A road repair state monitoring method and system based on edge computing

Abstract

The present application relates to the technical field of road repair state monitoring, and particularly relates to a road repair state monitoring method and system based on edge computing. The method comprises the following steps: acquiring planning information of a repaired road, and deploying 3D laser scanning vibration meters at multiple repair points to collect road surface vibration signals at different time periods, perform unordered distribution time domain feature analysis, and obtain an unordered distribution time domain graph of the vibration signals; then, based on the graph, structure weakening behavior simulation is performed, periodic structure weakening data is quantified, and road surface water damage cumulative gradient is obtained through water damage cumulative gradient estimation; finally, the repair state is evaluated in terms of service life by combining the periodic structure weakening data and the water damage cumulative gradient, and the evaluation result is sent to a terminal. The present application optimizes the road repair state monitoring technology, and makes the road repair state monitoring technology more perfect.

Description

Technical Field

[0001] This invention relates to the field of road repair status monitoring technology, and in particular to a road repair status monitoring method and system based on edge computing. Background Technology

[0002] Intelligent sensing-based road repair status monitoring methods are gradually becoming an important means to improve road management efficiency. Edge computing, as an emerging computing paradigm, can process and analyze data at the source, near the data source, avoiding the latency and bandwidth pressure caused by large-scale data transmission to remote servers, thus achieving efficient, low-latency real-time monitoring and early warning. Combining edge computing with road vibration monitoring technology, various sensors deployed in the road repair area (such as 3D laser scanning vibrometers and accelerometers) can collect data on road vibration signals, deformation, and environmental changes in real time. After processing and preliminary analysis by edge computing nodes, this data can be quickly fed back to relevant management personnel to help assess the effectiveness and quality of road repair in real time, promptly identify potential problems, and implement effective intervention and repair. However, traditional edge computing-based road repair status monitoring methods suffer from inaccurate analysis of road loss status, resulting in large errors in the assessment of road repair life. Summary of the Invention

[0003] Therefore, it is necessary to provide a road repair status monitoring method and system based on edge computing to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a road repair status monitoring method based on edge computing is provided, the method comprising the following steps:

[0005] Step S1: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information; deploy 3D laser scanning vibration meters at the locations of multiple repaired road defects to collect road surface vibration signals during multiple traffic periods, and then perform disordered distribution time-domain feature analysis to obtain a disordered distribution time-domain map of vibration signals.

[0006] Step S2: Based on the disordered distribution time-domain diagram of vibration signals, simulate and quantify the structural weakening behavior during the service cycle to obtain periodic structural weakening data; estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the road surface.

[0007] Step S3: Based on the reinforcement learning algorithm in the edge computing platform, assess the repair status service life of the periodic structure weakening data and the cumulative gradient of pavement water damage to obtain repair status service life assessment data; send the repair status service life assessment data to the terminal.

[0008] Preferably, step S1 includes the following steps:

[0009] Step S11: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information;

[0010] Step S12: Deploy 3D laser scanning vibration meters at multiple repaired road defect locations to collect road surface vibration signals during multiple traffic periods;

[0011] Step S13: Perform average filtering on the road vibration signal to obtain the road vibration filtered signal;

[0012] Step S14: Perform time-domain feature analysis on the disordered distribution of the road vibration filter signal to obtain the time-domain diagram of the disordered distribution of the vibration signal.

[0013] Preferably, step S2 includes the following steps:

[0014] Step S21: Extract quality design requirements for multiple repaired road defect locations based on the repair planning information of the repaired roads, and obtain the quality repair design requirements for multiple defect locations;

[0015] Step S22: Perform vertical load intensity transformation on the disordered distribution time domain diagram of the vibration signal to obtain vibration vertical load intensity data;

[0016] Step S23: Based on the vibration vertical load intensity data, simulate and quantify the structural weakening behavior during the service life of the quality repair design requirements for multiple defect locations to obtain periodic structural weakening data.

[0017] Step S24: Estimate the cumulative gradient of water loss based on the periodic structure weakening data and vibration vertical load intensity data, thereby obtaining the cumulative gradient of water loss on the road surface.

[0018] Preferably, step S22 includes the following steps:

[0019] Step S221: Analyze the amplitude change rate of the disordered distribution time-domain graph of the vibration signal to obtain the disordered amplitude change rate;

[0020] Step S222: Based on the disordered amplitude change rate, perform an approximate linear correlation analysis on the time-domain graph of the disordered distribution of the vibration signal from low to high to obtain approximate correlation data of the slope increment;

[0021] Step S223: Perform multi-scale spectral entropy decomposition on the disordered distribution time-domain map of the vibration signal based on the slope increment approximate correlation data to obtain time-series energy spectral entropy decomposition data;

[0022] Step S224: Perform equal-quantity vertical load mechanical state mapping on the time-series energy spectrum entropy decomposition data to obtain the vertical load mechanical state;

[0023] Step S225: Perform vertical load strength conversion based on the vertical load mechanical state to obtain vibration vertical load strength data.

[0024] Preferably, step S23 includes the following steps:

[0025] Step S231: Extract the subgrade bearing capacity, density, and thickness from the quality repair design requirements for multiple defect locations;

[0026] Step S232: Perform nonlinear effect coupling on the vibration vertical load intensity data to generate load intensity nonlinear coupling data;

[0027] Step S233: Based on the nonlinear coupling data of load strength, perform a simulation analysis of the mechanical state of base course shear deformation to obtain the mechanical state of base course shear deformation.

[0028] Step S234: Analyze the degree of adhesion failure at the joint caused by the mechanical state of shear deformation of the base layer;

[0029] Step S235: Based on the aforementioned mechanical state of the base layer shear deformation and the degree of adhesive failure at the joint, the structural weakening behavior during the cycle is simulated and quantified to obtain periodic structural weakening data.

[0030] Preferably, step S234 includes the following steps:

[0031] Obtain the basic properties of the repair material; perform interfacial adhesion strength analysis on the basic properties of the repair material to obtain the interfacial adhesion strength of the material;

[0032] Based on the mechanical state of shear deformation of the base layer, an oblique shear stress analysis is performed to generate the oblique shear stress intensity of the base layer;

[0033] The simulated inclined shear stress intensity of the coupled base layer exhibits a gradient decay along the joint depth;

[0034] Based on the gradient decay state, the surface interface slip pattern at the material joint is correlated and identified to obtain the surface interface slip pattern.

[0035] The bonding failure at the joint was analyzed based on the surface interface slip mode to obtain the degree of bonding failure at the joint.

[0036] Preferably, step S24 includes the following steps:

[0037] Step S241: Based on the vibration vertical load intensity data, perform multi-scale base layer crack evolution structure simulation on the periodic structure weakening data to obtain the base layer crack evolution structure.

[0038] Step S242: Perform an equal mapping of the maximum water intrusion value to the base layer crack evolution structure to obtain the maximum water intrusion value corresponding to the base layer crack evolution structure;

[0039] Step S243: Based on the maximum value of moisture intrusion and the evolution structure of base layer cracks, perform base layer expansion / softening coupling effect analysis to obtain base layer expansion / softening coupling effect data;

[0040] Step S244: Based on the data of maximum water intrusion, base layer crack evolution structure and base layer expansion / softening coupling effect, estimate the cumulative water loss gradient to obtain the pavement water loss gradient.

[0041] Preferably, step S3 includes the following steps:

[0042] Step S31: Normalize the periodic structure weakening data and the cumulative gradient of pavement water loss respectively to obtain the normalized data of periodic structure weakening and the normalized gradient of pavement water loss.

[0043] Step S32: Based on the reinforcement learning algorithm in the edge computing platform, the repair status service life is evaluated by the periodic structure weakened normalized data and the pavement water loss cumulative normalized gradient, and the repair status service life evaluation data is obtained.

[0044] Step S33: Send the repair status lifespan assessment data to the terminal.

[0045] Preferably, the present invention also provides a road repair status monitoring system based on edge computing, used to execute the road repair status monitoring method based on edge computing as described above, the road repair status monitoring system based on edge computing includes:

[0046] The vibration signal analysis module is used to acquire road repair planning information that has been repaired; extract the locations of multiple road defects that have been repaired from the planning information; deploy 3D laser scanning vibration meters at the locations of multiple road defects to collect road surface vibration signals during multiple traffic periods, and then perform disordered distribution time-domain feature analysis to obtain a disordered distribution time-domain map of vibration signals;

[0047] The damage analysis module is used to simulate and quantify the structural weakening behavior during the service cycle based on the disordered distribution time-domain diagram of vibration signals to obtain periodic structural weakening data; and to estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the pavement.

[0048] The repair service life assessment module is used to assess the repair service life based on the reinforcement learning algorithm provided in the edge computing platform, which is used to assess the periodic structural weakening data and the cumulative gradient of pavement water damage, and obtain the repair service life assessment data; the repair service life assessment data is then sent to the terminal.

