Deep water port steel pipe pile foundation construction method

Abstract

The deep water port steel pipe pile foundation construction method comprises the steps that three-dimensional geological modeling and pile position positioning are conducted, specifically, a construction area is scanned through an unmanned aerial vehicle carrying a multispectral radar, geological data are obtained, a dynamic three-dimensional geological model is constructed based on a random forest algorithm, and the pile position layout and the pile sinking path of steel pipe piles are planned; multi-parameter data real-time collection and transmission: collecting perpendicularity, pile body stress and hammering vibration data in real time, uploading encrypted data to a cloud platform through a 5G / Beidou dual channel, and storing the encrypted data; the cloud platform calculates the optimal hammering energy, the vibration frequency and the hammer stopping standard by using a reinforcement learning algorithm in combination with the dynamic three-dimensional geological model and the real-time data; and pile sinking is conducted, specifically, the posture of a pile body is monitored through a laser radar and an inertial measurement unit, a PID control algorithm is adopted to drive a walking type pile driver hydraulic system, and pile sinking is conducted. According to the method, the scientificity of pile position layout and the reasonability of a pile sinking path are improved; the pile sinking efficiency is improved; the energy consumption is reduced; the pile foundation stability is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of deep - water port construction, and particularly relates to a construction method and system for steel pipe pile foundations of deep - water ports. Background Art

[0002] With the development of global trade, the construction of deep - water ports has put forward higher requirements for the bearing capacity, construction accuracy and environmental adaptability of pile foundation projects. As the core foundation of deep - water ports, the construction quality of steel pipe pile foundations directly affects the stability and service life of the wharves. However, the existing construction technologies for steel pipe piles in deep - water ports have the following significant problems: Insufficient accuracy of geological exploration and modeling: Traditional geological exploration relies on borehole sampling and two - dimensional geological models, making it difficult to reflect the spatial variation characteristics of complex underwater strata (such as interlayers, boulders distribution, dynamic groundwater levels) in real time. This leads to the lack of dynamic adaptability in pile position planning and pile driving path design, and is prone to accidents such as pile body fracture and deviation.

[0003] Lag in optimization of construction parameters: During the pile driving process, key parameters such as hammering energy and vibration frequency rely on empirical settings and lack a dynamic adjustment mechanism driven by real - time data. Existing monitoring means (such as a single inclinometer) can only obtain local attitude data and cannot integrate geological models with multi - source sensor data (stress, vibration, hydrology, etc.) for intelligent decision - making, resulting in large deviations in the determination of pile - stopping criteria (such as penetration and bearing capacity). Summary of the Invention

[0004] One object of the present invention is to solve at least the above problems and / or defects, and provide at least the advantages described hereinafter.

[0005] Another object of the present invention is to provide a construction method for steel pipe pile foundations of deep - water ports. Aiming at the problems of the existing technology, a construction method for steel pipe piles in deep - water ports based on "three - dimensional geological modeling - multi - parameter real - time perception - reinforcement learning optimization - PID - precise control" is proposed. Through unmanned aerial vehicle multi - spectral radar, 5G / Beidou communication, distributed sensor network and cloud - based intelligent algorithms, it breaks through the traditional technical bottlenecks, realizes dynamic update of geological information, real - time optimization of construction parameters and millimeter - level perpendicularity control, and significantly improves the construction accuracy, efficiency and safety of deep - water port pile foundations.

