshold is defined as the hot spot region, i.e., the extraction time point. Corresponding hot spot set: , For time frames Hot spot clusters / hot spot regions These are three-dimensional spatial coordinates, representing the position of a voxel, node, or discrete element in the structural model. For time frames The fatigue loss value, The threshold value for the fatigue loss field represents the stress concentration region or potential crack accumulation region of the structure under the current fatigue loading state. The system performs boundary extraction operations on the spatial contour of each hot spot region. Edge detection algorithms (such as 3D Sobel filtering) or voxel envelope construction (such as...) are employed. The system employs shape-based methods and convex hull algorithms to extract the external geometric boundaries of the hotspot. It tracks the position of the geometric center (centroid) of the hotspot region in each time frame, recording the sequence of changes in this center across all time frames to form a continuous hotspot trajectory line. After acquiring the trajectory, the system performs dynamic characteristic analysis on the hotspot motion based on the trajectory line, including calculating the velocity vector (displacement per unit time), acceleration vector (rate of change of velocity), and trajectory perturbation amplitude (measuring the extent of deviation from the main trend line) of the hotspot's geometric center. The system outputs a hotspot dynamic evolution dataset including the hotspot's spatial extent, boundary contour information, the hotspot's geometric center trajectory line, and corresponding dynamic characteristic indices for each time frame.

[0138] S35. Based on the crack evolution slope data, entropy growth rate data, and hot spot data in the risk area, a reduced-order modeling spatial expansion simulation is performed to obtain loss simulation data.

[0139] Specifically, the system constructs a state feature vector for each structural mesh element. This vector includes the crack evolution slope (representing the local crack growth rate), the fracture entropy growth rate (reflecting the intensity of material energy state changes), and a hotspot path association identifier (used to indicate whether the element is within the hotspot trajectory range, taking a Boolean value or a normalized value of the distance function). Specifically, the state vector of a mesh element can be represented as: ,in This indicates the crack evolution slope of the unit. This represents the local growth rate of fracture entropy. This indicates whether the unit falls within the vicinity of the hotspot's dynamic evolution path, serving as a dynamic marker of structural evolution. The system performs reduced-order modeling on the aforementioned state vector data. This process can be implemented in various ways, including principal component analysis (PCA) for linear feature compression, autoencoders for nonlinear representation learning, or multi-channel attention mechanisms based on graph structures to enhance the structural awareness between features. The reduced-order model outputs compressed low-dimensional latent variables. During the modeling phase, the system uses time-series neural networks (such as long short-term memory networks) or graph neural networks (such as temporal graph convolutional networks) to dynamically learn the aforementioned reduced-order features and simulate the state changes of the structure over several future time frames. If combined with physical mechanism constraints, an energy release rate threshold can be set to treat cracks exceeding the threshold as continuing to propagate, thus embedding them into subsequent processing modules to achieve an evolution simulation process that combines physical and data-driven approaches. The system remaps the latent loss representation data corresponding to the predicted time frames to the structural space grid, restoring it to a three-dimensional loss field form. This mapping process supports trilinear interpolation, nearest-neighbor interpolation, or spatial reconstruction of high-dimensional prediction results through a trained back-projection decoder. The output is the simulation data of structural fatigue loss at the predicted time point.

[0140] Optionally, the entropy growth rate extraction includes:

[0141] The global entropy growth rate and the neighborhood entropy growth rate are calculated based on the fracture entropy field data corresponding to the multi-time frame data, and the global entropy growth rate data and the neighborhood entropy growth rate data are distributed.

