Asphalt pavement construction quality control method based on multi-source data fusion

Abstract

The application discloses a kind of asphalt pavement construction quality control methods based on multi-source data fusion, comprising: obtaining temperature field monitoring and the motion state of equipment and other multi-source heterogeneous data, extract the physical characteristic elements of temperature field including rolling heat footprint or topological skeleton, and construct motion consistency constraint;Using probability correlation model, the physical characteristics and motion constraints are jointly solved, and the registration confidence and unified space-time reference data representing space-time alignment uncertainty are generated;Based on this, the physical consistent intra-layer temperature state is reconstructed, the effective operation time window is predicted in combination with the confidence, and an adaptive remedial strategy is generated to identify quality abnormalities.The application realizes cross-device accurate registration self-healing, and solves the data fusion problem caused by positioning degradation and temperature time variation.

Description

Technical Field

[0001] This invention relates to the field of digital construction technology for road engineering, and in particular to a method for quality control of asphalt pavement construction based on multi-source data fusion. Background Technology

[0002] The construction quality of asphalt pavement directly determines the service life and performance of the road. Among these factors, segregation at paving temperature and insufficient compaction are the main causes of early water damage and cracking. By integrating multi-source heterogeneous data from pavers, rollers, and environmental monitoring, a full-process, full-section digital monitoring system can be constructed to achieve real-time perception and closed-loop feedback of the construction status. This system has significant engineering application value for ensuring the uniformity of pavement compaction and optimizing construction process parameters.

[0003] Current digital construction technologies typically utilize GNSS (Global Navigation Satellite System) receivers installed on construction equipment to acquire movement trajectories, combined with infrared temperature sensors to collect road surface temperature data. In the data processing stage, a direct matching method based on absolute geographic coordinates is mainly employed. This involves mapping temperature data and compaction trajectories to the same geographic grid using GNSS timestamps to determine the number of compaction passes and operating temperature at a specific location, thereby assessing construction quality. Some solutions attempt to introduce image feature matching, using road surface texture or obvious temperature boundaries as landmarks to help correct GNSS positioning errors and improve the collaborative accuracy of multi-machine operation data.

[0004] However, existing technologies suffer from severe limitations in cross-device spatiotemporal registration accuracy under complex working conditions. The main issues are data alignment failures due to positioning signal degradation and feature matching drift caused by time-varying temperature fields. Specifically, construction sites often face signal obstruction, and relying solely on GNSS can lead to inaccurate spatial coordinate correspondence between paving and compaction data, resulting in misjudgments of quality. Furthermore, asphalt temperature decays rapidly over time, with significant differences in absolute temperature values ​​at the same location at different times. This makes it difficult for existing feature matching methods based on static temperature thresholds to establish stable correlation benchmarks during dynamic cooling. In addition, the lack of a direct physical causal mapping between continuous temperature field images and discrete equipment trajectories makes self-healing of multi-source data fusion difficult in the absence of high-precision positioning. Summary of the Invention

[0005] The purpose of this invention is to provide a method for controlling the construction quality of asphalt pavement based on multi-source data fusion, in order to solve at least one of the aforementioned problems in the existing technology.

[0006] According to one aspect of this application, a method for quality control of asphalt pavement construction based on multi-source data fusion includes:

[0007] Acquire multi-source heterogeneous data from asphalt construction sites, including at least temperature field monitoring data and construction equipment motion status data;

[0008] Physical characteristic elements of the temperature field are extracted from temperature field monitoring data, and motion consistency constraints are constructed based on the motion state data of construction equipment.

[0009] By using a probabilistic correlation model, the physical characteristics of the temperature field and the motion consistency constraints are jointly solved to generate unified spatiotemporal reference data and registration reliability characterizing the spatiotemporal alignment uncertainty.

[0010] Based on unified spatiotemporal reference data, the temperature state within the layer is reconstructed using a physically consistent intra-layer state evolution model, and the effective working time window of the construction grid is predicted by combining the registration confidence.

[0011] Based on unified spatiotemporal reference data and effective operation time windows, quality anomalies or interface risks during construction are identified, and adaptive remedial control strategies are generated accordingly.

[0012] According to another aspect of this application, an asphalt pavement construction quality control system is provided, comprising:

[0013] Memory, used to store computer programs;

[0014] A processor is used to execute a computer program to implement the steps of any one of the methods for quality control of asphalt pavement construction based on multi-source data fusion.

[0015] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of any one of the methods for asphalt pavement construction quality control based on multi-source data fusion.

[0016] Beneficial effects: Through the above technical solutions, this invention achieves accurate cross-device registration and self-healing, and solves the data fusion problem caused by positioning degradation and temperature variations. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the process for a method of controlling the construction quality of asphalt pavement based on multi-source data fusion provided in the embodiments of this application.

[0018] Figure 2 This is a schematic diagram of the dynamic differential feature extraction process based on the compaction thermal footprint provided in the embodiments of this application.

[0019] Figure 3 This is a schematic diagram of the static topological feature extraction process based on the isotherm topological skeleton provided in the embodiments of this application.

[0020] Figure 4 This is a schematic diagram of the process of jointly solving the physical characteristics of the temperature field and the motion consistency constraints using a probabilistic correlation model, as provided in the embodiments of this application. Detailed Implementation

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] Example 1 details the overall architecture of an asphalt pavement construction quality control method based on multi-source data fusion, as well as the preliminary data access and spatiotemporal benchmark unification process. This addresses the technical problem of the inability to fuse heterogeneous multi-source data due to inconsistencies in temporal and spatial benchmarks, laying a data foundation for subsequent accurate registration and quality control. Figure 1 As shown.

[0023] Step 101: Obtain multi-source heterogeneous data from the asphalt construction site. The multi-source heterogeneous data shall include at least temperature field monitoring data and construction equipment motion status data.

[0024] In this embodiment, the acquisition of multi-source heterogeneous data relies on various sensors and monitoring devices deployed at the construction site. Specifically, temperature field monitoring data mainly comes from an infrared thermal imager installed behind the paver's crossbeam, which can collect the surface temperature distribution of freshly laid asphalt mixture in real time in an array format, forming a continuous temperature field image sequence. Additionally, it includes single-point temperature data collected by point-type infrared temperature sensors installed on the roller. The motion status data of the construction equipment mainly comes from GNSS receivers installed on the paver and roller, such as high-precision positioning modules supporting RTK (Real-Time Kinematic) differential positioning, which can provide centimeter-level spatial coordinate data; and operational status parameters such as travel speed, vibration frequency, and amplitude read through the equipment controller local area network (CAN) bus interface.

[0025] In addition, multi-source heterogeneous data can also include environmental parameters such as temperature, wind speed, and humidity collected by environmental monitoring stations installed at the construction site, as well as spraying volume records uploaded by tack coat spraying trucks. All of the above data is aggregated in real time to the data processing center via the construction site's wireless communication network (such as 4G / 5G or a mesh self-organizing network).

[0026] In one specific implementation, the infrared thermal imager has a spatial resolution of 640×480 pixels or higher, a temperature measurement range of -20℃ to 350℃, a temperature measurement accuracy better than ±2℃ or ±2% (whichever is greater), and a frame rate of 10 to 30 frames per second. The GNSS receiver in RTK differential mode has a positioning accuracy of ±2cm horizontally and ±5cm vertically, with a data output frequency of 1 to 10Hz. The inertial measurement unit has an acceleration measurement range of ±16g, an angular velocity measurement range of ±2000 degrees per second, and a sampling frequency of 100 to 200Hz.

[0027] Step 102: Extract the physical characteristics of the temperature field from the temperature field monitoring data, and construct motion consistency constraints based on the motion state data of the construction equipment.

[0028] In this step, the core feature extraction will be detailed in subsequent embodiments. This embodiment focuses on the preprocessing work before the data enters the core algorithm, namely, the unification of time reference and the normalization of coordinate reference. Specifically, since the sampling frequencies of sensors in different devices are inconsistent (e.g., GNSS may be 1Hz, while thermal imagers may be 10Hz), and network transmission has random delays, time alignment is required first.

[0029] The system calculates and compensates for transmission delays based on hardware-generated timestamps in the data packets reported by each device, combined with the reference time synchronized by the Network Time Protocol (NTP). It then uses linear interpolation to align all data to a preset global time axis, for example, a sampling point every 100 milliseconds. Coordinate reference normalization is performed, converting the GNSS latitude and longitude coordinates (WGS-84, World Geodetic System 1984) of all devices to the local engineering coordinate system of the construction site (such as Gauss-Kruger projection coordinates). Based on the vehicle's geometry and sensor installation offset, the positioning center is corrected to the actual operating center of the equipment, such as the center of a road roller's steel wheel or a paver's screed.

