A soft soil road anti-disturbance construction quality control system and method thereof

Abstract

The application relates to the technical field of soft soil road construction, and discloses a soft soil road anti-disturbance construction quality control system and a method thereof. The system comprises a multi-source geological data acquisition and fusion module, an anti-disturbance intelligent design optimization module, a construction process real-time monitoring module, a full-wave field intelligent detection module, a multi-parameter early warning and AI decision module and a quality tracking module, and the data of the modules are connected to form a closed-loop management and control link. According to the method, a three-dimensional geological digital model is generated through the spatio-temporal fusion of multi-source heterogeneous data, a multi-objective anti-disturbance processing scheme is optimized through a genetic algorithm, and the influence range and safety distance of construction disturbance are quantitatively predicted. In the construction stage, multiple types of parameters are collected in real time and dynamically controlled through a dynamic threshold, and millimeter-level detection of pile foundation and solidified soil is realized in combination with full-wave field elastic wave inversion. Through LSTM time series prediction and XGBoost risk grading, grading early warning and intelligent disposal are completed. The application realizes full-process anti-disturbance closed-loop management and control, and effectively solves the problems of traditional construction monitoring lag and uncontrollable disturbance.

Description

Technical Field

[0001] This invention relates to the field of soft soil road construction technology, and in particular to a quality control system and method for soft soil road construction that resists disturbance. Background Technology

[0002] Soft soil refers to fine-grained soil with high natural water content, high compressibility, and low shear strength. It is widely distributed in coastal areas of my country, the middle and lower reaches of the Yangtze River Plain, and inland lake sedimentary areas. Soft soil's high compressibility, low bearing capacity, and strong thixotropy have always made it a challenge for road engineering construction. Traditional soft soil road construction faces the following key problems:

[0003] 1. The disconnect between monitoring and construction: Current construction methods mostly use static monitoring (such as settlement plates and inclinometers), which result in delayed data feedback and an inability to respond in real time to soft soil deformation caused by construction disturbances, leading to the risk of excessive settlement or instability of the roadbed.

[0004] 2. Limitations of construction based on experience: Construction parameters (such as filling rate and loading method) are mostly determined based on geological reports and experience, lacking a dynamic optimization mechanism, and are difficult to adapt to the spatial variability of soft soil properties and the dynamic changes in the construction process.

[0005] 3. Uncontrollability of disturbance response: Soft soil is prone to thixotropic failure due to vibration of construction machinery and loading rate. Traditional methods lack quantitative assessment and active intervention means for disturbance risk, and often cause engineering accidents due to "over-construction" or "insufficient response". Summary of the Invention

[0006] To overcome the above deficiencies, the present invention provides a construction quality control system and method for soft soil road anti-disturbance, which solves the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a soft soil road anti-disturbance construction quality control system, comprising:

[0008] The multi-source geological data acquisition and fusion module is used to acquire multi-source heterogeneous geological data of soft soil sites, and to preprocess, spatially interpolate, perform confidence-weighted fusion, and perform anti-disturbance sensitivity zoning on the multi-source heterogeneous geological data through a multi-source heterogeneous data spatiotemporal fusion algorithm to generate a three-dimensional geological digital model and anti-disturbance sensitivity zoning parameters.

[0009] The anti-disturbance intelligent design optimization module is connected to the multi-source geological data acquisition and fusion module. It is used to automatically generate the optimal combination processing scheme based on the three-dimensional geological digital model through the intelligent selection algorithm of anti-disturbance processing scheme, and to calculate the disturbance impact range and safe construction distance of each processing method through the construction disturbance impact prediction algorithm.

[0010] The real-time monitoring module for the construction process is connected to the anti-disturbance intelligent design optimization module. It is used to receive the optimal combination processing scheme, the disturbance influence range and the safe construction distance as the construction control benchmark, and to collect the filling rate, compaction degree, settlement, pore water pressure and lateral displacement parameters in real time during the construction process. The parameters are denoised by Kalman filtering, compared with multi-level early warning thresholds and made dynamic rate control decisions through a multi-parameter dynamic threshold control algorithm.

[0011] The full-wave field intelligent detection module is used to perform millimeter-level detection of pile foundation integrity and solidified soil strength based on the full-wave field elastic wave testing theory and the full-wave field elastic wave inversion imaging algorithm, and outputs high-resolution imaging results and defect quantitative assessment.

[0012] The multi-parameter early warning and AI decision-making module is connected to the real-time construction process monitoring module and the full-wavelength intelligent detection module, respectively. It is used to integrate the real-time monitoring data of the real-time construction process monitoring module and the detection data of the full-wavelength intelligent detection module, and to conduct a comprehensive risk assessment through a multi-source fusion risk early warning algorithm, and issue early warnings and automatically generate disposal suggestions. The multi-parameter early warning and AI decision-making module is also connected to the real-time construction process monitoring module to feed back control commands to the real-time construction process monitoring module, and to coordinate the adjustment of operating parameters of on-site construction equipment.

[0013] The quality tracking module communicates with the aforementioned modules and is used to aggregate data from the entire process. It uses a full-process quality tracking and evaluation algorithm to divide the project quality into inspection batches and assign weighted scores, and outputs quality evaluation reports and quality archives.

[0014] Furthermore, the spatiotemporal fusion algorithm for multi-source heterogeneous data in the multi-source geological data acquisition and fusion module includes the following:

[0015] Step 1: Data Acquisition and Preprocessing: Obtain the probe resistance through a static cone penetration test. and sidewall friction Undrained shear strength was obtained through vane shear test. residual value and shear modulus Moisture content was obtained through indoor experiments. Liquid limit Plastic Limit Compression coefficient and consolidation coefficient and adopt The criteria remove outliers from all data. , For the first One original observation data value, The arithmetic mean of all data. If the value is less than the standard deviation, it is considered an outlier and removed.

[0016] Step 2: Spatial Interpolation and Geological Profile Generation: Spatial interpolation of discrete borehole data is performed using ordinary kriging interpolation to establish a semi-variogram function. ,in Describe the spatial correlation of data with distance Changes, The distance interval is Number of data point pairs For position Measured geological parameter values ​​at the location For position Measured geological parameter values ​​at the location The spatial distance between two sampling points;

[0017] Fitting a spherical model ,in The distance between data points For sill values, For variance, For variable range;

[0018] Solving for the weighting coefficients using the Kriging equations , for the first The contribution weights of each known point to the interpolation point, and the interpolation result. ,in, Given the number of discrete points used for interpolation. For position The Kriging interpolation estimates at the location are used to generate continuous geological profiles;

[0019] Step 3: Weighted Fusion of Multi-Source Data Confidence: Assigning Weights to Static Cone Probes The cross-plate shearing imparts weight. Indoor tests assign weights Geophysical exploration assigns weight The fusion value is calculated as follows: ,in For the kth method Independent measurements, These are the weighted and integrated geological parameter values;

[0020] Step 4: Disturbance Sensitivity Zoning: Calculate the disturbance sensitivity index for each region. ,in, , The compression modulus of soil. Based on the characteristic value of foundation bearing capacity, the site is divided into high-sensitivity zones. Medium-sensitive area Low-sensitivity areas The sampling frequency for construction monitoring in highly sensitive areas is once every 15 minutes, in medium-sensitive areas it is once every 30 minutes, and in low-sensitive areas it is once every 60 minutes.

[0021] Furthermore, the intelligent optimization algorithm for disturbance rejection processing schemes in the disturbance rejection intelligent design optimization module includes the following:

[0022] Step 1: Establish a candidate solution library: candidate solution set This includes one or more combinations of bagged sand well surcharge preloading method, cement mixing pile composite foundation, replacement method, solid waste-based green gel material solidification, drainage sand cushion layer dynamic compaction method, and lightweight solidified soil filling.

[0023] Step 2: Multi-objective optimization evaluation: Establishing the evaluation function ,in, , These are the normalized project cost and construction period, normalized to... interval; For the disturbance resistance safety factor (target) ≥1.3), The green index ranges from 0 to 1, with higher values ​​indicating greater environmental friendliness; weighting... ;

[0024] Step 3: Genetic Algorithm Optimization: Encode the combination schemes into chromosomes and establish a fitness function. Evaluation value The smaller the value, the greater the fitness. A roulette wheel selection method is used, and crossover is performed using a single-point crossover with a high probability of crossover. The mutation is uniform and the mutation probability is... The iteration termination condition is the maximum number of iterations. Or fitness change <10 -6 Output the optimal combination of solutions. .

[0025] Furthermore, the construction disturbance impact prediction algorithm in the anti-disturbance intelligent design optimization module includes the following:

[0026] Step 1: Calculation of the disturbance range based on the theory of circular hole expansion: Treat the pile or sand well as a problem of cylindrical hole expansion, and define the radius of the failure zone. radius of yield zone ,in, The initial radius of the pile or well. For the undrained shear strength of the soil, To obtain the ultimate lateral pressure of the pile material, For initial stress, The internal friction angle of the soil. This is the sine value of the internal friction angle;

[0027] Step 2, Finite Element Disturbance Simulation: Establish a two-dimensional axisymmetric finite element model and use the modified Cambridge model as the constitutive relation: The built-in constitutive equations of this model are automatically solved by the finite element software to simulate the pile penetration process;

[0028] Step 3: Determine the safe construction distance, minimum safe distance , The disturbance influence radius is defined; output the disturbance influence radius and safety distance diagram for each processing method.