[0049] The beneficial effects of this invention lie in its ability to accurately extract multiple defect locations in the repaired area by acquiring repair planning information of the repaired road and deploying a 3D laser scanning vibration meter to collect road surface vibration signals at different times. This method enables comprehensive monitoring of the dynamic response of the repaired road, especially the impact of road defects on traffic periods. By performing time-domain feature analysis on the disordered distribution of the collected vibration signals, a time-domain map of the disordered distribution of vibration signals is obtained, which helps to accurately identify the spatiotemporal characteristics of road vibration and the evolution trend of defects, providing an important basis for subsequent structural weakening and repair status assessment. This process improves the real-time performance and accuracy of post-repair road monitoring. Based on the time-domain map of the disordered distribution of vibration signals, further simulation and quantification of structural weakening behavior over the service life are performed to obtain periodic structural weakening data. This step can quantify the weakening of the road under the influence of traffic load, climate change, and other environmental factors during long-term use. Combining the periodic structural weakening data, the cumulative gradient of water loss is estimated to obtain the cumulative gradient of road surface water loss. This process can effectively reflect the changes in road durability under water erosion, further reveal the cumulative effect of water damage on repaired roads, help to detect potential water damage problems early, take timely repair measures, and extend the service life of the road. On an edge computing platform, reinforcement learning algorithms are used to assess the repair status and lifespan of roads based on periodically weakened data and water loss accumulation gradients. Reinforcement learning can autonomously learn and optimize the repair status assessment model, accurately predicting the long-term performance of roads under different environmental and traffic conditions. This process not only improves the accuracy of repair status assessment but also allows the assessment process to be gradually optimized with data accumulation, providing dynamic support for road repair decisions. The assessment results are ultimately fed back to management personnel in real time through terminal devices, ensuring that road management can promptly grasp the repair effects and lifespan, providing a scientific basis for future road maintenance and resource allocation, reducing maintenance costs, and improving the overall efficiency of road management. Therefore, this invention is an optimization of a traditional edge computing-based road repair status monitoring method, solving the problem that traditional edge computing-based road repair status monitoring methods have inaccurate analysis of road loss status, resulting in large errors in road repair status and lifespan assessment. It improves the accuracy of road loss status analysis and reduces the error in road repair status and lifespan assessment. Attached Figure Description

[0050] Figure 1 This is a schematic diagram illustrating the steps of a road repair status monitoring method based on edge computing;

[0051] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0052] Figure 3 for Figure 1 A detailed flowchart illustrating the implementation steps of step S3. Detailed Implementation

[0053] Please see Figures 1 to 3 A road repair status monitoring method based on edge computing, the method comprising the following steps:

[0054] Step S1: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information; deploy 3D laser scanning vibration meters at the locations of multiple repaired road defects to collect road surface vibration signals during multiple traffic periods, and then perform disordered distribution time-domain feature analysis to obtain a disordered distribution time-domain map of vibration signals.

[0055] Step S2: Based on the disordered distribution time-domain diagram of vibration signals, simulate and quantify the structural weakening behavior during the service cycle to obtain periodic structural weakening data; estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the road surface.

[0056] Step S3: Based on the reinforcement learning algorithm in the edge computing platform, assess the repair status service life of the periodic structure weakening data and the cumulative gradient of pavement water damage to obtain repair status service life assessment data; send the repair status service life assessment data to the terminal.

[0057] In this embodiment of the invention, reference is made to Figure 1 The diagram shown illustrates the steps of a road repair status monitoring method based on edge computing according to the present invention. In this example, the road repair status monitoring method based on edge computing includes the following steps:

[0058] Step S1: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information; deploy 3D laser scanning vibration meters at the locations of multiple repaired road defects to collect road surface vibration signals during multiple traffic periods, and then perform disordered distribution time-domain feature analysis to obtain a disordered distribution time-domain map of vibration signals.

[0059] In this embodiment of the invention, repair planning information in GIS format is first extracted from the road repair completion archives issued by the municipal road maintenance unit. This information should include structural parameters such as the start and end latitude and longitude of the road repair, construction repair time, defect type, repair process type, repair thickness, and structural layer material type. After loading the repair rule information layer in ArcGIS 10.7, multiple independent defect repair sections are extracted by the road chain node number, and their geographical location information is exported as CSV format for subsequent equipment deployment. RIEGL VZ-400i 3D laser scanning vibration meters are deployed in multiple defect repair sections. Anti-vibration tripods are set up at the deployment points, and the level is used to ensure the stability of the equipment. The 3D laser scanning vibration meters are fixed to the ground with high-strength spiral expansion bolts to prevent displacement or vibration interference during the measurement process. Each 3D laser scanning vibration meter needs to be angled so that it is perpendicular to the road surface and the optical axis points to the main direction of vehicle operation. After initial alignment using a built-in laser collimator and IMU sensor, the angle is fine-tuned to the optimal level by real-time feedback of the echo intensity of the acquired signal. To achieve three-dimensional spatial uniformity of multi-point data, GNSS reference points are set up at both ends of each repair section before deployment, and coordinate calibration is performed using a total station or RTK to record the precise three-dimensional coordinates of the vibration meter deployment points. The device has a spatial resolution of 5 mm and a sampling frequency of 5000 Hz. Road surface vibration signal data is continuously collected for 30 minutes each during three traffic periods: 7:00-9:00 AM, 12:00-2:00 PM, and 5:00-7:00 PM. The raw vibration data is stored locally as a three-dimensional acceleration sequence in the XYZ directions. Each acquisition is completed... The data was then uploaded to a local edge server via a high-speed LAN interface for preprocessing. During the preprocessing stage, a Butterworth low-pass filter was used to filter the signal with a frequency cutoff of 100 Hz to remove high-frequency noise. After that, a non-uniform sampling time window was constructed using the NumPy library in the Python environment. Each window was 1 second wide and had a step size of 0.2 seconds. The signal segments within each window were normalized, and then the disorder feature value was extracted using the local maximum difference sequence. Finally, Matplotlib was used to plot the disorder distribution time domain diagram of the vibration signal. This diagram uses the time axis as the horizontal axis and the disorder index as the vertical axis to characterize the irregular distribution characteristics of the vibration response under traffic load.

[0060] Step S2: Based on the disordered distribution time-domain diagram of vibration signals, simulate and quantify the structural weakening behavior during the service cycle to obtain periodic structural weakening data; estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the road surface.

[0061] In this embodiment of the invention, based on the obtained disordered distribution time-domain plot of the vibration signal, wavelet packet decomposition is used in the MATLAB R2021a environment to extract the energy distribution of different frequency bands. Five sub-frequency bands are selected within the frequency range of 5 Hz to 50 Hz. The rate of change of energy density in each sub-frequency band is statistically analyzed, and the periodic fitting residual of its change trend is calculated. The specific formula for calculating the energy density change is as follows: ,in, This represents the energy density of the j-th sub-band at time t; Indicates the number of sampling points. Indicates the index of the sampling point, ranging from 1 to N. The index represents the sub-band, ranging from 1 to 5 (5 sub-bands were selected). This represents the amplitude of the j-th sub-band at the i-th sampling point; the variance of the periodic fitting residual serves as the structural periodic weakening index. This index is calculated separately at each defect repair location, forming periodic structural weakening data. The data structure includes fields such as location number, frequency band number, residual variance value, and timestamp. Next, in the Python environment, the inverse distance weighted interpolation method is used to perform spatial interpolation processing on the periodic structural weakening data to construct a complete spatial heat map of road segment structural weakening. Based on this map, the cumulative gradient of water loss is further estimated. The water loss estimation adopts a joint analysis method of geotechnical parameters and structural response. In the section where the structural layer type is cement-stabilized crushed stone layer, the permeability coefficient of this structural layer is set to be... Meters per second, and a water accumulation function is constructed based on the joint rate of change of residual variance and time, where the function expression is: ,in, This represents the cumulative water content at location x at time t; This indicates the initial moisture content, with a range of 3-5%. The residual variance-moisture content conversion factor is an empirical value. ; Indicates the time accumulation coefficient. This represents the permeability correction factor, which is related to the material's permeability coefficient (1.0 × 10⁻⁻⁴). 5 Calibration coefficients related to (m / s) The derivative of the thickness influence index reflects the nonlinear effect of thickness on moisture accumulation. α represents the coefficient of moisture migration efficiency, reflecting the material's responsiveness to moisture flow, which is related to soil physical properties and environmental conditions (such as rainfall or evaporation). This function is used to inversely deduce the moisture content change trend at different time points. Then, combined with the structural layer thickness and drainage slope parameters, a layer-by-layer moisture accumulation distribution model is constructed. Finally, the water loss accumulation gradient data of each defect location under different service cycles is obtained. The data structure includes parameter fields such as spatial location number, time point, layer moisture content, and total moisture gradient.

[0062] Step S3: Based on the reinforcement learning algorithm in the edge computing platform, assess the repair status service life of the periodic structure weakening data and the cumulative gradient of pavement water damage to obtain repair status service life assessment data; send the repair status service life assessment data to the terminal.