[0030] In some optional implementations, the system identifies potential positioning degradation windows for GNSS signals, such as signal loss due to obstruction or multipath effects, based on the intensity of trajectory jumps. For example, when the rate of displacement change of a device exceeds its maximum physical speed limit (e.g., a road roller's maximum speed is 2 m / s, but a sudden change in displacement to 10 m / s is detected), or when the positioning status flag shows a non-differential fixed solution, this time period is marked as a positioning degradation window. This marking serves as a condition for subsequently triggering cross-device registration self-healing, ensuring that positioning can still be performed using physical features even when GNSS positioning degrades.

[0031] Step 103: Use the probabilistic correlation model to jointly solve the physical characteristics of the temperature field and the motion consistency constraints to generate unified spatiotemporal reference data and registration credibility that characterizes the spatiotemporal alignment uncertainty.

[0032] In this embodiment, this step involves generating gridded fusion base data. Specifically, the system divides the construction area into regular two-dimensional grids with a preset spatial resolution (e.g., 0.1m × 0.1m). For each grid, its corresponding unified spatiotemporal reference data is associated. The unified spatiotemporal reference data can be represented as a dataset D containing multidimensional attributes. ref :

[0033] D ref ={(x,y,t,v,status,T surf )},

[0034] Where (x,y) are the coordinates of the grid center, t is the time when the grid is operated on (e.g., paving or compaction), v is the speed of the equipment passing by, status is the operation status, such as vibration on / off, and T surf This represents the surface temperature value of the grid.

[0035] Registration reliability is an important basis for subsequent quality control decisions. It is a quantitative indicator with a value between 0 and 1. The closer the value is to 1, the higher the certainty of the spatiotemporal alignment of the data in the grid. Conversely, it indicates that there is a greater risk of error.

[0036] Step 104: Based on unified spatiotemporal reference data, reconstruct the temperature state within the layer using a physically consistent intra-layer state evolution model, and combine the registration confidence level to predict the effective operation time window of the construction grid.

[0037] In this embodiment, the system utilizes a heat conduction physical model, with the measured temperature T of the grid surface as the reference value. surf As boundary conditions, combined with parameters such as ambient temperature and wind speed, the temperature distribution T inside the asphalt layer (such as the core or bottom) is calculated. core Based on this, the remaining time required for the grid temperature to drop to the minimum compactable temperature (e.g., 80°C) is predicted, which is the effective working time window. Registration confidence level acts as a safety valve here: when the confidence level is low, the system automatically shortens the predicted effective time window, leaving a larger safety margin to prevent forced compaction due to positioning errors when the temperature is too low.

[0038] Step 105: Identify quality anomalies or interface risks during the construction process based on unified spatiotemporal reference data and effective operation time windows, and generate adaptive remedial control strategies based on the quality anomalies or interface risks.

[0039] Specifically, the system scans the status data of all grids in real time. If an abnormal sudden change in equipment speed is detected in a certain area, such as abrupt stops and starts, or failure to complete the prescribed number of compaction passes after the effective time window has expired, it is identified as a quality anomaly. For interface risks, the system comprehensively analyzes the amount of tack coat applied and the waiting time to identify areas with potential incomplete demulsification or contamination. Based on the identification results, the system automatically generates remedial control strategies, such as sending instructions to the roller operator to prioritize compaction of area A, or sending a warning message to site managers that area B requires additional tack coat application. These remedial control strategies are distributed through the construction management terminal, forming a closed-loop control system.

[0040] In practical implementation, the detection of compaction thermal footprints is suitable for the initial and secondary compaction stages when the surface temperature of the asphalt mixture is relatively high. When the surface temperature drops to the final compaction temperature range (e.g., 80–90°C), the signal-to-noise ratio of the thermal footprint signal may decrease due to the weakening of frictional heating effect and the reduction of ambient temperature contrast. At this time, the system can automatically increase the dependence weight on static topological features, or maintain registration accuracy by lowering the temperature rise detection threshold and using stricter morphological filtering constraints.

[0041] In other words, the detection performance of dynamic differential features is affected by the asphalt surface temperature and environmental conditions. When the asphalt surface temperature is high (e.g., above 120℃), the mixture is in the initial or intermediate compaction stage. At this time, the natural cooling rate is relatively fast, the background temperature drop between adjacent sampling frames is relatively uniform, and the local temperature rise generated by rolling can be effectively separated by differential calculation, resulting in a high signal-to-noise ratio. When the asphalt surface temperature approaches the final compaction temperature (e.g., 80-90℃), the viscosity of the mixture increases, the frictional heating effect weakens, and the natural cooling rate slows down, leading to a decrease in the signal amplitude of the rolling thermal footprint and increasing the detection difficulty.

[0042] In some preferred embodiments, before performing the threshold detection in step 202, the system further includes a signal-to-noise ratio pre-judgment step: based on the average temperature level of the current temperature field and the noise equivalent temperature difference parameter of the infrared thermal imager, the theoretically detectable minimum temperature rise threshold is calculated; when the ratio of the temperature rise threshold to the preset typical temperature rise amplitude of the rolling thermal footprint is lower than the preset signal-to-noise ratio threshold (e.g., 3:1), the system automatically switches to a registration mode based on static topological features to ensure the stability of registration accuracy.

[0043] According to one aspect of this application, a method for controlling the construction quality of asphalt pavement based on multi-source data fusion may further include the following steps:

[0044] Acquire multi-source heterogeneous data from the asphalt construction site. The multi-source heterogeneous data includes at least temperature field monitoring data, construction equipment movement status data, environmental monitoring data, and tack coat operation data. Among them, the construction equipment movement status data includes equipment positioning data and corresponding positioning status flags.

[0045] Physical characteristic elements of the temperature field are extracted from temperature field monitoring data, and motion consistency constraints are constructed based on the motion state data of construction equipment.

[0046] By using a probabilistic correlation model, the physical characteristics of the temperature field and the motion consistency constraints are jointly solved to generate unified spatiotemporal reference data and registration reliability characterizing the spatiotemporal alignment uncertainty.

[0047] Based on unified spatiotemporal reference data and environmental monitoring data, the temperature state within the layer is reconstructed using a physically consistent intralayer state evolution model, and the effective working time window of the construction grid is predicted by combining the registration reliability.

[0048] Based on unified spatiotemporal reference data, tack coat operation data, and effective operation time windows, quality anomalies or interface risks during the construction process are identified, and adaptive remedial control strategies are generated based on these anomalies or interface risks.

[0049] Example 2 further refines the steps for extracting physical features of the temperature field from temperature field monitoring data. It elaborates on how to extract two different physical features from continuous temperature field data: the compaction thermal footprint based on dynamic thermal effects and the topological skeleton based on static distribution structure. It also provides specific calculation methods and noise reduction logic to solve the problem of insufficient robustness of single features in complex construction environments.

[0050] The physical characteristics of a temperature field include at least one of the following:

[0051] Dynamic differential features are extracted based on time-series differential operations of temperature field monitoring data and are used to characterize the transient thermal effects generated during the operation of construction equipment.

[0052] Static topological features are extracted based on the spatial distribution structure of temperature field monitoring data and are used to characterize the spatiotemporal invariance of the temperature field distribution structure during the cooling process.

[0053] Dynamic differential feature extraction based on compaction thermal footprints, such as Figure 2 As shown, the specific process includes the following:

[0054] The transient temperature rise effect (i.e., rolling thermal footprint) generated during road roller compaction is used as the registration feature.

[0055] Step 201: Calculate the temperature field difference sequence of the temperature field monitoring data at adjacent sampling times to eliminate the overall cooling attenuation trend of the temperature field.

[0056] Specifically, since freshly laid asphalt mixtures cool naturally over time, directly using absolute temperature values ​​for feature extraction is easily affected by the overall cooling process. This embodiment employs the time-difference method to eliminate this low-frequency background trend and highlight high-frequency local variations. The system calculates the temperature field difference sequence according to the following formula:

[0057] ΔT(x,y,t)=T(x,y,t+δt)-T(x,y,t);

[0058] Where ΔT(x,y,t) represents the temperature change at position (x,y) and time t; T(x,y,t) represents the original surface temperature measurement at that position at time t; and δt represents the time interval between two adjacent sampling frames, for example, 1 second.

[0059] The calculation shows that the natural cooling of asphalt is a small negative value, while the frictional heat and internal heat release from the roller compaction are significant positive values ​​(temperature rise).

[0060] Step 202: Detect regions in the temperature field difference sequence where the local temperature rise exceeds a preset threshold.

