Highway pavement disease estimation method and device

Abstract

The invention provides a highway pavement disease estimation method and device, and belongs to the technical field of highway disease detection. Comprising a modeling stage and an application stage, wherein the modeling stage specifically comprises the steps of collecting three-dimensional point cloud data of a standard road section asphalt pavement for a long period and preprocessing, identifying a subsidence area by using a deep learning network to obtain morphological characteristic data, collecting monitoring data of the subsidence area, establishing a training data set, and applying the training data set to the subsidence area; training a subsidence disease prediction equation set by using the training data set to obtain a subsidence disease trend analysis model; the application stage specifically comprises the steps of collecting and preprocessing three-dimensional point cloud data of a to-be-evaluated road asphalt pavement in a short period, identifying a subsidence area by using a deep learning network to obtain morphological characteristic data, collecting monitoring data of the subsidence area, obtaining a subsidence prediction result according to the data, and generating an evaluation report. The technical problems that in the prior art, only local road sections can be statically evaluated, the subsidence disease trend of kilometers cannot be comprehensively and dynamically mastered, and appropriate maintenance measures cannot be selected are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of highway disease detection, and in particular, relates to a method and device for predicting highway pavement diseases. Background Art

[0002] As a basic transportation infrastructure, highways play an important role in the development of human society. In the highway system, asphalt pavement is undoubtedly one of the most common pavement forms. Compared with cement concrete pavement, asphalt pavement has the advantages of good compression resistance, flexibility and low construction cost, so it is widely used in various highway projects. However, long-term traffic loads and complex and changing environmental conditions also bring severe challenges to the service life and performance of asphalt pavement.

[0003] Among them, the subsidence disease of asphalt pavement is a prominent problem. The so-called subsidence refers to the permanent sinking and deformation of the pavement structure under the action of external forces, which is usually manifested as distortion phenomena such as dents and potholes. The occurrence of this disease not only affects driving comfort, but also may cause safety hazards such as vehicle damage and pedestrian injuries. Severe subsidence may also cause structural problems such as roadbed damage and base instability, which will bring great difficulties to subsequent repairs and maintenance. Existing technologies often use manual inspections, mechanical measurements and other methods to perform static assessments on local sections of the road. It is impossible to fully and dynamically grasp the subsidence situation of the entire route and implement corresponding maintenance measures. Summary of the invention

[0004] In view of this, the present invention provides a method and device for predicting highway pavement defects, which can solve the technical problem that the prior art can only perform static evaluation on local sections, cannot comprehensively and dynamically grasp the subsidence disease trend of kilometers, and thus cannot perform accurate maintenance.

[0005] The present invention is achieved in that: The present invention provides a method for predicting highway pavement defects, comprising a modeling stage and an application stage: the modeling stage specifically comprises: long-term collection and preprocessing of three-dimensional point cloud data of asphalt pavement of a standard section, using a deep learning network to identify subsidence areas to obtain morphological feature data, collecting monitoring data of subsidence areas, establishing a training data set, and using the training data set to train a subsidence defect prediction equation group to obtain a subsidence defect trend analysis model; the application stage specifically comprises: short-term collection and preprocessing of three-dimensional point cloud data of asphalt pavement of a highway to be evaluated, using a deep learning network to identify subsidence areas to obtain morphological feature data, collecting monitoring data of the subsidence areas, obtaining subsidence prediction results based on the data and generating an evaluation report, and executing corresponding maintenance methods based on the evaluation report.

[0006] In the further lightweight neural network error correction function, the lightweight neural network specifically adopts the MobileNetV3 structure. This structure has the characteristics of high computational efficiency and small number of parameters, and is suitable for rapid deployment and application in practical engineering. The structure of the lightweight neural network is: 1. Input layer: Determine the input dimension according to the number of input variables of each prediction equation; 2. Feature extraction layer: contains 3 depth-wise separable convolution modules, each of which contains a depth-wise convolution layer (kernel_size=3×3) and a point-wise convolution layer (kernel_size=1×1); 3. Squeeze-and-Excitation (SE) module: used to adaptively adjust feature weights; 4. Non-linear activation layer: using h-swish activation function to improve network expression ability; 5. Fully connected layer: It adopts a two-layer structure, the output dimension of the first layer is 32, and the output dimension of the second layer is 1.

[0007] The structure of the lightweight neural network error correction term function is: 1. Error correction term of the time series evolution equation : Input feature vector Contains depth time series data, subsidence area data, subsidence edge slope data, subsidence shape symmetry data, subsidence profile curvature data, time increment parameters and historical subsidence rate parameters; 2. Error correction term of environmental impact equation : Input feature vector Contains annual average temperature cycle data, temperature gradient accumulation data, annual average relative humidity change data, humidity gradient accumulation data, strain accumulation data, seasonal temperature variation coefficient, daily temperature variation coefficient, rainfall data and groundwater level data; 3. Error correction term of load response equation : Input feature vector Contains annual average axle load distribution data, axle load cumulative action times data, annual average vehicle speed distribution data, vehicle braking frequency distribution data, asphalt content data, void ratio data, gradation data, strength data, anti-deformation data, traffic flow data, vehicle type distribution data, tire ground contact pressure data and road surface temperature data; 4. Error correction term of comprehensive prediction equation : Input feature vector It includes the time evolution trend value of subsidence depth, environmental impact coefficient, load effect value, pavement structure layer parameters, base material parameters, and subgrade soil parameters. is the weight.

[0008] The model or equation involved in the present invention is described in detail below: 1. In S102, the calculation process of the subsidence morphological characteristic data: the calculation expression of the subsidence depth is: ;Where: is the subsidence depth, in mm; is the elevation of the highest point in the subsidence area, in mm; It is the elevation of the lowest point in the subsidence area, in mm.

[0009] The calculation expression of subsidence area is: ; Where: is the subsidence area in m²; is the grid point The subsidence discrimination value at the location is 1 in the subsidence area and 0 in the non-subsidence area; are the grid spacing in the x and y directions, in m; are the number of grids in the x and y directions respectively.

[0010] The calculation expression of the subsidence edge slope is: ; Where: is the average edge slope; is the number of edge sampling points; is the elevation difference of the kth sampling point, in mm; is the horizontal distance of the kth sampling point, in mm.

[0011] The calculation expression of the symmetry of the subsidence shape is: ; Where: is the symmetry coefficient, ranging from [0,1]; For point The elevation value at the point is in mm.

[0012] The calculation expression of the curvature of the subsidence profile is: ; Where: is the curvature value; and are the first and second derivatives of elevation with respect to distance, respectively.

[0013] 2. In S103, historical subsidence rate parameter calculation: ; Where: is the historical subsidence rate, in mm / year; It is the increment of subsidence depth within one year, in mm; is the time interval in years.

[0014] 3. In S104-S105, statistical parameters of environmental and traffic monitoring data are calculated: Calculation of temperature cycles: ; Where: is the average annual temperature cycle number; is the number of temperature cycles on the i-th day.

[0015] Temperature gradient accumulation calculation: ; Where: is the accumulated value of temperature gradient, in °C·h / mm; is the temperature difference at the jth hour on the i-th day, in °C; is the depth difference, in mm; is the time interval in h.