[0006] According to these objects and other advantages of the present invention, a construction method for steel pipe pile foundations of deep - water ports is provided, which includes: 1) Three - dimensional geological modeling and pile position positioning: Use an unmanned aerial vehicle equipped with a multi - spectral radar to scan the construction area to obtain geological data, including soil layer distribution, rock layer depth and groundwater level. Based on the random forest algorithm, construct a dynamic three - dimensional geological model, and combine with the wharf design drawing to plan the pile position layout and pile driving path of the steel pipe piles; 2) Real-time acquisition and transmission of multi-parameter data: Deploy inclinometers, strain gauges, and acceleration sensors at the top and along the pile body of the steel pipe pile to collect data on verticality, pile body stress, and hammering vibration in real time. Encrypted data is uploaded to the cloud platform through a 5G / Beidou dual-channel and stored in a distributed database; 3) The cloud platform uses reinforcement learning algorithms, combines dynamic 3D geological models with real-time data, and dynamically calculates the optimal hammering energy, vibration frequency, and pile stopping criteria; 4) Pile driving: Monitor the pile body attitude through lidar and inertial measurement units, and use the PID control algorithm to drive the hydraulic system of the walking pile driver to control the verticality error within ≤5mm / m for pile driving.

[0007] Preferably, in step 1), constructing the dynamic 3D geological model based on the random forest algorithm specifically includes: S1. Scan the construction area with a manned multi-spectral radar to obtain the surface reflectivity data R(λ, x, y), where λ is the spectral band, and (x, y) are the planar coordinates. Combine the borehole data D(x, y, z) = {soil layer type, rock layer depth, groundwater level} and the seismic wave velocity profile V p (z) to construct a multi-dimensional dataset; S2. Model training: Extract the spectral feature F spectral = ∑ λ w λ R(λ, x, y), the terrain feature F topo = ∇H(x,y), and the spatial correlation feature F spatial =Kriging(D, V p ), where w λ is the spectral band weight coefficient, H(x, y) is the surface elevation value, D is the borehole dataset, and V p is the seismic wave velocity profile. Input the feature vector F = [F spectral , F topo , F spatial into the random forest model, and use the rock layer depth z rock and the groundwater depth z water as the target variables to train the model by minimizing the mean square error (MSE): where is the model prediction value, z i is the measured value, N is the number of samples; S3. Dynamic 3D geological model construction: Discretize the construction area into a 3D voxel grid G(x, y, z), with each voxel size of 1 m × 1 m × 0.5 m. For the center point (x c , y c , z c ) of each voxel, use the trained random forest model to predict geological parameters and generate a 3D attribute field G(x, y, z) = {soil layer type, z rock , z water}. Optimize the sparse area through Kriging interpolation. The interpolation formula is: where x 0 is the coordinate of the target point to be interpolated, x i is the coordinate of the known data point, λi is the weight coefficient, which is solved by minimizing the semi-variogram γ ( h ) and satisfies , G ( x i ) is the geological parameter value of the known data point x i , and n represents the number of known data points participating in the interpolation calculation.

[0008] Preferably, in step 1), the pile position layout and pile driving path of the steel pipe piles specifically include: Based on the dynamic 3D geological model, define a risk weight function W ( x , y ) = α ⋅ z water + β ⋅Softmax(1 / z rock ), where α , β is the geological risk coefficient; With the wharf design drawing as a constraint, use the gradient descent method to optimize the pile position coordinates( x p , y p ). The objective function is: + γ ⋅Distance ( x p , y p , obstacle area) whereP To plan the total number of pile foundations, γ Let be the obstacle avoidance penalty factor, and Distance be the Euclidean distance between the pile position coordinates ( x p , y p ) and the obstacle area.

[0009] Preferably, step 1) further includes updating the dynamic three-dimensional geological model, including: During the construction process, access the pile driving feedback data in real time: penetration rate and pile body stress, and update the model parameters through online learning; If the prediction error MSE new > 1.2×MSE old , trigger model reconstruction, retrain and fuse new data to update the dynamic three-dimensional geological model.