[0029] Furthermore, the multi-parameter dynamic threshold control algorithm in the real-time monitoring module for the construction process includes the following:

[0030] Step 1: Set multi-level early warning thresholds:

[0031] The green threshold for soil filling rate is ≤ design value, the yellow threshold is ~1.2 times the design value, and the red threshold is >1.2 times the design value.

[0032] Daily settlement threshold: green threshold ≤10mm, yellow threshold 10-20mm, red threshold >20mm;

[0033] Lateral displacement threshold: green ≤ 5 mm / d, yellow 5–10 mm / d, red > 10 mm / d;

[0034] Pore ​​water pressure threshold: green ≤ 80% of hydrostatic pressure, yellow 80%–95%, red > 95%;

[0035] Compaction degree threshold: green threshold ≥96%, yellow threshold 93%~96%, red threshold <93%;

[0036] Step 2: Real-time Data Acquisition and Kalman Filtering for Noise Reduction: State Equation ,in, Here is the state transition matrix. To control the input matrix, The control input vector at time k (such as the soil filling operation command). For process noise, the observation equation Prediction Step Kalman gain during update step Status update ,in, for The observation vector at any given time (the actual value collected by the sensor). The observation matrix (maps the state space to the observation space). To observe the noise covariance matrix;

[0037] Step 3: Dynamic Rate Control Decision: Calculate the current settlement rate ,like If the rate of soil filling shows an upward trend for three consecutive observation periods and exceeds the yellow threshold, the maximum allowable filling rate will be automatically calculated. And send a speed reduction command, where, To design the filling rate, The yellow threshold represents the settling rate. The red threshold represents the settlement rate; if the red threshold is reached, a stop command is sent.

[0038] Step 4: Automatic generation of construction logs: Every 24 hours, a daily construction quality report is automatically generated, which includes time history curves of various parameters, number of times exceeding limits and handling records.

[0039] Furthermore, the full-wavefield elastic wave inversion imaging algorithm in the full-wavefield intelligent detection module includes the following:

[0040] Step 1: Elastic wave data acquisition: Excitation is performed using a hammer impact combined with a swept-sine frequency signal, with a frequency range of 1kHz to 50kHz. The signal is received by a 24-channel high-density detector array with a detector spacing of 0.15m, acquiring full-field signals including longitudinal waves (P-waves), transverse waves (S-waves), and surface waves (R-waves).

[0041] Step 2, Dispersion Curve Extraction: Perform dispersion curve extraction on the surface wave signal. Transformation ,in The displacement field (the time-domain signal received by the detector). The wave number in the x-direction. Given the angular frequency, extract the dispersion curve of the fundamental Rayleigh wave. High-resolution Radon transform is used to separate different modes. Let be the area element of the double integral, representing the area element first applied to the double integral. Points, then check Integration, that is, in Integrating over a two-dimensional domain;

[0042] Step 3: Tomographic inversion imaging: Establishing the initial velocity model In the iterative inversion, the forward modeling calculation uses the fast multipole boundary element method to calculate the theoretical travel time. Wave propagation time obtained from forward modeling; travel time residual Sensitivity matrix Model update Regularization parameters The L-curve method is used to determine the iteration termination condition. Or a maximum of 50 iterations, where, For observation travel time: the actual arrival time of the wave collected; For the velocity model of the nth iteration, Regularization parameters: determined using the L-curve method to prevent instability during inversion; It is the identity matrix. This is the transpose of the sensitivity matrix;

[0043] Step 4, Defect Identification and Quantification: Determination of Pile Integrity: If the area of ​​the abnormal wave velocity zone is less than 5%, it is classified as Class I pile with good condition; if it is 5% to 15%, it is classified as Class II pile with minor defects; and if it is greater than 15%, it is classified as Class III pile with serious defects.

[0044] Strength determination of solidified soil: elastic modulus ≥15MPa is considered qualified, 10-15MPa is barely qualified, and <10MPa is unqualified.

[0045] Furthermore, the multi-source fusion risk warning algorithm in the multi-parameter early warning and AI decision-making module includes the following:

[0046] Step 1: Feature Vector Construction: Constructing Real-Time Feature Vectors ,in These are the current filling rate, daily settlement, and cumulative settlement, respectively. and These are the pore water pressure ratio and the lateral displacement rate, respectively. These are compaction degree, moisture content deviation, and elastic modulus of the solidified soil, respectively. These are the time since the last warning, the rate of change of construction load, the change of groundwater level, and temperature, respectively.

[0047] Step 2, LSTM Time Series Prediction: A two-layer LSTM network is used to predict the subsidence trend over the next 72 hours. The first LSTM layer has 128 hidden units with dropout=0.2, and the second LSTM layer has 64 hidden units with dropout=0.2. The input is the X sequence of the past 72 hours with a window length of 144 and a sampling interval of 30 minutes. The loss function is... ,in For the first Predicted settlement values ​​for each sample For the first The actual settlement value of each sample The total number of training samples, This is the L2 regularization coefficient (penalizing model complexity). This represents all trainable parameters (weights + biases) of the LSTM network. The square of the L2 norm of the parameter;

[0048] Step 3: XGBoost Risk Classification: Concatenate the LSTM predictions with the current feature vector and input the concatenation into the XGBoost classifier. The classifier uses 500 trees, a maximum depth of 6, a learning rate of 0.05, and classifies the target values ​​as {Safe, Attention, Warning, Danger}. The output is the risk level. ;

[0049] Step 4, Early Warning Response: At T+0s, XGBoost outputs the risk level R. If R≥2 at T+5s, an early warning is triggered. The early warning response is completed within 30 seconds, with the early warning triggered at T+5s, the emergency response plan matched according to the risk level at T+10s, and the response issued simultaneously through all three channels at T+30s.

[0050] Specifically, the risk level matching and response plan includes: risk level R=1 corresponds to reducing the filling rate by 20% and increasing monitoring frequency to 15 minutes / time; risk level R=2 corresponds to immediately reducing the speed by 50% and notifying the supervising engineer; risk level R=3 corresponds to immediately stopping work and issuing a warning simultaneously through APP, SMS, and sound and light channels at T+30s.

[0051] Step 5: Feedback on the handling effect: Record the handling measures and subsequent monitoring data for each warning, and update the XGBoost model weights monthly.

[0052] Furthermore, the end-to-end quality tracking and evaluation algorithm in the quality tracking module includes the following:

[0053] Step 1: Division of Inspection Batches: According to the Highway Engineering Quality Inspection and Evaluation Standards, the project is divided into sub-item projects, sub-section projects, and unit projects. Each inspection batch includes the main control items of compaction degree and bearing capacity, as well as general items of flatness and width.

[0054] Step 2: Quality Score Calculation: Using the weighted scoring method ,in For the first Item weight, For the number of qualified points, The total score is determined by the following criteria: For excellent To be qualified, It is unqualified;

[0055] Step 3: Generate Quality Files: Each inspection batch is automatically associated with the inspection time, GPS coordinates of the inspection location, inspection value, inspection equipment, and operator to generate an unalterable quality file.

[0056] A method for quality control in the construction of soft soil roads to resist disturbance includes the following steps:

[0057] Step S1, Multi-source geological data acquisition and fusion: Multi-source heterogeneous geological data are acquired through static cone penetration testing, vane shear testing, indoor geotechnical tests, and geophysical exploration. Criterion outlier removal, Kriging interpolation spatial interpolation, confidence-weighted fusion, and disturbance-sensitive zoning are used to generate a three-dimensional geological digital model;

[0058] Step S2, Anti-disturbance intelligent design optimization: Based on the three-dimensional geological digital model, a candidate library of treatment schemes is established. Multi-objective optimization is performed through genetic algorithm to find the optimal combination of treatment schemes. The disturbance influence range and safe construction distance of each construction method are predicted by the circular hole expansion theory and the modified Cambridge model finite element simulation.

[0059] Step S3: Real-time monitoring of the construction process: Real-time collection of filling rate, compaction degree, settlement, pore water pressure and lateral displacement parameters, after Kalman filtering to remove noise, and dynamic comparison with multi-level early warning thresholds. Based on the settlement rate trend, the maximum allowable filling rate is automatically calculated and a speed-down or work stoppage command is sent. At the same time, a daily construction quality report is automatically generated every 24 hours.

[0060] Step S4, Full-field Intelligent Detection: A full-field elastic wave is excited using a hammer impact combined with a swept-sine frequency signal. The dispersion curve is extracted by transformation, and the pile foundation integrity and solidified soil strength are output as millimeter-level detection results and defect quantitative assessment by fast multipole boundary element method tomographic inversion imaging.

[0061] Step S5, Multi-parameter Early Warning and AI Decision-Making: Construct a multi-dimensional real-time feature vector, predict the settlement trend in the next 72 hours through a two-layer LSTM network, concatenate the predicted value with the feature vector and input it into the XGBoost classifier for risk classification, complete the early warning response within 30 seconds and automatically generate disposal suggestions, and update the model weights monthly to achieve self-evolution.