[0016] Humidity variation coefficient calculation: ; Where: is the humidity variation coefficient; is the standard deviation of relative humidity; is the mean relative humidity.

[0017] 4. Mathematical expression of the time evolution equation: ; Where: is the predicted value of the subsidence depth at time t, in mm; is the initial subsidence depth, in mm; is the fitting coefficient; is the neural network error correction term; is the input feature vector; is the network weight matrix.

[0018] 5. Mathematical expression of environmental impact equation: ; Where: is the environmental impact coefficient; is the weight coefficient; is the annual rainfall in mm; is the groundwater level, in m; is the cumulative strain; is the neural network error correction term.

[0019] 6. Mathematical expression of load response equation: ; Where: is the load effect value; is the weight coefficient; is the i-th level axle load size, in kN; is the corresponding number of actions; is the vehicle speed influence coefficient; is the braking influence coefficient; is the comprehensive performance coefficient of the material; is the traffic volume, in vehicles per day; is the tire ground contact pressure, in MPa; is the neural network error correction term.

[0020] 7. Mathematical expression of comprehensive prediction equation: ; Where: They are respectively used to predict the depth, area, development rate and stability of subsidence; is the weight coefficient matrix element; is the structural influence coefficient; is the layer influence coefficient; are the strength coefficients of the surface layer, base layer and roadbed respectively; is the neural network error correction term.

[0021] The derivation process of each equation is explained in detail below.

[0022] 1. Derivation process of time evolution equation: First, establish the basic evolution model: ;This is a linear evolution model that expresses the basic development trend of subsidence over time.

[0023] Considering the influence of morphological characteristics, nonlinear terms are introduced: ; The nonlinear term The following conditions must be met: (1) Subsidence area The larger the value, the faster the development rate, and they are positively correlated; (2) Edge slope The larger the value, the more obvious the stress concentration is and the faster it develops; (3) Symmetry The better, the more stable the development, so adopt ; (4) Curvature The larger it is, the greater the deformation gradient is and the faster it develops.

[0024] After data fitting and theoretical analysis, the nonlinear term expression is obtained: ; in Through experimental data regression, we can obtain: Experimental steps: (1) Select multiple standard road sections and record the subsidence development data within one year; (2) Extract the characteristic parameters of each morphology; (3) Use nonlinear least squares method to determine value.

[0025] Finally, the neural network error correction term is introduced: .

[0026] 2. Derivation process of environmental impact equation: First consider the basic effects of temperature and humidity: ; in is the influence coefficient. Consider the coupling effect between factors: ; Determined by multi-factor orthogonal test. Introducing the influence of moisture and strain: ; The cumulative strain Calculation method: ; is the strain increment of the ith measurement.

[0027] Finally, the complete expression is: .

[0028] 3. Derivation process of load response equation: First, establish a single-level load basic response model: ; Consider the impact of vehicle speed: ; Where: is the actual vehicle speed; is the standard vehicle speed; is the speed influence coefficient.

[0029] Consider the braking effect: ; Where: is the braking frequency; is the braking influence coefficient.

[0030] The cumulative effect of multi-level loads is obtained by combination: .

[0031] Introducing the influence of material properties: ; Where: is the cohesiveness coefficient; is the void ratio influence coefficient; is the strength influence coefficient.

[0032] Finally, the complete expression is: .

[0033] 4. Derivation process of comprehensive prediction equation: First, establish the basic prediction matrix: .

[0034] Consider the impact of pavement structure: is the structural strength matrix; is the layer thickness matrix; is the structural influence coefficient matrix.

[0035] Finally, the complete prediction equation is obtained: .

[0036] The effects of each equation are as follows: 1. The time-series evolution equation can accurately describe the nonlinear development process of subsidence, especially the influence of subsidence morphology on the development rate through the coupling of morphological characteristic parameters; 2. The environmental impact equation comprehensively considers the combined effects of environmental factors such as temperature cycles, humidity changes, and precipitation, and can quantify the role of the environment in promoting subsidence development; 3. The load response equation not only considers the load size and frequency, but also introduces dynamic effects such as vehicle speed and braking, which improves the prediction accuracy; 4. The comprehensive prediction equation realizes the coupling calculation of multiple parameters in matrix form, which not only takes into account the effects of environment and load, but also reflects the influence of pavement structure, making the prediction results more comprehensive and accurate.

[0037] The environmental monitoring system specifically comprises: 1. Temperature monitoring equipment: (1) Platinum resistance temperature sensor, installation depth is 0mm, 50mm, 100mm, 200mm; (2) Infrared temperature sensor, used for real-time monitoring of road surface temperature; (3) Temperature data collector, sampling frequency is 1 time / hour; 2. Humidity monitoring equipment: (1) Soil moisture sensor, installation depth is 300mm, 500mm, 800mm; (2) Air relative humidity sensor, installation height is 1.5m; (3) Humidity data collector, sampling frequency is 1 time / hour; 3. Precipitation monitoring equipment: (1) Rain gauge, installation height is 1.5m above the ground; (2) Precipitation data collector, real-time recording; 4. Groundwater level monitoring equipment: (1) Water level sensor, installation depth is 1m below the groundwater level; (2) Water level data collector, sampling frequency is 1 time / day; 5. Strain monitoring equipment: (1) Optical fiber strain sensor, arranged in each layer of the pavement structure; (2) Strain data collector, sampling frequency is 100Hz.

[0038] The specific steps of deployment are: 1. Site survey: (1) Determine the scope of the monitoring area; (2) Select the sensor installation location; (3) Evaluate the power supply conditions; 2. Equipment installation: (1) Dig the sensor installation groove; (2) Lay the sensor connection line; (3) Install the sensor; (4) Backfill the installation groove; 3. Data acquisition system construction: (1) Install the data collector; (2) Connect the sensor and the collector; (3) Configure the acquisition parameters; 4. System debugging: (1) Check the working status of the sensor; (2) Calibrate the collected data; (3) Test the data transmission; 5. Environmental protection: (1) Install the rain cover; (2) Lay out the lightning protection facilities; (3) Set up warning signs.

[0039] The traffic load monitoring system specifically includes: 1. Axle load monitoring equipment: (1) Capacitive weighing sensor; (2) Piezoelectric weighing sensor; (3) Axle load data collector; 2. Vehicle speed monitoring equipment: (1) Laser speed meter; (2) Radar speed sensor; (3) Vehicle speed data collector; 3. Braking monitoring equipment: (1) Infrared thermal imager; (2) Braking status recognition system; (3) Braking data collector; 4. Traffic flow monitoring equipment: (1) Coil detector; (2) Video detector; (3) Flow data collector; 5. Vehicle type recognition equipment: (1) Laser profiler; (2) Vehicle type classification system; (3) Classification data collector; 6. Tire pressure monitoring equipment: (1) Pressure sensor pad; (2) Pressure data collector.