[0142] Specifically, the system input is a series of fracture entropy field data sequences arranged in chronological order, denoted as... ,in The structure represents the first Spatial location in each time frame The dataset represents the local fracture entropy values ​​at a given location. It reflects the changes in the local microscopic energy states of the structure during loading cycles. At a global scale, the system calculates the voxel mean of the entropy field for each frame. This allows us to estimate the rate of increase of the average entropy over time. The specific calculation formula is as follows: ,in This represents the growth rate of global entropy. This represents the total number of time frames. The current time frame number, For the first The voxel mean of the frame fracture entropy field across the entire spatial domain. For the first Spatial mean of the frame fracture entropy field In the first Spatial location within +1 time frame The local fracture entropy value at the location, In the first Spatial location within a time frame The local fracture entropy value at the location, This represents the time interval between adjacent time frames. At a local scale, the system targets each grid point. We extract the local entropy increment between adjacent time frames and calculate the neighborhood entropy growth rate accordingly. The definition of the local entropy growth rate is as follows: , This represents the rate of increase in local entropy. In the first Spatial location within +1 time frame The local fracture entropy value at the location, In the first Spatial location within a time frame The local fracture entropy value at the location, The time interval between adjacent frames. The calculation of the neighborhood entropy growth rate also includes calculating the local entropy increment of the adjacent space to obtain the local spatial entropy increment of the current frame; the neighborhood entropy growth rate is calculated based on the local spatial entropy increment between adjacent time frames to obtain the neighborhood entropy growth rate data. The output results include two parts: (1) the global entropy growth rate value, which is a single scalar index; (2) the neighborhood entropy growth rate map, which is three-dimensional spatial distribution data, representing the spatial distribution map of the growth trend of the unit or neighborhood difference energy state at each position inside the structure.

[0143] The entropy increase time-varying variance is calculated based on the global entropy growth rate data and the neighborhood entropy growth rate data to obtain the entropy increase time-varying variance data;

[0144] Specifically, the system is based on the generated neighborhood entropy growth rate map sequence. A sliding time window structure is constructed to extract local temporal fluctuation features. The time window size can be set to ±w time frames (e.g., w=3), then at each sliding moment... The corresponding window is This represents the local entropy acceleration data over 20,000+1 time frames. The system analyzes the local entropy acceleration data for each spatial location. Perform local time series variance statistics and define the entropy increase time-varying variance index for that position within the window. Specifically, set the entropy growth rate sequence for that point within the current window as... Then its time-varying variance is defined as: ,in Indicates spatial location The variance of the entropy growth rate within the current window. The system may optionally employ an exponentially weighted moving variance approach, assigning higher weights to time points closer to the current frame. The output is an entropy-increment time-varying variance plot, which identifies the strength of energy state fluctuations at each location in three-dimensional space.

[0145] Local entropy flow graphs are constructed based on entropy-increasing time-varying variance data to obtain local entropy flow graph data;

[0146] Specifically, based on the generated time-varying entropy variance map, the system extracts the spatial gradient of the entropy variance at each location in three-dimensional space. Specifically, for each spatial grid point, its gradient vector consists of partial derivatives in three directions, i.e., the rate of change of entropy variance in the x, y, and z directions is calculated respectively, thus obtaining a local gradient vector field. This gradient vector can be regarded as a local directional index characterizing the trend of entropy disturbance diffusion, i.e., the entropy flow direction. The system constructs a local streamline structure based on the above gradient vector field. At each initial point, using the current gradient direction as the initial velocity, the next location point is calculated through a fourth-order integral process, and the direction vector is updated until a preset termination condition is reached (such as reaching a boundary, the magnitude of the directional gradient approaching zero, or the trajectory length exceeding a threshold). This path is the local entropy flow path reflecting the trend of entropy disturbance propagation. As an alternative, the system can also construct a graph structure based on grid nodes: each three-dimensional grid point is used as a node in the graph, and directed edges are established between neighboring nodes. The direction of the edges is indicated by the gradient vector, and the weight of the edges is set to the gradient magnitude or its normalized value. The system forms a complete local entropy flow graph structure, denoted as ,in This represents a set of nodes, corresponding to discrete grid points in space. Let represent the set of directed edges, and let represent the direction of propagation of local entropy perturbation; Let represent the set of edge weights.

[0147] Entropy growth rate data is obtained by performing cross-resolution entropy fusion based on local entropy flow graph data.

