Roadbed compaction quality intelligent monitoring method and system based on multi-source data fusion

Abstract

The invention relates to a roadbed compaction quality intelligent monitoring method and system based on multi-source data fusion, and the method comprises the steps: collecting a compaction surface image, extracting image features, synchronously collecting a vibration signal, calculating the time-frequency energy, and constructing a nonlinear regression model to dynamically output the compaction degree; establishing a filler parameter database; calculating flatness and gradient indexes; calculating flatness and gradient indexes; a non-contact laser range finder is adopted to measure the filling thickness variation in real time; calculating the elasticity modulus of the roadbed; and integrating multi-source spatio-temporal data and generating a compaction quality three-dimensional thermodynamic diagram, and optimizing a compaction strategy based on reinforcement learning. The method has the advantages that image recognition, three-dimensional modeling, non-contact measurement and a traditional detection method are combined, multi-dimensional data complementation is formed, and monitoring precision and reliability are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of road construction, and in particular relates to a method and system for intelligently monitoring roadbed compaction quality based on multi-source data fusion. Background Art

[0002] Traditional roadbed compaction quality testing relies on point sampling methods like sand filling and nuclear density meters, which suffer from low efficiency and poor real-time performance. Existing compaction assessment methods based on vibration signals fail to consider the impact of filler properties and dynamic changes in temperature and humidity on the number of compaction passes, and lack direct mapping modeling of compacted thickness and elastic modulus. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and system for intelligent monitoring of roadbed compaction quality based on multi-source data fusion.

[0004] First, an intelligent monitoring method for roadbed compaction quality based on multi-source data fusion is provided, including:

[0005] Step 1: Collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output the compaction degree;

[0006] Step 2: Establish a filler parameter database;

[0007] Step 3: Generate a 3D point cloud model of the roadbed based on LiDAR scanning and SLAM algorithm, and calculate the flatness and slope indicators;

[0008] Step 4: Use multi-layer probes to monitor the conductivity, temperature, and moisture content of soil layers at different depths in real time, and dynamically adjust the number of compactions;

[0009] Step 5: Use a non-contact laser rangefinder to measure the change in fill thickness in real time;

[0010] Step 6: Calculate the elastic modulus of the roadbed based on the roller wheel load, filler Poisson's ratio, and dynamic settlement;

[0011] Step 7: Integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

[0012] Preferably, step 1 comprises:

[0013] Step 1.1: Capture and preprocess the compacted surface image using a binocular camera, and then extract deep features of the image using an improved ResNet-50 model; the preprocessing includes adaptive median filtering and contrast-limited adaptive histogram equalization; the improved ResNet-50 model removes the original fully connected layer and is equipped with an attention module;

[0014] Step 1.2, collecting vibration signals through an acceleration sensor and performing wavelet packet decomposition to calculate time-frequency energy;

[0015] Step 1.3: Construct a nonlinear regression model to dynamically output the compaction degree, and optimize the nonlinear regression model through a loss function.

[0016] Preferably, step 2 comprises:

[0017] Step 2.1, select the measuring points and associate the compaction degree of the measuring points with the image features and vibration energy of the corresponding positions to construct a training data set;

[0018] Step 2.2, performing physical and mechanical tests on the filler to obtain key database parameters to construct a filler parameter database; the physical and mechanical tests on the filler include a CBR test, a direct shear test, a compression test, and a permeability test; the key database parameters include Poisson's ratio, maximum dry density, elastic modulus reference value, internal friction angle, cohesion, and permeability coefficient;

[0019] Step 2.3: Upload the filler parameter database to the cloud platform and associate it with the real-time monitoring data.

[0020] Preferably, step 3 includes:

[0021] Step 3.1: Install the 3D LiDAR and dual-antenna RTK module on top of the roller and perform LiDAR scanning. Combined with the SLAM algorithm, generate a 3D point cloud model of the roadbed.

[0022] Step 3.2: Use moving least squares to perform surface fitting and dynamically evaluate flatness and slope, where the slope includes longitudinal slope and transverse slope.