[0040] The specific steps of deployment are: 1. System planning: (1) Determine the location of the monitoring section; (2) Select the equipment installation point; (3) Plan the cable routing; 2. Equipment installation: (1) Dig grooves to bury sensors; (2) Install the support frame; (3) Fix the monitoring equipment; (4) Lay out the connecting cables; 3. System integration: (1) Install the control cabinet; (2) Connect the power supply system; (3) Configure the data logger; (4) Debug the communication module; 4. Calibration: (1) Perform static calibration; (2) Perform dynamic calibration; (3) Verify the monitoring accuracy; 5. System protection: (1) Sealing and waterproofing; (2) Installing the protective cover; (3) Setting up warning signs.

[0041] The steps of collecting material parameters of the subsidence area are specifically as follows: 1. Road surface core sampling: (1) determine the sampling point location; (2) drill the asphalt mixture core sample; (3) cut the core sample in layers; 2. Asphalt content determination: (1) determine the asphalt content by combustion method; (2) calculate the asphalt content percentage; 3. Void ratio determination: (1) determine the bulk density; (2) determine the theoretical maximum density; (3) calculate the void ratio; 4. Gradation analysis: (1) perform a screening test; (2) draw a gradation curve; (3) calculate grading index; 5. Strength test: (1) conduct Marshall test; (2) determine dynamic stability; (3) determine splitting strength; 6. Deformation performance test: (1) conduct rutting test; (2) determine dynamic modulus; (3) determine fatigue life; 7. Base material inspection: (1) sample drilling core; (2) determine strength; (3) determine modulus; 8. Roadbed soil inspection: (1) sample drilling core; (2) determine water content; (3) determine bearing ratio; (4) determine compression modulus.

[0042] The present invention also provides a highway pavement disease prediction device, comprising: a processor, used to read a computer program in a storage medium, the computer program stores the code of the above method, and the processor executes the method in the above technical solution.

[0043] Compared with the prior art, the highway pavement maintenance method and device provided by the present invention have the following beneficial effects: 1. Comprehensive monitoring: The use of high-precision three-dimensional laser scanning technology can regularly and comprehensively obtain road surface status data for standard sections and even entire highways, making up for the limitations of manual inspections.

[0044] 2. Depth of analysis: The present invention establishes a prediction system including multiple analysis modules such as time series evolution, environmental impact, and load response by comprehensively considering multiple factors such as environment, transportation, and materials, which better reflects the complex mechanism of the development of subsidence diseases.

[0045] 3. Prediction accuracy: Incorporating an error correction strategy based on a lightweight neural network into the prediction model greatly improves the ability to learn and infer actual road conditions, making the prediction results closer to reality.

[0046] In general, the highway asphalt pavement subsidence disease trend assessment method of the present invention makes full use of advanced three-dimensional laser scanning technology, deep learning algorithm and neural network model, comprehensively considers the influencing factors, can accurately predict the development trend of subsidence, and solves the technical problem that the existing technology can only perform static assessment on local sections and cannot comprehensively and dynamically grasp the subsidence disease trend of kilometers. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of the steps of the modeling stage of the present invention; Figure 2 It is a flow chart of the application phase steps of the present invention; Figure 3 It is a subsidence disease prediction diagram in the embodiment. DETAILED DESCRIPTION

[0048] like Figure 1-2 FIG. 1 is a flowchart of the steps in the modeling phase and the application phase included in the present invention. The specific implementation method of each step is described in detail below: The specific implementation method of step S101 is: first, select a representative standard road section as the monitoring object. The standard road section should be a road section with relatively typical asphalt pavement conditions, which can better reflect the common asphalt pavement disease characteristics in the highway network. Use a high-precision three-dimensional laser scanner to conduct a comprehensive scan of the standard road section every year to obtain the three-dimensional point cloud data of the asphalt pavement of the road section. To ensure data quality, the scanning equipment needs to be calibrated before and after scanning. Subsequently, the acquired three-dimensional point cloud data is preprocessed. This includes operations such as denoising, scaling and registration, and coordinate system conversion, with the aim of eliminating interference from environmental factors and improving data accuracy and consistency. Through this step, preprocessed data reflecting the asphalt pavement conditions of the standard road section is obtained.

[0049] The specific implementation method of step S102 is: use an advanced deep learning network to segment and analyze the pre-processed three-dimensional point cloud data. Specifically, the point cloud data is processed using a semantic segmentation network based on U-Net, which can accurately identify the various subsidence areas in the asphalt pavement. Through this segmentation process, rich morphological feature data such as the location information, area, depth, edge slope, shape symmetry and contour curvature of the subsidence area are obtained. These feature data will become an important basis for subsequent subsidence trend analysis. Among them, the calculation formula for the subsidence depth is ,in and They represent the highest and lowest elevations of the subsidence area respectively. The calculation formula for the subsidence area is: ,in is the grid point The subsidence discrimination value of and are the grid spacing in the x and y directions respectively. The edge slope is Calculate, where and are the elevation difference and horizontal distance of the kth edge sampling point respectively. Shape symmetry index , which describes the degree of symmetry of the subsidence area. The profile curvature is calculated from the first and second derivatives of the elevation Through the extraction of these characteristic parameters, the morphological characteristics of each subsidence area in the asphalt pavement of the standard road section are fully described.

[0050] The specific implementation method of step S103 is: collect monitoring record data of the standard road section subsidence area in the past year. These historical data mainly include the subsidence depth value at each monitoring. Using these depth data, the historical subsidence rate parameters of the subsidence area can be calculated. ,in represents the increment of subsidence depth within one year, This historical subsidence rate parameter reflects the development trend of the subsidence of the road section and is an important basis for predicting the future development of subsidence.

[0051] The specific implementation method of step S104 is: an environmental monitoring system is deployed around the subsidence area of ​​the standard road section. The system mainly includes temperature monitoring equipment, humidity monitoring equipment, rainfall monitoring equipment, groundwater level monitoring equipment and strain monitoring equipment. The temperature monitoring equipment includes a platinum resistance temperature sensor and an infrared temperature sensor, which are installed at different depths to obtain time series data of the road surface temperature. The humidity monitoring equipment includes a soil moisture sensor and an air relative humidity sensor, which are used to monitor the humidity changes of the road surface and the roadbed. The rainfall monitoring equipment is a rain gauge that can record rainfall information in real time. The groundwater level monitoring equipment is used to measure the dynamic changes of the groundwater level. The strain monitoring equipment uses an optical fiber strain sensor, which is deployed in each structural layer of the road surface to monitor the strain accumulation. By monitoring the changes in these environmental parameters for a long time, the degree of influence of temperature, humidity, rainfall, groundwater level and strain on the development of subsidence can be quantified. With one year as a cycle, these environmental monitoring data are statistically analyzed to obtain the average annual temperature cycle number. , Temperature gradient accumulation , Humidity variation coefficient , annual rainfall , groundwater level and strain accumulation And other parameters.