[0062] Step S6, Quality Tracking and Record Generation: Perform weighted scoring quality assessment by inspection batch and generate quality records.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. This invention constructs a closed-loop quality control system for the entire process of soft soil road construction, covering geological exploration, scheme design, construction management, physical testing, risk warning, and quality traceability. Based on multi-source geological data and with intelligent algorithms at its core, the system uses real-time perception and closed-loop control to upgrade the traditional post-event rectification and passive monitoring mode to a proactive anti-disturbance management mode that includes pre-event prediction, in-process control, and full-process traceability. On the one hand, it accurately constrains the impact of construction disturbances through quantitative prediction and dynamic control, effectively reducing engineering risks such as excessive settlement of soft soil subgrade, lateral instability, and thixotropic damage of soil. On the other hand, it improves construction efficiency, optimizes engineering costs, and ensures physical quality through full-process digitalization and AI decision-making, providing a systematic quality control solution for road construction in soft soil areas. It solves the core pain points of traditional soft soil road construction, such as the disconnect between exploration and design, the separation between monitoring and construction, and the imbalance between disturbance and control.

[0065] 2. The multi-source geological data acquisition and fusion module of this invention integrates four heterogeneous data sources: static cone penetration test, vane shear test, indoor geotechnical test, and engineering geophysical exploration. Combined with outlier removal and confidence-weighted fusion algorithms, it effectively compensates for the accuracy limitations of single exploration methods and significantly improves the reliability and comprehensiveness of geological parameters. It uses ordinary kriging interpolation to transform discrete borehole data into a continuous three-dimensional geological digital model, accurately characterizing the spatial heterogeneity of soft soil physical and mechanical parameters, and providing refined geological boundary conditions for subsequent scheme design and disturbance prediction.

[0066] 3. In the real-time monitoring module of the construction process of this invention, five core construction parameters are collected simultaneously: filling rate, compaction degree, settlement, pore water pressure, and lateral displacement. This enables all-weather, high-frequency real-time sensing of key indicators of construction disturbance, eliminating the data feedback lag problem of traditional manual monitoring. The Kalman filter algorithm is used to remove random noise from the sensors, improving the smoothness and accuracy of the monitoring data. Three-level warning thresholds (green, yellow, and red) are set, and the maximum allowable filling rate is automatically calculated based on the settlement rate change trend, and control commands are issued to achieve dynamic adaptive control of the construction loading rate, shifting from passive alarm to active constraint of disturbance intensity.

[0067] 4. The anti-disturbance intelligent design optimization module in this invention establishes a candidate library of treatment schemes combining multiple construction methods. With multiple objectives such as project cost, construction period, anti-disturbance safety factor, and green index, it achieves global intelligent optimization through genetic algorithm, breaking through the limitations of traditional scheme selection relying on personal experience, and outputs the foundation treatment combination scheme with the best comprehensive benefits, taking into account technical safety, economy, and environmental protection. Combining the analytical calculation of circular hole expansion theory and the corrected Cambridge model finite element numerical simulation, it quantitatively outputs the disturbance influence radius and minimum safe construction distance of each treatment method, realizing the pre-emptive and quantitative prediction of the construction disturbance impact, providing clear quantitative design basis for construction layout, process interval arrangement, and protection of adjacent structures, and avoiding the risk of foundation instability caused by the superposition of disturbances from the design source. Attached Figure Description

[0068] Figure 1 A flowchart illustrating the data flow of each module in the quality control system for soft soil road disturbance mitigation construction.

[0069] Figure 2 A flowchart illustrating the quality control method for disturbance-resistant construction of roads in soft soil. Detailed Implementation

[0070] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments take a section from K1+200 to K1+800 of a soft soil road project in a coastal area as an application scenario. The section is 600m long and 32m wide, and is designed as a two-way four-lane urban arterial road. The embodiments fully demonstrate the closed-loop management process of the present invention from geological exploration to final acceptance.

[0071] Example 1

[0072] like Figure 1 As shown, the soft soil road anti-disturbance construction quality control system includes:

[0073] The multi-source geological data acquisition and fusion module is used to acquire multi-source heterogeneous geological data of soft soil sites, and to preprocess, spatially interpolate, perform confidence-weighted fusion, and anti-disturbance sensitivity partitioning of multi-source heterogeneous geological data through a multi-source heterogeneous data spatiotemporal fusion algorithm to generate a three-dimensional geological digital model.

[0074] The anti-disturbance intelligent design optimization module is used to automatically generate the optimal combination treatment scheme based on the three-dimensional geological digital model through the intelligent selection algorithm of anti-disturbance treatment scheme, and to calculate the disturbance impact range and safe construction distance of each treatment method through the construction disturbance impact prediction algorithm.

[0075] The real-time monitoring module for the construction process is used to collect parameters such as filling rate, compaction degree, settlement, pore water pressure and lateral displacement during the construction process in real time. It also uses a multi-parameter dynamic threshold control algorithm to perform Kalman filtering for noise reduction, multi-level early warning threshold comparison and dynamic rate control decision-making on the parameters.

[0076] The full-wave field intelligent detection module is used to perform millimeter-level detection of pile foundation integrity and solidified soil strength based on the full-wave field elastic wave testing theory and the full-wave field elastic wave inversion imaging algorithm, and outputs high-resolution imaging results and defect quantitative assessment.

[0077] The multi-parameter early warning and AI decision-making module is used to integrate real-time data from the construction process real-time monitoring module and detection data from the full-wave intelligent detection module. It performs comprehensive risk assessment through the LSTM time series prediction model and XGBoost risk classification model in the multi-source fusion risk early warning algorithm, and issues early warning and automatically generates disposal suggestions within 30 seconds.

[0078] The quality tracking module is used to divide and weight the engineering quality into inspection batches through a full-process quality tracking and evaluation algorithm, and output a quality evaluation report.

[0079] Among them, the input end of the multi-source geological data acquisition and fusion module receives data from static cone penetration test equipment, vane shear tester, geotechnical testing system and engineering geophysical equipment, and outputs a three-dimensional geological digital model to the anti-disturbance intelligent design optimization module, providing refined geological parameters and spatial boundary conditions for the whole site for the selection of anti-disturbance treatment scheme and quantitative prediction of construction disturbance impact. At the same time, it outputs anti-disturbance sensitivity zoning parameters to the real-time monitoring module of the construction process for differentiated configuration of on-site monitoring sampling frequency.

[0080] The anti-disturbance intelligent design optimization module receives the three-dimensional geological digital model output by the multi-source geological data acquisition and fusion module, outputs the optimal processing scheme, disturbance impact range, and safe construction distance as construction control benchmarks, and sends them to the real-time monitoring module of the construction process, providing a unified control basis at the design level for parameter management, early warning threshold setting, and work layout during the construction phase.

[0081] The input end of the real-time construction monitoring module receives data from various sensors and construction equipment acquisition terminals on site. It also receives anti-disturbance sensitivity zoning parameters output by the multi-source geological data acquisition and fusion module and construction control benchmarks issued by the anti-disturbance intelligent design optimization module. It outputs real-time construction monitoring data to the multi-parameter early warning and AI decision-making module, providing a core time-series monitoring data source for risk trend prediction and comprehensive risk classification. The module also receives control instructions and early warning information fed back from the multi-parameter early warning and AI decision-making module, and links on-site construction equipment to adjust operating parameters, serving as the on-site execution terminal for closed-loop management.

[0082] The input end of the full-wave intelligent detection module is connected to the elastic wave acquisition equipment, and the output pile foundation and solidified soil detection data is connected to the multi-parameter early warning and AI decision-making module. This module is used to supplement the dimensions of entity quality characteristics, correct risk assessment parameters, and improve the accuracy of risk classification. At the same time, the detection results are output to the quality tracking module as the core judgment basis for the main control project of quality assessment.

[0083] The input terminals of the multi-parameter early warning and AI decision-making module receive real-time construction monitoring data from the real-time construction process monitoring module and pile foundation and solidified soil detection data from the full-wave field intelligent detection module, respectively. The module outputs control commands and early warning information to the real-time construction process monitoring module, which then coordinates with on-site construction equipment to adjust operating parameters, thereby achieving proactive intervention in construction disturbances and forming a closed-loop control of "design, construction, early warning, and control". At the same time, the module outputs early warning handling records to the quality tracking module as a basis for evaluating the quality of process control.

[0084] The quality tracking module aggregates all the data from the aforementioned modules throughout the entire process, outputs quality assessment reports and archives, and enables traceability and immutability of the quality throughout the entire life cycle of the project. The accumulated full-process data can be used to support the iteration of system algorithms and the optimization of similar project designs.

[0085] Example 2

[0086] Specific implementation of the spatiotemporal fusion algorithm for multi-source heterogeneous data:

[0087] This embodiment takes the section from K1+200 to K1+800 of a coastal soft soil road project as an example to explain in detail the process of multi-source geological data acquisition and fusion.

[0088] Step 1: Data Acquisition and Preprocessing

[0089] A total of 12 static cone penetration test boreholes were laid out on the site, with a depth of 20-25m and a spacing of 50m; 8 vane shear test points were distributed in different geological units; 24 sets of indoor geotechnical test soil samples were taken from each borehole, covering different depths; 3 seismic geophysical profiles were arranged longitudinally along the road, with a total length of 600m.