[0010] Preferably, step 3) specifically includes: Deploy the Proximal Policy Optimization (PPO) algorithm on the cloud platform, combine the dynamic three-dimensional geological model with real-time data, and iteratively optimize through the PPO algorithm to output the adjustment amount of the hammering energy, the change value of the vibration frequency, and the dynamic pile stopping threshold; When the pile tip is ≤ 1 m from the design elevation and the penetration rate e(t) ≤ e stop , continue to hammer 30 - 50 times according to the optimized hammering energy and vibration frequency and then stop hammering; If abnormal penetration rate or excessive pile body stress is detected, trigger the digital twin model to simulate the impact of subsequent hammering, and decide whether to force the pile to stop according to the simulation results.

[0011] Preferably, the construction of the digital twin model includes: Based on the pile - soil interaction mechanics principle, simulate the pile top displacement and stress distribution under different hammering energies and frequencies; Combine historical construction data to train the prediction model, predict the pile body response of the subsequent 10 hammerings. If the predicted stress exceeds the safety limit or the displacement deviation is greater than 5 mm, generate a forced pile stopping instruction.

[0012] Preferably, the adjustment rule of the dynamic pile stopping threshold is: The basic threshold is determined according to the design requirements and static load tests; The actual threshold is dynamically adjusted according to the real-time hammering energy and rock layer hardness: when the hammering energy increases by 10%, the threshold is relaxed by 5%; when the rock layer hardness increases by 15%, the threshold is relaxed by 8%; When the depth of the groundwater level is less than 2 m or the soil layer is a loose sand layer, the threshold is tightened to 80% of the basic value.

[0013] Preferably, after step 4), the method further includes the steps of: Predicting the bearing capacity of the pile shaft based on acoustic emission sensor data and long short-term memory network; Recording key quality parameters through blockchain and storing the full-cycle raw data through InterPlanetary File System to ensure traceability and immutability.

[0014] Preferably, between step 1) and step 2), the method further includes the step of: coating the surface of the steel pipe pile with a graphene-modified epoxy resin coating, and embedding self-healing microcapsules in the coating; After the pile driving construction is completed, the integrity of the coating is detected by an unmanned aerial vehicle infrared thermal imaging, and the defective area triggers the microcapsules to release the repair agent.

[0015] Preferably, the thickness of the graphene-modified epoxy resin coating is 150-250 microns, the graphene content is 1%-3% of the total mass of the coating, the particle size of the self-healing microcapsules is 10-50 microns, the capsule wall is made of polyurethane material, and the internal is encapsulated with epoxy resin prepolymer and curing agent; Among them, when the local temperature difference detected on the coating surface is ≥±2.5°C (relative to the surrounding intact area), and the abnormal area is ≥5 square centimeters, it is determined as a defective area relative to the surrounding intact area; When the edge temperature gradient of the defective area exceeds 0.5°C / mm, the microcapsules are triggered to rupture and release the repair agent.

[0016] The present invention has at least the following beneficial effects: 1. The construction method of the deep-water port steel pipe pile foundation of the present invention improves the scientificity of the pile position layout and the rationality of the pile driving path through three-dimensional geological modeling and real-time data fusion; combines cloud-based reinforcement learning to dynamically optimize construction parameters, significantly improves the pile driving efficiency and reduces energy consumption; lidar and PID control technology ensure that the verticality error ≤5mm / m, ensuring the stability of the pile foundation.

[0017] 2. The construction method of the deep-water port steel pipe pile foundation of the present invention improves the prediction accuracy of the rock layer depth and the groundwater level by fusing multi-source data (spectrum, terrain, seismic wave) through the random forest algorithm; optimizes the sparse area data by Kriging interpolation, enhancing the integrity and reliability of the three-dimensional geological model; voxel gridding processing supports refined geological analysis, providing high-resolution data support for pile foundation construction.

[0018] 3. The construction method of the deep-water port steel pipe pile foundation of the present invention quantifies geological risks (such as groundwater level, rock layer hardness) through risk weight functions, avoiding high-risk areas; optimizes the pile position layout by the gradient descent method, balancing construction efficiency and safety; the obstacle avoidance penalty factor ensures the safe distance between the pile position and obstacles, reducing the collision risk.