[0061] In this step, the system performs thresholding on the calculated difference image ΔT (temperature change). The preset temperature rise threshold θ... rise The setting can be adjusted based on the ambient temperature and asphalt type, typically between 0.5℃ and 2.0℃. Only when ΔT(x,y,t) > θ rise Only when this condition is met will the grid be marked as a candidate thermal footprint region. This step effectively filters out minute temperature fluctuations caused by ambient winds or random sensor noise.

[0062] Step 203: Use the preset constraints of roller wheel width and working length to perform morphological filtering on the region, and determine the sequence of center positions of the region that meets the constraints as dynamic differential features.

[0063] The actual thermal footprint is generated by the steel wheel of the road roller, and its spatial shape must conform to the geometric characteristics of the steel wheel. The system performs connectivity analysis on the candidate regions, calculating the lateral width w and longitudinal length L of each connected region, and pre-setting a width constraint range [W] that matches the construction equipment. min W max ](W min W is a preset minimum allowable width threshold, representing the lower limit of the width of the steel wheel of construction equipment (such as a road roller). max This is a preset maximum allowable width threshold, representing the upper limit of the steel wheel width. For example, if the steel wheel width is 2 meters, the setting range is 1.8 meters to 2.2 meters, and a minimum length constraint L. min For example, 0.5 meters. Only when W is satisfied... min <=w<=Wmax And L>=L min Only then was the area confirmed as a valid compaction thermal footprint. The system extracted the geometric center coordinates p of this area. t This forms a time-varying sequence of central positions, which serves as a dynamic difference feature.

[0064] Furthermore, to enhance noise immunity, the system also performs temporal continuity verification, calculating the displacement velocity of the center position of the region within consecutive time frames; if the displacement velocity exceeds the preset maximum travel speed threshold of the road roller, the corresponding region is identified as noise and removed.

[0065] Based on the physical properties of the road roller, its moving speed cannot exceed its maximum travel speed v. max (For example, 6 km / h). The system uses the following logical formula to verify the detected thermal footprint sequence:

[0066] ;

[0067] Where C(t) is the continuity score at time t; K is the number of historical frames traced back (e.g., 5 frames); p t It is the location of the center of the thermal footprint detected at the current moment; p t-k*δt δt represents the position of the previous k frames, and dist represents the time interval between two adjacent sampling frames; dist(...) represents the Euclidean distance between the two points; 1[...] is the indicator function, which takes the value 1 when the condition is met, and 0 otherwise. If the calculated continuity score C(t) is lower than the preset threshold (e.g., 0.8), it indicates that the thermal footprint has an unreasonable jump in time sequence, such as moving too far in an instant. The system judges it as false noise and removes it.

[0068] Static topological feature extraction based on isotherm topological skeleton, such as Figure 3 As shown, the details are as follows:

[0069] The spatiotemporal invariance of the temperature field distribution structure during the cooling process (i.e., the topological skeleton) is used as a registration feature. This method is independent of the roller's movement and is suitable for registration in areas where the roller is not operating.

[0070] Step 204: Calculate the gradient field and Hessian matrix of the temperature field monitoring data using the discrete difference operator, and identify the gradient critical points where the gradient is zero. Gradient critical points include local maxima, local minima, and saddle points.

[0071] Specifically, for a temperature field image T(x,y) at a certain time t, the system uses the Sobel operator or the central difference method to calculate its first derivative (gradient) and second derivative (Hessian matrix). Points where the gradient is zero are critical points. The Hessian matrix is ​​classified by the sign of its eigenvalues: if both eigenvalues ​​are negative, it is a local maximum (hot spot); if both are positive, it is a local minimum (cold spot); if one is positive and the other negative, it is a saddle point. These points typically correspond to physical structural features such as segregation zones in the mixture and joints between paving operations of different batches.

[0072] Step 205: Construct isotherm paths connecting gradient critical points to form the topological framework of the temperature field.

[0073] In this step, the system establishes connections between critical points based on Morse-Smale complex theory and an integral line tracing algorithm. For example, it traces paths from saddle points along the gradient ascent direction to maxima, or along the gradient descent direction to minima. These connections constitute the skeleton structure of the temperature field.

[0074] Step 206: Calculate the position drift variance of the gradient critical point or topological skeleton within the time window, and select skeleton structures with position drift variance less than the preset stability threshold as static topological features.

[0075] Because the temperature field may deform during cooling, the skeleton structure is tracked over a time window (e.g., 30 seconds) to extract stable landmarks. The degree of positional drift of each critical point in the time series is calculated using a stability score formula.

[0076] ;

[0077] Among them, S i It is the stability score of the i-th critical point; N is the number of sampling frames within the time window; p i (t n ) is the point at t n Position at time; p avg_i This represents the average position of the point within the window; σ is the allowable positional deviation, typically ranging from 0.3 to 0.5 meters. The system filters out S... i Critical points exceeding a preset threshold (e.g., 0.8) and their connections are used as the final static topological features.

[0078] In addition, a multi-scale topological feature vector f is constructed. i =[d (i) ,θ (i) ,τ i ],

[0079] Where, d(i) Including the relative distance from the point to the nearest critical point, θ (i) Includes relative azimuth, τ i It uses type encoding (maximum / minimum / saddle point). Descriptors that do not rely on absolute coordinates improve the accuracy of cross-time period matching.

[0080] After filtering out static topological features, the following steps are also included:

[0081] For each selected gradient critical point, a multi-scale topological feature vector is constructed. The multi-scale topological feature vector contains the relative distance, relative azimuth angle and critical point type encoding between the critical point and its nearest neighbor critical points to support cross-time period topological structure matching.

[0082] According to one aspect of this application, as a supplement to or alternative to the two methods described above, a process for boundary feature extraction based on temperature gradient is provided.

[0083] Specifically, temperature abrupt change boundaries, cold joint zones, or continuous temperature gradient zones in temperature field monitoring data are identified; the spatial coordinates of temperature abrupt change boundaries, cold joint zones, or continuous temperature gradient zones are extracted as static topological features.

[0084] In certain construction scenarios, such as when cold joints are very obvious, the system can directly identify temperature abrupt change boundaries (i.e., edge lines with the largest gradient magnitude) or continuous temperature gradient bands in temperature monitoring data. The pixel coordinate sequence on the boundary line is extracted and used as a simplified form of static topological features for registration.

[0085] According to one aspect of this application, the detailed process of constructing the topology skeleton is as follows:

[0086] For a discrete temperature field image with a resolution of M rows × N columns, let the temperature value at pixel coordinates (i,j) be T(i,j), where i represents the row index and j represents the column index. The central difference method is used to calculate the temperature gradient vector at this point.

[0087] The gradient component along the horizontal direction is calculated using the following formula:

[0088] G x (i,j)=(T(i,j+1)-T(i,j-1)) / (2×Δx);

[0089] Among them, G x (i,j) represents the horizontal temperature gradient component at pixel (i,j), in degrees Celsius (°C) per meter; T(i,j+1) represents the temperature value of the pixel to the right of this pixel; T(i,j-1) represents the temperature value of the pixel to the left of this pixel; Δx represents the horizontal physical spacing between pixels, in meters.

[0090] The gradient component along the vertical direction is calculated using the following formula:

[0091] G y (i,j)=(T(i+1,j)-T(i-1,j)) / (2×Δy);

[0092] Among them, G y (i,j) represents the temperature gradient component along the vertical direction at pixel (i,j); T(i+1,j) represents the temperature value of the adjacent pixel below this pixel; T(i-1,j) represents the temperature value of the adjacent pixel above this pixel; Δy represents the physical spacing between pixels in the vertical direction.

[0093] For pixels located at image boundaries, a one-sided difference formula is used instead of a central difference formula. For example, for pixels on the left boundary, the horizontal gradient is calculated using the following formula: G x (i,0)=(T(i,1)-T(i,0)) / Δx;

[0094] The gradient vector at this point can be expressed as: ▽T(i,j)=(G x (i,j),G y (i,j));

[0095] Where ▽T(i,j) represents the temperature gradient vector at pixel (i,j), which points in the direction of the fastest temperature increase, and its magnitude represents the degree of temperature change.

[0096] A critical point is a pixel whose gradient vector magnitude approaches zero, i.e., a pixel that satisfies the following condition:

[0097] |▽T(i,j)|=sqrt(G x (i,j) 2 +G y (i,j) 2 )<ε g ;

[0098] Where |▽T(i,j)| represents the magnitude of the gradient vector; sqrt represents the square root operation; ε g This indicates the preset gradient magnitude threshold, typically ranging from 0.01 to 0.05 degrees Celsius per meter.

[0099] For the detected candidate critical points, the type of critical point is determined by calculating the Hessian matrix and analyzing its eigenvalues. The Hessian matrix is ​​a symmetric matrix composed of the second-order partial derivatives of the temperature function, and is calculated as follows.