[0090] The raw data includes:

[0091] Static cone penetration test: probe resistance (0.8~3.2MPa), sidewall friction (0.02~0.12MPa), a total of 288 sets of data;

[0092] Vane shear: Undrained shear strength (12~38kPa), residual value shear modulus A total of 96 sets of data;

[0093] Indoor test: Natural moisture content (35%~68%), liquid limit (42%~72%), plastic limit (20%~35%), compression ratio (0.8~2.2MPa) -1 ), consolidation coefficient (0.8~3.5×10) -3 (cm² / s), a total of 144 sets of data;

[0094] Geophysical exploration: P-wave velocity (1200~2100m / s), shear wave velocity (80~180m / s), a total of 752 data collection points;

[0095] use The criterion is to remove outliers from all raw data: calculate the arithmetic mean of all data. with standard deviation If a certain observation value satisfy If so, it is considered an outlier and removed;

[0096] In this example, the initial probe resistance at a certain measuring point (K1+280, depth 6m) during static cone penetration testing was 2.8 MPa. The average value of the data from 48 measuring points in this batch was... =1.6MPa, standard deviation =0.38MPa, |2.8-1.6|=1.2>3×0.38=1.14, which is judged as an outlier and removed; after checking the original records, this point is caused by the gravel interlayer and does not represent the true parameters of the soft soil, so the removal is reasonable;

[0097] After item-by-item elimination, a total of 18 sets of abnormal data were removed, leaving 1262 sets of valid data, with a data validity rate of 98.6%.

[0098] Step 2: Spatial Interpolation and Geological Profile Generation

[0099] Spatial interpolation of discrete borehole data is performed using ordinary kriging interpolation:

[0100] Establish semi-mutation function ,in The spatial distance between two points The distance interval is Number of data point pairs For position Geological parameter values ​​at the location, For position Geological parameter values ​​at the location;

[0101] With undrained shear strength Fit a spherical model to the target parameters. The fitting obtained in this embodiment =0.2, =1.8, =120;

[0102] By solving the weight coefficients of each known point using the Kriging equations, the parameter estimates of the interpolation points are calculated, generating a continuous geological profile along the longitudinal direction of the road. A three-dimensional geological digital model with 600×32×25 grid nodes (600 in the X direction, 32 in the Y direction, and 25 in the Z direction) is output, with a grid accuracy of 1m×1m×0.5m, including undrained shear strength. (Range 12~38kPa), Compression modulus (Range 1.8~5.2MPa), foundation bearing capacity (Range 55–110 kPa), Natural water content The three-dimensional spatial distribution of parameters such as (range 35%–68%);

[0103] Step 3: Confidence-weighted fusion of multi-source data

[0104] Setting confidence weights for four types of data sources: static probing cross blade shearing Indoor test Geophysical exploration ;

[0105] The fusion value is calculated as follows ,in For the kth method Independent measurements, These are the weighted and integrated geological parameter values;

[0106] The compression modulus measured by various methods at section K1+350 and depth 4m in this example. The values ​​are as follows: static cone penetration test back-calculated value 3.2 MPa, vane shear conversion value 2.8 MPa, laboratory test value 3.5 MPa, and geophysical inversion value 3.0 MPa. The calculated values ​​are:

[0107] =(0.35×3.2+0.30×2.8+0.25×3.5+0.10×3.0)=3.11MPa;

[0108] The integrated geological parameters of all points across the entire site are combined to form a complete three-dimensional geological digital model;

[0109] Step 4: Disturbance Sensitivity Zone

[0110] Calculate the disturbance immunity sensitivity index ,in ,

[0111] Taking section K1+350 as an example: natural water content =52%, liquid limit =68%, compression modulus =3.1MPa, foundation bearing capacity =72kPa, then: =0.4×(52 / 68)+0.35×(1 / 3.1)+0.25×(1 / 72)=0.306+0.113+0.003=0.422;

[0112] according to Value division: K1+200~K1+380 segment Values ​​of 0.55–0.85 indicate a medium to high sensitivity zone; the range K1+380–K1+600. Values ​​of 0.35–0.52 indicate a low to medium sensitivity range; the range K1+600–K1+800. Values ​​of 0.28 to 0.38 indicate a low-sensitivity region.

[0113] Highly sensitive areas: >0.8 corresponds to a construction monitoring sampling frequency of once every 15 minutes; Medium-sensitive area: 0.5 < ≤0.8, corresponding to 1 time / 30min; Low sensitivity zone: ≤0.5 corresponds to 1 time / 60min;

[0114] The final zoning results for this project are as follows: the high-sensitivity area accounts for approximately 15% (mainly the section from K1+250 to K1+330, where the deep silty soil layer is thick), the medium-sensitivity area accounts for approximately 55%, and the low-sensitivity area accounts for approximately 30%.

[0115] The zoning parameters will be sent to the real-time monitoring module of the construction process for differentiated configuration of the monitoring sampling frequency of each section; the three-dimensional geological digital model and the sensitivity zoning results will serve as the basic input for the scheme design of Example 3 and the construction monitoring of Example 5.

[0116] Example 3

[0117] This embodiment uses the three-dimensional geological digital model and sensitivity zoning results generated in Embodiment 2 to intelligently optimize the anti-disturbance processing scheme:

[0118] Step 1: Establish a candidate solution library

[0119] The candidate foundation treatment scheme for this project includes six foundation construction methods: bagged sand well surcharge preloading method, cement mixing pile composite foundation, replacement method, solid waste base green gel material solidification, drainage sand cushion dynamic compaction method, and lightweight solidified soil filling; single construction method or combination of multiple construction methods are supported.

[0120] Based on the geological conditions of the K1+200~K1+800 section, the main stress-bearing layer is silty clay with a thickness of 8~18m and a bearing capacity of 55~110kPa, generating 3 sets of candidate combination schemes:

[0121] Option A: Bagged sand wells + surcharge preloading + drainage sand cushion layer;

[0122] Option B: Cement mixing pile composite foundation + drainage sand cushion layer;

[0123] Option C: Solid waste-based green gel solidification + lightweight solidified soil filling;

[0124] Step 2: Multi-objective optimization evaluation

[0125] Establish evaluation function Weight ;

[0126] The parameters for each scheme are as follows:

[0127] Option A: Bagged sand wells + surcharge preloading + drainage sand cushion layer:

[0128] The estimated project cost is 6.5 million yuan, normalized. =0.65; Construction period 120 days, normalized =0.80; Disturbance immunity safety factor =1.25, numerical analysis shows that it does not meet the requirements. ≥1.3 constraint; Green Index =0.70 (Sand mining has a certain environmental impact); because <1.3, apply penalty +M (M=100). The value is extremely large, so it is excluded.

[0129] Option B: Cement mixing pile composite foundation + drainage sand cushion layer:

[0130] The estimated project cost is 9.8 million yuan, normalized. =1.00; Construction period 85 days, normalized =0.53; Disturbance immunity safety factor =1.42, which satisfies the condition. ≥1.3 constraint; Green Index =0.65 (Cement production has high carbon emissions); ;

[0131] Option C: Solid waste-based green gel solidification + lightweight solidified soil filling:

[0132] The estimated project cost is 8.2 million yuan, normalized. =0.86; Construction period 68 days, normalized =0.40; Disturbance immunity safety factor =1.35, which satisfies the condition. ≥1.3 constraint; Green Index =0.85 (Industrial solid waste resource utilization has significant environmental benefits);

[0133] ;

[0134] Evaluation value of Option C =0.584 is the optimal value;

[0135] Step 3: Optimization using genetic algorithm

[0136] The scheme combination is encoded as a binary chromosome of length 6, with each bit corresponding to the activation status of a construction method. Scheme C is encoded as [0,0,0,1,0,1]; the fitness function is set to The smaller the evaluation value, the greater the fitness.

[0137] Genetic algorithm parameters: population size 50, roulette wheel selection, single-point crossover ( ), uniform variation ( Maximum number of iterations: 200 generations; fitness change: <10. -6 ;

[0138] The iteration converged at generation 47, and the optimal solution combination was: solid waste-based green gel solidification + lightweight solidified soil filling + drainage sand cushion, which had the best fitness value. =1 / 0.584=1.712, corresponding to a disturbance rejection safety factor of 1.38, which satisfies the condition. With a constraint of ≥1.3, the total cost is reduced by about 16% compared to the traditional cement mixing pile scheme, the construction period is shortened by about 20%, and the solid waste resource utilization rate reaches 85%. This optimal scheme will be used as the subsequent construction control benchmark and input into the real-time monitoring module of the construction process.

[0139] Example 4

[0140] This embodiment is based on the optimal scheme selected in Embodiment 3: solid waste-based green gel solidification + lightweight solidified soil filling + drainage sand cushion layer. The construction disturbance impact prediction algorithm is used to calculate the disturbance impact range and minimum safe construction distance of each construction method, providing a quantitative basis for construction layout.

[0141] Step 1: Calculation of the disturbance range based on the theory of circular hole expansion

[0142] Taking cement mixing piles as an example, the designed pile diameter is 0.5m, the pile length is 12m, and they are arranged in a quincunx pattern with a pile spacing of 1.4m. The disturbance range of a single pile is calculated using the circular hole expansion theory.