[0052] The specific implementation method of step S105 is: a traffic load monitoring system is deployed in the subsidence area of ​​the standard road section. The system mainly includes axle load monitoring equipment, vehicle speed monitoring equipment, brake monitoring equipment, traffic flow monitoring equipment, vehicle type identification equipment and tire pressure monitoring equipment. The axle load monitoring equipment adopts capacitive weighing sensors and piezoelectric weighing sensors, which can obtain the axle load size and number of actions of various types of vehicles. The vehicle speed monitoring equipment includes a laser speed meter and a radar speed sensor, which can record the driving speed of the vehicle in real time. The brake monitoring equipment includes an infrared thermal imager and a brake state identification system, which is used to judge the braking condition and frequency of the vehicle. The traffic flow monitoring equipment adopts a coil detector and a video detector to provide traffic flow data for each time period. The vehicle type identification equipment includes a laser profiler and a vehicle type classification system, which can automatically identify the type of passing vehicles. The tire pressure monitoring equipment is a pressure sensor pad that measures the tire ground pressure. By long-term monitoring of the changes in these traffic parameters, the influence of load on the development of subsidence can be fully understood. These parameters include the average annual axle load distribution , cumulative number of axle load actions , Annual average vehicle speed distribution , vehicle braking frequency , Traffic flow and tire contact pressure wait.

[0053] The specific implementation method of step S106 is: collecting material parameter data of the subsidence area of ​​the standard road section. First, core sampling of the pavement is carried out in the subsidence area. The asphalt content is determined by the combustion method, and the porosity is calculated by the volume density and the theoretical maximum density. The core samples are screened, and the gradation curve is drawn and the relevant indicators are calculated. Then, the Marshall test is carried out to determine the dynamic stability and splitting strength, and the rutting test, dynamic modulus and fatigue life test are carried out to obtain the strength and deformation performance parameters of the pavement material. In addition, it is also necessary to sample and test the base and roadbed materials to determine their strength modulus, bearing ratio and other indicators. Through the above series of tests and inspections, the material parameter data of each layer of the pavement structure in the subsidence area are obtained, including the asphalt content. , porosity , gradation index, strength , anti-deformation and roadbed bearing Etc. These material characteristic parameters will play an important role in the subsequent comprehensive analysis.

[0054] The specific implementation method of step S107 is: using a high-precision three-dimensional laser scanner to monitor the subsidence area of ​​the standard road section every day. The complete three-dimensional point cloud data of the area is obtained by laser scanning, and the time series data of the subsidence depth is extracted from it. At the same time, the road surface temperature sensor is used to record the daily changes in the road surface temperature. After long-term monitoring, detailed data on the changes in the subsidence depth over time and the corresponding road surface temperature field information are accumulated. These time series data provide a basis for subsequent disease trend prediction.

[0055] The specific implementation method of step S108 is: integrate the various types of data obtained above into a training data set. First, it includes pre-processed three-dimensional point cloud data, morphological feature data of the subsidence area, historical subsidence rate parameters, environmental monitoring data, traffic monitoring data and material parameters. Secondly, the subsidence depth time series data and road surface temperature data obtained from daily monitoring are also added. Through the integration of these data, a training sample library that fully reflects the development law of subsidence diseases in standard road sections is constructed.

[0056] The specific implementation method of step S109 is: using the training data set, the pre-established subsidence disease prediction equation group is trained. This prediction equation group includes four parts: time series evolution equation, environmental impact equation, load response equation and comprehensive prediction equation. A lightweight neural network is set inside each equation as an error correction term function.

[0057] When training these prediction equations, the input data is first normalized, time series features are constructed, and data augmentation is performed to improve the generalization ability of the network. Then a phased training strategy is adopted, where the basic parameters of the prediction equation are fixed and only the neural network part is trained, and then joint optimization is performed. A weighted multi-task loss function is used to comprehensively consider the accuracy of each prediction indicator. The model performance is evaluated on the test set, and the root mean square error (RMSE) and mean absolute error (MAE) of each prediction indicator are calculated. Finally, the trained model parameters are saved, and the model is compressed and quantized to optimize the deployment efficiency of practical applications.

[0058] Through the above modeling process, an analytical model that can accurately predict the development trend of highway asphalt pavement subsidence disease is obtained. This model fully integrates the historical monitoring data of standard road sections, environmental parameters, traffic loads and pavement material characteristics, and comprehensively considers various influencing factors, providing strong support for the prediction and prevention of subsidence diseases.

[0059] The specific implementation of step S201 is: take the asphalt pavement of the highway to be evaluated as the monitoring object, use the same three-dimensional laser scanner as step S101 to fully scan it every 7 days, and obtain the three-dimensional point cloud data of the road section. Similarly, pre-process the collected point cloud data to eliminate environmental interference, and obtain pre-processed data reflecting the actual condition of the road section to be evaluated.

[0060] The specific implementation of step S202 is similar to step S102, using the trained deep learning network to segment and analyze the preprocessed 3D point cloud data, accurately identify the various subsidence areas in the road section to be evaluated, and extract their location information, area, depth, edge slope, shape symmetry, contour curvature and other morphological feature data. These feature data will be used as input parameters for subsequent comprehensive prediction.

[0061] The specific implementation of step S203 is similar to that of steps S104 and S105. An environmental monitoring system and a traffic load monitoring system are deployed around the subsidence area of ​​the road section to be evaluated for a period of 7 days to collect environmental parameter data and traffic load data of the road section. Environmental parameters include annual average temperature cycle times, temperature gradient accumulation, humidity variation coefficient, annual rainfall, groundwater level and strain accumulation. Traffic load parameters include annual average axle load distribution, axle load cumulative action times, annual average vehicle speed distribution, vehicle braking frequency, traffic flow, vehicle model distribution and tire grounding pressure. At the same time, based on step S106, the material parameter data of the road section to be evaluated are recollected, including asphalt content, void ratio, grading index, strength characteristics and anti-deformation performance.

[0062] The specific implementation of step S204 is similar to step S107. A high-precision laser scanner is used to monitor the subsidence area of ​​the road section to be evaluated every day to obtain time series data of the subsidence depth. At the same time, a road surface temperature sensor is used to record the daily changes in the road surface temperature in the area. Through 7 days of continuous monitoring, detailed data on the subsidence depth and road surface temperature of the road section to be evaluated are accumulated.

[0063] The specific implementation method of step S205 is: the above-obtained morphological feature data of the subsidence area, environmental monitoring data, traffic monitoring data, material parameters, subsidence depth time series data and road surface temperature data are input into the trained highway asphalt pavement subsidence disease trend analysis model. The model includes four parts: time series evolution equation, environmental impact equation, load response equation and comprehensive prediction equation. A lightweight neural network is integrated in each equation as an error correction term. Through the calculation and processing of the model, the subsidence prediction results of the road section to be evaluated within a certain period of time in the future are obtained, including indicators such as predicted subsidence depth, predicted subsidence area, predicted subsidence development rate and predicted subsidence stability.

