A damage evolution prediction system for heavy-duty traffic asphalt pavement

Abstract

This invention discloses a damage evolution prediction system for heavy-duty traffic asphalt pavement, relating to the field of road engineering technology. The system includes: a data acquisition module for acquiring continuous multi-frame time-series images of pavement cracks, vehicle axle load spectra, and environmental temperature and humidity data; an image processing and spatiotemporal modeling module for performing multi-scale fractal processing and dynamic time warping on the time-series images to calculate the fractal dimension sequence across scales; a material property correlation module for establishing and updating a probabilistic correlation model between fractal dimension and shear modulus in real time using dynamic shear rheometer test data; a multi-physics data fusion module for fusing multi-physics features through a fractal attention network to generate a multi-physics collaborative damage correction factor; and a damage prediction and early warning module for predicting the fractal dimension growth trend using a fractal domain adaptive transfer learning algorithm and outputting a damage evolution early warning based on viscoelastic phase consistency verification results.

Description

Technical Field

[0001] This application relates to the field of road engineering technology, and in particular to image recognition and image classification, specifically a heavy-load traffic asphalt pavement damage evolution prediction system. Background Technology

[0002] Heavy-duty asphalt pavements are subjected to a complex dynamic environment over long periods, experiencing the combined effects of high-frequency vehicle loads, changes in ambient temperature and humidity, and material aging. These factors interact and cause gradual accumulation of damage to the pavement structure, leading to increasingly severe problems such as crack propagation and material performance degradation. The dynamic loads of vehicles, especially the axle load spectrum of heavy traffic, exert periodic impacts on the pavement, causing changes in the stiffness and relaxation modulus of the asphalt material, further accelerating crack propagation. Simultaneously, changes in ambient temperature and humidity affect the viscoelastic properties of the asphalt material, resulting in decreased crack resistance and the gradual macroscopic evolution of cracks. Furthermore, over time, asphalt pavement materials inevitably age, leading to fatigue damage in the internal structure. Cracks gradually develop from initial microcracks into main cracks and branches, eventually forming large-area damage to the pavement. These cracks not only reduce the service life of the pavement but also significantly increase maintenance costs and may even jeopardize driving safety.

[0003] The analysis of damage evolution in heavy-duty traffic asphalt pavement based on fractal damage theory aims to deeply study the development process of asphalt pavement damage and quantify the degree and trend of damage by utilizing the self-similarity principle of fractal geometry. The application of this theory plays a crucial role in scientifically predicting pavement service life, rationally formulating maintenance strategies, and reducing the total life-cycle cost of roads.

[0004] In existing research, patent CN119830662A proposes a method and system for crack resistance of segmental joints in asphalt concrete pavement based on odor-neutralizing technology. This method involves fitting a segmental joint model, analyzing the damaged and fractured zones, constructing a viscoelastic damage model of the asphalt concrete, calculating the viscoelastic damage coefficient, and simulating the crack evolution process of the segmental joint. The method also optimizes the elastic modulus and crack resistance of geotextiles by mixing chemical additives, ultimately implementing a crack resistance scheme for the segmental joint. While this patent has made some progress in optimizing the local crack resistance of segmental joints, the method primarily targets local damage to the segmental joint and fails to comprehensively consider the dynamic evolution process of pavement cracks, especially the crack propagation patterns at different temporal and spatial scales.

[0005] To address the aforementioned issues, this application proposes a heavy-duty traffic asphalt pavement damage evolution prediction system. This system quantifies asphalt pavement damage based on fractal theory, thereby enabling the prediction of pavement damage evolution. Summary of the Invention

[0006] This application provides a heavy-duty traffic asphalt pavement damage evolution prediction system to solve the above-mentioned technical problems.

[0007] This application provides a heavy-duty traffic asphalt pavement damage evolution prediction system, including:

[0008] The data acquisition module collects continuous multi-frame time-series images of road surface cracks, vehicle axle load spectrum, and environmental temperature and humidity data.

[0009] The image processing and spatiotemporal modeling module is used to perform multi-scale fractal processing and dynamic time warping on time-series images, calculate the fractal dimension sequence across scales, and construct a spatiotemporal fractal evolution model. Based on the multi-scale fractal dynamic time warping algorithm, it performs cross-scale fractal dimension alignment on time-series images, extracts multi-level self-similar patterns of crack branch number and length, and constructs a spatiotemporal fractal evolution model.

[0010] The material property correlation module, based on the fractal dimension sequence output by the spatiotemporal fractal evolution model, updates the probabilistic correlation model between fractal dimension and shear modulus in real time through dynamic shear rheometer experiments. When the fractal dimension exceeds the adaptive threshold, it triggers an accelerated expansion warning. It is used to receive the fractal dimension sequence output by the spatiotemporal fractal evolution model and establish and update the probabilistic correlation model between fractal dimension and shear modulus in real time through dynamic shear rheometer experimental data.

[0011] The multi-physics data fusion module is used to integrate the fractal dimension sequence, vehicle axle load spectrum and environmental temperature and humidity data. It performs multi-physics feature fusion through a fractal attention network to generate a multi-physics collaborative damage correction factor. Based on the fractal attention fusion network, and taking the multi-level self-similar mode of the spatiotemporal fractal evolution model as a benchmark, it dynamically couples the vehicle axle load spectrum and environmental temperature and humidity data to generate a multi-physics collaborative damage correction factor.

[0012] The damage prediction and early warning module is used to input the damage correction factor into the high-frequency vibration stress wave attenuation model, predict the fractal dimension growth trend through the fractal domain adaptive transfer learning algorithm, and output damage evolution early warning based on the viscoelastic phase consistency verification results.