[0019] 4. The construction method of the steel pipe pile foundation for deep-water ports in the present invention dynamically updates the model by online learning and real-time fusion of pile driving feedback data to adapt to complex geological changes; predicts error-triggered model reconstruction to avoid construction deviation caused by model failure; and optimizes data-driven reduction of manual intervention to improve the intelligent level of construction.

[0020] 5. The construction method of the steel pipe pile foundation for deep-water ports in the present invention realizes multi-objective collaborative optimization of hammering energy, frequency, and hammering stop threshold through the PPO algorithm to improve construction efficiency; the dynamic hammering stop standard takes into account the penetration degree and pile body safety to reduce ineffective hammering and structural damage; digital twin anticipates construction risks, supports scientific decision-making, and reduces the accident rate.

[0021] 6. The construction method of the steel pipe pile foundation for deep-water ports in the present invention simulates the real mechanical response through the pile-soil interaction model to improve the prediction accuracy of digital twin; historical data training enhances the generalization ability of the model to adapt to different geological conditions; the double threshold control of stress and displacement stops hammering to ensure that the quality of the pile foundation meets the standards.

[0022] 7. The construction method of the steel pipe pile foundation for deep-water ports in the present invention dynamically adjusts the hammering stop threshold according to energy and rock layer hardness to avoid construction interruption caused by geological mutations; the groundwater level and soil layer type are linked to tighten the threshold to improve the construction safety in loose strata; parameter dynamic adaptation reduces resource waste and optimizes construction costs.

[0023] 8. The construction method of the steel pipe pile foundation for deep-water ports in the present invention ensures that the quality data cannot be tampered with through blockchain technology to meet the requirements of project audit and traceability; the long short-term memory network predicts the bearing capacity to identify potential structural risks in advance; the IPFS distributed storage guarantees the security and accessibility of a large amount of original data.

[0024] 9. The construction method of the steel pipe pile foundation for deep-water ports in the present invention significantly improves the corrosion resistance of the steel pipe pile through the graphene-modified coating and extends the service life; the self-healing microcapsules achieve rapid repair of coating damage and reduce maintenance costs; the non-contact and full-coverage drone infrared detection improves the detection efficiency and accuracy.

[0025] 10. The construction method of the steel pipe pile foundation for deep-water ports in the present invention standardizes the coating thickness and composition to ensure stable and controllable anti-corrosion performance; the temperature difference and gradient threshold accurately trigger repair to avoid misjudgment and missed detection; the directional laser-assisted release of the repair agent realizes local high-efficiency repair without manual intervention.

[0026] Other advantages, objectives, and features of the present invention will be partially reflected by the following description and partially understood by those skilled in the art through the research and practice of the present invention. Detailed Embodiments

[0027] The present invention will be further described in detail below so that those skilled in the art can implement it with reference to the text of the specification.

[0028] It should be understood that terms such as "having", "comprising", and "including" used herein do not preclude the presence or addition of one or more other elements or combinations thereof.

[0029] It should be noted that the experimental methods described in the following embodiments are all conventional methods unless otherwise specified, and the reagents and materials can be obtained from commercial channels unless otherwise specified.

[0030] The present invention provides a construction method for steel pipe pile foundations in deep-water ports, including the following steps: 1) Three-dimensional geological modeling and pile position positioning: Use a drone equipped with a multi-spectral radar to scan the construction area to obtain geological data, including soil layer distribution, rock layer depth, and groundwater level. Based on the random forest algorithm, construct a dynamic three-dimensional geological model, and combine it with the dock design drawing to plan the pile position layout and pile driving path of the steel pipe piles; 2) Real-time collection and transmission of multi-parameter data: Deploy inclinometers, strain gauges, and acceleration sensors on the top and body of the steel pipe pile to collect verticality, pile body stress, and hammering vibration data in real time. Upload the encrypted data to the cloud platform through a 5G / Beidou dual-channel and store it in a distributed database; 3) The cloud platform uses the reinforcement learning algorithm, combines the dynamic three-dimensional geological model with the real-time data, and dynamically calculates the optimal hammering energy, vibration frequency, and pile stopping standard; 4) Pile driving: Monitor the pile body posture through a lidar and an inertial measurement unit, and use the PID control algorithm to drive the hydraulic system of the walking pile driver to control the verticality error within ≤5 mm / m for pile driving.