[0100] Second partial derivative H xx Calculate using the following formula:

[0101] H xx(i,j)=(T(i,j+1)-2×T(i,j)+T(i,j-1)) / (Δx 2 );

[0102] Among them, H xx (i,j) represents the second-order partial derivative of the temperature function in the horizontal direction at (i,j).

[0103] Second partial derivative H yy Calculate using the following formula:

[0104] H yy (i,j)=(T(i+1,j)-2×T(i,j)+T(i-1,j)) / (Δy 2 );

[0105] Among them, H yy (i,j) represents the second-order partial derivative of the temperature function at (i,j) along the vertical direction.

[0106] Mixed partial derivative H xy Calculate using the following formula:

[0107] H xy (i,j)=(T(i+1,j+1)-T(i+1,j-1)-T(i-1,j+1)+T(i-1,j-1)) / (4×Δx×Δy);

[0108] Among them, H xy (i,j) represents the mixed second-order partial derivative of the temperature function at (i,j).

[0109] The Hessian matrix H can be expressed as: H = [[H xx H xy ],[H xy H yy ]];

[0110] Here, H is a 2x2 symmetric matrix.

[0111] The two eigenvalues ​​λ1' and λ2' of the Hessian matrix are calculated using the following formula:

[0112] λ1'=((H xx +H yy )+sqrt((H xx -H yy ) 2 +4×H xy 2 )) / 2;

[0113] λ2'=((H xx +H yy )-sqrt((H xx -Hyy ) 2 +4×H xy 2 )) / 2;

[0114] Where λ1' represents a larger eigenvalue and λ2' represents a smaller eigenvalue.

[0115] Based on the combination of the signs of the two eigenvalues, critical points are classified into the following three types:

[0116] When λ1'<0 and λ2'<0, that is, both eigenvalues ​​are negative, the critical point is a local maximum point, which corresponds to the center of the local high-temperature region in the temperature field.

[0117] When λ1'>0 and λ2'>0, that is, both eigenvalues ​​are positive, the critical point is a local minimum point, which corresponds to the center of the local low temperature region in the temperature field.

[0118] When λ1' and λ2' have opposite signs, that is, one eigenvalue is positive and the other is negative, the critical point is called a saddle point, which corresponds to the transition position in the temperature field where the temperature increases in one direction and decreases in another orthogonal direction.

[0119] An integral line is a curve that starts from a saddle point and traces along the gradient direction or the negative gradient direction until it reaches an extremum. The integral line connecting the saddle point and the maximum point is called the ascending line, and the integral line connecting the saddle point and the minimum point is called the descending line. The network formed by all integral lines constitutes the one-dimensional skeleton of the Morse-Small complex.

[0120] Integral line tracking is implemented using a fourth-order Runge-Kutta iterative algorithm. Let the position of the current tracking point be P. k =(x k ,y k When tracing along the gradient ascent direction, the next position P k+1 Calculate using the following steps.

[0121] First, calculate the four intermediate increments: K1 = h × ▽T(x k ,y k ) / |▽T(x k ,y k )|;

[0122] Where K1 represents the first increment vector; h represents the tracking step size, typically ranging from 0.5 to 1.0 pixel spacing; ▽T(x k ,y k ) indicates at point (x k ,y k The gradient vector obtained by bilinear interpolation at position ); |▽T(x) k ,y kThe )| represents the magnitude of the gradient vector, used to normalize the gradient to a unit vector.

[0123] K2=h×▽T(x k +K 1x / 2,y k +K 1y / 2) / |▽T(x k +K 1x / 2,y k +K 1y / 2)|;

[0124] Where K2 represents the second increment vector; K 1x and K 1y These represent the horizontal and vertical components of K1, respectively.

[0125] K3=h×▽T(x k +K 2x / 2,y k +K 2y / 2) / |▽T(x k +K 2x / 2,y k +K 2y / 2)|;

[0126] Here, K3 represents the third increment vector.

[0127] K4=h×▽T(x k +K 3x ,y k +K 3y ) / |▽T(x k +K 3x ,y k +K 3y )|;

[0128] Here, K4 represents the fourth increment vector.

[0129] The position of the next tracking point is updated according to the following formula:

[0130] P k+1 =P k +(K1+2×K2+2×K3+K4) / 6;

[0131] The termination conditions for integral line tracking include the following three cases:

[0132] The current tracking point enters a preset neighborhood of a certain extreme point, and the radius of this neighborhood typically takes the value of 2 to 3 pixels;

[0133] The gradient magnitude at the current tracking point is less than the preset threshold ε g This indicates that the critical point is approaching.

[0134] The number of tracking steps exceeds the preset maximum number of iterations N. max The typical value is 500 to 1000 steps, used to prevent infinite loops caused by numerical errors.

[0135] For a descent line traced along the negative gradient direction, simply replace the gradient vector ▽T in the above formula with its negative value.

[0136] In a given scenario, suppose there exists a 5x5 temperature field region, with the temperature value matrix shown below, in degrees Celsius (°C):

[0137] Line 1: 120.0,122.5,125.0,122.5,120.0;

[0138] Line 2: 117.5,120.0,122.5,120.0,117.5;

[0139] Line 3: 115.0,117.5,120.0,117.5,115.0;

[0140] Line 4: 117.5,120.0,122.5,120.0,117.5;

[0141] Line 5: 120.0,122.5,125.0,122.5,120.0;

[0142] Assume that the physical pixel spacing Δx and Δy are both 0.01 meters.

[0143] For the center pixel (3,3), i.e., the 3rd row and 3rd column, calculate the gradient components according to the center difference formula:

[0144] G x (3,3)=(T(3,4)-T(3,2)) / (2×0.01)=(117.5-117.5) / 0.02=0;

[0145] G y (3,3)=(T(4,3)-T(2,3)) / (2×0.01)=(122.5-122.5) / 0.02=0;

[0146] If the gradient magnitude is zero, this point is a candidate critical point.

[0147] Calculate the elements of the Hessian matrix:

[0148] H xx (3,3)=(117.5-2×120.0+117.5) / (0.01 2 = -5.0 / 0.0001 = -50000;

[0149] H yy (3,3)=(122.5-2×120.0+122.5) / (0.01 2 =5.0 / 0.0001 = 50000;

[0150] H xy (3,3)=(120.0-120.0-120.0+120.0) / (4×0.01×0.01)=0;

[0151] Eigenvalue calculation:

[0152] λ1'=((-50000+50000)+sqrt((-50000-50000) 2 +0)) / 2=50000;

[0153] λ2'=((-50000+50000)-sqrt((-50000-50000) 2 +0)) / 2=-50000;

[0154] Since λ1'>0 and λ2'<0, the two eigenvalues ​​have opposite signs, so the critical point is determined to be a saddle point.

[0155] In practical applications, integral line tracing is performed on all detected saddle points, and the resulting skeleton network can be used as the topological features for registration.

[0156] Example 3: A detailed explanation of the process of jointly solving the physical characteristics of the temperature field and the consistency constraints of motion using a probabilistic correlation model, such as... Figure 4 As shown, the probabilistic association model is a Bayesian inference model. This embodiment elaborates on how to use the Bayesian probabilistic framework to fuse heterogeneous physical features (thermal footprint / topology) with kinematic constraints, and to solve the technical problem of traditional rigid registration failing in a positioning degradation environment by quantifying the credibility of registration through posterior variance.

[0157] Step 301: Take the physical characteristics of the temperature field as the observation variables, and transform the motion consistency constraint into a state transition constraint to construct a likelihood function that reflects the observation likelihood.

[0158] In this embodiment, the system models the registration problem as an estimation problem of the actual trajectory state x of the device. For a certain time t, assume the system observes a valid physical feature of the temperature field, such as the thermal footprint center p. foot Simultaneously, relative displacement was estimated using wheel and axle encoders or inertial navigation (motion consistency constraints). The system constructs a likelihood function p(z|x), representing the probability of observing feature z given the true position x. Specifically, this likelihood function consists of two parts:

[0159] p(z|x)=p(z foot |x)*p(z cons |x);

[0160] Wherein, p(z) foot |x) represents the likelihood of the thermal footprint observation, which is usually assumed to follow a Gaussian distribution:

[0161] p(z foot |x)=N(z foot ;x,σ foot 2 );

[0162] Here, z foot It is the position of the thermal footprint in the image coordinate system, σ foot This is the standard deviation of the observed noise, for example, 0.5 meters;

[0163] p(z cons |x) represents the likelihood of the motion consistency constraint, that is, the current position of the device should satisfy the constraint of the previous position plus the relative displacement:

[0164] p(z cons |x)=N(x;x t-1 +Δx odo ,σ cons 2 );

[0165] Here, x t-1 It is the estimated position at the previous moment, Δx odo It is the displacement calculated by the odometer, σ cons It is the calculation of cumulative error.

