Road hidden damage inversion and evolution prediction method based on physical constraint network
By introducing a physical constraint network and using the heat conduction equation and generative adversarial network to reconstruct the surface benchmark, the shortcomings of existing technologies in road damage depth inversion and evolution prediction are solved, achieving non-destructive detection and accurate prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JILIN JIANZHU UNIVERSITY
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-19
AI Technical Summary
Existing technologies cannot effectively address specific problems in the three-dimensional automated detection and prediction of road damage.
By using a physical constraint network based on the physical laws of heat conduction to invert and predict the hidden road damage, embedding the Fourier heat conduction equation into the neural network loss function to invert the depth, introducing a generative adversarial network to reconstruct the surface benchmark, and establishing microscopic dynamic evolution rules of water-temperature coupling, a leap from static appearance recognition to dynamic three-dimensional mechanism quantification is achieved.
It enables non-destructive detection of hidden road damage, eliminates volumetric calculation errors under complex road alignments, and achieves a leap from macroscopic statistical prediction to microscopic mechanism deduction, providing accurate damage prediction and early warning capabilities.
Smart Images
Figure CN122244359A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of digital twin and artificial intelligence technology for transportation infrastructure, and in particular relates to a method for inverting and predicting the evolution of hidden road damage based on physical constraint networks. Background Technology
[0002] Currently, the technology for automated 3D detection and prediction of road damage mainly relies on surface geometric data acquired through photogrammetry or laser scanning, combined with deep learning algorithms for identification, and supplemented by time series models for trend prediction. Specifically, existing technologies typically include the following steps: collecting point cloud or image data of the road surface through drones or vehicle-mounted platforms to construct a 3D model; using deep learning models such as convolutional neural networks to perform semantic segmentation and identification of apparent damage such as cracks and potholes in the 3D model; for identified potholes, using a random sampling consensus algorithm to fit a plane in the neighborhood of the damage boundary as a reference surface, and calculating the pothole volume by comparing the measured surface with the reference surface; and for the evolution trend of damage, constructing an autoregressive integral moving average model or recurrent neural network based on historical detection data to perform macroscopic statistical prediction of the future degree of damage.
[0003] However, the aforementioned existing technologies have the following significant structural and methodological shortcomings in the three-dimensional automated analysis and prediction of road damage: The physical inversion of subsurface damage depth is lacking. Existing 3D analysis methods mostly remain within the geometrically visible range of the surface. For hidden damage such as frost heave cracks with small openings but large interiors or voids in the subsurface, traditional data-driven convolutional neural networks can only make empirical estimations based on geometric features such as the width of the two-dimensional opening on the surface. Due to the lack of mathematical modeling of the thermodynamic conduction processes of pavement materials, the inherent physical correlation between surface temperature field changes and subsurface medium properties cannot be utilized, resulting in the inability to accurately invert the true physical penetration depth of cracks and the distribution of internal voids.
[0004] The absolute reference surface for volume calculation fails under complex road alignments. Calculating pothole volume requires a reference surface based on an intact road surface. Current technologies generally employ random sampling consensus algorithms to fit an absolute plane as a reference surface within a local area surrounding the pothole. However, when the road has a design camber, superelevation, or is located on a curve with varying slope, the road surface itself is a complex spatial surface. Forcibly using the assumption of plane fitting will result in a systematic deviation from the actual surface, leading to significant errors in the calculation of the depression volume and failing to meet the requirements for precise earthwork calculation in repair projects.
[0005] Evolutionary predictions lack microscopic physical driving mechanisms. Current research on the evolution of damage with freeze-thaw cycles largely relies on historical data regression or time series analysis methods based on macroscopic statistics, such as the ARIMA model. These models treat roads as homogeneous bodies, neglecting the microscopic core variables that lead to frost heave damage, such as local differences in material moisture content. Two cracks with identical surface morphology can exhibit drastically different frost heave cracking rates under the same sudden temperature drop due to their different internal moisture contents. Existing "black box" prediction models cannot incorporate such physical and mechanical essences into their prediction mechanisms, resulting in insufficient predictive ability for the evolutionary behavior of individual diseases under specific meteorological conditions.