[0013] Furthermore, the multi-scale fractal processing performed by the image processing and spatiotemporal modeling module includes:

[0014] Wavelet fractal decomposition is performed on time-series images to generate fractal dimension sequences including main branches of cracks and microcracks at multiple decomposition scales;

[0015] Based on the directional propagation characteristics of longitudinal wheel track cracks, an asymmetric wavelet basis function is used to match the geometric self-similarity of crack branches. The waveform parameters of the asymmetric wavelet basis function are dynamically adjusted based on the spatiotemporal gradient of the wheel track coverage. The waveform parameters include a waveform asymmetry factor α, which is dynamically calculated based on the product of the crack orientation angle and the wheel track coverage.

[0016] The cross-scale alignment path is calculated based on the damage contribution weight, which is dynamically adjusted according to the spatiotemporal correlation of the fractal dimension and the second derivative of the crack propagation rate.

[0017] The spatiotemporal fractal evolution model is generated by integrating multi-level fractal features, and the spatiotemporal coordinates of the inflection point of crack acceleration propagation are output.

[0018] Furthermore, the material property correlation module updates the probabilistic correlation model, including:

[0019] A Gaussian process regression model with fractal dimension and time derivative as covariates is constructed, and the weights of the covariates are dynamically adjusted according to the axial weight distribution of high-frequency loads.

[0020] The shear modulus decay curve was obtained through the dynamic shear rheometer test, and the model hyperparameters were updated by combining the Bayesian optimization algorithm. The model hyperparameters include the material relaxation time.

[0021] When the fractal dimension exceeds the adaptive threshold, a hierarchical warning is triggered. The adaptive threshold is set in segments according to the wheel track coverage.

[0022] Furthermore, the multiphysics data fusion module performs multiphysics feature fusion through a fractal attention network, including:

[0023] A fractal gradient enhancement module is constructed to extract the first-order and second-order gradient features of the crack branch length change rate;

[0024] The fractal dimension sequence is spliced ​​with vehicle axle load spectrum and environmental temperature and humidity data to serve as a multi-physics collaborative damage correction factor. Through a multi-head self-attention mechanism, the fractal dimension sequence output by the spatiotemporal fractal evolution model is used as a query vector to perform cross-modal correlation with the main frequency band of the vehicle axle load spectrum and environmental temperature and humidity data.

[0025] Attention weights are dynamically adjusted based on the first- and second-order gradient features of the crack branch length change rate.

[0026] Furthermore, the fractal domain adaptive transfer learning performed by the damage prediction and early warning module includes:

[0027] Adversarial training is performed on the fractal dimension distributions of the source and target domains in the feature space, and the discriminator of the adversarial network is constrained by the crack branch topology structure.

[0028] A viscoelastic phase consistency loss function is introduced and calibrated using measured ultrasonic phase difference data.

[0029] The fractal dimension growth trend under high-frequency vibration is predicted as a result of pavement damage evolution prediction. An emergency warning is triggered when the fractal dimension growth trend exceeds the growth rate threshold.

[0030] Furthermore, the waveform parameter adjustment of the asymmetric wavelet basis function includes:

[0031] The propagation direction angle θ of the longitudinal wheel track crack is extracted, and θ is calculated using the Hough transform algorithm;

[0032] The waveform asymmetry factor α is dynamically adjusted. When the wheel track coverage is >80%, α increases exponentially.

[0033] Multiscale fractal features matching the directional propagation of cracks are generated, with the decomposition scale of the main crack branches set to 0.1 to 2.0 mm to match the features of heavy-load wheel track compaction.

[0034] Furthermore, the triggering conditions for the tiered early warning include:

[0035] When the fractal dimension exceeds the first threshold, it is marked as a potential damage area;

[0036] When the fractal dimension exceeds the second threshold and the wheel track coverage is greater than 75%, an accelerated expansion warning is triggered.

[0037] The first threshold and the second threshold are dynamically calculated using the product function of the aggregate angularity index and the asphalt viscosity value.

[0038] Furthermore, the gradient feature extraction of the fractal gradient enhancement module includes:

[0039] The first-order gradient of the rate of change of crack branch length is calculated using the adaptive step-size finite difference method.

[0040] Second-order gradient features were extracted, and the inflection point determination condition was set as second-order gradient > 0.03 / mm².

[0041] Gradient features are mapped to a high-dimensional space, and the embedding dimension is dynamically expanded according to the damage complexity.

[0042] Furthermore, the calculation of the viscoelastic phase consistency loss function includes:

[0043] Stress wave phase delay data were obtained by ultrasonic flaw detection test, with the sampling frequency set to 10MHz to match the microcrack scale.

[0044] The phase difference predicted by the fractal features after migration was calculated, and the consistency was verified by the complex domain cross-correlation coefficient.

[0045] By combining the relaxation time parameter of the viscoelastic constitutive equation and adjusting the weighting coefficients according to the temperature gradient, the viscoelastic changes of asphalt materials are quantified using the viscoelastic phase consistency loss function.

[0046] Furthermore, the dynamic adjustment rule for the waveform asymmetry factor α includes:

[0047] When the coverage of the wheel track is less than 60%, α = 1.0;

[0048] When the coverage rate is between 60% and 80%, α = 1.0 + 0.02 × (coverage rate - 60);

[0049] When the coverage rate is greater than 80%, α = 1.4 + 0.1 × ln(coverage rate - 79).