[0023] Preferably, step 4 includes:

[0024] Step 4.1: Use multi-layer probes to monitor the conductivity, temperature, and moisture content of soil layers at different depths in real time, and perform temperature compensation on the moisture content.

[0025] Step 4.2: Dynamically adjust the number of rolling times according to the maximum moisture content threshold.

[0026] Preferably, step 5 includes:

[0027] Step 5.1. Install the non-contact laser rangefinder on the bracket behind the roller's steel wheel and aim it vertically downward at the compacted surface to measure the distance.

[0028] Step 5.2: Obtain the initial thickness according to the construction drawings;

[0029] Step 5.3: Conduct real-time thickness monitoring and compare the real-time thickness with the designed thickness to provide early warning of insufficient or excessively thick fill areas.

[0030] Preferably, step 6 includes:

[0031] Step 6.1, measuring dynamic subsidence;

[0032] Step 6.2: Based on the Hertz contact theory, combined with the roller wheel load, Poisson's ratio of the fill material, and dynamic settlement, calculate the elastic modulus of the contact area between the roller wheel and the fill material.

[0033] Step 6.3: Generate a compaction quality heat map and dynamically adjust the compaction strategy.

[0034] Preferably, step 7 includes:

[0035] Step 7.1: Perform multi-source data synchronization and timing alignment;

[0036] Step 7.2: Generate a visual 3D thermal map of compaction quality;

[0037] Step 7.3: Execute dynamic control strategy, including local pressure replenishment and slope deviation compensation;

[0038] Step 7.4: Co-optimize the objective function and reinforcement learning.

[0039] In a second aspect, a roadbed compaction quality intelligent monitoring system based on multi-source data fusion is provided, which is used to execute any of the methods described in the first aspect, including:

[0040] Image-vibration fusion compaction analysis module, which is used to collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output compaction;

[0041] A packing parameter database establishment module is used to establish a packing parameter database;

[0042] The LiDAR 3D modeling module is used to generate a 3D point cloud model of the roadbed based on LiDAR scanning and SLAM algorithms, and calculate flatness and slope indicators;

[0043] Multi-layer soil parameter monitoring module, which is used to monitor the conductivity, temperature and moisture content of soil layers at different depths in real time through multi-layer probes, and dynamically adjust the number of rolling times;

[0044] Fill thickness monitoring module, used to measure the change of fill thickness in real time using a non-contact laser rangefinder;

[0045] The dynamic monitoring module of roadbed elastic modulus is used to calculate the roadbed elastic modulus by combining the roller wheel load, filler Poisson's ratio and dynamic settlement;

[0046] A collaborative control module is used to integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

[0047] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any one of the methods described in the first aspect.

[0048] The beneficial effects of the present invention are:

[0049] 1. The present invention combines image recognition, three-dimensional modeling, non-contact measurement and traditional detection methods to form multi-dimensional data complementarity and improve monitoring accuracy and reliability.

[0050] 2. The present invention dynamically adjusts the number of compaction passes based on real-time monitored environmental parameters (temperature, humidity) to ensure that the roadbed compaction quality meets standard requirements.

[0051] 3. The present invention uses 3D laser scanning and RTK technology to establish a 3D model of roadbed compaction, intuitively evaluate geometric indicators such as roadbed flatness, longitudinal slope, and transverse slope, and then determine whether the roadbed compaction meets the design requirements.

[0052] 4. The present invention generates a three-dimensional compaction quality heat map by real-time integration of multi-source data (compaction degree, elastic modulus, thickness, flatness, slope, etc.), and combines reinforcement learning to dynamically optimize construction parameters (such as rolling times and paths), forming a "perception-analysis-control" closed loop, significantly improving construction efficiency and quality consistency. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A module diagram of a roadbed compaction quality intelligent monitoring system based on multi-source data fusion provided by the present invention;

[0054] Figure 2 This is a diagram of an intelligent monitoring system for roadbed compaction quality based on multi-source data fusion provided by the present invention. DETAILED DESCRIPTION

[0055] The present invention will be further described below with reference to the following examples. The following examples are provided only to facilitate understanding of the present invention. It should be noted that, without departing from the principles of the present invention, it is possible for a person skilled in the art to make various modifications to the present invention, and such improvements and modifications fall within the scope of the claims of the present invention.