[0006] To address the aforementioned problems in existing technologies, there is an urgent need to propose a method for road hidden damage inversion and evolution prediction based on physical constraint networks. Summary of the Invention
[0007] To address the aforementioned technical problems, this invention provides a method for road hidden damage inversion and evolution prediction based on physical constraint networks. By embedding the Fourier heat conduction equation into the inversion depth of the neural network loss function, introducing a generative adversarial network to reconstruct the surface benchmark, and establishing water-temperature coupled micro-dynamic evolution rules, a leap from static appearance recognition to dynamic three-dimensional mechanism quantification is achieved.
[0008] This invention provides a method for road hidden damage inversion and evolution prediction based on physical constraint networks, comprising the following steps: Based on the acquired multimodal 3D real-world model of the road, a 3D tensor field containing geometric, spectral, and thermal infrared temporal features is constructed, and the damaged areas are identified. Based on the temporal thermal infrared characteristics of the damaged area and a physical information neural network embedded with the physical laws of thermal conduction, the subsurface physical morphology parameters of the damaged area are obtained by inversion. Based on the identified damage area type, for pit-type damage, a virtual intact surface is reconstructed using a generative adversarial network based on the geometric features of the surrounding intact area as the calculation reference surface, and the pit volume is calculated by combining the measured pit surface. Based on the subsurface physical morphology parameters of the damaged area obtained by inversion and the material properties extracted from spectral features, combined with external meteorological driving data, the evolution of the pit volume in time and space is simulated by cellular automata to obtain the predicted morphology of the damage at future moments.
[0009] Optionally, based on the acquired multimodal 3D real-world model of the road, a 3D tensor field containing geometric, spectral, and thermal infrared temporal features is constructed, and damaged areas are identified, including: The multimodal 3D real-world model of the road is discretized into a voxel mesh. For each voxel, a feature tensor containing geometric symbolic distance field values, multispectral reflectance vectors, and dual-temporal surface temperature vectors is constructed. The voxel field is then processed using a 3D semantic segmentation network to output a 3D mask with different damage category labels.
[0010] Optionally, the subsurface physical morphological parameters of the damaged area are obtained by inversion, including: A deep neural network is constructed with surface coordinates and time as inputs and the distribution of subsurface thermal diffusivity coefficient and damage depth as outputs. The partial differential equation of thermal conduction of pavement material is embedded as a physical constraint into the loss function of the deep neural network. By minimizing the sum of the data fitting residual and the physical differential constraint residual, the output of the deep neural network is trained to conform to the physical laws of thermal conduction and the true three-dimensional depth of the crack.
[0011] Optionally, a virtual intact surface is reconstructed using a generative adversarial network as the computational reference surface, and combined with the measured pit surface, the pit volume is calculated, including: The generator of the adversarial network is trained using a preset number of intact road models; the identified pothole areas are used as masks to be hollowed out, and the geometric mesh of the surrounding intact areas is input into the trained generator to generate a virtual intact surface that is consistent with the original road surface design; Boolean difference operations are performed on the generated virtual intact surface and the measured pothole surface to calculate the envelope volume between the two, which is taken as the pothole volume.
[0012] Optionally, the evolution of the pit volume in time and space can be simulated using cellular automata to obtain the predicted damage morphology at future time points, including: Each voxel in the three-dimensional voxel model is defined as a cell and assigned attributes including damage state and local water content inverted from spectral features. A local evolution rule is established driven by freeze-thaw cycle parameters in meteorological data. The local evolution rule determines the state transition probability under future meteorological conditions based on the damage state of adjacent cells and the water content threshold of the cell itself. Input time series meteorological forecast data, drive the cellular automaton to perform iterative calculations, and output the three-dimensional road damage prediction morphology at a specified future time point.
[0013] Optionally, the freeze-thaw cycle parameters include daily temperature difference and cooling rate; The local evolution rule is configured as follows: when the temperature in the meteorological data is below zero and the cooling rate exceeds a preset threshold, if the adjacent cells are in a damaged state and the water content of the current cell exceeds a preset water-induced swelling threshold, the probability of the current cell transforming into a damaged state in the next iteration increases.