[0063] In this embodiment of the invention, the periodic structural weakening data and pavement water loss cumulative gradient data are first normalized using the Pandas library. The normalization method adopts the maximum-minimum standardization method, compressing each data set to the interval between 0 and 1. The normalized data is input into the reinforcement learning analysis module deployed on the edge server in the form of a CSV file. This module is built based on TensorFlow and uses the DQN reinforcement learning algorithm to evaluate the service life of the road structure. The state space is defined as the normalized periodic weakening value and water loss gradient value at each defect location. The action space is set as the state adjustment strategy with repair levels from 1 to 5. The reward function is designed to minimize the pre-repair level. The difference between the measured repair life and the actual monitored life is obtained from historical working condition backtesting analysis. In each training round, the number of iterations is set to 1000, the learning rate is set to 0.01, and the discount factor is set to 0.9. After training, the trained policy network is used to perform batch inference output on all defective sections. The predicted output is the number of remaining service cycles for each section in days. This output data is the repair status life assessment data. Finally, the assessment data is packaged into JSON format and uploaded to the municipal road management terminal through a 4G communication module. The terminal's visualization interface displays the life data of each section in the form of heat map and time series graph for subsequent road maintenance decision-making.

[0064] Step S1 includes the following steps:

[0065] Step S11: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information;

[0066] Step S12: Deploy 3D laser scanning vibration meters at multiple repaired road defect locations to collect road surface vibration signals during multiple traffic periods;

[0067] Step S13: Perform average filtering on the road vibration signal to obtain the road vibration filtered signal;

[0068] Step S14: Perform time-domain feature analysis on the disordered distribution of the road vibration filter signal to obtain the time-domain diagram of the disordered distribution of the vibration signal.

[0069] In this embodiment of the invention, road repair planning information is extracted from the "Road Repair Project Completion Information Registration Form" authorized and archived by the road maintenance department and the supporting road management information system. This information is exported in XML format and includes road name, starting station number, ending station number, defect type number, repair time, structural layer type, repair method code, and structural thickness parameters. The lxml library in Python is used to parse the fields of this XML file, and regular expressions are used to identify all "structural repair completed" identifier fields and locate the starting and ending station numbers of the corresponding road sections. Then, it is converted into latitude and longitude format of WGS84 coordinate system and exported in CSV format to guide the positioning and deployment of vibration measurement equipment. Each extracted defect location must have a clear structural type label, such as asphalt concrete or cement-stabilized crushed stone, which serves as a prerequisite for structural response analysis in subsequent steps.

[0070] A RIEGL VZ-400i 3D laser scanning vibration meter was deployed at the identified defect locations. This device has a spatial sampling frequency of 5000 Hz and a positioning error of less than 5 mm. Before deployment, the coordinates of the points were calibrated twice using an RTK GPS receiver to ensure that the deployment accuracy was no less than 0.2 meters. The data collection period was set to 30 minutes of continuous collection during each traffic peak period, with three time periods collected daily: 7:00-9:00, 12:00-14:00, and 17:00-19:00. The raw data was stored in the form of a three-axis acceleration time series in TXT format, with 5000 acceleration values ​​collected per second. Data transmission was carried out via a wired LAN connection to an edge server located on-site for preprocessing. Data for each defect point was collected for at least 3 consecutive days to ensure that the data covered the complete short-term traffic disturbance characteristics in the time domain. After the data collection was completed, the raw data was named and archived according to the rule of "defect number_date_time period".

[0071] The original vibration signal was averaged and filtered using a Python data processing script deployed on an edge server. The filtering operation used a sliding window averaging method with a length of 50, that is, averaging and smoothing was performed once every 50 consecutive sampling points. The sliding window step size was set to 1. The processed acceleration signal removed sudden high-frequency noise from the original data while retaining the true response of the main vibration frequency band. The processed data was stored in a structured JSON format, including fields such as timestamp, triaxial filtered acceleration value, and the corresponding number of the original file. All time period data corresponding to each defect location were saved separately and marked as filtered versions to distinguish the signal source used in subsequent processing steps. No frequency domain processing such as Fourier transform or wavelet transform was performed in this process to ensure the integrity of the time domain signal. All data was processed on the local edge node without the need for remote cloud transmission. First, a time window was constructed for the filtered vibration signal in NumPy, with a window length of 1 second, corresponding to 5000 sets of data. The window sliding step was set to 0.2 seconds. Within each window, the standard deviation of the difference between the local maximum peaks was used as an indicator to describe the disorder. Local peaks were extracted using the find_peaks function in SciPy, and the minimum peak height was limited to the mean plus 0.5 times the standard deviation to remove weak disturbances. One disorder index value was obtained for each time window, and they were arranged in chronological order to form a disorder degree sequence. Then, Matplotlib was used to plot the disorder distribution in the time domain, with the horizontal axis representing time in seconds and the vertical axis representing the disorder index value of the corresponding window.

[0072] Step S2 includes the following steps:

[0073] Step S21: Extract quality design requirements for multiple repaired road defect locations based on the repair planning information of the repaired roads, and obtain the quality repair design requirements for multiple defect locations;

[0074] Step S22: Perform vertical load intensity transformation on the disordered distribution time domain diagram of the vibration signal to obtain vibration vertical load intensity data;

[0075] Step S23: Based on the vibration vertical load intensity data, simulate and quantify the structural weakening behavior during the service life of the quality repair design requirements for multiple defect locations to obtain periodic structural weakening data.

[0076] Step S24: Estimate the cumulative gradient of water loss based on the periodic structure weakening data and vibration vertical load intensity data, thereby obtaining the cumulative gradient of water loss on the road surface.

[0077] As an example of the present invention, reference is made to... Figure 2 As shown, in this example, step S2 includes:

[0078] Step S21: Extract quality design requirements for multiple repaired road defect locations based on the repair planning information of the repaired roads, and obtain the quality repair design requirements for multiple defect locations;

[0079] In this embodiment of the invention, road repair records with a repair completion date of no more than 180 days are retrieved from the road facility archive database. All project records with defect type identification codes are extracted from the search results. Construction design data is located based on the repair process field and the structural design document number. Contents from the structural layer design specification table and construction quality standard template are called, and the quality repair design requirements for each defect location are determined. These requirements include the total thickness of the structural layer, the thickness of each structural layer, the standard value of the subgrade bearing capacity, the target values ​​of the compaction degree of the surface layer and base layer, the standard value of the joint bonding strength, and the design axle load passage limit. During the extraction process, a Python dictionary mapping table based on field mapping rules is used to correspond the repair method code to the structural parameter fields one-to-one. Taking cement-stabilized crushed stone base layer as an example, its design requirements include a base layer thickness ≥ 180 mm, compaction degree ≥ 98%, unconfined compressive strength ≥ 2 MPa, and interlayer bond strength ≥ 0.3 MPa. This type of structure is commonly used in urban main roads or heavy-duty logistics channels. The quality repair design requirements for each defect section are ultimately structured into an array data format and cached in edge nodes for subsequent data comparison and analysis.

[0080] Step S22: Perform vertical load intensity transformation on the disordered distribution time domain diagram of the vibration signal to obtain vibration vertical load intensity data;

[0081] In this embodiment of the invention, the disordered distribution time-domain image of the vibration signal obtained in the previous stage is deconstructed frame by frame, and the timestamp and disorder value sequence corresponding to each frame image are extracted. The OpenCV library in Python is used to extract pixels from the curve trajectory in the image. The number of sampling points for each curve is set to 100 points at equal intervals, and the sampling frequency is 50Hz. Then, the amplitude change rate calculation method is adopted, taking the absolute value of the change in the vertical coordinate of each two adjacent points divided by the time interval of the horizontal coordinate (0.02 seconds) as the amplitude change rate per unit time. After constructing the change rate sequence of all points, a smoothing process is performed using a moving average window with a window length of 5 points (corresponding to a 0.1-second time window). Then, an approximate slope increment analysis is performed, selecting every 10 consecutive points (0.2 seconds) to fit a linear trend line and recording its slope change increment. The resulting slope sequence is normalized in the [0,1] interval and used as input parameters for multi-scale spectral entropy decomposition. EMD algorithm was used for intrinsic mode decomposition. The screening stopping criterion was set as an SD value less than 0.3. The energy spectra of the first to third modes were selected and their average spectral entropy values ​​were calculated. The average spectral entropy value was used as an indicator of the vibration energy disturbance level. Next, an equal mechanical state mapping operation was performed to divide the spectral entropy value range into 5 levels: 0-0.2 corresponds to load level 1 (10kN), 0.2-0.4 corresponds to load level 2 (25kN), 0.4-0.6 corresponds to load level 3 (40kN), 0.6-0.8 corresponds to load level 4 (55kN), and 0.8-1.0 corresponds to load level 5 (70kN). A mapping reference table extracted from a dataset based on historical traffic test site loading tests (containing 2000 sets of calibration samples) was used to map the average spectral entropy value to the load level one by one. Finally, the vibration vertical load intensity data corresponding to each time point was obtained. The data update frequency was 10Hz, and the defect location coordinates were labeled.

[0082] Step S23: Based on the vibration vertical load intensity data, simulate and quantify the structural weakening behavior during the service life of the quality repair design requirements for multiple defect locations to obtain periodic structural weakening data.