[0056] Example 1:

[0057] In order to solve the problems of the prior art, Example 1 of the present invention provides an intelligent monitoring method for roadbed compaction quality based on multi-source data fusion, which integrates image recognition, three-dimensional laser modeling, multi-physics field sensing and deep learning to realize full-process monitoring of parameters such as fill compaction thickness changes, compaction degree, roadbed elastic modulus, and roadbed soil properties during the roadbed compaction process.

[0058] Specifically, the intelligent monitoring method for roadbed compaction quality based on multi-source data fusion provided by the present invention includes:

[0059] Step 1: Collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output the compaction degree.

[0060] Step 1 includes:

[0061] Step 1.1: Use a binocular camera to collect and preprocess the compaction surface image, and then extract the deep features of the image based on the improved ResNet-50 model.

[0062] For example, a binocular camera (resolution 3840×2160, frame rate 60fps, CMOS size 1 / 1.8", aperture f / 2.0) is installed under the roller's vibrating wheel bracket, covering a 1.8m×1.5m compaction surface area at a 45° tilt angle. A ring light is also included to ensure clear images in low-light environments.

[0063] Image preprocessing includes: using adaptive median filtering (window size 7×7) to remove salt and pepper noise; using CLAHE (contrast-constrained adaptive histogram equalization, block size 32×32, contrast limit 2.0) to enhance texture details.

[0064] The improved ResNet-50 model removes the original fully connected layer and is equipped with an attention module (SENet structure) to output a 256-dimensional feature vector F image ={f1,f1,Λf 256}, the feature weighting formula is:

[0065]

[0066] Among them, W i ,b i is a trainable parameter, f i is the feature vector.

[0067] In addition, the improved ResNet-50 model was trained and optimized. The training dataset was 120,000 labeled images (including roughness grade, filler particle size, filler specification, filler shape, and moisture content labels). The loss function used for optimization was Focal Loss (α=0.25, γ=2), and the optimizer was Nadam (learning rate 0.0005, decay rate 0.9).

[0068] Step 1.2: Collect vibration signals through the acceleration sensor and perform wavelet packet decomposition to calculate the time-frequency energy.

[0069] For example, the acceleration sensor is a piezoresistive MEMS three-axis acceleration sensor (measuring range ±20g, bandwidth 0-1kHz, non-linearity <0.5%).

[0070] Wavelet packet decomposition includes: using Daubechies-6 wavelet basis, decomposing 8 layers, extracting energy components of frequency bands 0-125Hz, 125-250Hz, ..., 875-1000Hz, where the energy calculation formula is:

[0071]

[0072] Total vibration energy

[0073] Step 1.3: Construct a nonlinear regression model to dynamically output the compaction degree, and optimize the nonlinear regression model through a loss function.

[0074] Specifically, dynamic calibration was performed in step 1.3:

[0075] The measured value K of the synchronous sand filling method every 24 minutes sand , construct the objective function:

[0076]

[0077] Furthermore, the optimization algorithm is: quantum particle swarm optimization (population size 50, number of iterations 200), with the constraint condition α+β+γ=1.

[0078] Where λ is the loss function, which is used to measure the difference between the predicted compaction degree and the true value. The smaller the value, the more accurate the model. N is the number of samples, which represents the total number of data points involved in the calibration, such as the number of measured values ​​of the sand filling method. Tanh is the hyperbolic tangent function, which means compressing the linear combination results to the interval [-1, 1] to enhance the nonlinear expression ability of the model while suppressing the influence of outliers. α is the image feature weight, which represents the control of the image feature F. image The contribution ratio to the compaction prediction. β is the vibration energy weight, which represents the control vibration energy E vibThe contribution ratio to the compaction prediction. γ is the weight of the measured value of the sand filling method, which represents the control of the measured value K of the sand filling method. sand Correction effect on the forecast. K sand The compaction value measured by the traditional method is used as the calibration benchmark. true The actual compaction degree.