[0014] This invention also proposes a road hidden damage inversion and evolution prediction system based on a physical constraint network, used to implement the method, including: The data acquisition and preprocessing module is used to acquire a multimodal 3D real-world model of the road and construct a 3D tensor field containing geometric, spectral, and thermal infrared temporal features to identify the damaged area. The physical information inversion module is used to invert the subsurface physical morphology parameters of the damaged area based on the surface temporal thermal infrared characteristics of the damaged area and by using a physical information neural network embedded with the physical laws of thermal conduction. The reference surface reconstruction and volume calculation module is used to reconstruct a virtual intact surface as the calculation reference surface based on the identified damage area type. For pit-type damage, it uses the geometric features of the surrounding intact area to reconstruct the virtual intact surface as the calculation reference surface, and combines it with the measured pit surface to calculate the pit volume. The damage evolution prediction module is used to simulate the evolution of the pit volume in time and space using cellular automata based on the subsurface physical morphology parameters and material properties extracted from spectral features, combined with external meteorological driving data, to obtain the predicted damage morphology at future moments.
[0015] The present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.
[0016] The present invention also proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.
[0017] The present invention also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.
[0018] Compared with the prior art, the present invention has the following advantages and technical effects: This invention achieves non-destructive detection of hidden damage by constructing a three-dimensional tensor field containing geometric, spectral, and thermal infrared temporal features, and by using a physical information neural network embedded with the physical laws of thermal conduction to invert the physical morphological parameters of the subsurface. Existing technologies can only make empirical estimations based on the width of the crack opening on the surface, and cannot obtain the true depth of frost heave cracks with a small opening but a large interior, or the true depth of voids beneath the road surface. This invention embeds the partial differential equation of thermal conduction as a constraint into the neural network, forcing the network to simultaneously satisfy the requirements of data fitting and physical conservation. Therefore, without destructive drilling, the true three-dimensional penetration depth of the crack can be inverted using the surface thermal infrared response.
[0019] This invention eliminates systematic errors in volume calculations for complex alignments by using a generative adversarial network (GAN) to reconstruct a virtual intact surface as a computational reference for pothole-type damage. Existing technologies employ plane fitting methods as a reference, which deviate significantly from the actual surface when road camber, superelevation, or curve slopes are present. This invention utilizes a GAN to learn the spatial distribution manifold of intact road surfaces, adaptively generating a virtual surface that fits the original design based on the geometric trends surrounding potholes, and achieving unbiased volume calculations for arbitrary alignments through Boolean difference operations.
[0020] This invention combines inverted physical morphological parameters with spectral inversion of material properties, and utilizes cellular automata to simulate damage evolution under external meteorological conditions, achieving a leap from macroscopic statistical prediction to microscopic mechanism deduction. Existing time series models cannot distinguish the differentiated evolution rates of cracks with the same appearance but different water contents under freeze-thaw conditions. This invention inverts local water content from spectral characteristics as a cellular intrinsic property, constructs water-temperature coupled microscopic dynamic rules, and achieves precise deduction of the spatial expansion direction and velocity of damage. Attached Figure Description
[0021] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation
[0022] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.
[0023] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.
[0024] Example 1 like Figure 1 As shown, this embodiment provides a method for road hidden damage inversion and evolution prediction based on physical constraint networks, including the following steps: Based on the acquired multimodal 3D real-world model of the road, a 3D tensor field containing geometric, spectral, and thermal infrared temporal features is constructed, and the damaged areas are identified. Based on the temporal thermal infrared characteristics of the damaged area and a physical information neural network embedded with the physical laws of thermal conduction, the subsurface physical morphology parameters of the damaged area are obtained by inversion. Based on the identified damage area type, for pit-type damage, a virtual intact surface is reconstructed using a generative adversarial network based on the geometric features of the surrounding intact area as the calculation reference surface, and the pit volume is calculated by combining the measured pit surface. Based on the subsurface physical morphology parameters of the damaged area obtained by inversion and the material properties extracted from spectral features, combined with external meteorological driving data, the evolution of the pit volume in time and space is simulated by cellular automata to obtain the predicted morphology of the damage at future moments.
[0025] Feasible method: Based on the acquired multimodal 3D real-world model of the road, construct a 3D tensor field containing geometric, spectral, and thermal infrared temporal features, and identify the damaged areas, including: The multimodal 3D real-world model of the road is discretized into a voxel mesh. For each voxel, a feature tensor containing geometric symbolic distance field values, multispectral reflectance vectors, and dual-temporal surface temperature vectors is constructed. The voxel field is then processed using a 3D semantic segmentation network to output a 3D mask with different damage category labels.