[0083] In this embodiment of the invention, in step S23, the vibration vertical load intensity data is organized into a three-dimensional data matrix according to spatial location and timestamp. The matrix size is the number of time periods (48-hour time periods) × the number of measurement points (15-25 points per segment) × the load intensity value (range 10-70kN). The data matrix is ​​subjected to time-frequency analysis using Hilbert-Huang transform. During the transform process, empirical mode decomposition technology is used (stopping condition is residual less than 0.05 or the number of iterations reaches 12) to extract the instantaneous energy curves of each mode. The sampling frequency of the instantaneous energy curves is set to 200 Hz, the window width is 5 seconds, and the overlap rate is 75%. Nonlinear fitting (Levonburg-Marquardt algorithm, convergence threshold 0.001) is performed on the instantaneous energy and structural parameters to calculate the elastic modulus attenuation rate. The parameters of the attenuation rate formula are set as follows: initial elastic modulus 300-500MPa, attenuation coefficient 0.02-0.08, and power exponent 0.35-0.65. The cumulative shear strain curve is calculated using the fourth-order Runge-Kutta numerical integration method (step size 0.001 seconds). The portion of the cumulative shear strain exceeding the 0.6% threshold is defined as the structural weakening index value, and a structural weakening data table is generated. The usage cycle refers to one cycle in actual use after the road repair is completed. The collection cycle described in the embodiment of step S12 is set to continuously collect data for 30 minutes during each traffic peak period, collecting data in three time periods per day: 7:00 to 9:00, 12:00 to 14:00, and 17:00 to 19:00, for a period of 3 consecutive days.

[0084] Step S24: Estimate the cumulative gradient of water loss based on the periodic structure weakening data and vibration vertical load intensity data, thereby obtaining the cumulative gradient of water loss on the road surface.

[0085] In this embodiment of the invention, based on the obtained periodic structure weakening data and vibration vertical load intensity data, a two-dimensional finite difference mesh is constructed for a typical structural layer profile using the PDE toolbox in MATLAB. The mesh size is set to 5 mm × 5 mm. The structural profile is divided into four parts: asphalt surface layer, cement-stabilized crushed stone base layer, gravel cushion layer, and undisturbed soil layer. Shear modulus attenuation parameters are applied to the central nodes of the base layer according to the periodic structure weakening data. A concentrated load simulation is applied at the time point corresponding to the vibration vertical load intensity. The load application range is set to a 0.3 m × 0.3 m square area, and the vertical load intensity is taken at the corresponding position. Estimated values ​​were set, and moisture diffusion analysis was performed using Fick's second law under a simulation period of 72 hours. The initial moisture content was set to 8%, and the critical moisture content to 12%. A permeability boundary condition was applied to the bottom of the structure. Moisture gradient distribution data within the structural layer were output at each time period. After integrating the moisture gradient growth rate at each time period, a cumulative moisture intrusion layer was constructed. The data at each grid point in this layer is the cumulative water loss per unit area. Then, the cumulative water loss gradient value at each defect location was obtained by integrating the grid across the entire section. This data was recorded as pavement water loss cumulative gradient data to support subsequent service life assessment tasks.

[0086] Step S22 includes the following steps:

[0087] Step S221: Analyze the amplitude change rate of the disordered distribution time-domain graph of the vibration signal to obtain the disordered amplitude change rate;

[0088] Step S222: Based on the disordered amplitude change rate, perform an approximate linear correlation analysis on the time-domain graph of the disordered distribution of the vibration signal from low to high to obtain approximate correlation data of the slope increment;

[0089] Step S223: Perform multi-scale spectral entropy decomposition on the disordered distribution time-domain map of the vibration signal based on the slope increment approximate correlation data to obtain time-series energy spectral entropy decomposition data;

[0090] Step S224: Perform equal-quantity vertical load mechanical state mapping on the time-series energy spectrum entropy decomposition data to obtain the vertical load mechanical state;

[0091] Step S225: Perform vertical load strength conversion based on the vertical load mechanical state to obtain vibration vertical load strength data.

[0092] In this embodiment of the invention, the disordered distribution time-domain image data of the vibration signal corresponding to each defect location is first read in the edge computing platform. This image data comes from the disordered degree sequence image extracted from the three-dimensional acceleration signal after filtering and window decomposition in the previous stage. Each frame of the image corresponds to the vibration state change curve of a time window. The image is converted into two-dimensional coordinate data using the image grayscale extraction method in OpenCV. The horizontal axis is the time index point, and the vertical axis is the disorder value. Then, the amplitude change rate per unit time is calculated by dividing the difference of the vertical axis of adjacent points by the horizontal axis step size. The horizontal axis step size is 0.2 seconds, and the sample interval is fixed. After performing this operation on all points, a complete amplitude change rate sequence is formed. The amplitude change rate, as the speed index of disordered degree change, reflects the intensity of short-term vibration energy disturbance under traffic load. In this process, the NumPy library is used to perform differential operations and uniform standardization. The standardization adopts the range normalization method to compress all change rate values ​​to between 0 and 1. The resulting data is called disordered amplitude change rate data. This processing flow is executed independently for each defect location and each collection period. The obtained data is used as the input basis for the next slope increment analysis.

[0093] The processing flow of step S222 is as follows: The disordered amplitude change rate data obtained in step S221 is called, and the data points are divided into segments of 100. Within each segment, a linear regression curve is fitted using the least squares method to obtain the fitted slope value for each segment. After extracting the slope for all data segments, a continuous slope sequence is formed. Then, the difference operation between adjacent segments is performed on this slope sequence to obtain the incremental slope data, which represents the linear transition change characteristics of the vibration disturbance response. Subsequently, the incremental slope data is subjected to trend correlation detection using Pearson correlation. The numerical method performs correlation analysis between slope increment data and time period numbers, selects sequence segments with correlation coefficients greater than 0.6 as significantly abruptly correlated segments, and marks their start and end time window numbers. Further, local slope trend fitting is performed on these segments to form approximate correlation data of slope increments. This data structure includes fields such as time period number, starting slope, ending slope, average increment value, and correlation coefficient value. The results reflect the temporal relationship between the nonlinear intensification trend of vibration signals and the peak change of traffic load, and provide basic variable inputs for spectral entropy decomposition to estimate the degree of energy disturbance aggregation.

[0094] Based on the obtained slope increment approximate correlation data, multi-scale spectral entropy decomposition processing is performed on the disordered distribution time-domain graph of the corresponding vibration signal. The processing method uses the Empirical Mode Decomposition (EMD) algorithm. The amplitude change rate sequence of each time period is decomposed by calling the PyEMD module in Python. The number of decomposition layers is set to 3, that is, the first to third order intrinsic mode functions IMF1 to IMF3 are extracted, which represent high-frequency short-period energy disturbance, mid-frequency dynamic change, and low-frequency long-term trend, respectively. The energy spectral density function of each order IMF function is obtained by short-time Fourier transform. After performing probability normalization processing on the function, its spectral entropy value is calculated as spectral entropy data. The higher the obtained spectral entropy value, the more uniform the energy distribution and the stronger the irregularity of the disturbance, reflecting the degree of dispersion of traffic disturbance in time. The spectral entropy values ​​of all time periods are integrated to form complete time-series energy spectral entropy decomposition data. The data structure includes fields such as time number, IMF order, energy density, spectral entropy value, and frequency domain dominant frequency information. This process is repeated for each defect location, and the result data is passed to the next stage of mechanical mapping analysis.

[0095] Based on time-series energy spectral entropy decomposition data, a vertical load mechanical state mapping rule table was established. This rule table establishes a mapping relationship between the typical spectral entropy response of the pavement structure under traffic disturbance loading and the vertical load mechanical response obtained from in-situ platen loading tests. During the data acquisition period, standard platen loading tests were conducted in areas with the same defect type. The tests used a 300 mm diameter circular steel plate to apply concentrated loads at loading levels of 20 kN, 40 kN, 60 kN, and 80 kN, each lasting 10 seconds. Acceleration response signals were recorded simultaneously, and spectral entropy data were constructed. Finally, the corresponding values ​​for each loading level were statistically analyzed. The distribution range of spectral entropy values ​​is defined. For example, a spectral entropy value between 0.1 and 0.2 corresponds to a low disturbance and low load state, i.e., the 20kN level, while a spectral entropy value between 0.6 and 0.8 corresponds to a 60kN level disturbance. The mapping process from spectral entropy values ​​to vertical load mechanical states is achieved through table lookup operations. In actual data processing, each spectral entropy data is classified and labeled according to its corresponding range. The labeling result is the vertical load mechanical state value. The types are divided into 5 categories: extremely low load, low load, medium load, high load, and extremely high load. A corresponding state number sequence is generated for each time period and used as input data for vertical load intensity conversion.