[0079] Step 2: Establish a filler parameter database.

[0080] Step 2 includes:

[0081] Step 2.1: Select the measuring points and associate the compaction degree of the measuring points with the image features and vibration energy of the corresponding positions to construct a training data set.

[0082] Specifically, the measurement point arrangement rule is: the measurement points are arranged in a "plum blossom" grid in the compaction area, with a spacing of 2m×2m, covering the entire working surface.

[0083] In addition, the compaction degree K is measured according to the sand filling method. sand , divide the measuring points into three categories: Better: K sand ≥95%; general: 90%≤K sand <95%; poor: K sand <90%. And record the RTK coordinates (accuracy ±5cm) and compaction value of each measuring point to generate a calibration map.

[0084] Data association includes: associating the compaction degree of the measuring point with the image features and vibration energy of the corresponding position to construct a training data set.

[0085] Step 2.2, conduct physical and mechanical tests on fillers to obtain key database parameters to construct a filler parameter database; the physical and mechanical tests on fillers include CBR test, direct shear test, compression test and penetration test; the key database parameters include Poisson's ratio, maximum dry density, elastic modulus reference value, internal friction angle, cohesion and permeability coefficient.

[0086] Specifically, the test types and purposes are as follows.

[0087] CBR test: Determines the penetration strength of filler under standard load, reflecting its anti-deformation ability.

[0088] Direct shear test: Obtain the internal friction angle φ and cohesion c of the roadbed filler to calculate the shear strength.

[0089] Compression test: Determine the compression modulus E s , used to evaluate the compression characteristics of roadbed fillers.

[0090] Permeability test: measure the filler permeability coefficient k and analyze the effect of moisture content on compaction.

[0091] The construction of parameter database includes: establishing indoor parameter database according to different filler categories, where the key parameters of the database include Poisson's ratio ν, maximum dry density ρ max , Elastic modulus reference value E base , internal friction angle φ, cohesion c, permeability coefficient k, etc.

[0092] Step 2.3: Upload the filler parameter database to the cloud platform and associate it with the real-time monitoring data.

[0093] Test data input: upload the filler parameter database to the cloud platform and associate it with real-time monitoring data (images, vibrations, moisture content) to achieve real-time adjustment and optimization of the number of filling and compaction passes.

[0094] Step 3: Based on LiDAR scanning and SLAM algorithm, generate a three-dimensional point cloud model of the roadbed and calculate the flatness and slope indicators.

[0095] Step 3 includes:

[0096] Step 3.1: Install the 3D LiDAR and dual-antenna RTK module on top of the roller and perform LiDAR scanning. Combined with the SLAM algorithm, generate a 3D point cloud model of the roadbed.

[0097] Specifically, the hardware configuration is: a 128-line three-dimensional laser radar (scanning frequency 20Hz, ranging accuracy ±1cm, vertical field of view ±15°) and a dual-antenna RTK module (positioning error <1cm) installed on the top of the roller.

[0098] The SLAM algorithm includes: based on the LIO-SAM framework, fusing laser point cloud, IMU (zero bias stability 0.5° / h) and wheel speed meter data; generating dense point cloud P(x, y, z) in real time with a resolution of 1cm×1cm and a global error of less than 2cm.

[0099] Step 3.2: Use moving least squares to perform surface fitting and dynamically evaluate flatness and slope, where the slope includes longitudinal slope and transverse slope.

[0100] Specifically, the moving least squares (MLS) surface fitting is used: the block size is 0.5m×0.5m, and the radial basis function is: The fitting surface equation is: Residual sum of squares calculation:

[0101] Dynamic flatness assessment includes:

[0102] L is the evaluation path length (unit: m), z i are the elevation values ​​of consecutive points along the path.