[0026] As a specific implementation method, the input terminal of this embodiment is a high-precision "3D real scene model of road with multi-dimensional attributes" generated in the early stage by UAV photogrammetry, which includes X, Y, Z geometric coordinates, multispectral albedo and time series thermal infrared temperature and other information.
[0027] The process of constructing and initially screening the three-dimensional tensor field of heterogeneous multimodal data includes: (1) Discretize the three-dimensional multimodal model space into a voxel mesh with a fixed resolution; (2) Construct the feature tensor V=[G,S,T] for each voxel seq ].
[0028] Where G is the geometric symbolic distance field value (SDF), S is the multispectral reflectance vector (indicating water content characteristics), and T is the geometric symbolic distance field value (SDF). seq This is the two-phase surface temperature vector.
[0029] (3) Use a three-dimensional semantic segmentation network to preprocess the voxel field and output an initial three-dimensional mask with labels of “cracks”, “pits” and “intact”.
[0030] Feasible inversion yields subsurface physical morphological parameters of the damaged region, including: A deep neural network is constructed with surface coordinates and time as inputs and the distribution of subsurface thermal diffusivity coefficient and damage depth as outputs. The partial differential equation of thermal conduction of pavement material is embedded as a physical constraint into the loss function of the deep neural network. By minimizing the sum of the data fitting residual and the physical differential constraint residual, the output of the deep neural network is trained to conform to the physical laws of thermal conduction and the true three-dimensional depth of the crack.
[0031] As a specific implementation method, the process of hidden crack depth inversion based on physical information neural networks includes: (1) Construction of physical constraints: According to the laws of thermodynamics, the diffusion of the internal temperature field T(x,y,z,t) of the road surface satisfies the partial differential equation: ; Here, α is the thermal diffusivity of the material. The α of water-containing cracks or air cavities differs significantly from that of dense asphalt.
[0032] (2) Network computation: Construct a deep neural network with surface coordinates (x, y) and time t as inputs and the predicted subsurface thermal diffusivity distribution α as output. pred And depth D.
[0033] (3) Joint loss function optimization: The total loss function (Loss) of the network consists of the data fitting residual (Lossdata) and the physical differential constraint residual (Lossphysics): ; By continuously minimizing the physical violation penalty through backpropagation, the network is forced to converge to a unique solution that conforms to the heat conduction law, thereby outputting the true three-dimensional depth of the crack under lossless conditions.
[0034] Feasible method: A virtual intact surface is reconstructed using a generative adversarial network as the computational reference surface, and combined with the measured pit surface, the pit volume is calculated, including: The generator of the adversarial network is trained using a preset number of intact road models; the identified pothole areas are used as masks to be hollowed out, and the geometric mesh of the surrounding intact areas is input into the trained generator to generate a virtual intact surface that is consistent with the original road surface design; Boolean difference operations are performed on the generated virtual intact surface and the measured pothole surface to calculate the envelope volume between the two, which is taken as the pothole volume.
[0035] As a specific implementation method, the process of surface benchmark reconstruction and volume integration based on conditional generative adversarial networks (cGANs) includes: (1) Surface manifold learning: The generator of cGAN is trained using a large number of intact road models so that it learns and masters the spatial distribution manifold of road longitudinal slope, cross slope and design curvature.
[0036] (2) Virtual Repair Generation: The area labeled "pothole" is hollowed out (input as Mask), and the intact geometric mesh around it is input into cGAN. The network automatically generates a continuous "virtual intact surface" (ideal reference surface) that conforms to the original road surface design trend based on the smooth transition characteristics of the surrounding curvature.
[0037] (3) Boolean difference integral: Perform spatial Boolean operations on the generated ideal surface model and the measured pit bottom model, and calculate the precise envelope volume between the two surfaces by using the tetrahedral infinitesimal decomposition method.
[0038] It is feasible to simulate the evolution of the pit volume in time and space using cellular automata to obtain the predicted damage morphology at future moments, including: Each voxel in the three-dimensional voxel model is defined as a cell and assigned attributes including damage state and local water content inverted from spectral features. A local evolution rule is established driven by freeze-thaw cycle parameters in meteorological data. The local evolution rule determines the state transition probability under future meteorological conditions based on the damage state of adjacent cells and the water content threshold of the cell itself. Input time series meteorological forecast data, drive the cellular automaton to perform iterative calculations, and output the three-dimensional road damage prediction morphology at a specified future time point.