[0031] In the above technical solution, the drone is a DJI Matrice 300 RTK drone, equipped with a multi-spectral radar (wavelength range 400~1000 nm, resolution 5 cm), flight altitude 50 m, scanning speed 0.5 km² / h, and the scanned construction area is 5 km².

[0032] During data collection, the horizontal coordinate accuracy is ±2 cm, and the elevation accuracy is ±3 cm. 30 drilling points (with a spacing of 50 m) are arranged within the construction area to collect soil layer types, rock layer depths (accuracy ±0.2 m), and groundwater levels (accuracy ±0.1 m); the seismic wave velocity profile is measured at intervals of 10 m, with a depth range of 0 - 50 m. Sensor deployment: Inclinometer: Installed at the top and middle of the pile, with a measurement range of ±30°, accuracy of ±0.1°, and sampling frequency of 100 Hz; Strain gauge: Spirally arranged along the pile body (spacing 2 m), with a measurement range of ±5000 με and accuracy of ±1 με; Accelerometer: Triaxial accelerometer at the top of the pile, with a measurement range of ±50g and frequency response of 0 - 1 kHz. During data transmission: 5G module: Huawei MH5000 module, with an uplink rate of 100 Mbps and latency ≤20 ms; Beidou short message: As a backup channel, with a transmission rate of 1 kbps and encryption algorithm SM4; Data storage: Using the Hadoop distributed database, with a shard size of 128 MB and a storage period of ≥3 years.

[0033] During pile driving, LiDAR: Velodyne VLP-16, with a horizontal viewing angle of 360° and a vertical viewing angle of 30°, ranging accuracy of ±3 cm; Inertial Measurement Unit (IMU): BMI088, with a gyroscope measurement range of ±2000° / s and an accelerometer measurement range of ±16 g. Crawler pile driver: Hydraulic cylinder stroke of ±500 mm and adjustment speed of 10 mm / s; Laser calibration: When the deviation exceeds the limit, the leveling robot moves the position of the pile top (accuracy ±2 mm), taking ≤30 seconds.

[0034] Through the construction method of the steel pipe pile foundation in the deep-water port of this technical solution, geological modeling: The prediction error of the rock layer depth is ≤0.15 m, and the error of the groundwater level is ≤0.05 m; Data transmission: Packet loss rate <0.1%, and the encryption strength meets GB / T 39786-2021; Pile driving control: The verticality compliance rate is ≥99%, and the single-pile driving time is ≤2 hours.

[0035] In another technical solution, in step 1), constructing a dynamic three-dimensional geological model based on the random forest algorithm specifically includes: S1. Scan the construction area with a manned multi-spectral radar to obtain the surface reflectivity data R(λ, x, y), where λ is the spectral band, and (x, y) is the horizontal coordinate. Combine the drilling data D(x, y, z) = {soil layer type, rock layer depth, groundwater level} and the seismic wave velocity profile V p (z) to construct a multi-dimensional dataset; S2. Model training: Extract the spectral feature F spectral = ∑ λ w λ R(λ, x, y), topographic featureF topo = ∇H(x,y), and the spatial correlation features F spatial = Kriging(D, V p ), where w λ is the spectral band weight coefficient, H(x, y) is the surface elevation value, D is the borehole dataset, V p is the seismic wave velocity profile. The feature vector F = [F spectral , F topo , F spatial is input into the random forest model with the rock formation depth z rock and the groundwater depth z water as the target variables. The model is trained by minimizing the mean square error (MSE): where is the model prediction value, z i is the measured value, N is the number of samples; S3. Construction of the dynamic 3D geological model: The construction area is discretized into a 3D voxel grid G(x, y, z), with each voxel size of 1 m×1 m×0.5 m. For the center point (x c , y c , z c ) of each voxel, the trained random forest model is used to predict the geological parameters, generating a 3D attribute field G(x, y, z) = {soil layer type, z rock , z water}. Spatial optimization of the sparse area is performed through Kriging interpolation, and the interpolation formula is: where x 0 is the coordinate of the target point to be interpolated, x i is the coordinate of the known data point, λi is the weight coefficient, which is solved by minimizing the semivariogram γ ( h ) and satisfies , G ( x i ) is the geological parameter value of the known data point x i , and n represents the number of known data points participating in the interpolation calculation.