[0050] The beneficial effects of this application are as follows: By constructing a spatiotemporal fractal evolution model and a multi-physics field collaborative correction mechanism, the technical problem that traditional static fractal analysis cannot capture the dynamic evolution of damage under heavy traffic is solved. Specifically, this includes: based on a multi-scale fractal dynamic time warping algorithm, real-time tracking of multi-level self-similar pattern changes in crack branches, accurately identifying abrupt changes in damage rate caused by overloaded vehicle crushing, such as the inflection point of accelerated crack propagation caused by axle load > 30 tons; through a fractal tensor direction weighting strategy, distinguishing the damage characteristics of longitudinal wheel track zones and transverse shear zones, quantifying the directional load effect where longitudinal cracks account for more than 70% within the wheel track zone; combining a stress wave attenuation model and viscoelastic phase verification, capturing the microcrack activation mechanism caused by vibration loads above 10Hz, avoiding the omission of dynamic scattering effects by traditional methods; and through fractal domain adaptive transfer learning, using small sample data to quickly adapt to the damage prediction of new materials such as nano-modified asphalt, solving the bottleneck of traditional models relying on massive training data. Attached Figure Description

[0051] Figure 1 This is a structural diagram of an optional heavy-duty traffic asphalt pavement damage evolution prediction system according to an embodiment of this application;

[0052] Figure 2 This is a flowchart illustrating the execution of multi-scale fractal processing by an optional image processing and spatiotemporal modeling module according to an embodiment of this application.

[0053] 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

[0054] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0055] This application provides a heavy-duty traffic asphalt pavement damage evolution prediction system. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Optionally, such as Figure 1 As shown, this application provides a heavy-duty traffic asphalt pavement damage evolution prediction system, comprising:

[0057] The data acquisition module 101 is used to acquire continuous multi-frame time-series images of road surface cracks, vehicle axle load spectrum, and environmental temperature and humidity data.

[0058] The image processing and spatiotemporal modeling module 102 is used to perform multi-scale fractal processing and dynamic time warping on time-series images, and to calculate the fractal dimension sequence across scales in order to construct a spatiotemporal fractal evolution model.

[0059] The material property correlation module 103 is used to receive the fractal dimension sequence output by the spatiotemporal fractal evolution model, and to establish and update the probabilistic correlation model between fractal dimension and shear modulus in real time using dynamic shear rheometer test data.

[0060] In some embodiments, an accelerated expansion warning is triggered when the fractal dimension exceeds an adaptive threshold.

[0061] Optionally, in this embodiment, the adaptive threshold does not use a fixed fractal dimension warning standard. Instead, it automatically adjusts the fractal dimension warning threshold based on real-time monitored information such as truck load, ambient temperature and humidity, and crack growth rate, making the warning more accurate and more consistent with the actual damage situation. Specifically, the greater the load, the harsher the environment (high temperature / high humidity), and the faster the cracks grow, the lower the warning threshold, thus enabling more timely detection of road surface damage.

[0062] In one specific implementation, the adjustment of the adaptive threshold and the early warning may include the following steps:

[0063] 1. Real-time Data Acquisition: Truck Load Capacity: Real-time axle load weight of passing trucks is acquired through dynamic weighing equipment installed on the road, distinguishing between standard trucks under 20 tons, overloaded trucks of 20-30 tons, and severely overloaded trucks over 30 tons. Environmental Conditions: Temperature and humidity are monitored in real-time through embedded sensors, identifying conditions with temperatures above 30℃ and humidity levels above 80%. Crack Changes: Road surface cracks are periodically photographed by drones, and the growth rate of the crack fractal dimension is calculated by analyzing multiple consecutive images, e.g., the growth rate of the fractal dimension is calculated weekly.

[0064] 2. Dynamic Adjustment of Early Warning Thresholds: Basic Threshold Setting: Through laboratory tests, the early warning threshold under standard conditions (trucks under 20 tons, normal temperature of 20℃, 50% humidity, slow crack growth) is determined to be 2.7 (the higher the fractal dimension, the more complex the crack). Load Impact: When encountering overloaded trucks, the threshold is automatically lowered: 20-30 ton overloaded trucks: the threshold drops from 2.7 to 2.6 (early warning); over 30 ton severely overloaded trucks: the threshold is further lowered to 2.5 (emergency warning). Environmental Impact: When encountering harsh environments, the threshold is further lowered: High temperature (above 30℃): the threshold is lowered by 0.05 (e.g., from 2.7 to 2.65); High humidity (above 80%): the threshold is lowered by 0.03 (e.g., from 2.65 to 2.62). Crack Growth Rate Impact: If the crack fractal dimension grows rapidly (e.g., more than 0.02 per week), indicating accelerated damage, the threshold is lowered by 0.05 (e.g., from 2.62 to 2.57).

[0065] 3. Triggering early warning: Real-time comparison of the fractal dimension of the current crack with the calculated dynamic threshold: When the fractal dimension reaches or exceeds the dynamic threshold, an early warning is triggered immediately. For example, when overloaded by 35 tons and at a high temperature of 35℃, the threshold drops to 2.5. At this time, an alarm will be triggered as soon as the fractal dimension reaches 2.5, which is earlier than the traditional fixed threshold of 2.7.

[0066] Based on this embodiment, truck load, environmental conditions, and crack growth rate are integrated into a single threshold adjustment mechanism, overcoming the shortcomings of traditional methods that only use fixed values ​​for early warning. The warning conditions are automatically "relaxed" according to actual working conditions: the greater the overload, the worse the weather, and the faster the crack grows, the more "sensitive" the mechanism becomes, allowing for earlier detection of damage risks, for example, providing warnings 1-3 months earlier than traditional methods. All threshold adjustment ranges, such as a 0.1 reduction for a 10-ton overload and a 0.05 reduction for a 10°C high temperature, are determined through real truck crushing tests and aging tests, ensuring scientific reliability.

[0067] The multi-physics data fusion module 104 is used to integrate fractal dimension sequence, vehicle axle load spectrum and environmental temperature and humidity data. It performs multi-physics feature fusion through fractal attention network to generate a multi-physics collaborative damage correction factor. Based on the fractal attention fusion network, and taking the multi-level self-similar mode of the spatiotemporal fractal evolution model as the benchmark, it dynamically couples vehicle axle load spectrum and environmental temperature and humidity data to generate a multi-physics collaborative damage correction factor.