[0148] Specifically, the system constructs a multi-resolution representation of the entropy increase time-varying variance map obtained in the previous step. A Gaussian pyramid construction method is used to downsample the entropy increase variance map at multiple scales, generating a series of entropy variance layer-level sequences at different spatial scales. Each layer represents the observation results of local energy state fluctuations at different spatial resolutions. The system introduces a structural guiding factor as a weighted reference in the fusion process, forming a spatially adaptive entropy variance fusion strategy. This structural guiding factor can be an indicator closely related to the structural stress state, such as a local shear stress gradient map, a historical stress fluctuation map, or a weighted map based on the node centrality of the graph structure. The system normalizes the above candidate indicators and uses a weighted combination strategy to form a spatial guiding weight map. Weighted summation is performed on all scale layers at corresponding points to generate the fused entropy increase rate map. The system maps the fused entropy increase rate map values ​​to the graph nodes of the local entropy flow graph structure as the updated attribute information for each node. The output is a fusion result map, representing the fused entropy increase rate data distributed in three-dimensional space.

[0149] Optionally, the evolution of hotspots in the risk area includes:

[0150] Fatigue concentration areas are selected based on multi-time frame data to obtain hot spot data;

[0151] Specifically, the system input is a sequence of fatigue loss images across multiple time frames, where each frame represents the spatial distribution of fatigue intensity of the structure under a specific loading cycle. Each image corresponds to a two-dimensional spatial field, representing the local fatigue accumulation at various locations on the structure surface, denoted as the fatigue loss image sequence. The system sets a fatigue intensity threshold for each time frame. This threshold can be adaptively set in two ways: first, by using the percentile of the fatigue intensity value across the entire image, for example, the upper 15th percentile; second, by using the mean fatigue intensity of each frame plus twice the standard deviation as a dynamic threshold. Based on the threshold set above, the system marks pixels in each frame that satisfy a fatigue intensity greater than or equal to the threshold. The system uses an 8-neighborhood connectivity criterion to merge adjacent high-fatigue-intensity pixels into a hotspot region, and performs numbering and boundary filling processing to extract continuous fatigue concentration regions with certain area and shape characteristics. The system outputs a hotspot mask image for each frame, as a binary image with the same size as the original fatigue image, where pixels in the hotspot region have a value of 1, and the rest have a value of 0.

[0152] Specifically, further, hot spot regions are selected based on multi-time frame data using preset filtering parameters to obtain preliminary hot spot data; elastic hot spot migration is performed on the preliminary hot spot data to obtain elastic hot spot data; and multi-frame collaborative potential domain processing is performed on the elastic hot spot data to obtain hot spot data.

[0153] The system acquires stress response data (such as equivalent stress, fracture entropy, crack density, etc.) of each mesh element in the structural model across consecutive time frames, and constructs the mechanical response tensor of the 3D structure over time. For each time frame, the system sets the following hotspot screening parameters: when the equivalent stress value of a mesh element is higher than 1.2 times the average stress of the entire structure, it is identified as an abnormal region; when the fracture entropy value of a mesh element is higher than 1.5 times the average fracture entropy of the structure, it is considered a region with active local failure evolution; when the crack density exceeds 0.8, it is considered a crack aggregation region. Mesh elements that meet any of the conditions are identified as potential hotspots, and a preliminary set of hotspot regions is output.

[0154] For each initial hotspot unit, its spatial neighborhood is constructed. The neighborhood can be set as a local cubic region with a radius of 1 to 2 grid units centered on the unit. The frequency of hotspot units appearing in the unit and its neighborhood is counted within several frames before and after (e.g., k frames before and after) to form the temporal active frequency index of the unit. If a unit does not meet the hotspot judgment criteria in the current time frame, but hotspot active signals frequently appear in its neighborhood, the system will compensate by including it in the hotspot region. If the hotspot state of a unit only appears in less than 3 time frames, it is considered to be caused by occasional disturbances and will be removed by the system and not included in the hotspot judgment.