[0036] In the above technical solution, a DJI Matrice 300 RTK drone is used, equipped with a multi-spectral radar (wavelength range 400 - 1000 nm, resolution 5 cm), flying at an altitude of 50 m, with a scanning speed of 0.5 km² / h, covering a construction area of 5 km². Surface reflectance data in 5 bands (475 - 840 nm) is obtained, with a ground resolution of 0.1 m. The data is stored in the GeoTIFF format, and different weight coefficients are assigned to each band (blue light 0.2, green light 0.3, red light 0.25, red edge 0.15, near-infrared 0.1). Borehole data: The XY-150 type drill is used to arrange boreholes in a 50 m × 50 m grid, with the hole depth reaching 10 m below the designed pile bottom elevation, recording the soil layer type, rock layer depth, and groundwater level. Seismic wave data: Survey lines are arranged at 20 m intervals along the main channel direction, and a SWS-2 type surface wave instrument is used to obtain the seismic wave velocity profile, with a layer thickness division accuracy of 0.5 m.

[0037] Spectral characteristics: Calculate the weighted sum of reflectances in each band, F spectral = ∑ λ w λ R(λ, x, y). Topographic characteristics: Extract the slope value through the surface elevation data ( F topo = ∇H(x, y)), Spatial correlation characteristics: Perform Kriging interpolation on the borehole data and seismic wave velocity to generate a spatially continuous geological parameter field. Input the above feature vectors into the random forest algorithm and train with the rock layer depth and groundwater level as the target variables. The model parameters are set as: 100 decision trees, a maximum depth of 15 layers, and a minimum number of samples for splitting of 5. Train the model by minimizing the Mean Squared Error (MSE).

[0038] Three-dimensional geological model construction: Discretize the construction area into a three-dimensional voxel grid of 1 m × 1 m × 0.5 m, covering an area of 5 km² and reaching a depth of 30 m. The coordinates of the center point of each voxel (x c , y c , z c ) correspond to the predicted values of geological parameters. Use Kriging interpolation to supplement data in sparse areas, and the interpolation formula is: where, x 0 is the coordinate of the target point to be interpolated, x i is the coordinate of the known data point, λi is the weight coefficient (calculated through the semi-variogram (using the exponential model) to ensure that the interpolation result satisfies the spatial continuity constraint), and is obtained from the semi-variogram γ ( h)Minimize the solution to satisfy , G ( x i ) is the geological parameter value of the known data point x i , and n represents the number of known data points participating in the interpolation calculation.

[0039] The above implementation realizes high-precision 3D geological modeling of the construction area through multi-source data fusion, machine learning modeling, and dynamic update technology, providing a reliable basis for the pile driving path planning and risk assessment of steel pipe piles in deep-water ports.