[0166] Step 302: Based on the historical motion state or reliable positioning interval of the construction equipment extracted from the motion state data of the construction equipment, construct a prior distribution that reflects the distribution characteristics of the trajectory to be estimated.

[0167] Specifically, based on the positioning status flags in the motion status data of the construction equipment, the reliable positioning intervals in which the positioning signal quality meets the preset requirements during the construction process are identified; the motion status parameters of the construction equipment within the reliable positioning intervals are statistically analyzed to generate initial prior distribution parameters; for the positioning signal degradation intervals, based on the initial prior distribution parameters and the diffusion process over time, a prior distribution with time-varying covariance is generated, wherein the time-varying covariance monotonically increases with the time interval moving away from the reliable positioning interval.

[0168] This step details how to construct a priori information using historical data during periods of lost (degraded) positioning signals:

[0169] The uncertainty in the position estimate at the end of the reliable positioning interval is determined as the initial covariance benchmark;

[0170] A diffusion coefficient that varies with time interval is constructed using an exponential growth function, where the time interval is the difference between the current moment and the end moment of the reliable positioning interval;

[0171] The product of the initial covariance benchmark and the diffusion coefficient is determined as the time-varying covariance at the current time, which characterizes the drift uncertainty that accumulates over time.

[0172] Specifically, the system first identifies reliable positioning intervals, i.e., periods with good GNSS signal quality, and then calculates the time T at the end of these intervals. end The position estimation uncertainty is σ0. For any time t (t>T) after entering the degradation interval. end The system constructs a prior distribution p(x) that diffuses over time:

[0173] p(x) = N(x; μ prior ,Σ prior );

[0174] Wherein, mean μ prior The covariance matrix Σ is predicted from the state at the previous time step. prior The construction of this model reflects the accumulating uncertainty in location estimation over time. An exponential growth model is used to describe this diffusion process:

[0175] Σ prior =σ0 2 *exp((tT end ) / τ)*I;

[0176] Where Δt=tT end This represents the time interval between the current moment and the end of the reliable positioning interval; τ is a preset extrapolation decay time constant, such as 60 seconds, used to control the diffusion rate of uncertainty; I is the identity matrix. This formula shows that the farther away from the reliable interval, the flatter the prior distribution, that is, the more uncertain the prediction of the location, and the greater the weight given to real-time observation data (such as thermal footprints) in subsequent fusion.

[0177] Step 303: Calculate the posterior distribution based on the likelihood function and the prior distribution, and determine the mean or maximum posterior estimate of the posterior distribution as unified spatiotemporal reference data.

[0178] The posterior distribution is calculated based on the likelihood function and the prior distribution. Specifically, a regularized objective function is constructed that includes observation residuals, constraint satisfaction terms, and trajectory smoothing terms. The unified spatiotemporal reference data is then solved by minimizing the regularized objective function.

[0179] In this step, the system uses Bayes' theorem p(x|z)∝p(z|x)*p(x) for fusion solving. To achieve efficient engineering implementation, the Maximum A posteriori (MAP) estimation method is employed, transforming it into a minimization problem of a regularized objective function:

[0180] minimize x (λ1*||xz foot || 2 +λ2*||x-(x t-1 +Δx odo )|| 2 +λ3*R(x));

[0181] The first term is the residual of the thermal footprint observation, the second term is the residual of the motion constraint, and the third term R(x) is the trajectory smoothness regularization term, such as the norm of the second derivative of the trajectory, which restricts abrupt acceleration changes. The coefficients λ1, λ2, and λ3 correspond to the reciprocals of the variances of the respective likelihood functions, automatically adjusting the weights of different information sources. For example, when the thermal footprint observation is very clear (observation noise standard deviation σ...), the regularization term... foot When λ1 increases, the solution will be forcibly pulled towards the thermal footprint position, correcting the cumulative drift of the odometer. The obtained x is the optimally estimated unified spatiotemporal reference data.

[0182] Step 304: Extract the posterior covariance matrix of the posterior distribution; calculate the trace or determinant of the posterior covariance matrix to quantify the spatiotemporal uncertainty of the registration result; generate a normalized numerical score based on the spatiotemporal uncertainty as the registration confidence level.

[0183] This step enables a quantitative assessment of registration quality. After completing the Bayesian update described above, the system can obtain the posterior covariance matrix Σ. post The diagonal elements of this matrix represent the variance of the position estimate in the x and y directions. The system calculates the trace of this matrix, which is the sum of the main diagonal elements, as a measure of total uncertainty. To generate an intuitive registration confidence Conf(t), the system uses the following normalization formula:

[0184] Conf(t)=exp(-tr(Σ post ) / σ ref 2 );

[0185] Where, σ ref It refers to the standard deviation of uncertainty, for example, 1.0 meter; tr(Σ) post ) represents the posterior covariance matrix Σ post The traces.

[0186] For example, suppose that after 30 seconds of localization degradation, the prior variance spreads significantly, and no features are observed at this point, resulting in a large posterior variance. The calculated registration confidence Conf(t) may drop to 0.2. Then, at the 31st second, the system captures a clear crushing thermal footprint (observation variance only 0.2), and the Bayesian updated Σ... post Rapid contraction, tr(Σ) post As the value decreases, the registration confidence Conf(t) may instantly rebound to 0.9. This dynamically changing confidence provides an accurate basis for subsequent control degradation.

[0187] To facilitate understanding, a specific numerical calculation example is given below.

[0188] Suppose a road roller is in a position degradation state at time t, and 30 seconds have passed since it reached the end of the nearest reliable positioning interval. According to the diffusion model in step 302, assuming the initial covariance reference σ0 is 0.1 meters and the time constant τ is 60 seconds, the prior covariance at the current time is:

[0189] Σ prior =0.1 2 ×exp(30 / 60)=0.01×1.649=0.0165 (square meters).

[0190] That is, the standard deviation of the prior distribution is approximately 0.128 meters.

[0191] At time t, the system detects a compacted thermal footprint, located at position z in the image coordinate system. foot =(100.5,50.2) meters, observation noise standard deviation σ foot =0.3 meters. Simultaneously, the odometer calculates that the road roller has traveled from the previous moment x... t-1 =(100.2,50.0) meters moved by Δx odo = (0.25, 0.15) meters, calculate the cumulative error σ cons =0.15 meters.

[0192] According to the Bayesian fusion formula in step 303, the mean of the posterior distribution (i.e., the optimal estimated location) can be approximately calculated as follows:

[0193] x post =(σ foot 2 ×z foot +σ cons -2 ×(x t-1 +Δx odo )+Σ prior -1 ×μ prior ) / (σ foot -2 +σcons -2 +Σ prior -1 );

[0194] Substituting the numerical values, the posterior location estimate is x. post ≈(100.38,50.12) meters.

[0195] Based on the confidence calculation formula in step 304, assume that the trace tr(Σ) of the posterior covariance matrix post = 0.045 square meters, with reference uncertainty σ ref =1.0 meter, then the registration reliability is:

[0196] Conf(t)=exp(-0.045 / 1.0)=0.956.

[0197] The confidence value is close to 1, indicating that with the assistance of thermal footprint observation, the system can still maintain high registration accuracy even if the GNSS signal degrades for 30 seconds.

[0198] Example 4 details how to use unified spatiotemporal reference data (location and surface temperature) and registration reliability, combined with a physical model, to extrapolate the internal state of the asphalt layer and predict time windows.

[0199] Step 401: Based on the surface temperature observation sequence in the unified spatiotemporal reference data, construct the time-varying gridded thermal boundary conditions.

[0200] In this embodiment, for each construction grid, the system extracts a series of discrete surface temperature observations T from the unified spatiotemporal reference data. surf (t k Since infrared cameras can only measure surface temperature, the observed surface temperature constitutes the top-level boundary conditions, Dirichlet boundary conditions, or Robin boundary conditions for the heat conduction equation. Simultaneously, the system utilizes environmental monitoring station data (temperature T). env Wind speed v wind Construct convection heat transfer boundary conditions.

[0201] Step 402: Obtain environmental monitoring data at the construction site, including air temperature, wind speed, and humidity; combine the environmental monitoring data to calibrate the heat dissipation constraints between the asphalt mixture and the environment, and construct the temperature evolution rules within the layer based on the physical laws of heat conduction.