[0155] A hotspot intensity map is constructed. All hotspot cells within the 3D space of the structure are assigned a value of 1, and non-hotspot cells are assigned a value of 0, forming a hotspot state tensor for each time frame. For each time frame, the system performs spatial weighting on the hotspot region based on a Gaussian kernel function to calculate the local hotspot density distribution function. The system then weights and accumulates the hotspot density data from all time frames to form a cooperative potential function. The time weighting factor is obtained using an exponential decay method, assigning higher weights to data closer to the current frame. The weight of a specific time frame can be set to... ,in Indicates the current time frame number. Indicates the historical time frame number. To control the decay rate, the system performs local maximum detection and threshold filtering on the cooperative potential function. Only spatial regions meeting the following two conditions are retained: they are local extrema in the cooperative potential function, and their potential value exceeds a preset threshold. These regions are then classified as the hotspot region set.

[0156] Contour extraction is performed on the hot spot data to obtain hot spot contour data;

[0157] Specifically, the system performs edge extraction on the hotspot mask map corresponding to each time frame. Edge detection algorithms (such as the Canny operator) are preferably used to identify the boundary lines between the hotspot region and the background. Contour extraction methods (such as the findContours function in OpenCV) are called to analyze the edge map and extract the set of boundary points for closed contours. Given that multiple spatially discontinuous hotspot regions exist in the actual structure, the system extracts the contours for each independently connected hotspot region and establishes an independent set of contour points for each region, forming a contour set. For contours with a large number of boundary points, the system takes the first and last two points from the contour point set as the initial line segment. The vertical distance from the remaining points to this line segment is calculated, and the point with the largest distance is found. If the maximum distance is greater than a preset threshold, this point is retained, and the current point set is divided into two segments, which are recursively compressed. If the maximum distance is less than or equal to the threshold, the intermediate points are considered redundant and can be approximated as straight lines; the intermediate points are removed, and only the first and last points are retained. All retained points will form a more concise sequence of boundary points than the original contour, achieving feature simplification of the original boundary point set. The system outputs a hotspot contour dataset for each frame. This dataset contains multiple contours, each consisting of an ordered set of boundary point pairs.

[0158] The centroid trajectory is extracted based on the hot spot contour data to obtain the hot spot centroid data;

[0159] Specifically, the system calculates the centroid of the extracted hotspot contours in each frame. For any hotspot contour, its boundary consists of several pixels. The system averages the x and y coordinates of all boundary points to obtain the geometric center position of the hotspot contour in that frame, denoted as the centroid coordinates of the hotspot. This centroid can be represented as a pair of numerical values, namely the center coordinates in the horizontal and vertical directions. To track the evolution trajectory of the same hotspot in different time frames, the system employs a cross-frame matching strategy. The system preferably uses the maximum intersection-over-union (IoU) matching method or the minimum Euclidean distance matching method to match the centroid positions between the current frame and the next frame, thereby establishing the hotspot identity association between consecutive frames. Throughout the entire time series, the system establishes a complete centroid trajectory sequence for each initial hotspot. This trajectory consists of centroid coordinates arranged in chronological order. When numerical values ​​are present, the system can choose to use linear interpolation to fill in missing positions in the centroid trajectory or mark it as a "trajectory interruption" state.

[0160] Dynamic evolution was performed based on hotspot centroid data to obtain hotspot data for the risk area.

[0161] Specifically, the system calculates the derivative of each hotspot trajectory to extract its motion characteristics. The system performs differential calculations on the changes in centroid coordinates between adjacent time frames to calculate the velocity vector at each frame. Based on the trend of the velocity vector change, the system calculates the acceleration vector. Based on the above velocity and acceleration data, the system constructs a dynamic evolution model of the hotspot region. If a hotspot region exhibits a significant velocity increase trend in consecutive time frames, or its centroid continuously accelerates in a certain direction, then the region is in a state of intensified fatigue evolution. The system sets thresholds for velocity or acceleration, or directly performs clustering calculations for classification; for example, when the hotspot velocity exceeds a preset upper limit, a risk marker is triggered for that region. The system classifies all hotspot regions into risk levels. Classification indicators include, but are not limited to, total trajectory length, average or maximum velocity, rate of change of acceleration, and historical entropy growth rate. The system can perform weighted calculations on hotspot regions based on the above multidimensional indicators (weights generated from empirical data for regression calculations), and classify them into high-risk, medium-risk, or low-risk regions accordingly, i.e., risk assessment information. The system outputs a hotspot data structure containing risk assessment information. Each risk hotspot object includes its geometric center coordinates, complete trajectory sequence, corresponding risk level, and other relevant attributes.