[0068] The damage prediction and early warning module 105 is used to input the damage correction factor into the high-frequency vibration stress wave attenuation model, predict the fractal dimension growth trend through the fractal domain adaptive transfer learning algorithm, and output damage evolution early warning based on the viscoelastic phase consistency verification results.

[0069] Based on the embodiments provided in this application, the technical problem that traditional static fractal analysis cannot capture the dynamic evolution of damage under heavy traffic is solved by constructing a spatiotemporal fractal evolution model and a multi-physics field collaborative correction mechanism. Specifically, this includes: based on a multi-scale fractal dynamic time warping algorithm, real-time tracking of multi-level self-similar pattern changes in crack branches to accurately identify abrupt changes in damage rate caused by overloaded vehicles, such as the inflection point of accelerated crack propagation caused by axle load > 30 tons; by using a fractal tensor direction weighting strategy to distinguish the damage characteristics of longitudinal wheel track zones and transverse shear zones, quantifying the directional load effect where longitudinal cracks account for more than 70% within the wheel track zone; by combining a stress wave attenuation model and viscoelastic phase verification to capture the microcrack activation mechanism caused by vibration loads above 10Hz, avoiding the omission of dynamic scattering effects by traditional methods; and by using fractal domain adaptive transfer learning to quickly adapt to damage prediction of new materials such as nano-modified asphalt using small sample data, solving the bottleneck of traditional models relying on massive training data.

[0070] Furthermore, such as Figure 2 As shown, the multi-scale fractal processing performed by the image processing and spatiotemporal modeling module includes:

[0071] S201, perform wavelet fractal decomposition on time-series images to generate a multi-scale fractal dimension sequence including crack main branches and microcracks;

[0072] S202, based on the directional propagation characteristics of longitudinal wheel track cracks, an asymmetric wavelet basis function is used to match the geometric self-similarity of crack branches. The waveform parameters of the asymmetric wavelet basis function are dynamically adjusted based on the spatiotemporal gradient of the wheel track coverage. Among them, the waveform parameters include the waveform asymmetry factor α, which is dynamically calculated based on the product of crack orientation angle and wheel track coverage.

[0073] It should be noted that, in this embodiment, by performing a two-dimensional discrete wavelet transform on the time series image, wavelet components of the time series image at multiple decomposition scales are obtained. The wavelet components include a low-frequency approximation image, a high-frequency horizontal detail image, a high-frequency vertical detail image, and a high-frequency diagonal detail image. The low-frequency approximation image reflects the global features of the time series image, while the high-frequency horizontal detail image, high-frequency vertical detail image, and high-frequency diagonal detail image reflect the edge and texture features of the cracks in the time series image.

[0074] Furthermore, high-frequency horizontal detail images, high-frequency vertical detail images, and high-frequency diagonal detail images are extracted from the wavelet components. Threshold segmentation is performed on the extracted images to obtain the crack regions in the images. A connected component labeling algorithm is used to label the main branches of cracks and microcracks in the crack regions. Crack regions with an area higher than a preset area threshold and a length higher than a preset length threshold are labeled as main branches of cracks.

[0075] Furthermore, images identifying the main crack branches from high-frequency horizontal detail images, high-frequency vertical detail images, and high-frequency diagonal detail images are overlaid on the mesh image. The number of meshes containing the main crack branches in the mesh image is counted, and the fractal dimension of the main crack branches in the time-series image at the current decomposition scale is calculated:

[0076] ;

[0077] Where G represents the fractal dimension, count represents the number of grids in the grid image that contain the main branches of the crack, and r represents the grid side length;

[0078] Images of microcracks identified in high-frequency horizontal detail images, high-frequency vertical detail images, and high-frequency diagonal detail images are overlaid on a mesh image. The number of meshes containing microcracks in the mesh image is counted, and the fractal dimension of microcracks in the time series image at the current decomposition scale is calculated.

[0079] The fractal dimension vectors of the main branch of the crack and the microcrack at the current decomposition scale in the time series image are constructed, and the fractal dimension vectors of all decomposition scales are constructed into a multi-scale fractal dimension sequence including the main branch of the crack and the microcrack.

[0080] Optionally, in this embodiment, the waveform asymmetry factor α is used as a core waveform parameter, and its value needs to conform to the coupling characteristics between the crack propagation direction and the load concentration of the wheel track. Specific dynamic adjustment methods include:

[0081] The crack orientation angle θ is defined as the angle between the crack principal axis and the vehicle's direction of travel (θ∈[0°,90°], the smaller θ is, the closer the crack is to the longitudinal expansion of the wheel track). θ directly affects the initial reference value of α: when θ=0°, α takes the base value of 1.0. For every 10° increase in θ, the reference value of α decreases by 0.1 (reflecting the reduced demand for asymmetric waveforms in transverse cracks).

[0082] When the wheel track coverage is greater than 70%, the accuracy of longitudinal crack fitting is enhanced based on the load concentration effect. For every 10% increase in coverage, α is increased by 0.3 on the basis of the benchmark value corresponding to θ. For example, when the coverage is 80% and θ is 15°, α is 1.0-0.15+0.3=1.15; when the coverage is 90%, α is increased by another 0.3 to 1.45.

[0083] By directly linking the waveform parameter α with the crack orientation angle θ and the wheel track coverage, the wavelet basis function waveform can be adaptively adjusted to an asymmetric shape that is closer to the longitudinal crack propagation characteristics, thus accurately capturing the directional self-similarity of cracks in the wheel track region.

[0084] S203, calculates the cross-scale alignment path based on damage contribution weight, and the damage contribution weight is dynamically adjusted according to the spatiotemporal correlation of fractal dimension and the second derivative of crack propagation rate;

[0085] Optionally, in this embodiment, the weight is increased to 1.5 times when the expansion rate changes abruptly;

[0086] S204 integrates multi-level fractal features to generate a spatiotemporal fractal evolution model, outputting the spatiotemporal coordinates of the inflection point of accelerated crack propagation.