[0143] Input parameter: Initial radius of the pile =0.25m; According to the geological model of Example 2, the undrained shear strength of the soil at a depth of 4m in section K1+350 is... =25kPa, initial stress =60kPa, soil internal friction angle =12°; pile side pressure =150kPa (the statistical value of the measured lateral pressure of the cement mixing pile at the site).

[0144] Calculate the radius of the damaged area With the radius of the yield zone :

[0145] ;

[0146] ;

[0147] Step 2: Finite element disturbance simulation

[0148] A two-dimensional axisymmetric finite element model was established, and the modified Cambridge model was used as the constitutive relation; the model parameters were taken from the geological model of Example 2: compressibility index. =0.35, void ratio =1.2, effective average stress =50kPa, specific volume on a normal compression line =0.42, initial deviatoric stress =15 kPa;

[0149] Model dimensions: radial 20m (boundary effect eliminated), depth 25m; boundary conditions: bottom fixed constraint, right radial constraint, left axis of symmetry; pile penetration rate 0.5m / min;

[0150] Simulation results show that the vertical normal strain of the soil around the pile is... The boundary of the affected area is 1.82m from the pile center (greater than the theoretical yield zone radius of 1.041m; the difference stems from the theoretical solution not considering soil dilatation). The boundary of the affected area with a horizontal displacement greater than 5mm is 2.05m from the pile center. The radius of disturbance influence is taken as a safety margin. =2.05m;

[0151] Step 3: Determine the safe construction distance

[0152] Calculate the radius of influence of disturbance for each construction method: Cement mixing pile: =2.05m; Bagged sand well: =1.35m; Lightweight solidified soil filling (layered compaction): =1.10m (per layer); Dynamic compaction method (drainage sand cushion layer): =3.20m (based on 3000kN·m impact energy);

[0153] Minimum safe construction distance = +1.5m;

[0154] Cement mixing piles: =2.05 + 1.5 = 3.55m, take 4.0m upwards; Bagged sand well: =1.35 + 1.5 = 2.85m, take 3.0m upwards; Lightweight solidified soil filling: =1.10 + 1.5 = 2.60m, take 3.0m upwards; Dynamic compaction method: =3.20 + 1.5 = 4.70m, round up to 5.0m;

[0155] Output a map showing the distribution of safe construction distances across the entire site for construction layout: the net distance between adjacent pile foundations should be no less than 4.0m, the edge of the dynamic compaction work area should be no less than 5.0m from the edge of the solidified area, and the filling and compaction equipment should maintain a safe distance of more than 3.0m from the edge of the completed structure; this safe distance parameter will be input into the real-time monitoring module of the construction process as a hard constraint for on-site equipment scheduling and work sequence arrangement.

[0156] Example 5

[0157] This embodiment is based on the sensitivity zoning results of embodiment 2: the monitoring frequency of the high-sensitivity zone K1+250~K1+330 is 15min / time, the safe construction distance of embodiment 4 is ≥4.0m for pile foundation construction, and real-time monitoring and dynamic control of the construction process are carried out at the K1+300 section, which is a typical section of the high-sensitivity zone;

[0158] Monitoring parameter acquisition method:

[0159] Filling rate: GNSS differential positioning terminals and operational data acquisition systems are installed on pavers and rollers at the filling work surface to upload real-time data on paving thickness, number of compaction passes, and work area. This data, combined with the daily fill volume, is used to calculate the layered filling rate. The designed filling rate in this example is... =0.5m / d;

[0160] Compaction degree: The continuous compaction monitoring system on the vibratory roller is used to calculate the roadbed compaction degree by back-calculating the acceleration response signal of the vibratory wheel. Full cross-section data is collected after each rolling pass; at the same time, a set of sand cone sampling data is used every 200m for calibration to ensure data accuracy.

[0161] Settlement: A monitoring section is set up every 50m along the longitudinal direction of the road. At the K1+300 section, one set of magnetostrictive settlement gauges is set up in the middle of the road and on the left and right shoulders. The accuracy is 0.01mm. The probes are buried in the roadbed surface to collect the cumulative settlement value in real time and upload it to the system every 15 minutes through the 4G wireless transmission module.

[0162] Pore ​​water pressure: Vibrating wire pore water pressure gauges are installed at depths of 2m, 5m and 8m below ground at section K1+300 to collect pore water pressure values ​​in real time. The system automatically calculates the pore water pressure ratio: pore water pressure ratio = pore water pressure / overlying effective stress.

[0163] Lateral displacement: At the K1+300 section, one inclinometer tube is installed at 1m and 3m outside the toe of the roadbed slope, with a depth of 15m. It is equipped with a fixed inclinometer probe and a measuring point every 0.5m. The horizontal displacement at different depths is collected in real time and the daily lateral displacement rate of the toe of the slope is calculated.

[0164] Step 1: Set multi-level early warning thresholds

[0165] Set three threshold levels (green, yellow, and red) for the five parameters:

[0166] Filling rate: Green ≤ 0.5 m / d, Yellow 0.5~0.6 m / d, Red > 0.6 m / d;

[0167] Daily settlement: Green ≤10mm, Yellow 10~20mm, Red >20mm;

[0168] Lateral displacement rate: green ≤5mm / d, yellow 5~10mm / d, red >10mm / d;

[0169] Pore ​​water pressure ratio: Green ≤ 0.8, Yellow 0.8~0.95, Red > 0.95;

[0170] Compaction degree: Green ≥96%, Yellow 93%~96%, Red <93%;

[0171] Step 2: Kalman filtering for noise reduction

[0172] Establish state equations Observation equation For the raw data from the settlement gauge at section K1+300, the original noisy data is smoothed and filtered through iterative calculations in the prediction and update steps. In this example, the original 72-hour settlement sequence is [5.2, 6.8, 8.5, 10.2, 12.8, 14.5, 16.3, 18.8, 17.5, 16.2, 18.3, 20.1, 21.5, 23.8, 26.2, 28.5, 30.1, 32.5, 35.2, 38.0, 40.5, 42.2, 44.0, 45.8] mm, and samples are taken every 3 hours.

[0173] After filtering, the output smooth sequence [5.1,6.6,8.3,10.1,12.5,14.4,16.1,18.5,17.3,16.0,18.0,20.0,21.2,23.5,26.0,28.2,29.8,32.2,34.8,37.5,40.0,42.0,43.5,45.2] mm effectively suppresses the sensor's random noise. The original noise standard deviation was about 0.5 mm, which was reduced to 0.15 mm after filtering.

[0174] Step 3: Dynamic Rate Control Decision

[0175] On the 15th day of construction, the cumulative filling height was 7.2m. The K1+300 section of the high-sensitivity area was continuously observed for 3 cycles, with sampling every 15 minutes and the average value within the cycle was taken. The settlement rates of the 3 observation cycles were 11mm / d, 14mm / d and 18mm / d, respectively, showing a continuous upward trend and exceeding the yellow threshold of 10mm / d.

[0176] at this time =10mm / d, =20mm / d, =18mm / d, =0.5m / d;

[0177] Automatically calculate the maximum allowable filling rate ,

[0178] ;

[0179] The system automatically issues a speed reduction command, adjusting the filling rate from 0.48 m / d to within 0.30 m / d. The actual implementation is achieved by automatically limiting the paving thickness through the paver's GNSS terminal. At the same time, the monitoring frequency of the K1+300 section is increased to once every 10 minutes.

[0180] On the 18th day after the rate reduction, the settlement rate dropped back to 8 mm / d, below the yellow threshold. The system automatically lifted the rate reduction restriction and restored the design rate of 0.5 m / d. If the settlement rate continues to rise and reaches the red threshold of 20 mm / d, the system will directly send a shutdown command. This was not triggered in this embodiment.

[0181] By the 45th day of construction, the filling was completed, and the cumulative settlement finally stabilized at 186mm (design allowable value ≤250mm); the maximum lateral displacement was 8.3mm / d, which did not exceed the red threshold; the maximum pore water pressure ratio was 0.89, which did not exceed the red threshold, and the compaction degree test results were all ≥96.5%.

[0182] Step 4: Automatic generation of construction logs

[0183] The daily construction quality report, automatically generated every 24 hours, has accumulated to 45 reports, providing a complete record of monitoring data throughout the entire process.

[0184] Example 6

[0185] This embodiment performs full-wavelength elastic wave non-destructive testing on the cement mixing piles completed in Embodiment 3. The cement mixing piles are located near the K1+300 section, with a total of 12 piles, each 12m long and 0.5m in diameter. The full-wavelength intelligent testing module of this invention is used.

[0186] Input: The object of detection is a cement mixing pile with a length of 12m and a pile diameter of 0.5m; a 24-channel high-density geophone array is arranged radially along the top of the pile with a geophone spacing of 0.15m; a hammer-excited swept-sine frequency signal is used with a frequency range of 1kHz to 50kHz and a sampling rate of 1MHz;

[0187] Step 1: Elastic Wave Data Acquisition

[0188] Test object: No. 3 cement mixing pile at section K1+300, pile length 12m, pile diameter 0.5m, age 28 days;

[0189] A 24-channel high-density geophone array is arranged radially along the top of the pile, with a geophone spacing of 0.15m and a total coverage area of ​​3.45m. A hammer-excited swept-sine frequency signal is used, with a frequency range of 1kHz to 50kHz, a sampling rate of 1MHz, and a recording duration of 20ms per channel. The full wavefield time domain signals of longitudinal wave (P-wave), transverse wave (S-wave), and surface wave (R-wave) are acquired simultaneously.