[0064] The specific implementation method of step S206 is: according to the above-mentioned subsidence prediction results, the subsidence disease level of the road section to be evaluated can be evaluated. Under normal circumstances, the following disease level classification standards can be set: minor disease: predicted subsidence depth is less than 50mm, predicted subsidence area is less than 3 square meters, predicted development rate is less than 10mm / year, and predicted stability is greater than 0.8; moderate disease: predicted subsidence depth is between 50-100mm, predicted subsidence area is between 3-10 square meters, predicted development rate is between 10-20mm / year, and predicted stability is between 0.5-0.8; severe disease: predicted subsidence depth is greater than 100mm, predicted subsidence area is greater than 10 square meters, predicted development rate is greater than 20mm / year, and predicted stability is less than 0.5. According to the actual subsidence prediction results of the road section to be evaluated, it is divided into minor, moderate or severe disease levels. Finally, a subsidence disease assessment report for the road section is generated to provide a basis for subsequent maintenance.

[0065] Specifically, the principle of the present invention is: it fully integrates advanced technical means such as three-dimensional laser scanning, deep learning and neural networks to form a highway asphalt pavement subsidence disease assessment system with comprehensive dynamic monitoring, in-depth impact analysis and accurate trend prediction. First, a high-precision three-dimensional laser scanner is used to regularly scan the standard road section, which can obtain detailed digital data of the road surface status and lay the foundation for subsequent analysis and modeling. This overcomes the limitations of traditional manual inspection methods and greatly improves the comprehensiveness and dynamics of monitoring. Secondly, the semantic segmentation network of deep learning is used to process the scanned data, which can accurately identify the subsidence area and extract its rich geometric features such as location information, area, depth, edge slope, shape symmetry and curvature. These feature data not only reflect the actual deformation of the road surface, but also lay the foundation for the subsequent establishment of an impact mechanism model. Thirdly, the present invention obtains the environmental parameters, traffic load and material performance data of the standard road section through long-term monitoring, and establishes a disease trend analysis model containing four modules: time series evolution, environmental impact, load response and comprehensive prediction. Among them, the time series evolution equation describes the change law of subsidence depth over time; the environmental impact equation quantifies the influence of factors such as temperature and humidity on the development of subsidence; the load response equation evaluates the contribution of traffic load to the deepening of subsidence; and the comprehensive prediction equation integrates various influencing factors to give the final subsidence development trend. This series of models effectively makes up for the shortcomings of existing analysis methods and better reflects the complex mechanism of subsidence disease development. Finally, the present invention incorporates a lightweight neural network as an error correction term into each prediction module. Such a network structure has the characteristics of high computational efficiency and small number of parameters, and can adaptively learn the actual change characteristics of road conditions and improve the accuracy of overall prediction. Through phased training and joint optimization strategies, the final prediction model can better fit the actual road conditions.

[0066] In summary, the highway asphalt pavement subsidence disease trend assessment method of the present invention fully utilizes the advantages of advanced technical means, comprehensively considers the influencing factors, and forms a systematic and complete monitoring and analysis system. This not only overcomes the limitations of existing methods, but also greatly improves the understanding and prediction capabilities of the development laws of subsidence diseases, providing strong support for more accurate and efficient maintenance and repair decisions.

[0067] In order to better understand and implement the present invention, an embodiment of a specific application scenario of the present invention is provided below: A provincial highway management office decided to use the highway asphalt pavement settlement disease trend assessment method proposed in the present invention to conduct a comprehensive assessment of an important highway section within its jurisdiction. The section is 80 kilometers long and adopts a two-lane two-way asphalt pavement structure. It was built in 2015. In the past few years of operation, the asphalt pavement has experienced many obvious settlement diseases, affecting driving safety and comfort. In order to discover problems in a timely manner and effectively manage and maintain them, the management office selected a typical section of the section, with a length of 5 kilometers, as a standard section to carry out subsequent monitoring and analysis work.

[0068] The modeling phase includes: S101, monitoring standard section First, the management office selected a 5-kilometer section in the middle of the highway as the standard section. The asphalt pavement condition of this section is relatively typical and can better represent the disease characteristics of the entire highway. Then, they used a Leica BLK360 3D laser scanner to conduct a comprehensive scan of the standard section once a year to obtain its 3D point cloud data. To ensure the accuracy of the data, the management office technicians calibrated the equipment before and after the scan.

[0069] Table 1 Leica BLK360 3D laser scanner parameters

[0070] The acquired 3D point cloud data was processed through denoising, registration and other preprocessing steps to eliminate the interference of environmental factors and improve the quality and consistency of the data. After preprocessing, the 5-kilometer-long asphalt pavement 3D point cloud data of the standard road section was successfully acquired, laying the foundation for subsequent analysis.

[0071] S102, Subsidence area analysis: For the pre-processed 3D point cloud data, the management office used a U-Net-based deep learning semantic segmentation network to process it, with the aim of accurately identifying the various subsidence areas on the road surface. Through this segmentation process, they obtained rich geometric feature data such as the location information, area, depth, edge slope, shape symmetry, and contour curvature of each subsidence area.

[0072] Table 2 Morphological characteristics of subsidence area in standard road sections

[0073] As can be seen from Table 2, there are 10 obvious subsidence areas in this standard section, and their geometric characteristics are somewhat different. For example, the subsidence depth is between 35-88 mm and the area is between 2.1-7.6 m The edge slope is between 4.2°-12.5°, the shape symmetry is between 0.72-0.91, and the curvature is between 0.027-0.142 mm. These differences reflect the distribution characteristics of subsidence diseases on the road surface and lay the foundation for the subsequent analysis of influencing factors.

[0074] S103. Analysis of historical subsidence rates: After obtaining the geometric characteristic data of the subsidence areas in the standard road sections, the management office also collected the monitoring records of these areas in the past year, mainly including the subsidence depth data of each detection. By comparing and analyzing these historical data, they calculated the average historical subsidence rate of each subsidence area. The results are shown in Table 3: Table 3 Historical subsidence rates of subsidence areas in standard road sections

[0075] As can be seen from Table 3, there are certain differences in the historical subsidence rates of the 10 subsidence areas in this standard section, with the lowest being 8.9 mm / year, the highest being 15.1 mm / year, and the average being 11.8 mm / year. This data reflects the development trend of the subsidence disease in this section and provides a key basis for the subsequent accurate prediction of subsidence changes.

[0076] S104, Environmental Parameter Monitoring In order to deeply analyze the impact of environmental factors on the development of subsidence, the management office has set up a comprehensive environmental monitoring system around the subsidence area of ​​the standard road section. The system mainly consists of the following parts: Temperature monitoring equipment: includes 4 platinum resistance temperature sensors installed on the road surface, at 50 mm depth, 100 mm depth and 200 mm depth, and 1 infrared temperature sensor for real-time monitoring of road surface temperature. Temperature data is recorded by the collector at hourly intervals.

[0077] Humidity monitoring equipment: including three soil moisture sensors installed at depths of 300 mm, 500 mm and 800 mm, and one air relative humidity sensor installed 1.5 m above the ground. Humidity data are recorded by the collector at hourly intervals.

[0078] Rainfall monitoring equipment: 1 rain gauge, installed at a height of 1.5 m above the ground, records rainfall conditions in real time.

[0079] Groundwater level monitoring equipment: 1 water level sensor, installed 1 m below the groundwater level, records groundwater level data once a day.