[0087] Optionally, in this embodiment, the inflection point determination condition is that the fractal dimension change rate is >0.05 / frame;

[0088] Based on the embodiments provided in this application, the problem of insufficient sensitivity of traditional symmetric wavelet decomposition to longitudinal wheel track cracks is solved by dynamically matching the asymmetric wavelet basis function with the crack orientation angle. For example, when the wheel track coverage is >80%, the asymmetric factor α increases exponentially, accurately quantifying the compaction concentration effect of heavy traffic and avoiding feature ambiguity caused by homogenization decomposition.

[0089] Furthermore, the material property correlation module updates the probabilistic correlation model, including:

[0090] A Gaussian process regression model with fractal dimension and time derivative as covariates is constructed, and the weights of the covariates are dynamically adjusted according to the axial weight distribution of high-frequency loads.

[0091] Optionally, in this embodiment, the weight is increased by 20% when the axle load is >30 tons;

[0092] The shear modulus decay curve was obtained by dynamic shear rheometer test, and the model hyperparameters were updated by Bayesian optimization algorithm. The model hyperparameters include material relaxation time.

[0093] Optionally, in this embodiment, the material relaxation time τ = asphalt viscosity / aggregate angularity index × 0.8;

[0094] When the fractal dimension exceeds the adaptive threshold, a hierarchical warning is triggered. The adaptive threshold is set in segments according to the wheel track coverage.

[0095] Optionally, in this embodiment, the threshold is 2.5 when the coverage is ≤60%, and the threshold is reduced by 0.1 for every 10% increase when the coverage is >60%.

[0096] As a specific implementation method, the probabilistic correlation model between fractal dimension and shear modulus is as follows:

[0097] ;

[0098] in, This represents the probability that the fractal dimension exceeds the critical value, ranging from [0,1], and is obtained through Gaussian process regression modeling. It is used to reflect the likelihood of damage occurring. It is a mean function, derived from the material relaxation time. (Unit: seconds, s) and axis weight distribution weight The decision was made jointly, and the calculation formula is as follows:

[0099] ,

[0100] in The dynamic shear rheometer was used to measure the viscoelastic properties of asphalt; the axle load distribution weight W was set at 1.0 when the axle load was ≤30 tons and 1.2 when the axle load was >30 tons, which was used to quantify the contribution of overload to the acceleration of damage. The standard deviation is dynamically updated through Bayesian optimization to describe the uncertainty of fractal dimension fluctuations. The initial value is 0.1, which increases with the degree of asphalt aging. For example, it may increase to 0.15 after 5 years of use. The fractal dimension at time t is an indicator describing the complexity of the crack. It should be noted that the fractal dimension at time t is calculated as follows:

[0101] ;

[0102] Where K represents the number of decomposition scales, and we set K to 3. The time series image at time t represents the time series at time t. The fractal dimension of the main branches of the crack at each decomposition scale. The time series image at time t represents the time series at time t. The fractal dimension of microcracks at each decomposition scale.

[0103] Based on the above formula, and based on the material relaxation time With axis weight distribution weight Dynamic coupling enables adaptive adjustment of the fractal dimension critical threshold. For example, when the axle weight is greater than 30 tons, the weight... Increased to 1.2, mean The offset lowers the threshold, providing early warning of the accelerated damage risk caused by overloaded vehicles; simultaneously, Bayesian optimization updates in real time. This enhances the model's adaptability to the asphalt aging process.

[0104] Based on the embodiments provided in this application, the influence of overloaded vehicles (axle load > 30 tons) on the fractal dimension threshold is enhanced by dynamically adjusting the axle load distribution weight W. For example, when the axle load exceeds the limit, the threshold is adaptively reduced by 0.1, providing an early warning of accelerated damage risk and overcoming the limitation that static thresholds cannot respond to sudden load changes.

[0105] Furthermore, the multiphysics data fusion module performs multiphysics feature fusion through a fractal attention network, including:

[0106] A fractal gradient enhancement module is constructed to extract the first-order and second-order gradient features of the crack branch length change rate;

[0107] Optionally, in this embodiment, the second-order gradient threshold can be set to 0.02 / mm² or 0.03 / mm² to filter noise interference.

[0108] By using a multi-head self-attention mechanism, the multi-level self-similar patterns of the spatiotemporal fractal evolution model are used as query vectors and cross-modal associations are performed with the main frequency band (5-15Hz) of the vehicle axle load spectrum.

[0109] The attention weight is dynamically adjusted based on the first and second gradient features of the crack branch length change rate. When the temperature is >40℃, the attention weight is increased by 1.2 times to enhance the sensitivity to high temperature damage.

[0110] As a specific implementation method, the fractal gradient attention weights can be calculated based on the following formula:

[0111] ;