[0096] The processing flow of step S225 is as follows: Based on the marked vertical load mechanical states, a vertical load strength conversion operation is performed. The conversion process uses a standard load level calibration table as the basic mapping basis. The calibration table is derived from the standard loading test results of highway engineering. Each mechanical state number has a one-to-one mapping relationship with the actual equivalent static load strength. Among them, state number 1 corresponds to an equivalent vertical load of 15kN, state number 2 corresponds to 30kN, state number 3 corresponds to 45kN, state number 4 corresponds to 60kN, and state number 5 corresponds to 75kN. In data processing, the mechanical state number of each time period is mapped to its vertical load strength value according to the calibration table. At the same time, combined with the actual time span of the disordered distribution time domain diagram, each load strength value is bound to a specific time period to achieve accurate numerical characterization of the mechanical action intensity generated by traffic disturbance in the vibration signal. Finally, a complete vibration vertical load strength data sequence is generated. The sequence structure includes fields such as time number, location number, mechanical state number, and equivalent vertical load strength value, which are used as boundary loading inputs in subsequent structural weakening and water loss simulation.

[0097] In another embodiment, in step S221, the time-domain image of the disordered distribution of the vibration signal is extracted using image processing technology. First, the Carney edge detection algorithm (low threshold 75, high threshold 150) is used to extract the vibration curve in the time-domain image, and the extracted image resolution is 1920×1080 pixels. After curve extraction, pixel coordinates are converted to physical quantities, with the horizontal axis mapped to time (accuracy 0.001 seconds) and the vertical axis mapped to disorder value (range 0-1). The curve is discretized using an equidistant resampling method, with the sampling interval set to 0.01 seconds, and each curve is resampled to a fixed 100 points. The amplitude change rate is calculated for the resampled point sequence, and the calculation formula is the absolute value of the difference in disorder value between two adjacent points divided by the time interval: amplitude change rate α = |disorder value (i+1) - disorder value (i)| / (time (i+1) - time (i)). The calculated amplitude change rate sequence was smoothed using a Hanning window function (window length 5 points, center weight 1.0, edge weight 0.25). The processed data was saved as an amplitude change rate matrix with dimensions [number of sampling points, number of defect locations].

[0098] In step S222, the unordered amplitude change rate data is first sorted according to amplitude magnitude using a quicksort algorithm with a time complexity of O(n log n). The sorted data is divided into 10 equally divided intervals, each containing an equal number of data points. A least-squares method is used to perform linear fitting on the data points within each interval, with the fitting formula being amplitude change rate = k × time + b, where k is the slope and b is the intercept. The convergence condition for linear fitting is set as the sum of squared residuals being less than 0.001 or the number of iterations reaching 50. The difference in slope values ​​between adjacent intervals is calculated to obtain slope increment data. A correspondence is established between the slope increment data and the original time series, forming a two-dimensional data table of time-slope increment. Each time point in the data table is associated with a slope increment value, reflecting the change in acceleration of the vibration signal at that moment. Missing values ​​in the slope increment data are filled using cubic spline interpolation, with an interpolation node interval of 0.05 seconds and natural boundary conditions (second derivative is zero).

[0099] In step S223, the slope increment approximate correlation data is used as input, and mode separation is performed using the Empirical Mode Decomposition (EMD) method. The EMD process is set to terminate when the envelope mean is less than 0.05 or the number of iterations reaches 12. Five intrinsic mode functions (IMFs) are obtained, each reflecting signal characteristics at different frequency scales. For each IMF, the instantaneous frequency and instantaneous amplitude are calculated using a Hilbert transform, implemented using a Fast Fourier Transform, with a sampling frequency of 200 Hz. The instantaneous energy is calculated by squared the instantaneous amplitude, forming a time-energy sequence. The spectral entropy value is calculated for the energy distribution within each time window (window width 5 seconds, overlap rate 75%), using the Shannon entropy formula: Where Pi is the normalized energy of the i-th frequency interval. Multi-scale spectral entropy is formed by calculating entropy values ​​at different scales, with scale factors ranging from 1 to 5, ultimately resulting in time-series energy spectral entropy decomposition data with dimensions of [number of time windows, number of scales, number of defect locations]. The time-series energy spectral entropy decomposition data is first normalized using a maximum-minimum normalization method, ensuring the spectral entropy values ​​are distributed between 0 and 1. A mapping relationship is established between the normalized spectral entropy values ​​and the vertical load mechanical state. This mapping is based on the cumulative load test database of the road engineering test field (containing 2500 test samples), which records the road vibration spectral entropy characteristics under different loads. The mapping relationship is defined by a piecewise function: spectral entropy values ​​of 0-0.2 correspond to mechanical state 1 (elastic deformation stage), 0.2-0.4 correspond to mechanical state 2 (elastoplastic transition stage), 0.4-0.6 correspond to mechanical state 3 (early stage of plastic deformation), 0.6-0.8 correspond to mechanical state 4 (mid-stage of plastic deformation), and 0.8-1.0 correspond to mechanical state 5 (late stage of plastic deformation). The mapping employs a hard classification method, with each spectral entropy value strictly mapped to a single mechanical state, eliminating state overlap. The mapping result forms a three-dimensional matrix of time-location-mechanical state, stored in a sparse matrix format with a compression rate of 85%.

[0100] In the process of converting vertical load strength based on the vertical load mechanical state, the load mechanical state level is first converted into a reference load value. The conversion relationship is as follows: 10kN for slight load, 25kN for light load, 40kN for moderate load, 55kN for heavy load, and 70kN for extremely heavy load. The reference load value is then corrected using factors including the pavement type coefficient, temperature influence coefficient, and humidity influence coefficient. The pavement type coefficient is determined based on the pavement material: 1.0 for asphalt concrete pavement, 1.2 for cement concrete pavement, and 1.1 for composite pavement. The temperature influence coefficient is calculated based on the ambient temperature, with a reference temperature of 20 degrees Celsius. For every 10 degrees Celsius increase, the coefficient decreases by 0.05; for every 10 degrees Celsius decrease, the coefficient increases by 0.08. The humidity influence coefficient is calculated based on relative humidity, with a reference humidity of 50%. For every 10% increase, the coefficient increases by 0.03; for every 10% decrease, the coefficient decreases by 0.02. The reference load value is multiplied by three correction factors to obtain the final vibration vertical load intensity value, and then organized into a vibration vertical load intensity data table according to the timestamp and location number. The data table contains four fields: location number, timestamp, load state level, and load intensity value.

[0101] Step S23 includes the following steps:

[0102] Step S231: Extract the subgrade bearing capacity, density, and thickness from the quality repair design requirements for multiple defect locations;

[0103] Step S232: Perform nonlinear effect coupling on the vibration vertical load intensity data to generate load intensity nonlinear coupling data;

[0104] Step S233: Based on the nonlinear coupling data of load strength, perform a simulation analysis of the mechanical state of base course shear deformation to obtain the mechanical state of base course shear deformation.

[0105] Step S234: Analyze the degree of adhesion failure at the joint caused by the mechanical state of shear deformation of the base layer;

[0106] Step S235: Based on the aforementioned mechanical state of the base layer shear deformation and the degree of adhesive failure at the joint, the structural weakening behavior during the cycle is simulated and quantified to obtain periodic structural weakening data.

[0107] In this embodiment of the invention, the processing flow of step S231 is as follows: During the quality control stage after road defect repair is completed, the quality repair design requirement data corresponding to each repaired area is extracted through the construction log system and the structural layer construction record table. This data includes three types of key indicators: the design value of subgrade bearing capacity, the design control value of compaction degree, and the design value of base course thickness. The design value of subgrade bearing capacity is expressed using standard static load test data, in kPa. The design control value of compaction degree is the percentage of relative density, in %. The design value of base course thickness is the vertical structural thickness from the surface layer to the top of the base course, in mm. During the data extraction process, a unique parameter is set for each defect location. A number is assigned, and data is extracted by field-level matching with construction quality control data. In this embodiment, data for 52 defect locations are extracted. The design value of the subgrade bearing capacity is between 140kPa and 180kPa, the design control value of compaction degree is between 93% and 98%, and the design value of base course thickness is between 220mm and 280mm. A unified verification logic is executed on the data of all defect locations. A dual verification mechanism is used to remove duplicate, missing, or excessive items in the data source and complete them. Under the premise of ensuring data consistency and integrity, all data is structured and stored as a standard three-field data table for use as input for subsequent nonlinear coupling processing and mechanical simulation. The processing flow of step S232 is as follows: The vibration vertical load intensity data obtained in step S22 is called, and this data is coupled with the design control parameters extracted in step S231 using nonlinear effects. The coupling process adopts a nonlinear fitting strategy based on piecewise regression. First, the load intensity sequence at each defect location is segmented into intervals, with each 20s serving as a time window, resulting in 480 window segments. Within each segment, statistical characteristics such as the mean, peak, and range of the load intensity are extracted and a feature vector is formed. Then, the design parameters at the same numbered location are normalized to normalize the values ​​of bearing capacity, compaction degree, and thickness to the [0,1] interval. Subsequently, multivariate piecewise regression is used. The method performs nonlinear fitting between each load characteristic and design parameters, uses stepwise regression to extract the most significant influencing factors, and uses F-value ranking to determine the significance of variables. For highly correlated variable combinations, principal component analysis is further performed, and the top two principal components are selected as mapping variables that comprehensively reflect the coupling effect between load disturbance and design structure. Finally, a coupled dataset for each time period is constructed. This dataset contains fields such as window number, principal component value 1, principal component value 2, load intensity extreme and mean values, fitting residuals, and correlation coefficients. This result is the nonlinear coupling data of load intensity. The data structure uses a dual index of position number and time window for the index admission mechanism of the next step of shear deformation analysis.