[0202] Establish a one-dimensional transient heat conduction model (assuming that heat conduction in the horizontal direction is much smaller than that in the vertical direction):

[0203] ;

[0204] Where T(z,t) is the temperature at depth z, ρ is the density, c is the specific heat capacity, and k' is the thermal conductivity.

[0205] To calibrate the heat dissipation constraints, the system calculates the surface heat flux q using the following formula. loss :

[0206] q loss =h conv *(T surf -T env )+q rad ;

[0207] Among them, h conv It is the convective heat transfer coefficient, not a constant, but a function of wind speed, such as h. conv =a+b*v wind 'a' is the baseline convective heat transfer coefficient for natural convection (no wind or zero wind speed), and 'b' is the enhancement coefficient of convective heat transfer by wind speed (i.e., the increment of the convective heat transfer coefficient caused by unit wind speed); T surf q represents the surface temperature value of the grid. rad It is a radiation heat dissipation item.

[0208] This physical model allows the system to describe how heat is transferred from the interior of the asphalt layer to the surface and base layer.

[0209] In one specific implementation, for a typical asphalt mixture AC-13 (dense-graded asphalt concrete mixture), the typical range for density ρ is 2300–2500 kg / m³, the typical range for specific heat capacity c is 850–950 joules / kg / Kelvin, and the typical range for thermal conductivity k is 1.0–1.5 watts / m / Kelvin. For modified asphalt mixture SMA-13 ​​(asphalt mastic aggregate mixture), due to the addition of mineral powder and fiber, the thermal conductivity k can be appropriately reduced to 0.9–1.3 watts / m / Kelvin. These parameters can be adjusted according to the specific asphalt mix proportions and construction specifications, or determined through on-site calibration experiments.

[0210] Step 403: Modify the temperature evolution rule within the layer using heat dissipation constraints, and solve the three-dimensional temperature field by combining the gridded thermal boundary conditions to obtain the temperature state within the layer that reflects the internal temperature distribution of the asphalt layer.

[0211] The system employs the finite difference method (FDM) to numerically solve the above equations. By continuously refining the model boundaries using measured surface temperature sequences, the temperature T of internal nodes (such as the geometric center (core) of the material layer) is obtained through inversion. core (t). This solves the problem that surface temperature alone cannot determine whether the interior has the conditions for compaction.

[0212] Step 404: Load the pre-configured compactibility criterion. The compactibility criterion is a pre-stored parameter used to set the lower limit of the temperature at which the asphalt material maintains compactibility.

[0213] Compactability criteria are typically provided by the materials laboratory. For example, for modified asphalt SMA-13, its minimum effective compaction temperature T... min It may be set to 85℃. Compaction below this temperature is not only ineffective, but may even lead to aggregate breakage.

[0214] Step 405: Perform forward time step extrapolation based on the temperature state within the layer, search for the critical moment when the temperature state within the layer drops below the lower limit defined by the compactability criterion, and calculate the remaining effective time.

[0215] Starting with the current intralayer temperature distribution, the system iteratively simulates future time points using the physical model from step 402 until the core temperature T is reached. core (t+Δt) <T min (Δt is the time step of the heat conduction model). The difference between this moment and the current moment is the theoretical remaining effective time T. remain _ theory .

[0216] Step 406: Use the registration confidence level to correct the confidence interval of the remaining valid time. When the registration confidence level decreases, generate an effective operation time window by narrowing the safety boundary of the remaining valid time.

[0217] This step reflects the risk-decision coupling; due to potential errors in the input surface temperature location (characterized by the registration confidence Conf(t)), the prediction results are uncertain. The system uses a descent correction formula to calculate the final effective operating time window T presented to the user. window :

[0218] T window =T remain _ theory -K safe *(1-Conf(t));

[0219] Among them, T remain _ theory K represents the theoretical remaining effective time. safe It is the preset maximum penalty factor, such as 5 minutes.

[0220] For example, if theoretical calculations show 20 minutes remaining for compaction, but the current mesh registration confidence level Conf(t) is only 0.6, indicating a possible location deviation leading to inaccurate associated surface temperatures, the system deducts a safety margin of 5*(1-0.6) = 2 minutes, ultimately outputting an effective window of 18 minutes. This mechanism ensures that the system tends to provide conservative recommendations when data is unreliable, preventing construction from exceeding the time limit.

[0221] Example 5 describes in detail how to generate specific construction control strategies based on the preceding analysis results, especially how to handle sudden anomalies and interface risks.

[0222] Step 501: Based on the equipment speed and operation status sequence in the unified spatiotemporal reference data, detect abnormal shutdown, sudden start-up, or speed oscillation events of pavers or rollers, and generate a list of transient events located to specific grids.

[0223] The system calculates the first-order difference (acceleration) of the velocity sequence v(t). When the absolute value of the acceleration exceeds a threshold a... limit (e.g., 0.5 m / s) 2 When the speed is zero for an extended period (e.g., more than 5 minutes), it is marked as a speed change event; when the speed is zero for an extended period (e.g., more than 5 minutes), it is marked as a shutdown event. The system maps these events to a spatial grid, forming a list of transient events. For example, an emergency stop occurred at grid (100, 200).

[0224] Step 502: Analyze the physical characteristics of the temperature field corresponding to the occurrence time of each event in the transient event list, identify the temperature segregation region caused by the sudden change in velocity, and mark it as a quality anomaly grid that needs to be dealt with first.

[0225] The system will further examine the temperature field at the location of the incident. If an abrupt stop causes the mix under the paver's screed to overheat, or a sudden start causes the pavement thickness to decrease and the temperature to drop sharply, the system will mark that grid as a temperature segregation risk zone.

[0226] Step 503: Generate a rolling priority queue for global coverage based on the effective operation time window and the current construction progress determined based on unified spatiotemporal reference data.

[0227] Under normal circumstances, the system adopts the Shortest Remaining Time First (SRPT) strategy, sorting all the grids to be compacted according to the time of their effective time window ending, and generating a global queue.

[0228] Step 504: Based on the remaining effective time of the quality anomaly grids in the transient event list, calculate the urgency of the remedial operation and generate a subset of transient remedial tasks.

[0229] For meshes marked as anomalous, the system calculates their urgency. Since anomalous regions (such as segregation) typically cool faster, their urgency weight is artificially increased.

[0230] Step 505: Under the premise of satisfying the motion constraints of construction machinery, dynamically insert the transient remedial task subset into the rolling priority queue to generate a synthetic remedial path that takes into account both global operations and local defect repair.

[0231] This is a dynamic scheduling process. The system treats transient remedial tasks as high-priority insertions. However, before insertion, reachability must be verified: will the road roller traveling from its current location to the remedial point and then returning along its original route cause timeouts in other critical areas? The system uses a greedy algorithm or a local search algorithm to insert the remedial point into the optimal position in the queue, forming a synthetic remedial path, without violating global constraints.

[0232] Step 506: Extract the distribution of tack coat application rate, the waiting time after application, and the driving trajectory traces of construction vehicles from the unified spatiotemporal reference data; combine the obtained environmental monitoring data to assess the integrity of tack coat demulsification and film formation, the degree of interlayer contamination, and the risk of moisture.

[0233] This step corresponds to the interface risk assessment. Image processing technology is used to analyze the sprayer's operation records and on-site photos to identify any leaks or traces of tack coat carried away by the vehicle's tires. Simultaneously, humidity sensor data is used to determine if there is a risk of the sprayer remaining wet.

[0234] Step 507: Perform multi-factor weighted fusion on the above assessment results, calculate the grid-level interface adhesion risk index, and add the treatment actions corresponding to the high-risk index grids to the adaptive remedial control strategy.

[0235] System Construction Comprehensive Risk Index (Risk):

[0236] Risk = w1 * I uniformity +w2*I pollution +w3*I moisture ;

[0237] Among them, I uniformity As a risk indicator for temperature / material uniformity, I pollution As an indicator of interface contamination risk, I moisture The interface moisture content / dampness risk index is represented by w1, w2, and w3, which are the weighting coefficients for each risk factor. When the risk exceeds the threshold, a response action is generated, such as paving suspension, manual respraying, or cleaning up the contamination. This is the highest priority instruction and is directly pushed to the construction handheld terminal.

[0238] Step 508: Set a registration confidence threshold; when the registration confidence is lower than the registration confidence threshold, trigger a degradation protection mechanism. The degradation protection mechanism includes: increasing the time safety margin when predicting the effective working time window, and / or limiting the maximum travel speed and vibration amplitude of the road roller when generating remedial control commands.

[0239] As a last line of defense, the degradation mechanism is implemented here. If the registration confidence level in a certain area remains below 0.5, the system will forcibly limit the maximum travel speed of the smart roller in that area, for example, to within 2 km / h, and increase the compaction overlap width, for example, from 20 cm to 40 cm, in order to offset the risk of under-compaction caused by positioning errors through oversaturation operation.