[0112] in, Indicates the first The query vector and the first The attention weights of each key-value vector, after being normalized by the Softmax function, are in the range of [0,1], and are used to focus on load-dominated damage such as longitudinal cracks in wheel tracks. For the query vector, crack self-similarity features are extracted from the spatiotemporal fractal evolution model, such as the number of crack branches and the fractal dimension sequence of their lengths. The extraction method is to extract the main crack branches and microcracks in the crack region after being labeled with a connected component labeling algorithm, calculate the length and number of the main crack branches and microcracks, and calculate the changes in the length of the main crack branches and microcracks as the first and second gradients of the crack branch lengths. Its dimension d=64, corresponding to the multi-level self-similarity pattern of the main crack branches (0.1-2.0 mm scale). The key-value vector is a feature encoding of vehicle axle load spectrum (such as high-frequency components above 10Hz) or environmental data (temperature and humidity); the vehicle axle load spectrum is obtained through fast Fourier transform, and usually the first 3 main frequency bands (such as 5Hz, 15Hz, and 25Hz) are retained. d is the feature dimension, which is the encoding dimension of the query / key value vector. It is used to scale the dot product result to stabilize training. In this scenario, the empirical value is d=64. This is the gradient enhancement coefficient, measured in square millimeters. It is used to amplify the effect of the second-order gradient on the attention. It is set according to the scale of the crack in the wheel track, with a default value of 0.03 square millimeters, corresponding to a sensitive range of 0.1 to 2.0 mm for the main crack branch scale. This represents the second-order gradient of the crack branch length, in millimeters⁻² (mm). −2 The second derivative, obtained through differential calculation of consecutive frame images, reflects the rate of change of crack branch length, i.e., the acceleration of crack propagation; when >0.03mm −2 When this occurs, it can be identified as a sudden inflection point, corresponding to scenarios such as overloading and sudden braking that cause a sudden increase in cracks.

[0113] Based on the above equation, using the second-order gradient term... The introduction of this feature enhances the focus on damage mutation characteristics. For example, when the second gradient of the rate of change of crack branch length is greater than 0.03 / mm, the gradient term significantly increases the corresponding attention weight, effectively filtering out spurious damage signals caused by material softening under high temperature conditions and improving the robustness of cross-modal fusion.

[0114] Based on the embodiments provided in this application, second-order gradient features are extracted by the fractal gradient enhancement module, which effectively identifies damage mutation signals caused by asphalt softening under high temperature conditions, suppresses noise interference caused by temperature fluctuations, and improves the stability of cross-modal fusion.

[0115] Furthermore, the fractal domain adaptive transfer learning performed by the damage prediction and early warning module includes:

[0116] Adversarial training is performed on the fractal dimension distributions of the source and target domains in the feature space. The discriminator of the adversarial network is constrained by a crack branch topology (branch angle error < 5°).

[0117] A viscoelastic phase consistency loss function is introduced and calibrated using measured ultrasonic phase difference data.

[0118] The fractal dimension growth trend under high-frequency vibration is predicted as a result of pavement damage evolution prediction. An emergency warning is triggered when the fractal dimension growth trend exceeds the growth rate threshold.

[0119] Optionally, in this embodiment, the growth rate threshold includes, but is not limited to, 0.08 / 10,000 loads, 0.07 / 10,000 loads, etc.

[0120] As a specific implementation method, the viscoelastic phase consistency loss function is:

[0121] ;

[0122] in, The total loss value incorporates stress wave phase difference and relaxation time constraints to ensure the transfer learning model conforms to the viscoelastic physical laws of asphalt. When the output value is <0.5, the model is considered converged, corresponding to a phase difference error <10% and a relaxation time error <20%.

[0123] This is the phase difference weighting coefficient, used to dynamically balance the weights of phase difference and relaxation time constraints. It is adjusted with the temperature gradient to adapt to changes in material viscosity; when the temperature is ≤20 degrees Celsius, =0.5, for every 10 degrees Celsius increase in temperature, Increase by 0.15, for example at 40 degrees Celsius. =0.8; The phase difference is predicted in radians (rad). It is the stress wave phase delay predicted by the transfer learning model, reflecting the scattering effect of microcracks on high-frequency loads (10MHz). It was obtained by actual measurement with a 10MHz ultrasonic flaw detector with a resolution of 0.1rad, corresponding to microcrack sizes of 0.1 to 1mm. The measured phase difference, in radians (rad), represents the actual stress wave phase difference detected and used as the true value for model calibration. It is positively correlated with the fractal dimension; for every 0.1 increase in fractal dimension, [the phase difference increases]. Increase by 0.2 rad; As a reference phase, it can be set to 1 rad. Used to normalize the phase difference squared term;

[0124] This is the theoretical relaxation time, expressed in seconds. It is the theoretical value from the viscoelastic constitutive equation, calibrated from the asphalt viscosity-temperature curve, and reflects the stress relaxation characteristics under ideal conditions. For example, at 15 degrees Celsius... =80s; The actual relaxation time, measured in seconds, is obtained in real-time through a dynamic shear rheometer test. It is affected by factors such as asphalt aging and load, reflecting the current viscoelastic state of the asphalt; [The text abruptly shifts to a seemingly unrelated topic:] aged pavement... The values ​​will increase by 30% to 50% compared to newer road surfaces, and the measurements are generally updated every quarter. For reference time, =1s or = , Used to normalize the relaxation time constraint term.

[0125] Based on the above formula, this loss function guides model training by balancing the normalized phase error and relaxation time error through weight coefficients. Through joint optimization of the phase difference term and relaxation time constraint, it ensures that the transfer learning model conforms to the viscoelastic physical laws of asphalt materials. For example, when temperature increases leading to asphalt softening (λ increases), the loss function focuses more on verifying phase difference consistency, avoiding prediction distortion caused by changes in material properties; the relaxation time constraint term prevents the transferred model from deviating from the actual material response.

[0126] Based on the embodiments provided in this application, the consistency between the transfer learning model and the physical laws of stress wave attenuation is constrained by combining the viscoelastic phase consistency loss function. For example, by verifying the phase difference tolerance (±0.1π), it is ensured that the damage prediction of nano-modified asphalt conforms to its viscoelastic relaxation characteristics, thus avoiding physical distortion of the data-driven model.

[0127] Furthermore, the waveform parameter adjustment of the asymmetric wavelet basis function includes:

[0128] The propagation direction angle θ of the longitudinal wheel track crack is extracted, and θ is calculated using the Hough transform algorithm;

[0129] The waveform asymmetry factor α is dynamically adjusted. When the wheel track coverage is >80%, α increases exponentially.