[0190] Data acquisition parameters: 5kg hammer impact energy, 0.5m drop height, 4 stacks per stage to improve signal-to-noise ratio;

[0191] Data quality assessment: The first arrival of the P wave is clearly distinguishable, and the signal-to-noise ratio is ≥35dB, which meets the inversion requirements;

[0192] Step 2: Extraction of dispersion curves

[0193] Front wave signal Transformation: Extracting the dispersion curve of the fundamental Rayleigh wave In this example, the Rayleigh wave phase velocity of pile foundation #3 is approximately 185 m / s at a frequency of 1 kHz, and gradually decreases to 65 m / s at 50 kHz as the frequency increases, exhibiting normal dispersion characteristics. After using high-resolution Radon transform to separate higher-order modes, the signal-to-noise ratio of the fundamental dispersion curve is improved by approximately 15 dB.

[0194] Step 3: Tomographic inversion imaging

[0195] Establish an initial velocity model The initial P-wave velocity was 1500 m / s, the S-wave velocity was 100 m / s, the layer thickness was 0.2 m, and there were a total of 60 layers. The forward modeling calculation used the fast multipole boundary element method to calculate the theoretical travel time. Calculate the travel time residual Construct a sensitivity matrix The velocity model is updated iteratively: , The regularization parameter is determined using the L-curve method. =0.35;

[0196] The termination condition is met on the 32nd iteration. It took approximately 4.5 minutes in total;

[0197] Output: Three-dimensional wave velocity imaging profile of the pile body and the solidified soil around the pile, with a spatial resolution of 0.15m (longitudinal wave) × 0.15m (transverse wave) × 0.5m (depth).

[0198] Wave velocity imaging results show that the normal longitudinal wave velocity of pile #3 is approximately 2650 m / s, but an abnormal wave velocity zone appears in the depth range of 4.2 to 4.8 m below the pile top, with a wave velocity of approximately 1850 m / s, which is about 70% of the normal wave velocity. Calculations show that the area of ​​this abnormal zone accounts for approximately 8.2% of the cross-sectional area of ​​the pile.

[0199] Pile integrity assessment: Based on the proportion of the wave velocity anomaly zone area: 8.2% is between 5% and 15%, and is classified as a Class II pile (minor defect).

[0200] Analysis revealed that the defect was caused by insufficient cement grouting due to the encounter of a local hard interlayer at this depth during construction. Cross-verification with data from two test piles symmetrically arranged on both sides of the pile showed that the defect was approximately 0.5m long and distributed in a fan-shaped area of ​​about 120° on the northwest side of the pile. Grouting reinforcement should be carried out on this part.

[0201] The strength of solidified soil is determined by converting the elastic modulus to wave velocity. The average elastic modulus of the intact section of the pile. =16.8MPa (≥15MPa, qualified); Elastic modulus of defective section =9.2MPa (<10MPa, unqualified); Pile perimeter solidified soil: average elastic modulus within a range of 0.8-1.2m from the pile center =12.3MPa (10-15MPa, barely acceptable);

[0202] Output: High-resolution imaging image and defect quantification assessment report; This report will be input into the quality tracking module as the core basis for quality assessment, and the defect data will be synchronized to the multi-parameter early warning and AI decision-making module and incorporated into the risk correction factor.

[0203] Example 7

[0204] This embodiment is based on the real-time monitoring data of embodiment 5: on the 15th day of construction, the settlement rate continued to rise, triggering a rate reduction. The pile foundation test results of embodiment 6: Class II defects in pile #3. This demonstrates the entire process of comprehensive risk assessment and intelligent early warning by the multi-parameter early warning and AI decision-making module.

[0205] Constructing 12-dimensional real-time feature vectors ,

[0206] The last sampling at 10:00 AM on the 15th day of construction, before the speed reduction command was issued, provided the real-time data for section K1+300:

[0207] =0.48m / d; =16mm (cumulative over the past 24 hours); =128mm (cumulative from start of work to the day); =0.88 (at a depth of 5m); =6.2mm / d (maximum daily rate at the foot of the slope); =94.5% (recently tested value); =+2.3% (relative to optimum moisture content); =12.3MPa (from the pile perimeter soil test results of Example 6); =12h (the last yellow alert was issued 12 hours ago); =8% (load increment in the last 24 hours / total load); =+0.1m (rise in the last 24 hours); =26℃;

[0208] Right now =[0.48m / d,16mm,128mm,0.88,6.2mm / d,94.5%,2.3%,12MPa,12h,8%,0.1m,26℃];

[0209] Step 1: LSTM Timing Prediction

[0210] A trained two-layer LSTM network is used to predict the subsidence trend over the next 72 hours: input the feature sequence of the past 72 hours, window length 144, sampling interval 30 min;

[0211] Model architecture: First layer: LSTM128 hidden units + dropout 0.2; Second layer: LSTM64 hidden units + dropout 0.2; Model training data: Approximately 200,000 sets of monitoring data from 12 soft soil road projects in China over the past 5 years; 5-fold cross-validation RMSE = 2.3mm.

[0212] Prediction results: The cumulative settlement in the next 72 hours will reach 45mm, increasing from the current 128mm to 173mm, and the settlement rate will continue to rise: the predicted settlement rate in the 72nd hour is 21mm / d, exceeding the red threshold of 20mm / d;

[0213] Forecast uncertainty range (95% confidence): 43–47 mm;

[0214] Step 2: XGBoost Risk Classification

[0215] The LSTM predictions—cumulative settlement prediction of 45 mm over the next 72 hours and final settlement rate prediction of 21 mm / d—are concatenated with the current 12-dimensional feature vector to obtain a 13-dimensional input vector, which is then input into the trained XGBoost classifier.

[0216] Classifier parameters: 500 trees, maximum depth 6, learning rate 0.05. Classification targets {safe (0), attention (1), warning (2), danger (3)};

[0217] Output: Probabilities for each level [0.02, 0.08, 0.78, 0.12], with the highest probability corresponding to risk level R=2 (warning level);

[0218] Step 3: Early Warning Response

[0219] T+0s (10:00:00): XGBoost outputs risk level R=2, and the system starts the early warning process;

[0220] T+5s (10:00:05): R≥2 triggers the early warning mechanism, and the system synchronously starts data recording and log storage;

[0221] T+10s (10:00:10): Match the emergency response plan—immediately reduce the backfilling rate by 50%, from the current 0.48m / d to 0.24m / d, increase the monitoring frequency of the K1+300 section to 15min / time (originally 30min / time), and simultaneously push early warning notifications to the mobile phones of the supervising engineer and the technical person in charge of the construction unit;

[0222] T+30s (10:00:30): Warning information is issued simultaneously through three channels: on-site operation APP pop-up, management personnel SMS, and construction site sound and light alarm, including risk level, predicted settlement curve, and suggested disposal measures;

[0223] Step 4: Feedback on the handling results

[0224] After the speed reduction command was executed, it was confirmed on-site at 10:05 that it had been implemented, and subsequent monitoring data continued to be tracked.

[0225] Day 1 after the rate reduction (Day 16): The settlement rate dropped to 14 mm / d; Day 3 after the rate reduction (Day 18): The settlement rate dropped to 8 mm / d, below the yellow threshold, and the system automatically lifted the rate reduction limit; Day 30: The cumulative settlement was 194 mm, and finally stabilized at 186 mm.

[0226] Meanwhile, in response to the Class II defect of pile #3, the system simultaneously issued reinforcement recommendations: high-pressure grouting reinforcement should be carried out on the defect section with a depth of 4.2 to 4.8m, with a grouting pressure of 1.5MPa and a grout water-cement ratio of 0.45; the construction unit completed the grouting reinforcement on the 20th day, and the retest on the 28th day showed that the wave velocity of this section recovered to 2350m / s, and the elastic modulus rose to 15.6MPa, indicating that the defect repair was qualified;

[0227] After all the warnings were handled, the complete handling measures and subsequent monitoring data were recorded. A total of 14 days of continuous monitoring data were collected, with more than 1,300 sampling points. The data was included in the monthly incremental training dataset. At the end of the month, all 8 warning samples of the month were summarized, and the XGBoost model was incrementally trained and the weights were updated, so that the accuracy of subsequent warnings gradually increased from 87% to 94%.

[0228] Example 8

[0229] This embodiment, based on the full-process data from embodiments 2 to 7, demonstrates the batch division, weighted quality scoring, and quality record generation of the quality tracking module;

[0230] Step 1: Lot Division

[0231] According to the "Standards for Quality Inspection and Evaluation of Highway Engineering", the section from K1+200 to K1+800 (600m) is divided into the following levels:

[0232] Unit Project: **City Coastal Avenue Project (Roadbed Engineering);

[0233] Sub-project: Soft soil treatment for section K1+200~K1+800, including pile foundation, filling and drainage;

[0234] Sub-project: Each 200m section is an inspection batch, for a total of 3 inspection batches;

[0235] This example uses the K1+200~K1+400 section (the first inspection batch) as an example for detailed evaluation; each inspection batch includes key control items (compaction degree, bearing capacity) and general items (flatness, width).