[0103] Dynamic slope assessment includes:

[0104] Longitudinal slope calculation:

[0105] Cross slope calculation:

[0106] In addition, step 3 also sets the judgment logic: when IRI>3mm / m 2 If or |i x -i y When |>0.3%, the spiral pressure compensation path is triggered (spacing 0.25m, speed 1.5km / h).

[0107] Step 4: Use multi-layer probes to monitor the conductivity, temperature and moisture content of soil layers at different depths in real time, and dynamically adjust the number of compactions.

[0108] Step 4 includes:

[0109] Step 4.1: Use multi-layer probes to monitor the conductivity, temperature, and moisture content of soil layers at different depths in real time, and perform temperature compensation on the moisture content.

[0110] Specifically, the probe deployment includes: insertion depth of 5m, monitoring conductivity J in 5 layers (1m per layer) j , temperature T j , moisture content w j (j=1,2,Λ5).

[0111] The temperature compensation formula is:

[0112] w' j =w j [1+0.015(T j -25)]

[0113] Step 4.2: Dynamically adjust the number of rolling times according to the maximum moisture content threshold.

[0114] According to the maximum moisture content threshold w max =18%, calculate the number of rolling times for each layer:

[0115]

[0116] Total number of crushes

[0117] Step 5: Use a non-contact laser rangefinder to measure the change in fill thickness in real time.

[0118] Step 5 includes:

[0119] Step 5.1. Install the non-contact laser rangefinder on the bracket behind the roller's steel wheel and aim it vertically downward toward the compacted surface to measure the distance.

[0120] Specifically, the non-contact laser rangefinder uses a phase-shifted laser rangefinder with a wavelength of 650nm, a measurement range of 0.1m to 30m, and an accuracy of ±0.5mm. It is installed on a bracket behind the roller drum, pointing vertically downward toward the compacted surface, with a coverage width of 1.5m and a sampling frequency of 100Hz.

[0121] In addition, the present invention also has an anti-interference design: equipped with a dust cover and a temperature control module (working temperature -20℃~60℃) to eliminate ambient light and vibration interference.

[0122] Step 5.2: Obtain the initial thickness according to the construction drawings.

[0123] For example, the initial thickness input is: the design thickness h0 is imported into the system from the construction drawing (such as the design value h0 = 30 cm).

[0124] Step 5.3: Conduct real-time thickness monitoring and compare the real-time thickness with the designed thickness to provide early warning of insufficient or excessively thick fill areas.

[0125] The formula for real-time thickness monitoring is:

[0126]

[0127] Where Δh i is the thickness change after the i-th rolling, and the calculation formula is:

[0128]

[0129] Among them, h pre,k is the thickness of the kth measuring point before rolling; h post,k is the thickness of the kth measuring point after rolling; N is the number of measuring points covered by a single rolling (default N = 50).

[0130] Step 6: Calculate the elastic modulus of the roadbed based on the roller wheel load, filler Poisson's ratio, and dynamic settlement.

[0131] Step 6 includes:

[0132] Step 6.1: Measure the dynamic subsidence.

[0133] Specifically, the sensor configuration is as follows: two sets of laser displacement sensors are installed symmetrically on both sides of the steel wheel to measure the wheel flange sinking amount δ l and δ r .

[0134] Dynamic subsidence average:

[0135]

[0136] Filtering: Kalman filtering is used to eliminate vibration noise. Smoothing formula:

[0137] δ filtered =0.8δ prev +0.2δ curent

[0138] Among them, δ filtered is the estimated value after filtering; δ prev is the filter value of the previous step; δ curent is the current measured value.

[0139] Step 6.2: Based on the Hertz contact theory, combined with the roller wheel load, Poisson's ratio of the fill material, and dynamic settlement, calculate the elastic modulus of the contact area between the roller wheel and the fill material.