[0108] The processing flow of step S233 is as follows: Based on the load intensity nonlinear coupling data generated in step S232, the mechanical state simulation of the base layer shear deformation is performed at each defect location within each time period. The simulation method uses the two-dimensional shear deformation derivation formula in the finite difference method to calculate the displacement-stress distribution. The material constitutive relation in the initial state is defined as a linear elastic model, where the design value of the subgrade bearing capacity is used as the vertical static load starting boundary, the compaction is used as the correction factor for the transverse deformation modulus, and the thickness is used as the interlayer exponent coefficient in the stress attenuation function. During the calculation, the nonlinear coupling principal component variables are applied as disturbance sources in the loading path by defining the period of the disturbance load term. The dynamic response of the disturbance deformation is controlled by the peak range. The differential step size is set to 0.05m. The calculation area is defined as a symmetrical arrangement of structural cells within a 3m×3m region, with 60×60 cells. The shear strain, shear stress, and principal strain direction changes are recorded in each cell. In particular, for the node cells near the disturbance center, the strain energy change rate is calculated as the criterion for whether they have entered the critical state of shear instability. Finally, time-series shear deformation mechanical state data is generated for each defect location. This data structure includes parameters such as time number, shear strain value, shear stress value, equivalent strain energy density, critical stability coefficient, and boundary displacement trend, which are stored in real time through edge nodes.

[0109] The processing flow of step S234 is as follows: Based on the mechanical state of the base layer shear deformation obtained in step S233, the spatial correspondence between the shear displacement abrupt change region and the known joint location is further analyzed. The spatial calibration of the joint location is achieved by matching the CAD coordinates in the repaired construction drawings with the LIDAR scan data recorded in the edge nodes. The distance threshold control method is used to determine the joint location point set. This point set extends 0.3m to both sides of each joint to form an analysis area, and a total of 68 joint areas are calibrated. Each joint area is divided into 12 equal-width profile units. The shear strain and the change in principal strain direction within each profile unit are analyzed using dual indicators. The analysis logic is that when the shear strain continuously exceeds 8‰ within 30 minutes and the cumulative change in principal strain direction exceeds 45 degrees, the profile area is determined to have a problem. In the case of potential adhesive slip, strain energy release rate is further extracted from the area with obvious slip trend and quantified using slip energy discrimination method. The threshold is set at 3.5 kJ / m². When this value continuously crosses 3 time windows and maintains an increasing trend, it is defined as moderate adhesive failure. When it exceeds 6.0 kJ / m² and fracture transfer occurs, it is defined as severe adhesive failure. All failure types are represented by adhesive failure level coding, with a coding range of 0 to 3. 0 represents no failure, 1 represents slight micro-slip, 2 represents moderate interlaminar slip, and 3 represents severe adhesive fracture. This failure level is mapped point by point to each joint profile unit to construct a two-dimensional spatial matrix, which serves as the initial condition input for the periodic structure weakening simulation. All analysis processes use strain time series and spatial location correspondence for automatic indexing to avoid human calibration errors.

[0110] The processing flow of step S235 is as follows: Using the shear deformation mechanical state obtained in step S233 and the degree of adhesive failure at the joint obtained in step S234, a quantitative analysis framework for the weakening behavior of the periodic structural response evolution is established. This framework adopts a time-layered advancement strategy, dividing the mechanical response window into 12-hour cycles. Within each cycle window, a sequence of key structural indicators is extracted as the quantitative basis. These indicators include five core data categories: maximum shear stress within the cycle, cumulative shear displacement, maximum joint failure level, spatial diffusion rate of joint failure level, and number of abrupt changes in local strain energy density. Grey relational analysis is used to extract the evolution trend of the indicator sequence during the cycle. The grey absolute correlation degree is calculated for the rate of change of indicators between any two cycles, and a threshold of 0.75 is selected as the strong correlation criterion. When the rate of change of three or more indicators exceeds this threshold consecutively between two cycles, a significant structural weakening event is determined to have occurred in that cycle. Furthermore, a periodic structural weakening index (WSI) is constructed. The WSI index is used to quantitatively express the degree of periodic weakening. This index is a dimensionless value, defined as 0 to 1, where 0 indicates no structural weakening and 1 indicates a strong weakening process. In the actual calculation process, the monitoring data of 52 defect areas over a continuous 72 hours were periodically analyzed, resulting in 144 periodic windows. Among them, 27 periodic windows had a WSI value greater than 0.8, accounting for 18.75%. These periodic windows are concentrated at the superposition of areas with frequent shear strain density fluctuations and areas with abrupt changes in failure level. The periodic structural weakening data is finally summarized into a structure with time index, spatial number, index sequence, and WSI value, which is used for subsequent edge rule matching to judge the structural over-limit trend and the input of maintenance intervention scheduling logic.

[0111] Step S234 includes the following steps:

[0112] Obtain the basic properties of the repair material; perform interfacial adhesion strength analysis on the basic properties of the repair material to obtain the interfacial adhesion strength of the material;

[0113] Based on the mechanical state of shear deformation of the base layer, an oblique shear stress analysis is performed to generate the oblique shear stress intensity of the base layer;

[0114] The simulated inclined shear stress intensity of the coupled base layer exhibits a gradient decay along the joint depth;

[0115] Based on the gradient decay state, the surface interface slip pattern at the material joint is correlated and identified to obtain the surface interface slip pattern.

[0116] The bonding failure at the joint was analyzed based on the surface interface slip mode to obtain the degree of bonding failure at the joint.

[0117] In this embodiment of the invention, the basic properties of the repair materials are obtained. The materials involved include three categories: C40 cement concrete repair material, modified asphalt emulsifier, and interface bonding resin. Their basic property parameters mainly include five items: elastic modulus, Poisson's ratio, tensile strength, shear strength, and bond strength. In-situ measurements are performed at edge nodes using an embedded multi-channel material property detection unit. Specifically, mechanical response curves are collected at 6h, 12h, 24h, 48h, and 72h after the initial setting of the repair material. The elastic modulus is obtained by recording the ratio of axial strain to stress using a triaxial compression sensor. The Poisson's ratio is calculated by the ratio of radial strain to axial strain. Tensile and shear strength tests are performed using a shear loading plate fixing method with progressively increasing load. The ultimate stress at fracture was used as a statistical value, while the adhesive force was obtained by performing a pull-out test in the vertical direction of the adhesive profile using an interface peeling unit. The load at the interface fracture point was converted into the unit adhesive strength. The average value was extracted from five sets of repeated experiments for subsequent analysis. Then, the interface adhesion strength analysis was performed on the basic properties of the repair materials. The layered stress analysis method was used to theoretically model the adhesion behavior of the three materials at the interface with the original substrate. During the modeling process, the minimum potential energy method was used to construct the interface force distribution and extract the average adhesion strength per unit contact area. The adhesion strength of the asphalt emulsifier was set to 0.85 MPa, C40 repair material to 1.62 MPa, and adhesive resin to 2.25 MPa, forming a dataset of material interface adhesion strength.

[0118] Based on the obtained mechanical state of shear deformation of the base layer, oblique shear stress analysis was performed. Oblique shear is defined as the shear stress response caused by the inconsistency between the principal strain direction and the loading direction in a non-orthogonal direction within the base layer. This shear stress response was reconstructed using the displacement tensor rotation method. In the operation, the shear strain principal vector in each 0.2m segment was extracted with the joint line as the axis of symmetry. The shear strain projection component along the 45° direction of the joint was rotated and multiplied with the Young's modulus of the material to obtain the oblique shear stress intensity. This operation was performed within a 0.6m extension on both sides of the joint line, and a total of 220 stress lines were completed. Force path extraction yielded a maximum oblique shear stress intensity of 3.42 MPa and a minimum of 0.96 MPa. Further simulation of the attenuation of this oblique shear stress intensity along the joint depth direction was conducted. An exponential fitting method was used to construct a stress depth attenuation function, with an initial attenuation coefficient of 0.015 m⁻¹. The coefficient was adjusted to an average error of less than 0.07 MPa using a fitting residual convergence method. This determined the attenuation trend of the oblique shear stress along the depth direction, forming a gradient distribution field. This distribution field was mapped onto the joint longitudinal section to form a three-dimensional tensor matrix, which was used for interface slip mode analysis.

[0119] Subsequently, the interfacial adhesion strength of the material interface was correlated and identified with the interfacial slip pattern at the joint surface. The slip threshold ratio method was used to calculate the ratio between the oblique shear stress intensity and the material interface adhesion strength at each depth point. A ratio greater than 1.0 was identified as an interfacial slip region, 0.8 to 1.0 as a transition zone, and below 0.8 as a stable adhesion region. Each joint was divided into 10mm depth units, and a total of 6800 points in the cross-sectional area were analyzed. After marking the slip type, the slip pattern was classified according to the slip region morphology. The specific patterns included three types: through-type, layered type, and point-type. The through-type is when the slip region is continuous along the depth for more than 80%, the layered type is when there is an intermediate slip zone in a local area, and the point-type is when there are scattered and discontinuous slip points. Each slip pattern formed a one-to-one corresponding data record under each joint number.