[0130] Multiscale fractal features matching the directional propagation of cracks are generated, with the decomposition scale of the main crack branches set to 0.1 to 2.0 mm to match the features of heavy-load wheel track compaction.

[0131] Based on the embodiments provided in this application, the Hough transform algorithm is used to accurately extract the crack orientation angle θ. Combined with the coverage-driven asymmetric factor adjustment rule, the directional propagation mode of cracks in the wheel track area is quantified, thus solving the modeling bias of directional damage in traditional algorithms.

[0132] Furthermore, the triggering conditions for tiered early warnings include:

[0133] When the fractal dimension exceeds the first threshold, it is marked as a potential damage area;

[0134] When the fractal dimension exceeds the second threshold and the wheel track coverage is greater than 75%, an accelerated expansion warning is triggered.

[0135] The first threshold and the second threshold are dynamically calculated using the product function of the aggregate angularity index (>45) and the asphalt viscosity value (>1.2kPa·s).

[0136] Based on the embodiments provided in this application, the graded early warning threshold is dynamically calculated using the product function of the aggregate angularity index and the asphalt viscosity, thereby achieving a coordinated response between material properties and load conditions. For example, highly angular aggregates can raise the threshold, delay the triggering of the early warning, and match their shear failure resistance.

[0137] Furthermore, the gradient feature extraction of the fractal gradient enhancement module includes:

[0138] The first-order gradient of the rate of change of crack branch length is calculated using the adaptive step-size finite difference method.

[0139] Optionally, in this embodiment, the step size = crack length × 0.1;

[0140] Second-order gradient features were extracted, and the inflection point determination condition was set as second-order gradient > 0.03 / mm².

[0141] Gradient features are mapped to a high-dimensional space, and the embedding dimension is dynamically expanded according to the damage complexity.

[0142] Optionally, in this embodiment, when the complexity is >0.7, the dimension is increased to 128;

[0143] Based on the embodiments provided in this application, the adaptive step size difference method (step size = crack length × 0.1) ensures that the gradient calculation matches the damage propagation scale, avoids the failure to detect gradients of microcracks with a fixed step size, and improves the detection sensitivity of abrupt change inflection points.

[0144] Furthermore, the calculation of the viscoelastic phase consistency loss function includes:

[0145] Stress wave phase delay data were obtained by ultrasonic flaw detection test, with the sampling frequency set to 10MHz to match the microcrack scale.

[0146] The phase difference predicted by the fractal features after migration was calculated, and the consistency was verified by the complex domain cross-correlation coefficient.

[0147] The weighting coefficients are adjusted according to the temperature gradient, taking into account the relaxation time parameter of the viscoelastic constitutive equation.

[0148] Optionally, in this embodiment, the coefficient increases by 0.15 for every 10°C increase in temperature;

[0149] Based on the embodiments provided in this application, the phase delay data of microcracks is accurately captured by a 10MHz ultrasonic sampling frequency, and the cross-correlation coefficient in the complex domain is used for verification to ensure the physical consistency between the fractal characteristics after migration and the microcrack propagation.

[0150] Furthermore, the dynamic adjustment rules for the waveform asymmetry factor α include:

[0151] When the wheel track coverage is less than 60%, α = 1.0;

[0152] When the coverage rate is between 60% and 80%, α = 1.0 + 0.02 × (coverage rate - 60);

[0153] When the coverage rate is greater than 80%, α = 1.4 + 0.1 × ln(coverage rate - 79).

[0154] Based on the embodiments provided in this application, the piecewise adjustment rule of the coverage-driven asymmetric factor α (linear + exponential increase) quantifies the load concentration effect of the wheel track zone. For example, when the coverage is >80%, α = 1.4 + 0.1 × ln(coverage - 79), which significantly enhances the adaptability to crack decomposition in the heavy-load compaction area.

[0155] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A heavy traffic asphalt pavement damage evolution prediction system, characterized in that, The method comprises the following steps: a data acquisition module is used to acquire continuous multi-frame time sequence images of pavement cracks, vehicle-mounted axle load spectrum and environmental temperature and humidity data; an image processing and space-time modeling module is used to perform multi-scale fractal processing and dynamic time warping on the time sequence images, calculate cross-scale fractal dimension sequences, and construct a space-time fractal evolution model; a material performance correlation module is used to receive the fractal dimension sequences output by the space-time fractal evolution model, and establish and update a probability correlation model of fractal dimension and shear modulus in real time through dynamic shear rheometer test data; the probability correlation model of fractal dimension and shear modulus is as follows: ； wherein, represents the probability that the fractal dimension exceeds the critical value, ranging between [0, 1]; is the mean function, and the calculation formula is: ， wherein, The viscoelastic properties of the asphalt are reflected by a dynamic shear rheometer; the axle load distribution weight W is 1.0 when the axle load is less than or equal to 30 tons, and is 1.2 when the axle load is greater than 30 tons, and is used to quantify the contribution of overloading to damage acceleration; is a standard deviation, which is dynamically updated by Bayesian optimization, and is used to describe the uncertainty of the fluctuation of the fractal dimension, and the initial value is 0.1 and increases with the increase of the aging degree of the asphalt; D(t) represents the fractal dimension at time t, and is an index for describing the complexity of the crack. the fractal dimension calculation method at time t is as follows: ； wherein K represents the number of decomposition scales, K is set to 3, a fractal dimension of a main branch of a crack in the time-series image at the first decomposition scale, a fractal dimension of a micro crack in the time-series image at the second decomposition scale; a multi-physical field data fusion module is used to integrate the fractal dimension sequences, vehicle-mounted axle load spectrum and environmental temperature and humidity data, perform multi-physical field feature fusion through a fractal attention network, and generate a multi-physical field collaborative damage correction factor; a damage prediction and early warning module is used to input the damage correction factor into a high-frequency vibration stress wave attenuation model, predict the fractal dimension growth trend through a fractal domain adaptive transfer learning algorithm, and output a damage evolution warning based on a viscoelastic phase consistency verification result.