[0140] Formula derivation: Based on Hertz contact theory, the elastic modulus E of the contact area between the roller wheel and the fill is calculated as follows:

[0141]

[0142] Among them, F is the roller wheel load: collected in real time by the built-in pressure sensor in the roller hydraulic system (range 0 to 500 kN, accuracy ±0.5%); ν is the Poisson's ratio of the filler: provided by the filler parameter database; R is the steel wheel radius: input by equipment parameters (such as R = 0.6m); δ is the dynamic settlement: the steel wheel settlement is measured by a laser displacement sensor with a sampling rate of 1kHz.

[0143] Step 6.3: Generate a compaction quality heat map and dynamically adjust the compaction strategy.

[0144] Perform real-time feedback on the elastic modulus through step 6.3. Specifically, combine the thickness h(t) and the elastic modulus E to dynamically generate a compaction quality heat map, and adjust the compaction strategy using the following logic:

[0145] If E<E min , triggering local pressure replenishment; if h(t)<h design -5mm, alarm prompts insufficient filling soil.

[0146] Step 7: Integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

[0147] Step 7 includes:

[0148] Step 7.1: Perform multi-source data synchronization and timing alignment.

[0149] The clock synchronization mechanism is:

[0150] An RTK timing module (accuracy ±1μs) provides a unified time base for all sensors (cameras, lidars, probes, etc.). The edge ARM motherboard synchronizes time with the cloud server via the NTP protocol, ensuring a global clock error of <10ms.

[0151] The spatiotemporal correlation database is designed as follows:

[0152] Data structure: using spatiotemporal four-dimensional index (longitude x, latitude y, elevation z, time t);

[0153] Data association rules: Fill in the blind spots of sensor coverage through spatiotemporal interpolation algorithms.

[0154] Step 7.2: Generate a visual 3D thermal map of compaction quality.

[0155] In step 7.2, parameter normalization is performed first:

[0156] The compaction degree K, elastic modulus E, and thickness h(t) are normalized to the interval [0,1]:

[0157] (The same goes for other parameters)

[0158] The heat map fusion formula is:

[0159] Q(x,y)=0.4K norm +0.3E norm +0.3h(t) norm

[0160] The basis for weight determination is: based on the regression analysis of historical data, the weight is optimized to make the correlation coefficient R between the thermal map and the measured value of the sand filling method 2 ≥0.95.

[0161] The 3D visualization is as follows:

[0162] The heat map is superimposed on the 3D roadbed model generated by the lidar using WebGL technology. The color mapping rules are as follows:

[0163] Red (Q<0.7): pressure needs to be replenished;

[0164] Yellow (0.7≤Q<0.85): warning;

[0165] Green (Q≥0.85): Meets the standard.

[0166] Step 7.3: Execute dynamic control strategy, including local pressure replenishment and slope deviation compensation.

[0167] Specifically, the local pressure replenishment triggering logic is as follows:

[0168] Path planning: Generate the optimal pressure replenishment path based on the A* algorithm, avoiding areas that have already met the standards, with a pressure replenishment interval of 0.25m and a speed of 1.5km / h.

[0169] Compaction times: Dynamically adjusted according to the gap between Q value and target value:

[0170]

[0171] Moisture content exceeding the limit: SMS / APP push notification is sent to construction management personnel via 4G gateway. Example content:

[0172] [Alarm] Water content exceeds the limit! Current position (116; 404.39; 915), w = 15.2% > w max =13.5%.

[0173] Adjustment of the number of crushes:

[0174]

[0175] The slope deviation compensation strategy is as follows:

[0176] Spiral compaction mode: The roller compacts along a spiral path (radius increment step 0.5m) to ensure that the slope deviation converges to within ±0.3% of the design value.

[0177] Dynamic parameter adjustment: adjust the spiral line spacing according to the real-time slope deviation ΔS:

[0178] Step 7.4: Co-optimize the objective function and reinforcement learning.