[0080] Strain monitoring equipment: includes several optical fiber strain sensors arranged in each structural layer of the road surface, and the collector collects strain data at a frequency of 100 Hz.

[0081] Through long-term monitoring, the management office obtained the environmental parameter data of the standard road section within one year, as shown in Table 4: Table 4 Monitoring results of environmental parameters of standard road sections

[0082] As can be seen from Table 4, the changes in environmental parameters of this standard section within one year are as follows: the average annual temperature cycle number is 72 times, the temperature gradient accumulation reaches 1458°C·h / mm; the average annual relative humidity change is 0.112, the humidity change coefficient is 0.062, and the humidity gradient accumulation is 4.81%·m / mm; the annual rainfall is 789 mm, the groundwater level is about 1.6 m; the strain accumulation value is 0.108%. These environmental parameter data lay the foundation for the subsequent quantitative analysis of the impact of temperature and humidity on the development of subsidence.

[0083] S105, Traffic load monitoring At the same time, the management office has also deployed a traffic load monitoring system around the subsidence area of ​​the standard road section to fully grasp the actual traffic conditions of the road section. The system mainly consists of the following parts: Axle load monitoring equipment: including 2 capacitive weighing sensors and 2 piezoelectric weighing sensors, which are used to measure the size of various axle loads of passing vehicles and the number of times they act. Vehicle speed monitoring equipment: including 1 laser speed meter and 1 radar speed sensor, which can measure the vehicle's driving speed. Braking monitoring equipment: including 1 infrared thermal imager and 1 brake state recognition system, which can monitor the braking frequency and intensity of the vehicle. Traffic flow monitoring equipment: including 2 sets of coil detectors and 1 set of video detectors, which record traffic flow data in real time. Vehicle type recognition equipment: including 1 laser profiler and 1 vehicle type classification system, which can automatically identify the type of passing vehicles. Tire pressure monitoring equipment: including 1 pressure sensor pad, which measures the ground pressure of the tire.

[0084] Through long-term monitoring, the management office obtained the traffic load data of the standard road section within one year, as shown in Table 5: Table 5 Traffic load monitoring results of standard road sections

[0085] As can be seen from Table 5, the traffic load characteristics of this standard road section within one year are as follows: the annual average axle loads of 5 tons, 10 tons and 15 tons account for 32.5%, 41.2% and 26.3% respectively; the number of times the axle loads of 5 tons, 10 tons and 15 tons acted reached 1.78 million times, 1.21 million times and 860,000 times respectively; the annual average vehicle speed was between 92-121 km / h, and the braking frequency was between 0.12-0.27; the annual average daily traffic flow was 18,500 vehicles, and the vehicle types were mainly small cars; the tire ground pressure was about 0.78 MPa. These traffic parameter data laid the foundation for the subsequent quantitative analysis of the impact of load on the deepening of subsidence.

[0086] S106. Obtaining material parameters In addition to environmental and traffic factors, the management office also conducted a comprehensive test on the performance of pavement materials in the subsidence area of ​​the standard road section. Specifically, the pavement core sampling was carried out in the subsidence area to obtain a physical sample of the asphalt mixture. The asphalt content was determined by the combustion method. The porosity of the road surface is calculated by measuring the volume density and the theoretical maximum density. The grading curve was drawn and the grading index was calculated based on the sieving test. The results showed that the grading performance of the pavement was good and could meet the requirements of the specification. Then, the Marshall test was carried out to determine the dynamic stability. At the same time, rutting tests were carried out to determine the dynamic modulus. The stress is 2800 MPa and the fatigue life is 1.2×10 Finally, the base and roadbed materials were sampled and tested. The strength modulus of the base material was determined. is 1800 MPa, the bearing capacity of the subgrade soil is It is 8.2%.

[0087] Through the above series of tests, the management office has fully mastered the material performance parameters of each layer of the asphalt pavement structure of this standard section, as shown in Table 6. These data lay the foundation for the subsequent analysis of the impact of material properties on the development of subsidence.

[0088] Table 6 Standard road section material parameters

[0089] S107, road condition monitoring: In addition to obtaining the geometric features of the subsidence area at one time, the management office also used the Leica BLK360 3D laser scanner to conduct continuous dynamic monitoring of the standard road section for one year. Every day, they scanned the subsidence area once to obtain its 3D point cloud data for the day. At the same time, they also used infrared temperature sensors installed on the road surface to monitor the temperature changes in the subsidence area. By comparing and analyzing these time series data, the management office mastered the evolution of the subsidence depth over time, laying the foundation for the subsequent establishment of a prediction model.

[0090] Table 7 Dynamic monitoring results of subsidence area in standard road section (partial)

[0091] Table 7 shows the change in the subsidence depth of the 1# and 2# subsidence areas of the standard road section within one year. The 1# area increased from the initial 35.2 mm to 47.1 mm, and the 2# area increased from 41.6 mm to 57.2 mm, which can clearly reflect the development trend of the subsidence. At the same time, these data also show that there is a certain correlation between the subsidence depth and the road surface temperature, which provides a reference for the subsequent establishment of the subsidence time series evolution model.

[0092] S108, Construction of training data set: With the various data obtained from the above monitoring, the technical staff of the management office began to build the training data set , in preparation for the subsequent prediction model training. The training data set includes the following: 1. Preprocessed 3D point cloud data , used for deep learning segmentation.

[0093] 2. Morphological characteristic data of subsidence area in standard road section , used to characterize geometric deformation.

[0094] 3. Historical subsidence rate parameters , used to analyze development trends.

[0095] 4. Environmental monitoring data , used to quantify environmental impacts.

[0096] 5. Traffic load data , used to evaluate the load effects.

[0097] 6. Material parameters , used to analyze structural performance.

[0098] 7. Subsidence depth time series and road surface temperature data , used for model training.

[0099] By integrating these monitoring data, the management office has built a training sample library that comprehensively reflects the development mechanism of subsidence diseases in standard road sections. , which laid the foundation for the subsequent prediction model training.

[0100] S109. Prediction model training With the above rich training data set , the technical staff of the management office began to train the highway asphalt pavement subsidence disease trend prediction model. The prediction model includes four modules: time series evolution equation, environmental impact equation, load response equation and comprehensive prediction equation. Each module incorporates a lightweight neural network as an error correction term.

[0101] The specific training steps are as follows: 1. Data preprocessing: (1) Normalize all input data and unify the numerical range to the interval [0, 1]; (2) Use the sliding time window method to construct time series features; (3) Perform data enhancement, including adding Gaussian noise, random scaling, and other methods to expand the number of samples.

[0102] 2. Training in stages: (1) First, fix the basic parameters of the prediction equation and only train the neural network part; (2) Use the Adam optimizer and set the initial learning rate to 0.001; (3) Use a mini-batch training strategy with a batch size of 32; (4) Use an early stopping strategy and stop training when the validation set loss does not decrease for five consecutive epochs.

[0103] 3. Model evaluation: (1) Evaluate model performance on the test set; (2) Calculate the root mean square error (RMSE) and mean absolute error (MAE) of each prediction indicator; (3) Analyze the stability and reliability of the prediction results.