2. The heavy traffic asphalt pavement damage evolution prediction system according to claim 1, characterized in that, The multi-scale fractal processing performed by the image processing and space-time modeling module comprises the following steps: wavelet fractal decomposition is performed on the time sequence images to generate fractal dimension sequences including crack main branches and microcracks at multiple decomposition scales; according to the directional expansion characteristics of longitudinal wheel track band cracks, an asymmetric wavelet basis function is used to match the geometric self-similarity of crack branches, the waveform parameters of the asymmetric wavelet basis function are dynamically adjusted based on the space-time gradient of the wheel track band coverage rate, wherein the waveform parameters include a waveform asymmetry factor α, which is dynamically calculated according to the product of the crack direction angle and the wheel track band coverage rate; a cross-scale alignment path is calculated based on damage contribution weights, and the damage contribution weights are dynamically adjusted according to the space-time correlation of the fractal dimension and the second derivative of the crack expansion rate; multi-level fractal features are fused to generate the space-time fractal evolution model, and the space-time coordinates of the crack accelerated expansion inflection point are output. 3.The heavy traffic asphalt pavement damage evolution prediction system according to claim 1, characterized in that, The material performance correlation module updates the probability correlation model, comprising the following steps: a Gaussian process regression model is constructed with the fractal dimension time derivative as the covariant, and the covariant weight is dynamically adjusted according to the axle load distribution of the high-frequency load; a shear modulus attenuation curve is obtained through the dynamic shear rheometer test, and the model hyperparameters including the material relaxation time are updated by combining the Bayesian optimization algorithm; when the fractal dimension exceeds an adaptive threshold, a hierarchical early warning is triggered, and the adaptive threshold is set in sections according to the wheel track band coverage rate.

4. The heavy traffic asphalt pavement damage evolution prediction system according to claim 1, characterized in that, The multi-physical field data fusion module performs multi-physical field feature fusion through a fractal attention network, comprising the following steps: a fractal gradient enhancement module is constructed to extract the first and second gradient features of the crack branch length change rate; the fractal dimension sequences, vehicle-mounted axle load spectrum and environmental temperature and humidity data are spliced to serve as a multi-physical field collaborative damage correction factor, and the fractal dimension sequences output by the space-time fractal evolution model are taken as query vectors through a multi-head self-attention mechanism to perform cross-modal correlation with the main frequency band of the vehicle-mounted axle load spectrum and the environmental temperature and humidity data. The first and second gradient features of the crack branch length change rate are used to dynamically adjust the attention weight.

5. The heavy traffic asphalt pavement damage evolution prediction system according to claim 1, wherein, The fractal domain adaptive transfer learning performed by the damage prediction and early warning module includes: The fractal dimension distributions of the source domain and the target domain are adversarially trained in the feature space, and the discriminator of the adversarial network adopts a crack branch topological structure constraint; An elastoviscous phase consistency loss function is introduced, and the elastoviscous phase consistency loss function is calibrated through ultrasonic phase difference measurement data; The fractal dimension growth trend under high-frequency vibration is predicted as the road damage evolution prediction result, and an emergency warning is triggered when the fractal dimension growth trend is greater than the growth rate threshold.

6. The heavy traffic asphalt pavement damage evolution prediction system according to claim 2, wherein, The waveform parameter adjustment of the asymmetric wavelet basis function includes: The expansion direction angle θ of the longitudinal wheel track band crack is extracted, and the Hough transform algorithm is used to calculate θ; The waveform asymmetry factor α is dynamically adjusted, and when the wheel track band coverage is greater than 80%, α is increased according to an exponential function; Multi-scale fractal features matching the directional expansion of the crack are generated, and the decomposition scale of the main crack branch is set to 0.1 to 2.0 mm to match the heavy load wheel track rolling characteristics.

7. The heavy traffic asphalt pavement damage evolution prediction system according to claim 3, characterized in that, The trigger conditions of the hierarchical early warning include: When the fractal dimension exceeds the first threshold, it is marked as a potential damage area; When the fractal dimension exceeds the second threshold and the wheel track band coverage is greater than 75%, an accelerated expansion warning is triggered. The first threshold and the second threshold are dynamically calculated by a product function of the aggregate angularity index and the asphalt viscosity value.

8. The heavy traffic asphalt pavement damage evolution prediction system according to claim 4, characterized in that, The gradient feature extraction of the fractal gradient enhancement module includes: The first gradient of the crack branch length change rate is calculated using an adaptive step difference method; The second gradient feature is extracted, and the mutation inflection point determination condition is set to the second gradient > 0.03 / mm²; The gradient features are mapped to a high-dimensional space, and the embedding dimension is dynamically expanded according to the damage complexity. 9.The heavy traffic asphalt pavement damage evolution prediction system of claim 5, wherein, The calculation of the elastoviscous phase consistency loss function includes: Stress wave phase delay data is obtained through ultrasonic flaw detection test, and the sampling frequency is set to 10MHz to match the microcrack scale; The phase difference predicted by the transferred fractal features is calculated, and the complex domain cross-correlation coefficient is used to verify the consistency; The relaxation time parameter of the elastoviscous constitutive equation is combined, the weighting coefficient is adjusted according to the temperature gradient, and the elastoviscous phase consistency loss function is used to quantify the elastoviscous change of the asphalt material.

10. The heavy traffic asphalt pavement damage evolution prediction system of claim 6, wherein, The dynamic adjustment rules of the waveform asymmetry factor α include: When the coverage of the wheel track band is less than 60%, α = 1.0; When the coverage is between 60% and 80%, α = 1.0 + 0.02 × (coverage-60%); When the coverage is greater than 80%, α = 1.4 + 0.1 × ln(coverage-79%).

Images