[0120] Finally, based on the aforementioned surface interface slip mode, the adhesive failure analysis at the joint was conducted. The segmented failure accumulation method was used to couple the slip mode with the actual shear stress fluctuation data. The slip state in each joint region was divided according to the cross-sectional layer. The number of slip points and the rate of change of area within each shear stress peak period were determined and attributed to three types of adhesive failure mechanisms: energy release type, interface degradation type, and stress accumulation type. The failure level was quantified according to the evolution trend of the slip area, and the level was divided into 0 to 3. Among them, the slip area of ​​the through-type slip mode with the peak stress period exceeding 90% was defined as severe failure, and the area of ​​the point-type slip mode less than 10% was defined as mild failure. Finally, a complete spatial distribution data of the adhesive failure degree at the joint was constructed for the input basic data structure of subsequent periodic structural weakening analysis.

[0121] Step S24 includes the following steps:

[0122] Step S241: Based on the vibration vertical load intensity data, perform multi-scale base layer crack evolution structure simulation on the periodic structure weakening data to obtain the base layer crack evolution structure.

[0123] Step S242: Perform an equal mapping of the maximum water intrusion value to the base layer crack evolution structure to obtain the maximum water intrusion value corresponding to the base layer crack evolution structure;

[0124] Step S243: Based on the maximum value of moisture intrusion and the evolution structure of base layer cracks, perform base layer expansion / softening coupling effect analysis to obtain base layer expansion / softening coupling effect data;

[0125] Step S244: Based on the data of maximum water intrusion, base layer crack evolution structure and base layer expansion / softening coupling effect, estimate the cumulative water loss gradient to obtain the pavement water loss gradient.

[0126] In this embodiment of the invention, multi-scale base layer crack evolution structural simulation is performed on periodic structural weakening data based on vibration vertical load intensity data. The simulation uses a graded crack propagation analysis method. First, a 1m×1m×0.5m three-dimensional profile unit block is constructed centered on the identified area in the periodic structural weakening data. Vibration vertical load intensity data is embedded in the primitive mesh units within the unit block, which are divided with a step size of 20mm. The data source is the vertical dynamic load obtained by converting the road surface vibration acceleration, with an amplitude range of 1.1kN to 5.8kN and a frequency range of 6Hz to 23Hz. Within each mesh unit, a... The fatigue strength threshold is 75% of the tensile strength of the repair material. The vibration energy accumulation rate is used as the time progression function to calculate the crack initiation time of each grid. Then, according to the strain concentration area between the crack initiation points, they are connected in the same direction to form the crack evolution path. All paths extend to the adjacent units and merge to form the crack network topology. At the microscale, the crack tip region is refined with a 5mm grid. The number of microcrack branches, branch length, extension angle and intersection frequency are recorded. The crack resistance coefficient and crack reconstruction geometry are adjusted in combination with the micro porosity distribution in the material to complete the simulation and construction of the base crack evolution structure.

[0127] A maximum water intrusion value mapping operation was performed on the constructed base layer crack evolution structure. Three parameters, crack channel length, depth, and aperture, were extracted in the crack profile in 10 mm units. The crack aperture was calculated from displacement difference sensing data and ranged from 0.2 mm to 3.4 mm. Based on the crack channel geometry parameters, the capillary-guided water migration method was used to analyze the water infiltration path. The initial soil moisture was set to 0.21 m³ / m³, and the capillary suction boundary at the bottom of the crack was set to 38 kPa. A 10-hour infiltration test simulation was conducted under the standard rainfall intensity of 35 mm / h. The maximum water intrusion depth and peak water volume concentration in each crack channel were recorded to generate a maximum water intrusion mapping field at the crack nodes. The water superposition effect at crack intersections and junctions was compared and analyzed with the original crack topology data to extract the spatial distribution results of the maximum water intrusion value in the crack structure.

[0128] Based on the obtained maximum water intrusion value and the evolution structure of base layer cracks, the coupling effect of base layer expansion and softening is analyzed. Within the crack path distribution area, 30mm×30mm coupling response cells are extracted sequentially. The number of cracks, water intrusion value and material hygroscopic expansion coefficient are read in this area. The expansion coefficient of asphalt stabilized crushed stone base is set to 0.014mm / mm·% moisture content. The volumetric strain is calculated using the expansion response transformation formula and mapped to the grid area. On this basis, a dynamic elastic modulus attenuation coefficient of 0.28 is introduced to describe the softening behavior. Then, a regional expansion-softening coupling tensor field is drawn based on each coupling response cell. The volume expansion ratio and elastic modulus attenuation percentage in the cell are recorded. This tensor field is used to reflect the composite deformation behavior caused by the combined action of water and cracks. The coupling data is accumulated step by step in the edge processing unit as input for the subsequent gradient analysis module. Finally, based on the data of the maximum water intrusion value, the evolution structure of base layer cracks, and the coupling effect of base layer expansion and softening, the cumulative gradient of water loss was estimated. This operation adopted a three-dimensional layered gradient weight superposition method, dividing the base layer thickness into 20 layers, each 10 mm thick. Three parameters were extracted from each layer: water intrusion concentration, expansion strain ratio, and elastic modulus attenuation rate. Weighting factors were set to 0.45, 0.35, and 0.20, respectively. The weighting factors of 0.45 (water intrusion concentration), 0.35 (expansion strain ratio), and 0.20 (elastic modulus attenuation rate) were based on a comprehensive weighing of the relative influence of each parameter on different physical effects in the base layer water loss evolution mechanism. Specifically, the water intrusion concentration directly determines the effective infiltration depth and range of water. The dominant factor causing the expansion and softening effect is assigned the highest weight (0.45); the expansion strain ratio reflects the degree of deformation caused by the material's moisture absorption and expansion, which has a significant impact on structural integrity and is assigned a medium weight (0.35); while the elastic modulus decay rate is an indirect reflection of material performance degradation, and its impact is lagging and cumulative, so it has a low weight (0.20). The cumulative impact value is calculated by assigning weights to each layer. The cumulative value is superimposed on the isosurface in the structural weakening path area and the gradient direction change rate is statistically analyzed according to the coordinate axis direction. Combined with the crack spatial path and the continuity of the moisture retention area, the water loss gradient is calculated layer by layer downward in 20mm steps to complete the derivation of the water loss cumulative gradient distribution data and store it in the monitoring task node structure for subsequent evaluation and alarm logic module call analysis.

[0129] Step S3 includes the following steps:

[0130] Step S31: Normalize the periodic structure weakening data and the cumulative gradient of pavement water loss respectively to obtain the normalized data of periodic structure weakening and the normalized gradient of pavement water loss.

[0131] Step S32: Based on the reinforcement learning algorithm in the edge computing platform, the repair status service life is evaluated by the periodic structure weakened normalized data and the pavement water loss cumulative normalized gradient, and the repair status service life evaluation data is obtained.

[0132] Step S33: Send the repair status lifespan assessment data to the terminal.

[0133] As an example of the present invention, reference is made to... Figure 3 As shown, step S3 in this example includes:

[0134] Step S31: Normalize the periodic structure weakening data and the cumulative gradient of pavement water loss respectively to obtain the normalized data of periodic structure weakening and the normalized gradient of pavement water loss.

[0135] In this embodiment of the invention, the periodic structural weakening data and the cumulative gradient of pavement water loss are normalized respectively. The periodic structural weakening data comes from four indicators in the evolution structure of base layer cracks: crack distribution density, crack length distribution, crack propagation direction consistency factor, and crack network connectivity. The crack distribution density is expressed in units of cracks / m², the length distribution ranges from 20mm to 410mm, the direction consistency factor is calculated as the ratio of the standard deviation of crack angle deviation to the maximum angle deviation, and the crack network connectivity is normalized to the maximum number of passable crack paths. The normalization process uses the min-max normalization method to map the original data to [0, 1]. In the operation, the maximum and minimum values ​​of the samples in the observation area are selected for standard linear normalization transformation. All parameters are fused with the same weight into periodic structure weakened normalized data. The cumulative gradient of pavement water loss comes from the maximum value of water intrusion, the depth of water retention layer, the degree of modulus decay caused by water, and the distribution of expansion strain gradient. The unit of water retention layer depth is mm, the degree of modulus decay is expressed as the ratio of the initial value to the current value of the material's elastic modulus, and the unit of expansion strain is %. The maximum-minimum normalization method is still used to linearly normalize each parameter. After normalization of each type of parameter, the first principal component is extracted by principal component analysis as the cumulative normalized gradient of pavement water loss.

[0136] Step S32: Based on the reinforcement learning algorithm in the edge computing platform, the repair status service life is evaluated by the periodic structure weakened normalized data and the pavement water loss cumulative normalized gradient, and the repair status service life evaluation data is obtained.