[0162] Optionally, this application also provides an artificial intelligence-based long-life pavement structure optimization system for executing the artificial intelligence-based long-life pavement structure optimization method described above, wherein the artificial intelligence-based long-life pavement structure optimization system includes:

[0163] The structural mechanics model building module is used to acquire pavement structure data and build a structural mechanics model based on the pavement structure data to obtain the structural mechanics model.

[0164] The fatigue loss simulation module is used to simulate fatigue loss based on the structural mechanics model and obtain fatigue loss data.

[0165] The rapid physical simulation module is used to perform rapid physical simulations based on fatigue loss data to obtain loss simulation data.

[0166] The physical prior lifetime prediction module is used to predict physical prior lifetime based on loss simulation data, and obtain lifetime prediction data.

[0167] Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended application documents rather than the foregoing description. Thus, it is intended that all variations falling within the meaning and scope of the equivalents of the application documents be incorporated into the invention.

[0168] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for optimizing long-life pavement structure based on artificial intelligence, characterized in that, The method comprises: S1, acquiring pavement structure data, constructing a structural mechanics model according to the pavement structure data, and obtaining the structural mechanics model; S2, simulating fatigue loss according to the structural mechanics model, and obtaining fatigue loss data; S3, performing rapid physical simulation according to the fatigue loss data, and obtaining loss simulation data; S4, performing physical a priori life prediction according to the loss simulation data, and obtaining life prediction data; S3 comprises: extracting multi-time frame data according to the fatigue loss data, and obtaining the multi-time frame data; According to the multi-time frame data, crack evolution slope calculation is performed to obtain crack evolution slope data. The system first selects an index field for describing the crack state, including crack length density and crack number density, which respectively represent the volume normalized values of crack total length and crack number at time moment and spatial position. On this basis, the system estimates the crack evolution speed of adjacent time frames in a time difference manner. The crack growth slope is defined as follows: wherein is the growth rate of crack length density at the point per unit time, is the crack length density of the spatial point at time frame, is the crack length density of the spatial point at time frame, is the loading time corresponding to the +1th time frame, is the spatial coordinate, is the spatial coordinate, is the spatial coordinate, is the spatial coordinate, is the crack length density of the spatial point at time frame, is the crack length density of the spatial point at time frame, is the time frame index number. The system performs spatial mean filtering processing on the obtained slope field. The system calculates the standard deviation of the slope change, and the output result is a crack evolution slope map. extracting entropy increase rate data from the fracture entropy field data corresponding to the multi-time frame data, wherein the system takes the fracture entropy field data corresponding to multiple loading time frames as input, including loading cycle time, spatial coordinates, for two continuous time frames, the system calculates the global and local entropy increase rate respectively, on the global scale, the system integrates the difference of the entropy field in the entire spatial domain and divides by the time interval to obtain the average entropy increase rate index, on the local scale, the system calculates the entropy value change rate per unit time point by point to form a local entropy increase rate field, the system calculates the time variance of the entropy increase rate of each point in the multi-time frame to obtain the local entropy growth fluctuation index; evolving a risk region hot spot according to the multi-time frame data, and obtaining risk region hot spot data, the system identifies a region where the fatigue loss value is significantly higher than the overall average level in each time frame, sets a threshold value of the fatigue loss field, defines a set of spatial position points that satisfy the local fatigue loss greater than the threshold value as a hot spot region, the system extracts the boundary of the spatial contour of each hot spot region, the system tracks the geometric center position of the hot spot region in each time frame, the system records the change sequence of the center in all time frames to form a continuous hot spot trajectory line, after obtaining the trajectory, the system analyzes the dynamics characteristics of the hot spot movement based on the trajectory line, including calculating the velocity vector, acceleration vector of the hot spot geometric center, and the trajectory disturbance amplitude index, the system outputs the hot spot dynamic evolution data set including the hot spot spatial range, boundary contour information, hot spot geometric center trajectory line of each time frame, and the corresponding dynamics characteristic index; performing reduced-order modeling space expansion simulation according to the crack evolution slope data, the entropy increase rate data, and the risk region hot spot data, and obtaining the loss simulation data, the system constructs a state feature vector for each structural grid element, the vector includes the crack evolution slope value, the fracture entropy increase rate, and the hot spot path association identifier, the system processes the above state vector data by reduced-order modeling, the system dynamically learns the reduced-order features by using a time series neural network or a graph neural network, and simulates the state change of the structure in the future time frames, the system maps the potential loss representation data corresponding to the prediction time frame to the structural spatial grid again to restore it to a three-dimensional loss field form.