[0236] Step 2: Weighted Quality Scoring

[0237] Weighted scoring method is used: ;

[0238] Weight settings: compaction degree 0.4, bearing capacity 0.3, flatness 0.15, width 0.15;

[0239] Summary of inspection data for batches K1+200 to K1+400:

[0240] Testing items Total points qualifying points pass rate Compaction degree (sand cone method) 32 points 31 points 96.9% Bearing capacity (static load test) 6 o'clock 6 o'clock 100% Flatness (using a 3m straightedge) 4 PM 3 PM 93.8% Width (steel ruler) 4 PM 4 PM 100%

[0241] Calculate the score:

[0242] Q=0.4×(31 / 32)×100+0.3×(6 / 6)×100+0.15×(15 / 16)×100+0.15×(16 / 16)×10 0=0.4×96.875+0.3×100+0.15×93.75+0.15×100=38.75+30+14.06+15=97.81 points

[0243] In this example, for a certain inspection batch: compaction was tested at 20 points, with 19 points passing; bearing capacity was tested at 3 points, with all 3 points passing; flatness was tested at 10 points, with 9 points passing; and width was tested at 10 points, with all 10 points passing.

[0244] Calculate the score: Q = 0.4 × (19 / 20) × 100 + 0.3 × (3 / 3) × 100 + 0.15 × (9 / 10) × 100 + 0.15 × (10 / 10) × 100 = 38 + 30 + 13.5 + 15 = 96.5 points;

[0245] Evaluation criteria: Q≥90 is excellent, 75≤Q<90 is qualified, and Q<75 is unqualified; this inspection report scored 97.81 points, and is rated as excellent.

[0246] Step 3: Quality Records

[0247] Each inspection batch is automatically associated with the following information to generate an unalterable quality file: inspection time, GPS coordinates of the inspection location (e.g., 120.352°E, 30.168°N), various test values, inspection equipment number, and operator ID. After being encrypted by hash, the information is stored on the blockchain to generate an unalterable quality file.

[0248] Example 9

[0249] Combination Figure 2 In accordance with Examples 1-8 above, the method for controlling the construction quality of soft soil roads to resist disturbance according to the present invention is performed according to the following steps:

[0250] Step S1, Multi-source geological data acquisition and fusion (see Example 2 for details): 1280 sets of data were collected through static cone penetration testing, vane shear testing, indoor geotechnical tests, and geophysical exploration. The criteria remove 18 outliers, and Kriging interpolation generates a 1m×1m×0.5m resolution three-dimensional geological digital model. After confidence weighted fusion, the comprehensive geological parameters of each point are output. The site is divided into high (15%), medium (55%), and low (30%) sensitive areas by anti-disturbance sensitivity zoning. The monitoring frequency of high sensitive areas is 15min / time.

[0251] Step S2, Anti-disturbance intelligent design optimization (see Examples 3 and 4 for details): Based on the three-dimensional geological model, the genetic algorithm outputs the optimal solution after 47 generations of iteration: Solution C: Solid waste-based green gel solidification + lightweight solidified soil filling + drainage sand cushion layer. =0.584, Disturbance Resistance Safety Factor =1.38; The radius of influence of disturbance for each construction method output by the circular hole expansion theory and finite element simulation: 2.05m for cement mixing piles and 3.20m for dynamic compaction; the safe construction distance: 4.0m for cement mixing piles and 5.0m for dynamic compaction, forming the construction control benchmark;

[0252] Step S3, Real-time monitoring and dynamic control of the construction process (see Example 5 for details): Five parameters are collected in real time by on-site sensors: filling rate, compaction degree, settlement, pore water pressure, and lateral displacement. After Kalman filtering to remove noise, the parameters are dynamically compared with the green, yellow, and red three-level thresholds. On the 15th day of construction, the settlement rate increases to 18 mm / d for three consecutive cycles. The system automatically calculates the maximum allowable filling rate of 0.30 m / d and issues a speed reduction command. 45 daily construction quality reports are automatically generated.

[0253] Step S4, Full-field Intelligent Detection (see Example 6): A hammer-impacted swept-sine frequency signal (1-50kHz) is used to excite the full-field, which is received by a 24-channel detector. The dispersion curve was extracted by transformation, and the fast multipole boundary element method tomographic inversion converged after 32 iterations. The output pile wave velocity image showed a spatial resolution of 0.15m. The wave velocity anomaly zone of pile #3 accounted for 8.2%, which was determined to be a Class II pile. The elastic modulus of the solidified soil was 12.3MPa, which was determined to be barely qualified.

[0254] Step S5, Multi-parameter Early Warning and AI Decision Making (see Example 7 for details): Construct a 12-dimensional real-time feature vector, use a two-layer LSTM (128+64 hidden units, dropout 0.2) to predict the subsidence trend in the next 72 hours: cumulative subsidence of 45mm, final rate of 21mm / d exceeding the red threshold, XGBoost (500 trees, depth 6) outputs risk level R=2 (warning), complete the early warning response within 30 seconds: reduce speed by 50%, increase monitoring density to 15min / time, and provide simultaneous early warning through three channels; monthly incremental training updates the model weights;

[0255] Step S6, Quality Tracking and File Generation (see Example 8 for details): Weighted scoring based on 3 inspection batches: K1+200~K1+400 97.81 points, K1+400~K1+600 96.52 points, K1+600~K1+800 98.13 points, all rated as excellent; Each inspection batch is automatically associated with the inspection time, GPS coordinates, inspection values, equipment and personnel information, and an unalterable quality file is generated after SHA-256 hash encryption.

[0256] The system and method of this invention can be executed by a computing device including a processor and a memory. The memory stores a computer program, and when the processor executes the computer program, it implements the functions and method steps of the above-mentioned modules. The processor includes one or more of a general-purpose processor, a digital signal processor, an application-specific integrated circuit, or a field-programmable gate array, and the memory includes a volatile or non-volatile storage medium.

Claims

1. A construction quality control system for soft soil roads to resist disturbance, characterized in that, include: The multi-source geological data acquisition and fusion module is used to acquire multi-source heterogeneous geological data of soft soil sites, and to preprocess, spatially interpolate, perform confidence-weighted fusion, and perform anti-disturbance sensitivity zoning on the multi-source heterogeneous geological data through a multi-source heterogeneous data spatiotemporal fusion algorithm to generate a three-dimensional geological digital model and anti-disturbance sensitivity zoning parameters. The anti-disturbance intelligent design optimization module is connected to the multi-source geological data acquisition and fusion module. It is used to automatically generate the optimal combination processing scheme based on the three-dimensional geological digital model through the intelligent selection algorithm of anti-disturbance processing scheme, and to calculate the disturbance impact range and safe construction distance of each processing method through the construction disturbance impact prediction algorithm. The real-time monitoring module for the construction process is connected to the anti-disturbance intelligent design optimization module. It is used to receive the optimal combination processing scheme, the disturbance influence range and the safe construction distance as the construction control benchmark, and to collect the filling rate, compaction degree, settlement, pore water pressure and lateral displacement parameters in real time during the construction process. The parameters are denoised by Kalman filtering, compared with multi-level early warning thresholds and made dynamic rate control decisions through a multi-parameter dynamic threshold control algorithm. The full-wave field intelligent detection module is used to perform millimeter-level detection of pile foundation integrity and solidified soil strength based on the full-wave field elastic wave testing theory and the full-wave field elastic wave inversion imaging algorithm, and outputs high-resolution imaging results and defect quantitative assessment. The multi-parameter early warning and AI decision-making module is connected to the real-time construction process monitoring module and the full-wavelength intelligent detection module, respectively. It is used to integrate the real-time monitoring data of the real-time construction process monitoring module and the detection data of the full-wavelength intelligent detection module, and to conduct a comprehensive risk assessment through a multi-source fusion risk early warning algorithm, and issue early warnings and automatically generate disposal suggestions. The multi-parameter early warning and AI decision-making module is also connected to the real-time construction process monitoring module to feed back control commands to the real-time construction process monitoring module, and to coordinate the adjustment of operating parameters of on-site construction equipment. The quality tracking module communicates with the aforementioned modules and is used to aggregate data from the entire process. It uses a full-process quality tracking and evaluation algorithm to divide the project quality into inspection batches and assign weighted scores, and outputs quality evaluation reports and quality archives.

2. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The spatiotemporal fusion algorithm for multi-source heterogeneous data in the multi-source geological data acquisition and fusion module includes the following: Step 1: Data Acquisition and Preprocessing: Probe resistance and sidewall friction are obtained through static cone penetration tests; undrained shear strength, residual value, and shear modulus are obtained through vane shear tests; water content, liquid limit, plastic limit, compressibility coefficient, and consolidation coefficient are obtained through laboratory tests. The criteria remove outliers from all data; Step 2: Spatial Interpolation and Geological Profile Generation: Spatial interpolation of discrete borehole data is performed using ordinary kriging interpolation. A semi-variogram is established to describe the change of spatial correlation of data with distance. A spherical model is fitted to reflect sampling error and micro-variation. The weighting coefficients are solved using the kriging equations to obtain the interpolation results and generate a continuous geological profile. Step 3: Confidence-weighted fusion of multi-source data: Assign weights to static cone penetration tests, vane shear tests, laboratory tests, and geophysical exploration, and calculate the fusion value; Step 4: Disturbance Sensitivity Zoning: Calculate the disturbance sensitivity index of each area and divide the site into high-sensitivity, medium-sensitivity and low-sensitivity zones.

3. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The intelligent optimization algorithm for disturbance rejection processing schemes in the disturbance rejection intelligent design module includes the following: Step 1: Establish a candidate treatment scheme library, including one or more combinations of the following: bagged sand well surcharge preloading method, cement mixing pile composite foundation, replacement method, solid waste-based green gel material solidification, drainage sand cushion dynamic compaction method, and lightweight solidified soil filling. Step 2: Multi-objective optimization evaluation: Establishing the evaluation function ,in , These are the normalized project cost and construction period, respectively. To ensure a safety factor against disturbances and a target value ≥ 1.3, For the green index; weight ; Step 3: Genetic Algorithm Optimization: Encode the combination schemes into chromosomes and establish a fitness function. Evaluation value The smaller the value, the greater the fitness. A roulette wheel selection method is used, and crossover is performed using a single-point crossover with a high probability of crossover. The mutation is uniform and the mutation probability is... The iteration termination condition is the maximum number of iterations. Or fitness change <10 -6 Output the optimal combination of solutions. .

4. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The construction disturbance impact prediction algorithm in the anti-disturbance intelligent design optimization module includes the following: Step 1: Calculation of the disturbance range based on the theory of circular hole expansion: Treat the pile or sand well as a problem of cylindrical hole expansion, and define the radius of the failure zone. radius of yield zone ,in, The initial radius of the pile or well. For the undrained shear strength of the soil, This refers to the lateral pressure of the pile. For initial stress, The internal friction angle of the soil. This is the sine value of the internal friction angle; Step 2, Finite Element Disturbance Simulation: Establish a two-dimensional axisymmetric finite element model, and use the modified Cambridge model as the constitutive relation to simulate the pile penetration process; Step 3: Determine the safe construction distance, minimum safe distance Output the disturbance influence radius and safety distance diagram for each processing method.

5. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The multi-parameter dynamic threshold control algorithm in the real-time monitoring module for the construction process includes the following: Step 1: Set multi-level early warning thresholds; Step 2: Real-time data acquisition and Kalman filtering for noise reduction; Step 3, Dynamic Rate Control Decision: Calculate the current settlement rate. If the current settlement rate shows an upward trend for three consecutive observation periods and exceeds the yellow threshold, automatically calculate the maximum allowable filling rate and send a deceleration command. Step 4: Automatic generation of construction logs: Every 24 hours, a daily construction quality report is automatically generated, which includes time history curves of various parameters, number of times exceeding limits and handling records.

6. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The full-wavelength elastic wave inversion imaging algorithm in the full-wavelength intelligent detection module includes the following: Step 1: Elastic wave data acquisition: Excitation is performed using a hammer impact combined with a swept-sine frequency signal, with a frequency range of 1kHz to 50kHz. The signal is received by a 24-channel high-density detector array with a detector spacing of 0.15m, acquiring full-field signals including longitudinal waves (P-waves), transverse waves (S-waves), and surface waves (R-waves). Step 2, Dispersion Curve Extraction: Perform dispersion curve extraction on the surface wave signal. The fundamental Rayleigh wave dispersion curve is extracted by transformation, and different modes are separated by high-resolution Radon transform. Step 3, Tomographic Inversion Imaging: Establish an initial velocity model. In the iterative inversion, the forward modeling calculation uses the fast multipole boundary element method to calculate the theoretical travel time, and the wave propagation time is obtained from the forward modeling calculation. Step 4: Defect Identification and Quantification Determination of pile integrity: If the area of ​​the abnormal wave velocity zone is less than 5%, it is classified as Class I pile with good integrity; if it is 5% to 15%, it is classified as Class II pile with minor defects; and if it is greater than 15%, it is classified as Class III pile with serious defects. Strength assessment of solidified soil: Elastic modulus ≥ 15 MPa is qualified, 10-15 MPa is barely qualified, and < 10 MPa is unqualified.

7. The soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The multi-source fusion risk warning algorithm in the multi-parameter early warning and AI decision-making module includes the following: Step 1: Feature Vector Construction: Constructing Real-Time Feature Vectors ,in These are the current filling rate, daily settlement, and cumulative settlement, respectively. and These are the pore water pressure ratio and the lateral displacement rate, respectively. These are compaction degree, moisture content deviation, and elastic modulus of the solidified soil, respectively. These are the time since the last warning, the rate of change of construction load, the change of groundwater level, and temperature, respectively. Step 2, LSTM Time Series Prediction: A two-layer LSTM network is used to predict the subsidence trend over the next 72 hours. The first LSTM layer has 128 hidden units with dropout=0.2, and the second LSTM layer has 64 hidden units with dropout=0.

2. The input is the X sequence of the past 72 hours with a window length of 144 and a sampling interval of 30 minutes. The loss function is... ,in For the first Predicted settlement values ​​for each sample For the first The actual settlement value of each sample The total number of training samples, The L2 regularization coefficient is... For all trainable parameters of the LSTM network, The square of the L2 norm of the parameter; Step 3: XGBoost Risk Classification: Concatenate the LSTM predictions with the current feature vector and input the concatenation into the XGBoost classifier. The classifier uses 500 trees, a maximum depth of 6, a learning rate of 0.05, and classifies the target values ​​as {Safe, Attention, Warning, Danger}. The output is the risk level. ; Step 4, Early Warning Response: At T+0s, XGBoost outputs the risk level R. If R≥2 at T+5s, an early warning is triggered. At T+10s, a response plan is matched according to the risk level. Step 5: Feedback on the handling effect: Record the handling measures and subsequent monitoring data for each warning, and update the XGBoost model weights monthly.

8. A soft soil road anti-disturbance construction quality control system according to claim 7, characterized in that: The fourth step, which involves matching the emergency response plan according to the risk level, includes: for risk level R=1, the filling rate should be reduced by 20% and monitoring should be intensified to once every 15 minutes; for risk level R=2, the filling rate should be reduced by 50% immediately and the supervising engineer should be notified; for risk level R=3, work should be stopped immediately and an early warning should be issued simultaneously through the APP, SMS, and sound and light channels at T+30s.

9. A soft soil road anti-disturbance construction quality control system according to claim 1, characterized in that: The end-to-end quality tracking and evaluation algorithm in the quality tracking module includes the following: Step 1: Divide the project into sub-items, sub-sections, and unit projects according to the highway engineering quality inspection and evaluation standards. Each inspection batch includes the main control items of compaction degree and bearing capacity, as well as general items of flatness and width. Step 2: Calculate the quality score using the weighted scoring method; Step 3: Each inspection batch is automatically associated with the inspection time, GPS coordinates of the inspection location, inspection value, inspection equipment and operator to generate an unalterable quality file.

10. A method for controlling the construction quality of soft soil roads to resist disturbance, applied to the construction quality control system for soft soil roads to resist disturbance as described in any one of claims 1 to 9, characterized in that, Includes the following steps: Step S1, Multi-source geological data acquisition and fusion: Multi-source heterogeneous geological data are acquired through static cone penetration testing, vane shear testing, indoor geotechnical tests, and geophysical exploration. Criterion outlier removal, Kriging interpolation spatial interpolation, confidence-weighted fusion, and disturbance-sensitive zoning are used to generate a three-dimensional geological digital model; Step S2, Anti-disturbance intelligent design optimization: Based on the three-dimensional geological digital model, a candidate library of treatment schemes is established. Multi-objective optimization is performed through genetic algorithm to find the optimal combination of treatment schemes. The disturbance influence range and safe construction distance of each construction method are predicted by the circular hole expansion theory and the modified Cambridge model finite element simulation. Step S3: Real-time monitoring of the construction process: Real-time collection of filling rate, compaction degree, settlement, pore water pressure and lateral displacement parameters, after Kalman filtering to remove noise, and dynamic comparison with multi-level early warning thresholds. Based on the settlement rate trend, the maximum allowable filling rate is automatically calculated and a speed-down or work stoppage command is sent. At the same time, a daily construction quality report is automatically generated every 24 hours. Step S4, Full-field Intelligent Detection: A full-field elastic wave is excited using a hammer impact combined with a swept-sine frequency signal. The dispersion curve is extracted by transformation, and the pile foundation integrity and solidified soil strength are output as millimeter-level detection results and defect quantitative assessment by fast multipole boundary element method tomographic inversion imaging. Step S5, Multi-parameter Early Warning and AI Decision-Making: Construct a multi-dimensional real-time feature vector, predict the settlement trend in the next 72 hours through a two-layer LSTM network, concatenate the predicted value with the feature vector and input it into the XGBoost classifier for risk classification, complete the early warning response within 30 seconds and automatically generate disposal suggestions, and update the model weights monthly to achieve self-evolution. Step S6, Quality Tracking and Record Generation: Perform weighted scoring quality assessment by inspection batch and generate quality records.

Images