[0240] Example 6 describes in detail the physical hardware architecture and data storage structure for implementing the above method, solving the problem that a simple method flow cannot be implemented on a physical entity, and further refining the implementation methods for generating closed-loop execution records and traceable evidence chains.

[0241] An asphalt pavement construction quality control system includes: a memory, a processor, and a communication interface.

[0242] The memory is used to store computer programs and massive amounts of process data generated during construction. Specifically, the memory may include high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device, flash memory, or other solid-state storage device. In this embodiment, to meet the requirement of solidifying the quality control evidence chain, a unified spatiotemporal database indexed by a grid is pre-built in the memory. The data structure stored in this database includes: input layer data (raw temperature image frame ID, raw GNSS message), intermediate layer data (unified spatiotemporal reference data, registration confidence Conf(t), and intra-layer temperature state T). core The system comprises three layers: decision-making layer data (effective operating time window, remedial control command ID) and receipt layer data (actual roller trajectory, compaction pass receipt). This hierarchical storage structure ensures that every quality judgment result can be traced back to the most original sensor observation, achieving full lifecycle traceability of construction quality.

[0243] The processor can be a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices. When executing a computer program, the processor implements the steps of the asphalt pavement construction quality control method based on multi-source data fusion as described in any of the above embodiments. Specifically, the processor internally runs multiple computing engines: a feature extraction engine for performing difference calculations and topology construction; a Bayesian registration engine for performing likelihood probability calculations and maximum a posteriori estimation; and a physics evolution engine for performing numerical solutions to the heat conduction equation.

[0244] The system also includes sensor components connected to the communication interface. Specifically, the sensor components include: an infrared thermal imager mounted on the paver for collecting temperature field monitoring data; a GNSS receiver and inertial measurement unit (IMU) mounted on the construction equipment (paver, roller) for collecting motion status data of the construction equipment; and an environmental monitoring station installed at the construction site for collecting real-time wind speed, air temperature, and humidity data. The communication interface (such as a 5G module or a Wi-Fi 6 module) is responsible for transmitting the above sensor data to the processor in real time.

[0245] In addition, this embodiment also provides a computer-readable storage medium storing a computer program, which, when running, controls the device where the computer-readable storage medium is located to execute the steps of the asphalt pavement construction quality control method based on multi-source data fusion in any of the above embodiments.

[0246] Storage media may include read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, etc.

[0247] According to one aspect of this application, during the multi-source construction data acquisition, unified spatiotemporal reference, and cross-device registration self-healing process, the cross-device spatiotemporal registration self-healing and registration reliability generation are specifically as follows:

[0248] By acquiring time-aligned data and location degradation window marker data, reproducible structures such as temperature abrupt boundary, stable cold joint zone, and continuous temperature gradient zone are extracted from the paving temperature profile sequence to obtain candidate temperature landmark data.

[0249] Acquire candidate temperature landmark data, perform stability screening on the candidate landmarks in the time dimension and establish multi-scale feature representation, remove candidates that are affected by short-term occlusion or noise, and obtain temperature landmark feature data.

[0250] By acquiring unified coordinate trajectory data and time-aligned data, wheel track continuity and accessibility constraints are constructed from the roller trajectory, speed, and operating status, forming a feasible domain constraint on spatial offset and temporal drift, thus obtaining wheel track continuity constraint data.

[0251] Acquire unified coordinate trajectory data and positioning degradation window marker data. Within the positioning degradation interval, generate short-term motion consistency supplementary constraints based on velocity and historical trajectory evolution to suppress registration solution drift and ensure trajectory continuity, thus obtaining degradation window motion constraint data.

[0252] Acquire temperature landmark feature data, wheel track continuity constraint data, and unified coordinate trajectory data. Establish cross-device alignment samples within the reliable positioning range. Solidify the relative relationship between temperature landmarks and compacted wheel tracks into alignment samples for training and verification, and obtain weakly supervised registration sample data.

[0253] The system acquires location degradation window marker data, weakly supervised registration sample data, temperature landmark feature data, wheel track continuity constraint data, and degradation window motion constraint data. Temporal drift correction and spatial offset correction are performed within the degradation window. A robust matching strategy is used to suppress abnormal landmarks and local missing data to obtain degradation window registration solution data.

[0254] We acquire weakly supervised registration sample data, degenerate window registration solution data, wheel track continuity constraint data, and degenerate window motion constraint data. We calculate the cross-device consistency residuals, constraint satisfaction degree, and sample coverage before and after registration. We map the credibility score to the grid and time slice and form a credibility label that can be used for degradation decision-making to obtain registration credibility data.

[0255] The process involves acquiring time alignment data, unified coordinate trajectory data, degradation window registration solution data, and registration reliability data. The alignment results within the degradation window are written back to the cross-device data index to form a unified reference consistent throughout the entire time period and generate a traceable alignment mapping. This yields unified spatiotemporal reference data, which is used to generate the gridded fusion base data process.

[0256] According to one aspect of this application, the physically consistent intra-layer temperature state estimation during the compressible state estimation and compressible real-time window prediction process is as follows:

[0257] The gridded thermal condition characteristic data and gridded fusion base data are obtained. The thermal boundary condition set of each grid is constructed from the temperature profile sequence, construction cycle and environmental records, and a grid-level initialization state description is formed to obtain thermal boundary condition data.

[0258] By acquiring thermal boundary condition data and gridded thermal condition characteristic data, online calibration of temperature decay behavior within a short time window is performed, and abnormal heat dissipation patterns caused by wind, shadow, and changes in the heat sink of the base layer are identified to obtain heat dissipation constraint data.

[0259] Acquire thermal boundary condition data and heat dissipation constraint data, generate intra-layer temperature evolution priors that satisfy the physical consistency of heat diffusion, constrain the rate of change and spatial smoothness of intra-layer states, and obtain intra-layer temperature evolution data.

[0260] By acquiring intralayer temperature evolution data and gridded thermal condition characteristic data, and integrating compaction side operation cycle, point temperature records, and surface temperature profile records, the intralayer temperature state is estimated and updated at the grid level to obtain intralayer temperature state data.

[0261] Acquire registration confidence data, heat dissipation constraint data, and intra-layer temperature state data. Perform a degradation estimation strategy on low-confidence grids and label the state reliability level to form a reliable label that can be directly used for subsequent time window predictions, thus obtaining intra-layer temperature reliable label data.

[0262] According to one aspect of this application, in the process of intralayer temperature estimation and compressibility real-time window prediction for compactable states, the compressibility real-time window prediction and the calculation of the remaining effective time are specifically as follows:

[0263] By acquiring compactability criterion data and gridded fusion base data, the compactability state boundaries corresponding to materials and processes are mapped to a unified grid scale, and the construction organization boundaries are standardized to obtain gridded compactability boundary data.

[0264] Acquire intralayer temperature state data, gridded compactable boundary data, and heat dissipation constraint data. Perform short-time forward extrapolation on the intralayer temperature of each grid to search for the critical moment when the compactable state changes from effective to ineffective, and obtain compactable failure time data.

[0265] The data on the moment of compaction failure is obtained and merged with the gridded base data. The current position of the roller, the reachable path, the construction organization constraints and the current work cycle are introduced to correct the reachability of the remaining compaction time, so as to obtain the remaining effective compaction time data.

[0266] The registration confidence data, intra-layer temperature confidence label data, and remaining effective compaction time data are obtained. The registration error, heat dissipation calibration error, and state estimation error are attributed to the grid-level output and a confidence label that can be used for subsequent control degradation is formed, resulting in compressible real-time window data.

[0267] According to one aspect of this application, in the process of identifying transient defects such as paving start-up and speed abrupt changes and generating closed-loop remedial control, the remedial priority and path generation based on a compressible real-time window are specifically as follows:

[0268] Acquire transient event list data, handling task list data, and gridded fusion base data. Merge the transient impact grid and the interface handling grid and establish avoidance and priority handling constraints to obtain remedial constraint grid data.

[0269] Obtain compressible real-time window data and remedial constraint grid data, calculate the urgency of remedial candidate grids and distinguish between those that must be dealt with immediately and those that can be dealt with later, and obtain remedial urgency data.

[0270] Obtain rolling priority queue data and remediation urgency data, insert transient remediation tasks without disrupting the global compaction organization, and maintain queue executability to obtain synthetic rolling queue data.

[0271] The synthetic rolling queue data and the meshed fusion base data are obtained. The path accessibility, construction boundary and conflict constraints are verified. Unreachable segments are locally rearranged and executable results are output to obtain remedial path planning data.