[0104] Through the above step-by-step training strategy, the management office finally obtained a prediction model that can accurately predict the development trend of subsidence disease in this standard road section. This model not only has good generalization ability, but also the stability and reliability of the prediction results have been verified.

[0105] The application phase is described in detail as follows: S201. Monitoring of road sections to be evaluated: After successfully establishing a subsidence disease prediction model for standard road sections, the management office decided to apply this method to the evaluation of the entire 80-kilometer highway. First, they selected a 5-kilometer section in the middle of the highway as the object to be evaluated, and used a Leica BLK360 3D laser scanner to conduct a comprehensive scan of the section every 7 days to obtain its 3D point cloud data. Similarly, the management office also pre-processed the scanned data, including denoising and registration, to obtain pre-processed data that reflects the actual conditions of the road section to be evaluated. .

[0106] S202, Subsidence area analysis: After obtaining the pre-processed data of the road section to be evaluated Afterwards, the management office used the previously trained deep learning semantic segmentation network By processing it, the subsidence area in the road section was accurately identified and its location information was extracted. ,area ,depth , Edge slope , shape symmetry and profile curvature And other morphological characteristics.

[0107] Through analysis, the management office found that there were 8 obvious subsidence areas in the road section to be evaluated, and their geometric characteristics are shown in Table 8: Table 8 Morphological characteristics of subsidence area in the road section to be evaluated

[0108] As can be seen from Table 8, the geometric features of the eight subsidence areas in the evaluated section are different from those of the standard section. For example, the subsidence depth is between 43 and 92 mm, and the area is between 3.2 and 8.5 m. The edge slope is between 6.1°-14.3°, the shape symmetry is between 0.68-0.87, and the curvature is between 0.034-0.162 mm. This difference reflects the uniqueness of the development of diseases in different sections and lays the foundation for subsequent trend prediction and analysis.

[0109] S203, Environmental parameters and load monitoring: In order to fully understand the various factors that affect the development of subsidence in the road section to be evaluated, the management office deployed the same environmental monitoring system and traffic load monitoring system around the subsidence area of ​​the road section for 7 days. After 7 days of monitoring, the management office obtained the environmental parameter data of the road section to be evaluated. and traffic load data , as shown in Table 9 and Table 10: Table 9 Monitoring results of environmental parameters of the road section to be evaluated

[0110] Table 10 Traffic load monitoring results of the road section to be evaluated

[0111] It can be seen from Tables 9 and 10 that the environmental parameters and traffic load characteristics of the road section to be evaluated are somewhat different from those of the standard road section. For example, the average annual temperature cycle number is 69 times, and the temperature gradient accumulation is 1378°C·h / mm; the average annual relative humidity change is 0.103, the humidity change coefficient is 0.057, and the humidity gradient accumulation is 4.39%·m / mm; the annual rainfall is 721 mm, and the groundwater level is about 1.8 m; the strain accumulation is 0.092%. In terms of traffic load, the 5-ton, 10-ton, and 15-ton axle loads account for 29.7%, 45.4%, and 24.9%, respectively, and the number of actions is 1.51 million times, 1.33 million times, and 740,000 times, respectively; the average annual vehicle speed is between 89-115 km / h, and the braking frequency is between 0.09-0.22; the average daily traffic flow throughout the year is 16,800.

[0112] S204, road condition monitoring: Like the standard road section, the management office also conducted a 7-day continuous monitoring of the subsidence area of ​​the road section to be evaluated. Every day, they used the Leica BLK360 3D laser scanner to scan these areas and obtain the 3D point cloud data of the day. At the same time, the infrared temperature sensor installed on the road surface also records the road surface temperature in the area. of changes.

[0113] By comparing and analyzing these time series data, the management office has mastered the dynamic evolution law of the subsidence area of ​​the road section to be evaluated, laying the foundation for the subsequent establishment of a prediction model. Table 11 gives some monitoring results: Table 11 Dynamic monitoring results of subsidence area in the road section to be evaluated (partial)

[0114] From Table 11, we can see the change of the settlement depth of the 1# and 2# settlement areas of the evaluated section within 7 days. The 1# area increased from the initial 43.2 mm to 46.9 mm, and the 2# area increased from 47.6 mm to 52.3 mm, which can clearly reflect the development trend of the settlement of this section. At the same time, these data also show that there is a certain correlation between the settlement depth and the road surface temperature, which provides a reference for the subsequent establishment of a prediction model.

[0115] S205, Trend prediction: With the rich data obtained above, including the morphological characteristics of the subsidence area , Environmental parameters , Traffic load , Material parameters , subsidence depth time series and road surface temperature data ,The technical personnel of the management office input these data into the ,trained highway asphalt pavement subsidence disease trend prediction model for ,calculation analysis.

[0116] The prediction model consists of four modules: time series evolution equation, environmental impact equation, load response equation and comprehensive prediction equation. Each module integrates a lightweight neural network as an error correction term. Through the calculation and processing of the model, the management office obtained the subsidence prediction results of the road section to be evaluated within the next year, as follows: Table 12 Prediction results of subsidence damage in the road section to be evaluated

[0117] As can be seen from Table 12, the subsidence damage of the road section to be evaluated in the next year will show the following trends: the predicted subsidence depth is between 69-107 mm, which is relatively serious; the predicted subsidence area is between 4.7-11.2 m The predicted settlement development rate is between 12.3-19.4 mm / year, showing an accelerating trend; the predicted settlement stability is between 0.52-0.75, indicating that the road bearing capacity is gradually decreasing. Combining these prediction indicators, it can be judged that the road section to be evaluated has moderate to severe settlement damage.

[0118] like Figure 3 As shown in the figure, the curve shows the predicted trend of the depth evolution of the 1# and 2# subsidence areas of the road section to be evaluated in the next year, including the confidence interval of the predicted value. The figure clearly shows the trend of the change of subsidence depth over time and the uncertainty range.

[0119] S206. Disease Assessment and Reporting Based on the above subsidence damage prediction results, the management office classified the subsidence condition of the road section to be evaluated as moderate damage. The specific reasons are as follows: the predicted subsidence depth is between 69-107 mm, exceeding the 50 mm threshold of minor damage; the predicted subsidence area is between 4.7-11.2 m The range is also greater than 3 m for minor diseases. Threshold; The predicted subsidence development rate is between 12.3-19.4 mm / year, which is between mild and severe damage; The predicted subsidence stability is between 0.52-0.75, which has not yet reached below 0.5 of severe damage.

[0120] Based on the above four indicators, the road section to be evaluated is of moderate subsidence. According to the evaluation report, corresponding maintenance measures are implemented, such as minor repairs, medium repairs, etc., which can be matched accordingly based on experience.

[0121] Through the above analysis and evaluation, the management office has mastered the development of subsidence diseases in the road section to be evaluated, providing a scientific basis for subsequent maintenance and repair decisions. This can not only extend the service life of the highway, but also effectively improve driving safety, which is of great significance to improving the management level.