[0179] The objective function is designed as:

[0180]

[0181] The physical meaning is: balancing the compaction prediction accuracy and geometric shape (slope) control requirements.

[0182] The dynamic parameter adjustment of reinforcement learning is as follows:

[0183] The algorithm framework is: using the deep deterministic policy gradient (DDPG) algorithm, the state space is (K pred ,S j ,w), the adjustment amount Δλ in the action space is λ.

[0184] The reward function is:

[0185] R=-(∑(K pred -K real ) 2 +10·∑|S j -S design |)

[0186] The training process is as follows: offline training is performed on a cloud GPU cluster, and after convergence, it is deployed to the edge for real-time inference.

[0187] Parameter initialization and constraints include:

[0188] The initial value λ=0.5, and the allowed range λ∈[0.1, 1.0].

[0189] λ is updated every hour to ensure that the system adapts to the needs of different construction stages.

[0190] Example 2:

[0191] Based on Example 1, Example 2 of the present invention provides an intelligent monitoring system for roadbed compaction quality based on multi-source data fusion, such as Figure 2 As shown in the figure, by integrating multiple technical means such as camera image recognition, three-dimensional laser scanning, non-contact displacement sensor, vibration acceleration monitoring and environmental parameter monitoring, the whole process monitoring of fill compaction thickness changes, compaction degree, subgrade elastic modulus and subgrade soil properties during subgrade compaction is achieved.

[0192] Specifically, such as Figure 1 As shown in the figure, the intelligent monitoring system for roadbed compaction quality based on multi-source data fusion includes:

[0193] Image-vibration fusion compaction analysis module, which is used to collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output compaction;

[0194] A packing parameter database establishment module is used to establish a packing parameter database;

[0195] The LiDAR 3D modeling module is used to generate a 3D point cloud model of the roadbed based on LiDAR scanning and SLAM algorithms, and calculate flatness and slope indicators;

[0196] Multi-layer soil parameter monitoring module, which is used to monitor the conductivity, temperature and moisture content of soil layers at different depths in real time through multi-layer probes, and dynamically adjust the number of rolling times;

[0197] Fill thickness monitoring module, used to measure the change of fill thickness in real time using a non-contact laser rangefinder;

[0198] The dynamic monitoring module of roadbed elastic modulus is used to calculate the roadbed elastic modulus by combining the roller wheel load, filler Poisson's ratio and dynamic settlement;

[0199] A collaborative control module is used to integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

[0200] It should be noted that the system provided in this embodiment is a system corresponding to the method provided in Example 1. Therefore, the parts in this embodiment that are the same or similar to those in Example 1 can be referenced to each other and will not be repeated in this application.

Claims

1. An intelligent monitoring method for roadbed compaction quality based on multi-source data fusion, characterized in that: include: Step 1: Collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output the compaction degree; Step 2: Establishing a filler parameter database; Step 3: Generate a 3D point cloud model of the roadbed based on LiDAR scanning and SLAM algorithm, and calculate the flatness and slope indicators; Step 4: Use multi-layer probes to monitor the conductivity, temperature, and moisture content of soil layers at different depths in real time, and dynamically adjust the number of compactions; Step 5: Use a non-contact laser rangefinder to measure the change in fill thickness in real time; Step 6: Calculate the elastic modulus of the roadbed based on the roller wheel load, filler Poisson's ratio, and dynamic settlement; Step 7: Integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

2. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 1 is characterized in that: Step 1 includes: Step 1.1: Capture and preprocess the compacted surface image using a binocular camera, and then extract deep features of the image using an improved ResNet-50 model; the preprocessing includes adaptive median filtering and contrast-limited adaptive histogram equalization; the improved ResNet-50 model removes the original fully connected layer and is equipped with an attention module; Step 1.2, collecting vibration signals through an acceleration sensor and performing wavelet packet decomposition to calculate time-frequency energy; Step 1.3: Construct a nonlinear regression model to dynamically output the compaction degree, and optimize the nonlinear regression model through a loss function.

3. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 2 is characterized in that: Step 2 includes: Step 2.1, select the measuring points and associate the compaction degree of the measuring points with the image features and vibration energy of the corresponding positions to construct a training data set; Step 2.2, performing physical and mechanical tests on the filler to obtain key database parameters to construct a filler parameter database; the physical and mechanical tests on the filler include a CBR test, a direct shear test, a compression test, and a permeability test; the key database parameters include Poisson's ratio, maximum dry density, elastic modulus reference value, internal friction angle, cohesion, and permeability coefficient; Step 2.3: Upload the filler parameter database to the cloud platform and associate it with the real-time monitoring data.

4. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 3 is characterized in that: Step 3 includes: Step 3.1: Install the 3D LiDAR and dual-antenna RTK module on top of the roller and perform LiDAR scanning. Combined with the SLAM algorithm, generate a 3D point cloud model of the roadbed. Step 3.2: Use moving least squares to perform surface fitting and dynamically evaluate flatness and slope, where the slope includes longitudinal slope and transverse slope.

5. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 4 is characterized in that: Step 4 includes: Step 4.1: Use multi-layer probes to monitor the conductivity, temperature, and moisture content of soil layers at different depths in real time, and perform temperature compensation on the moisture content; Step 4.2: Dynamically adjust the number of rolling times according to the maximum moisture content threshold.

6. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 5 is characterized in that: Step 5 includes: Step 5.

1. Install the non-contact laser rangefinder on the bracket behind the roller's steel wheel and aim it vertically downward at the compacted surface to measure the distance. Step 5.2: Obtain the initial thickness according to the construction drawings; Step 5.3: Conduct real-time thickness monitoring and compare the real-time thickness with the designed thickness to provide early warning of insufficient or excessive fill areas.

7. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 6 is characterized in that: Step 6 includes: Step 6.1, measuring dynamic subsidence; Step 6.2: Based on the Hertz contact theory, combined with the roller wheel load, Poisson's ratio of the fill material, and dynamic settlement, calculate the elastic modulus of the contact area between the roller wheel and the fill material. Step 6.3: Generate a compaction quality heat map and dynamically adjust the compaction strategy.

8. The intelligent monitoring method for roadbed compaction quality based on multi-source data fusion according to claim 7 is characterized in that: Step 7 includes: Step 7.1: Perform multi-source data synchronization and timing alignment; Step 7.2: Generate a visual 3D thermal map of compaction quality; Step 7.3: Execute dynamic control strategy, including local pressure replenishment and slope deviation compensation; Step 7.4: Co-optimize the objective function and reinforcement learning.

9. The intelligent monitoring system for roadbed compaction quality based on multi-source data fusion is characterized by: Used to perform the method according to any one of claims 1 to 8, comprising: Image-vibration fusion compaction analysis module, which is used to collect compaction surface images and extract image features, synchronously collect vibration signals and calculate time-frequency energy, and construct a nonlinear regression model to dynamically output compaction; A packing parameter database establishment module is used to establish a packing parameter database; The LiDAR 3D modeling module is used to generate a 3D point cloud model of the roadbed based on LiDAR scanning and SLAM algorithms, and calculate flatness and slope indicators; Multi-layer soil parameter monitoring module, which is used to monitor the conductivity, temperature and moisture content of soil layers at different depths in real time through multi-layer probes, and dynamically adjust the number of rolling times; Fill thickness monitoring module, used to measure the change of fill thickness in real time using a non-contact laser rangefinder; The subgrade elastic modulus dynamic monitoring module is used to calculate the subgrade elastic modulus by combining the roller wheel load, filler Poisson's ratio and dynamic settlement; A collaborative control module is used to integrate multi-source spatiotemporal data and generate a three-dimensional heat map of compaction quality, and optimize the compaction strategy based on reinforcement learning.

10. A computer storage medium, characterized in that The computer storage medium stores a computer program; when the computer program is run on a computer, the computer executes the method according to any one of claims 1 to 7.

Images