[0272] According to one aspect of this application, in the process of identifying transient defects such as paving start-up and speed change, and generating closed-loop remedial control, the generation of remedial control commands and adaptive setting of parameters are as follows:

[0273] Acquire transient event list data and remediation path planning data, select the corresponding control strategy template based on the event dominance type and generate a candidate parameter set to obtain candidate remediation strategy data.

[0274] Acquire intralayer temperature state data, compressible real-time window data, and candidate remedial strategy data. Perform grid-level adaptive tuning on parameters such as velocity, number of passes, and vibration mode to match the strategy with the remaining effective time, thereby obtaining adaptive remedial strategy data.

[0275] Acquire registration confidence data, intra-layer temperature confidence annotation data, and adaptive remediation strategy data. Inject degradation execution boundaries into low-confidence grids and restrict high-risk actions to form a set of instructions that can be directly executed and interpreted on-site, thus obtaining remediation control instruction data.

[0276] According to one aspect of this application, in the task scheduling process of short-term window risk assessment and treatment of tack coat oil interfaces, the calculation and reliable binding of the interface adhesion risk index are specifically as follows:

[0277] The interface state discrimination data and interface evidence feature data are obtained. The interface risk is decomposed into elements such as spraying stability, pollution risk, moisture risk and film formation process. Each element is then quantified at the grid level to obtain interface risk element data.

[0278] Obtain interface risk factor data, integrate multiple factors into a single risk level according to construction operability, and generate risk zones to obtain interface adhesion risk index data.

[0279] Obtain registration confidence data and interface adhesion risk index data, inject confidence annotations into the risk index, and give conservative output or trigger review mark for low confidence grids to obtain interface risk confidence annotation data.

[0280] This application employs a probabilistic correlation model based on Bayesian inference. It constructs a prior distribution that diffuses exponentially over time to describe the accumulation process of positioning uncertainty and combines this with physical feature observations for maximum a posteriori estimation. Within the degradation window of GNSS signal loss, it automatically corrects equipment trajectory drift, achieving accurate spatiotemporal registration and self-healing across all time periods. This solves the problem of data alignment failure caused by positioning signal degradation.

[0281] This application abandons the unstable absolute temperature threshold feature and extracts physical feature elements that are invariant to the cooling process: on the one hand, it uses the temperature field difference sequence to extract the dynamic rolling thermal footprint, eliminating background interference caused by overall cooling; on the other hand, it constructs an isotherm topological skeleton and uses the stability of the topological structure as a static registration benchmark, ensuring the robustness of the registration reference during the rapid cooling of asphalt. This solves the feature matching drift problem caused by time-varying temperature fields.

[0282] This application utilizes the transient temperature rise effect generated during compaction operations to establish an observation model, establishing a direct causal mapping between equipment movement trajectory and changes in road surface temperature field. This overcomes the ambiguity of traditional statistical matching and achieves deep fusion and closed-loop control of multi-source data at the physical mechanism level. It solves the problem of heterogeneous data lacking direct physical correlation.

[0283] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A method for quality control of asphalt pavement construction based on multi-source data fusion, characterized in that, include: Acquire multi-source heterogeneous data from asphalt construction sites, including at least temperature field monitoring data and construction equipment motion status data; Physical characteristic elements of the temperature field are extracted from temperature field monitoring data, and motion consistency constraints are constructed based on the motion state data of construction equipment. By using a probabilistic correlation model, the physical characteristics of the temperature field and the motion consistency constraints are jointly solved to generate unified spatiotemporal reference data and registration reliability characterizing the spatiotemporal alignment uncertainty. Based on unified spatiotemporal reference data, the temperature state within the layer is reconstructed using a physically consistent intra-layer state evolution model, and the effective working time window of the construction grid is predicted by combining the registration confidence. Based on unified spatiotemporal reference data and effective operation time windows, quality anomalies or interface risks during construction are identified, and adaptive remedial control strategies are generated accordingly.

2. The method according to claim 1, characterized in that, The physical characteristics of a temperature field include at least one of the following: The dynamic differential feature, extracted from the time-series differential operation of temperature field monitoring data, characterizes the transient thermal effects generated during the operation of construction equipment. Static topological features are extracted based on the spatial distribution structure of temperature field monitoring data, characterizing the spatiotemporal invariance of the temperature field distribution structure during the cooling process.

3. The method according to claim 2, characterized in that, Extracting dynamic difference features includes: Calculate the temperature field difference sequence of temperature field monitoring data at adjacent sampling times to eliminate the overall cooling decay trend of the temperature field; Detect regions in the temperature field difference sequence where the local temperature rise exceeds a preset threshold; The region is morphologically filtered by using preset constraints on the width of the roller drum and the working length, and the sequence of the center positions of the regions that meet the constraints is determined as dynamic differential features; Before determining the dynamic difference features, the following is also included: Perform temporal continuity verification and calculate the displacement velocity of the region center position within consecutive time frames; If the displacement speed exceeds the preset maximum travel speed threshold of the road roller, the corresponding area will be identified as noise and removed.

4. The method according to claim 2, characterized in that, Extracting static topological features includes: The gradient field and Hessian matrix of temperature field monitoring data are calculated using discrete difference operators, and the gradient critical points where the gradient is zero are identified, including local maxima, local minima and saddle points. Construct isotherm paths connecting gradient critical points to form the topological framework of the temperature field; Calculate the position drift variance of the gradient critical point or topological skeleton within the time window, and select skeleton structures with position drift variance less than the preset stability threshold as static topological features. After filtering out static topological features, the process also includes: For each selected gradient critical point, a multi-scale topological feature vector is constructed, which includes the relative distance, relative azimuth angle and critical point type encoding between the critical point and its nearest neighbor critical points, in order to support cross-time period topological structure matching.

5. The method according to claim 1, characterized in that, The probabilistic correlation model is a Bayesian inference model. It is used to jointly solve for the physical characteristics of the temperature field and the consistency constraints of motion, including: By taking the physical characteristics of the temperature field as observation variables and transforming the motion consistency constraint into a state transition constraint, a likelihood function reflecting the observation likelihood is constructed. Based on the historical motion state or the state distribution of the reliable positioning interval of the construction equipment, a prior distribution reflecting the distribution characteristics of the trajectory to be estimated is constructed. The posterior distribution is calculated based on the likelihood function and the prior distribution, and its mean or maximum a posteriori estimate is determined as unified spatiotemporal reference data.

6. The method according to claim 5, characterized in that, Generate registration confidence that represents spatiotemporal alignment uncertainty, including: Extract the posterior covariance matrix of the posterior distribution; Calculate the trace or determinant of the posterior covariance matrix to quantify the spatiotemporal uncertainty of the registration results; A normalized numerical score is generated based on spatiotemporal uncertainty, which serves as the registration reliability. Constructing the prior distribution includes: Based on the positioning status flags in the motion status data of construction equipment, identify the reliable positioning intervals in which the positioning signal quality meets preset requirements during construction. Statistically analyze the motion state parameters of construction equipment within the reliable positioning interval to generate initial prior distribution parameters; For the positioning signal degradation range, a prior distribution with time-varying covariance is generated based on the initial prior distribution parameters and the diffusion process over time, wherein the time-varying covariance monotonically increases with the time interval away from the reliable positioning range.

7. The method according to claim 1, characterized in that, Reconstructing the intra-layer temperature state using a physically consistent intra-layer state evolution model includes: Based on the surface temperature observation sequence in the unified spatiotemporal reference data, a time-varying gridded thermal boundary condition is constructed. By combining the acquired environmental monitoring data, the heat dissipation constraints between the asphalt mixture and the environment are calibrated, and the temperature evolution rules within the layer are constructed based on the physical laws of heat conduction. By modifying the temperature evolution rules within the layer using heat dissipation constraints and combining them with gridded thermal boundary conditions to solve the three-dimensional temperature field, the temperature state within the layer reflecting the internal temperature distribution of the asphalt layer is obtained.

8. The method according to claim 7, characterized in that, The effective operation time window for predicting construction grids based on registration reliability includes: Load the pre-configured compactibility criterion, which defines the lower limit of temperature for asphalt materials to maintain compactibility; Based on the intralayer temperature state, perform forward time step extrapolation, search for the critical moment when the intralayer temperature state drops below the compactability criterion, and calculate the remaining effective time; The confidence interval of the remaining effective time is corrected by using the registration confidence. When the registration confidence decreases, the effective operation time window is generated by narrowing the safety boundary of the remaining effective time.

9. A construction quality control system for asphalt pavement, characterized in that, include: Memory, used to store computer programs; A processor for executing a computer program to implement the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed, controls the device containing the computer-readable storage medium to perform the steps of any one of claims 1 to 8.

Images