2. The method of claim 1, wherein, S1 comprises: acquiring pavement structure data, dividing the structure into layers according to the pavement structure data, and obtaining structure layer data, wherein the structure layer data includes surface layer data, middle surface layer data, base layer data, and soil base data; setting interlayer conditions for the structure layer data, and obtaining an interlayer model; Physical condition setting is performed on the interlayer model to obtain a structure physical model; Material performance degradation condition setting is performed on the structure physical model to obtain a structure mechanics model.

3. The method of claim 1, wherein S2 It comprises: Fracture entropy field generation is performed according to the structure mechanics model to obtain fracture entropy field data; Crack density calculation is performed according to the structure mechanics model to obtain crack density data; Model labeling is performed on the structure mechanics model according to the fracture entropy field data and the crack density data to obtain fatigue loss data.

4. The method of claim 3, wherein, The fracture entropy field generation comprises: Local scanning is performed on stress distribution of structure layer data under external load disturbance according to the structure mechanics model to obtain local scanning data; Equivalent stress fluctuation amplitude calculation, shear stress gradient calculation and bonding interface instability index calculation are performed according to the local scanning data to obtain equivalent stress fluctuation amplitude data, shear stress gradient data and bonding interface instability data respectively; Discrete grid division is performed according to the equivalent stress fluctuation amplitude data, the shear stress gradient data and the bonding interface instability data to obtain grid microstate data; Local fracture entropy calculation is performed according to the grid microstate data to obtain fracture entropy field data.

5. The method of claim 3, wherein, The crack density calculation comprises: Crack criterion is performed according to the structure mechanics model to obtain crack data; Number density calculation and length density calculation are performed according to the crack data to obtain number density data and length density data; Spatial crack density map mapping is performed according to the number density data to obtain spatial crack density map data; Local density mutation extraction is performed according to the spatial crack density map data to obtain local density mutation data; The number density data, the length density data and the local density mutation data are integrated to obtain crack density data.

6. The method of claim 5, wherein, The model labeling comprises: Crack direction feature extraction is performed according to the crack data to obtain crack direction feature data; Anisotropy feature extraction is performed according to the fracture entropy field data and the crack direction feature data to obtain anisotropy feature data; The fracture entropy field data, the crack density data and the anisotropy feature data are labeled on the structure mechanics model to obtain fatigue loss data.

7. An artificial intelligence-based long-life pavement structure optimization system, characterized by, The AI-based long-life pavement structure optimization system for performing the AI-based long-life pavement structure optimization method of claim 1 comprises: A structure mechanics model construction module for obtaining pavement structure data, constructing a structure mechanics model according to the pavement structure data, and obtaining a structure mechanics model; A fatigue loss simulation module for simulating fatigue loss according to the structure mechanics model and obtaining fatigue loss data; A rapid physical simulation module for performing rapid physical simulation according to the fatigue loss data and obtaining loss simulation data; A physical prior life prediction module for predicting a physical prior life according to the loss simulation data and obtaining life prediction data.

Images