[0039] Furthermore, the freeze-thaw cycle parameters include daily temperature difference and cooling rate; The local evolution rule is configured as follows: when the temperature in the meteorological data is below zero and the cooling rate exceeds a preset threshold, if the adjacent cells are in a damaged state and the water content of the current cell exceeds a preset water-induced swelling threshold, the probability of the current cell transforming into a damaged state in the next iteration increases.
[0040] As a specific implementation method, the process of simulating the evolution of damage cellular automata driven by water-temperature coupling includes: (1) Cell state and attribute assignment: Define the model voxels as cells and assign a state set (0: intact, 1: damaged, 2: critical). Use the near-infrared spectral feature S to invert the local water content scalar for each cell.
[0041] (2) Construction of dynamic rules: Establish local evolution rules with meteorological data (diurnal temperature range, cooling rate) as environmental driving factors. Define the core rule equation: if the adjacent cell is in a damaged state and the water content of the current cell is > θ water (Water-induced swelling threshold) When the input weather forecast shows a sudden drop in temperature (<0℃ and ΔT / Δt>τ), the probability of the cell transitioning to the "damaged state" in the next iteration increases dramatically.
[0042] (3) Spatiotemporal simulation: Input the time series weather forecast of the future freeze-thaw cycle, drive the cellular automata to perform iterative calculations, and finally output the three-dimensional road damage morphology prediction model of a specific future time node to realize disaster early warning.
[0043] Compared with the prior art, the technical solution provided in this embodiment has the following advantages: This breakthrough overcomes surface geometric limitations, enabling physical-level non-destructive testing of road damage. By introducing a physical information neural network, pure image feature matching is transformed into a rigorous solution of partial differential equations. Without destructive drilling, the true depth and spatial media distribution of hidden cracks beneath the road surface can be accurately inverted using the surface's time-series thermodynamic response.
[0044] This technology eliminates volumetric calculation errors under complex road shapes. It replaces the traditional rigid plane fitting method with a generative adversarial mechanism. This technology can adapt to complex geometric surfaces such as road superelevation and camber, providing an unbiased dynamic calculation reference surface, ensuring that the calculation accuracy of pothole repair earthwork remains unaffected under any road alignment.
[0045] This approach upgrades the evolutionary prediction from empirical statistics to mechanistic dynamics. It uses "moisture content" derived from spectral remote sensing and "freeze-thaw cooling" from meteorological conditions as core variables, introducing cellular automata underlying rules. This enables precise spatial extrapolation of the expansion speed and direction of a specific road defect under specific climatic conditions, providing a rigorous scientific basis for preventative maintenance throughout the road's entire lifecycle.
[0046] On the other hand, this embodiment also proposes a road hidden damage inversion and evolution prediction system based on a physical constraint network, used to implement the method, including: The data acquisition and preprocessing module is used to acquire a multimodal 3D real-world model of the road and construct a 3D tensor field containing geometric, spectral, and thermal infrared temporal features to identify the damaged area. The physical information inversion module is used to invert the subsurface physical morphology parameters of the damaged area based on the surface temporal thermal infrared characteristics of the damaged area and by using a physical information neural network embedded with the physical laws of thermal conduction. The reference surface reconstruction and volume calculation module is used to reconstruct a virtual intact surface as the calculation reference surface based on the identified damage area type. For pit-type damage, it uses the geometric features of the surrounding intact area to reconstruct the virtual intact surface as the calculation reference surface, and combines it with the measured pit surface to calculate the pit volume. The damage evolution prediction module is used to simulate the evolution of the pit volume in time and space using cellular automata based on the subsurface physical morphology parameters and material properties extracted from spectral features, combined with external meteorological driving data, to obtain the predicted damage morphology at future moments.
[0047] On the other hand, this embodiment also proposes a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method.
[0048] On the other hand, this embodiment also proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the method.
[0049] On the other hand, this embodiment also proposes a computer program product, including a computer program that, when executed by a processor, implements the steps of the method.