[0137] In this embodiment of the invention, the reinforcement learning algorithm in the edge computing platform is used to evaluate the service life of the road repair state based on the normalized data of weakened periodic structure and the normalized gradient of cumulative road water loss. The discrete state space reinforcement training method in the Q-learning algorithm is used to estimate the optimal behavior strategy for the deterioration trend of the road repair state. Specifically, the normalized data of weakened periodic structure and the normalized gradient of cumulative road water loss are concatenated into a state vector input. The state vector is divided into 20×20 state nodes with a spacing of 0.05, totaling 400 state nodes. Each state node corresponds to a repair state level. The behavior selection includes continuing service and arranging secondary repair. The reward value is based on the rate of state deterioration per unit time. The training process involves inverse calculation, with a learning rate of 0.2, a discount factor of 0.9, and an initial Q-table of all zeros. An ε-greedy strategy is employed for exploration, with ε set to 0.1 and 10,000 iterations. The simulated road service time in each iteration ranges from 0 to 180 days. Historical sequences of periodic structural weakening and water loss gradient degradation collected from the observed road are used as training samples. In each iteration, the action with the maximum Q-value is selected from the Q-table based on the current state and behavior, and the strategy table is updated. This results in the expected remaining lifespan corresponding to each state node in the Q-table. The expected lifespan is extracted from all training samples according to the position of the state vector as the lifespan assessment data for the repair state.

[0138] Step S33: Send the repair status lifespan assessment data to the terminal.

[0139] In this embodiment of the invention, the obtained repair status lifespan assessment data is sent to the terminal. The sending operation is performed by packaging the assessment data into a binary array via the CAN bus module of the edge computing node. The structure includes four parts: data identifier, sampling timestamp, corresponding state vector index position, and lifespan assessment value. The data identifier is 8 bits, the timestamp is in 64-bit UNIX time format, the index position is encoded as two 8-bit unsigned integers representing the two-dimensional coordinates of the state vector, and the lifespan assessment value is represented as a 32-bit floating-point number. The overall length of the data packet is 20 bytes, the sending period is 60 seconds, and the sending path is forwarded to the roadside information interaction terminal via the local link relay station and cached in the terminal database. The terminal database uses a time-series-based structured index method to store and access the lifespan data. Each data entry is bound to a corresponding geographic location identifier and a unique identification code of the processing node to ensure the integrity and reliability of the data link.

[0140] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A road repair status monitoring method based on edge computing, characterized in that, Includes the following steps: Step S1: Obtain road repair planning information for the repaired roads; Extract the locations of multiple repaired road defects from the planning information; 3D laser scanning vibration meters were deployed at multiple repaired road defect locations to collect road surface vibration signals during multiple traffic periods. Then, the disordered distribution time-domain characteristics were analyzed to obtain a disordered distribution time-domain map of the vibration signals. Step S2: Based on the time-domain diagram of the disordered distribution of vibration signals, simulate and quantify the structural weakening behavior during the service cycle to obtain periodic structural weakening data; estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the pavement; specifically, step S2 is as follows: Step S21: Extract quality design requirements for multiple repaired road defect locations based on the repair planning information of the repaired roads, and obtain the quality repair design requirements for multiple defect locations; Step S22: Perform vertical load intensity transformation on the disordered distribution time domain diagram of the vibration signal to obtain vibration vertical load intensity data; Step S23: Based on the vibration vertical load intensity data, simulate and quantify the structural weakening behavior during the service life of the quality repair design requirements for multiple defect locations to obtain periodic structural weakening data. Step S24: Estimate the cumulative water loss gradient based on the periodic structure weakening data and vibration vertical load intensity data to obtain the pavement water loss cumulative gradient; specifically, step S24 involves: Step S241: Based on the vibration vertical load intensity data, perform multi-scale base layer crack evolution structure simulation on the periodic structure weakening data to obtain the base layer crack evolution structure. Step S242: Perform an equal mapping of the maximum water intrusion value to the base layer crack evolution structure to obtain the maximum water intrusion value corresponding to the base layer crack evolution structure; Step S243: Based on the maximum value of moisture intrusion and the evolution structure of base layer cracks, perform a coupling effect analysis of base layer expansion and softening to obtain data on the coupling effect of base layer expansion and softening; Step S244: Based on the maximum water intrusion value, the evolution structure of base layer cracks, and the coupling effect data of base layer expansion and softening, estimate the water loss accumulation gradient to obtain the pavement water loss accumulation gradient; Step S3: Based on the reinforcement learning algorithm in the edge computing platform, assess the repair status service life of the periodic structure weakening data and the cumulative gradient of pavement water damage to obtain repair status service life assessment data; send the repair status service life assessment data to the terminal.

2. The road repair status monitoring method based on edge computing according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain the repair planning information of the repaired roads; extract the locations of multiple repaired road defects from the planning information; Step S12: Deploy 3D laser scanning vibration meters at multiple repaired road defect locations to collect road surface vibration signals during multiple traffic periods; Step S13: Perform average filtering on the road vibration signal to obtain the road vibration filtered signal; Step S14: Perform time-domain feature analysis on the disordered distribution of the road vibration filter signal to obtain the time-domain diagram of the disordered distribution of the vibration signal.

3. The road repair status monitoring method based on edge computing according to claim 1, characterized in that, Step S22 includes the following steps: Step S221: Analyze the amplitude change rate of the disordered distribution time-domain graph of the vibration signal to obtain the disordered amplitude change rate; Step S222: Based on the disordered amplitude change rate, perform an approximate linear correlation analysis on the time-domain graph of the disordered distribution of the vibration signal from low to high to obtain approximate correlation data of the slope increment; Step S223: Perform multi-scale spectral entropy decomposition on the disordered distribution time-domain map of the vibration signal based on the slope increment approximate correlation data to obtain time-series energy spectral entropy decomposition data; Step S224: Perform equal-quantity vertical load mechanical state mapping on the time-series energy spectrum entropy decomposition data to obtain the vertical load mechanical state; Step S225: Perform vertical load strength conversion based on the vertical load mechanical state to obtain vibration vertical load strength data.

4. The road repair status monitoring method based on edge computing according to claim 1, characterized in that, Step S23 includes the following steps: Step S231: Extract the subgrade bearing capacity, density, and thickness from the quality repair design requirements for multiple defect locations; Step S232: Perform nonlinear effect coupling on the vibration vertical load intensity data to generate load intensity nonlinear coupling data; Step S233: Based on the nonlinear coupling data of load strength, perform a simulation analysis of the mechanical state of base course shear deformation to obtain the mechanical state of base course shear deformation. Step S234: Analyze the degree of adhesion failure at the joint caused by the mechanical state of shear deformation of the base layer; Step S235: Based on the aforementioned mechanical state of the base layer shear deformation and the degree of adhesive failure at the joint, the structural weakening behavior during the cycle is simulated and quantified to obtain periodic structural weakening data.

5. The road repair status monitoring method based on edge computing according to claim 4, characterized in that, Step S234 includes the following steps: Obtain the basic properties of the repair material; perform interfacial adhesion strength analysis on the basic properties of the repair material to obtain the interfacial adhesion strength of the material; Based on the mechanical state of shear deformation of the base layer, an oblique shear stress analysis is performed to generate the oblique shear stress intensity of the base layer; The simulated inclined shear stress intensity of the coupled base layer exhibits a gradient decay along the joint depth; Based on the gradient decay state, the surface interface slip pattern at the material joint is correlated and identified to obtain the surface interface slip pattern. The bonding failure at the joint was analyzed based on the surface interface slip mode to obtain the degree of bonding failure at the joint.

6. The road repair status monitoring method based on edge computing according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Normalize the periodic structure weakening data and the cumulative gradient of pavement water loss respectively to obtain the normalized data of periodic structure weakening and the normalized gradient of pavement water loss. Step S32: Based on the reinforcement learning algorithm in the edge computing platform, the repair status service life is evaluated by the periodic structure weakened normalized data and the pavement water loss cumulative normalized gradient, and the repair status service life evaluation data is obtained. Step S33: Send the repair status lifespan assessment data to the terminal.

7. A road repair status monitoring system based on edge computing, characterized in that, For executing the edge computing-based road repair status monitoring method as described in any one of claims 1-6, the edge computing-based road repair status monitoring system comprises: The vibration signal analysis module is used to acquire road repair planning information that has been repaired; extract the locations of multiple road defects that have been repaired from the planning information; deploy 3D laser scanning vibration meters at the locations of multiple road defects to collect road surface vibration signals during multiple traffic periods, and then perform disordered distribution time-domain feature analysis to obtain a disordered distribution time-domain map of vibration signals; The damage analysis module is used to simulate and quantify the structural weakening behavior during the service cycle based on the disordered distribution time-domain diagram of vibration signals to obtain periodic structural weakening data; and to estimate the cumulative water loss gradient based on the periodic structural weakening data to obtain the cumulative water loss gradient of the pavement. The repair service life assessment module is used to assess the repair service life based on the reinforcement learning algorithm provided in the edge computing platform, which is used to assess the periodic structural weakening data and the cumulative gradient of pavement water damage, and obtain the repair service life assessment data; the repair service life assessment data is then sent to the terminal.

Images