[0122] In general, the highway asphalt pavement subsidence disease trend assessment method proposed in this invention fully utilizes the advantages of advanced technical means such as three-dimensional laser scanning, deep learning and neural network models, comprehensively considers multiple factors such as environment, transportation and materials, and systematically analyzes the development law of subsidence diseases. Whether for standard sections or sections to be evaluated, this method can accurately predict the subsidence development trend in a certain period of time in the future, providing strong support for timely discovery and effective treatment of diseases.

[0123] It should be noted that the explanation of variables in the present invention is shown in Table 13 below.

[0124] Table 13 Variable explanation table

[0125] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A highway pavement disease prediction method, characterized in that: It includes a modeling stage and an application stage: the modeling stage specifically includes: long-term collection and preprocessing of three-dimensional point cloud data of asphalt pavement of standard road sections, use of deep learning networks to identify subsidence areas to obtain morphological feature data, collection of monitoring data of subsidence areas, establishment of training data sets, and use of the training data sets to train a group of subsidence disease prediction equations to obtain a subsidence disease trend analysis model; the application stage specifically includes: short-term collection and preprocessing of three-dimensional point cloud data of asphalt pavement of the highway to be evaluated, use of deep learning networks to identify subsidence areas to obtain morphological feature data, collection of monitoring data of the subsidence areas, obtaining subsidence prediction results based on the data and generating an evaluation report, and executing corresponding disease prediction measures based on the evaluation report.

2. A highway pavement damage prediction method according to claim 1, characterized in that: The modeling stage specifically includes the following steps: S101, collecting three-dimensional point cloud data of the asphalt pavement of a standard road section with a monitoring period of one year, and preprocessing the three-dimensional point cloud data to obtain preprocessed data; S102, segmenting the preprocessed data using a deep learning network, identifying a subsidence area in the asphalt pavement of the standard road section, and obtaining location information and morphological feature data of the subsidence area, wherein the morphological feature data includes subsidence depth data, subsidence area data, subsidence edge slope data, subsidence shape symmetry data, and subsidence profile curvature data; S103, collecting historical monitoring records of the subsidence area in the past year and establishing historical subsidence rate parameters; S104, deploying an environmental monitoring system to collect environmental monitoring data of the subsidence area within a one-year monitoring period, wherein the environmental monitoring data includes annual average temperature cycle data, temperature gradient accumulation data, seasonal temperature variation coefficient, daily temperature variation coefficient, annual average relative humidity variation data, humidity gradient accumulation data, rainfall data, groundwater level data, and strain accumulation data; S105, deploying a traffic load monitoring system to collect traffic monitoring data of the subsidence area within a one-year monitoring period, wherein the traffic monitoring data includes annual average axle load distribution data, axle load cumulative action times data, annual average vehicle speed distribution data, vehicle braking frequency distribution data, traffic flow data, vehicle type distribution data, and tire ground contact pressure data; S106, collecting material parameters of the subsidence area, wherein the material parameters include asphalt content data, void ratio data, gradation data, strength data, anti-deformation data, pavement structure layer parameters, base material parameters and subgrade soil parameters; S107, using a laser scanner to monitor the subsidence area daily during the one-year monitoring period to obtain depth time series data and road surface temperature data of the subsidence area; S108, establishing a training data set, including the pre-processed data, the morphological feature data, the historical subsidence rate parameters, the environmental monitoring data, the traffic monitoring data, the material parameters, the road surface temperature data and the depth time series data; S109, using the training data set to train the error correction term function of each equation in the pre-established subsidence disease prediction equation group containing a lightweight neural network error correction term function, and the trained subsidence disease prediction equation group is used as a highway asphalt pavement subsidence disease trend analysis model.

3. A highway pavement damage prediction method according to claim 2, characterized in that: The application stage specifically includes the following steps: S201, collecting three-dimensional point cloud data of the asphalt pavement of the highway to be evaluated with a monitoring period of 7 days, and preprocessing the three-dimensional point cloud data to obtain preprocessed data; S202, segmenting the preprocessed data using the deep learning network, identifying the subsidence area in the asphalt pavement of the highway to be evaluated, and obtaining morphological feature data of the subsidence area; S203, collecting environmental monitoring data, traffic monitoring data and material parameters of the asphalt pavement subsidence area of ​​the highway to be evaluated within the 7-day monitoring period; S204, collecting depth time series data and pavement temperature data of the asphalt pavement subsidence area of ​​the highway to be evaluated every day during the 7-day monitoring period; S205, inputting the morphological feature data, the environmental monitoring data, the traffic monitoring data, the material parameters, the depth time series data and the road surface temperature data into the highway asphalt pavement subsidence disease trend analysis model to obtain the subsidence prediction result; S206. Evaluate the subsidence damage grade of the asphalt pavement of the highway to be evaluated according to the subsidence prediction result, and generate a subsidence damage evaluation report.

4. A highway pavement damage prediction method according to claim 3, characterized in that: The subsidence prediction results specifically include predicted subsidence depth data, predicted subsidence area data, predicted subsidence development rate data and predicted subsidence stability data.

5. A highway pavement damage prediction method according to claim 4, characterized in that: The subsidence disease prediction equation group includes a time-series evolution equation, an environmental impact equation, a load response equation and a comprehensive prediction equation; each equation in the subsidence disease prediction equation group contains a lightweight neural network error correction term function.

6. A highway pavement damage prediction method according to claim 5, characterized in that: The time series evolution equation is used to calculate the change pattern of subsidence depth over time. The input includes the depth time series data, the subsidence area data, the subsidence edge slope data, the subsidence shape symmetry data, the subsidence profile curvature data, the time increment parameter and the historical subsidence rate parameter. The output is the time evolution trend value of the subsidence depth.

7. A highway pavement damage prediction method according to claim 6, characterized in that: The environmental impact equation is used to quantify the influence of temperature and humidity on the development of subsidence. The input includes the annual average temperature cycle data, the temperature gradient accumulation data, the annual average relative humidity change data, the humidity gradient accumulation data, the strain accumulation data, the seasonal temperature change coefficient, the daily temperature change coefficient, the rainfall data and the groundwater level data. The output is the environmental impact coefficient.

8. A highway pavement damage prediction method according to claim 7, characterized in that: The load response equation is used to evaluate the contribution of traffic load to the deepening of subsidence. The input includes the annual average axle load distribution data, the cumulative number of axle load actions data, the annual average vehicle speed distribution data, the vehicle braking frequency distribution data, the asphalt content data, the void ratio data, the grading data, the strength data, the anti-deformation data, the traffic flow data, the vehicle type distribution data, the tire ground contact pressure data and the road surface temperature data. The output is the load effect value.

9. A highway pavement damage prediction method according to claim 8, characterized in that: The comprehensive prediction equation is used to integrate various prediction parameters to calculate the final subsidence development trend. The input includes the subsidence depth time evolution trend value, the environmental impact coefficient, the load effect value, the pavement structure layer parameters, the base material parameters, and the roadbed soil parameters. The output is predicted subsidence depth data, predicted subsidence area data, predicted subsidence development rate data, and predicted subsidence stability data.

10. A highway pavement disease prediction device, characterized in that: include: A processor, configured to read a computer program in a storage medium and execute the method according to any one of claims 1 to 9.

Images