[0050] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A method for inverting and predicting the evolution of hidden road damage based on physical constraint networks, characterized in that, Includes the following steps: Based on the acquired multimodal 3D real-world model of the road, a 3D tensor field containing geometric, spectral, and thermal infrared temporal features is constructed, and the damaged areas are identified. Based on the temporal thermal infrared characteristics of the damaged area and a physical information neural network embedded with the physical laws of thermal conduction, the subsurface physical morphology parameters of the damaged area are obtained by inversion. Based on the identified damage area type, for pit-type damage, a virtual intact surface is reconstructed using a generative adversarial network based on the geometric features of the surrounding intact area as the calculation reference surface, and the pit volume is calculated by combining the measured pit surface. Based on the subsurface physical morphology parameters of the damaged area obtained by inversion and the material properties extracted from spectral features, combined with external meteorological driving data, the evolution of the pit volume in time and space is simulated by cellular automata to obtain the predicted morphology of the damage at future moments.
2. The method according to claim 1, characterized in that, Based on the acquired multimodal 3D real-world model of the road, a 3D tensor field containing geometric, spectral, and thermal infrared temporal features is constructed, and damaged areas are identified, including: The multimodal 3D real-world model of the road is discretized into a voxel mesh. For each voxel, a feature tensor containing geometric symbolic distance field values, multispectral reflectance vectors, and dual-temporal surface temperature vectors is constructed. The voxel field is then processed using a 3D semantic segmentation network to output a 3D mask with different damage category labels.
3. The method according to claim 1, characterized in that, The inversion yields the subsurface physical morphological parameters of the damaged area, including: A deep neural network is constructed with surface coordinates and time as inputs and the distribution of subsurface thermal diffusivity coefficient and damage depth as outputs. The partial differential equation of thermal conduction of pavement material is embedded as a physical constraint into the loss function of the deep neural network. By minimizing the sum of the data fitting residual and the physical differential constraint residual, the output of the deep neural network is trained to conform to the physical laws of thermal conduction and the true three-dimensional depth of the crack.
4. The method according to claim 1, characterized in that, A virtual intact surface is reconstructed using a generative adversarial network as the computational reference surface. Combined with the measured surface of the crater, the volume of the crater is calculated, including: The generator of the adversarial network is trained using a preset number of intact road models; the identified pothole areas are used as masks to be hollowed out, and the geometric mesh of the surrounding intact areas is input into the trained generator to generate a virtual intact surface that is consistent with the original road surface design; Boolean difference operations are performed on the generated virtual intact surface and the measured pothole surface to calculate the envelope volume between the two, which is taken as the pothole volume.
5. The method according to claim 1, characterized in that, By simulating the evolution of pit volume in time and space using cellular automata, the predicted damage morphology at future time points is obtained, including: Each voxel in the three-dimensional voxel model is defined as a cell and assigned attributes including damage state and local water content inverted from spectral features. A local evolution rule is established driven by freeze-thaw cycle parameters in meteorological data. The local evolution rule determines the state transition probability under future meteorological conditions based on the damage state of adjacent cells and the water content threshold of the cell itself. Input time series meteorological forecast data, drive the cellular automaton to perform iterative calculations, and output the three-dimensional road damage prediction morphology at a specified future time point.
6. The method according to claim 5, characterized in that, The freeze-thaw cycle parameters include daily temperature difference and cooling rate; The local evolution rule is configured as follows: when the temperature in the meteorological data is below zero and the cooling rate exceeds a preset threshold, if the adjacent cells are in a damaged state and the water content of the current cell exceeds a preset water-induced swelling threshold, the probability of the current cell transforming into a damaged state in the next iteration increases.
7. A road hidden damage inversion and evolution prediction system based on physical constraint networks, used to implement the method according to any one of claims 1-6, characterized in that, include: The data acquisition and preprocessing module is used to acquire a multimodal 3D real-world model of the road and construct a 3D tensor field containing geometric, spectral, and thermal infrared temporal features to identify the damaged area. The physical information inversion module is used to invert the subsurface physical morphology parameters of the damaged area based on the surface temporal thermal infrared characteristics of the damaged area and by using a physical information neural network embedded with the physical laws of thermal conduction. The reference surface reconstruction and volume calculation module is used to reconstruct a virtual intact surface as the calculation reference surface based on the identified damage area type. For pit-type damage, it uses the geometric features of the surrounding intact area to reconstruct the virtual intact surface as the calculation reference surface, and combines it with the measured pit surface to calculate the pit volume. The damage evolution prediction module is used to simulate the evolution of the pit volume in time and space using cellular automata based on the subsurface physical morphology parameters and material properties extracted from spectral features, combined with external meteorological driving data, to obtain the predicted damage morphology at future moments.
8. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-6.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1-6.