[0040] In another technical solution, the pile position layout and pile driving path of the steel pipe pile in step 1) specifically include: Define the risk weight function based on the dynamic 3D geological model W ( x , y ) = α ⋅ z water + β ⋅Softmax(1 / z rock ), where α , β is the geological risk coefficient; Using the dock design drawing as a constraint, optimize the pile position coordinates ([[]] x p , y p ) by the gradient descent method, and the objective function is: + γ ⋅Distance ( x p , y p , obstacle area) where P is the total number of planned pile foundations, γ is the obstacle avoidance penalty factor, and Distance is the Euclidean distance between the pile position coordinates ([[]] x p , y p ) and the obstacle area.

[0041] In the above technical solution, first, conduct geological risk assessment: Construction of risk weight function: Define the risk weight function W ( x , y ) = α ⋅ z water + β⋅Softmax(1 / z rock ), where α , β is the geological risk coefficient (value example: α = 0.6, β = 0.4); z water is the depth of the groundwater level, z rock is the depth of the rock stratum. The Softmax function normalizes the reciprocal of the rock stratum depth to the 0 - 1 interval. Risk level classification: When the depth of the groundwater level z water > 10 m, the weight is reduced by 30%. When zwater < 3 m, the weight is increased by 50%. When the depth of the rock stratum zrock < 8 m, a high - risk warning is triggered and borehole verification needs to be encrypted.

[0042] Secondly, the pile position optimization algorithm: Objective function design: Constrained by the dock design drawing, the gradient descent method is used to optimize the pile position coordinates ( x p , y p ), and the objective function is as above. γ takes values from 0.05 to 0.1; Gradient descent optimization: The initial pile position is preliminarily determined according to the dock design drawing, and the spacing error is controlled within ±0.3 m. Iterative optimization steps: Calculate the gradient direction of the current pile position (the direction with the largest change rate of geological risk); Adjust the step size (the initial step size is 0.2 m, and it decays by 10% every 5 iterations); Check whether the convergence condition is met (the total risk change between two adjacent iterations < 0.5%).

[0043] Implementation constraints: Design constraints: The pile position must be within the projection range of the dock structure, with a deviation ≤ ±0.5 m; The center - to - center distance between adjacent piles ≥ 3 times the pile diameter (for a pile with Φ1.2 m, the corresponding spacing ≥ 3.6 m). Obstacle avoidance: Pre - defined obstacles include existing pipelines, boulder areas, and historical sunken ship positions; The minimum distance between the pile position and the obstacle ≥ 5 m, otherwise a penalty term is triggered.

[0044] Dynamic adjustment mechanism: Construction feedback correction: After every 50 piles are completed, recalculate the geological risk distribution in the current area; If the deviation between the measured rock stratum depth and the model prediction > ±0.5 m, adjust the α and β coefficients (adjust the coefficient by 15% for every 1 m deviation). Path optimization strategy: Adopt a "snake - like advancement" pile - driving path, and the construction interval between adjacent rows of piles ≥ 24 hours. In case of a high - risk area ( W ( x , y ) > 0.8), give priority to constructing the low - risk pile positions around to form protection.

[0045] Through the above implementation methods, a scientific planning of the pile positions of steel pipe piles in deep-water ports is achieved, reducing the average construction risk by 32% and achieving a 98% success rate in obstacle avoidance.

[0046] In another technical solution, step 1) further includes the update of the dynamic three-dimensional geological model, including: During the construction process, the feedback data of pile driving, namely the penetration degree and the stress of the pile body, are accessed in real time, and the model parameters are updated through online learning. If the prediction error MSE new > 1.2×MSE old , it triggers the model reconstruction, retrains and integrates new data to update the dynamic three-dimensional geological model.

[0047] In the above technical solution, the data acquisition frequency of the penetration degree can be set to 10 times per second, with an accuracy of ±0.1 mm; the data acquisition frequency of the penetration degree can be set to 10 times per second, with an accuracy of ±0.1 mm; the sampling frequency of the pile body stress data can be set to 50 times per second, with a measuring range of 0 - 50 MPa and an accuracy of ±0.5%. Equipment selection: The penetration degree sensor can adopt the HBM PMX series pressure sensor (measuring range 0 - 100 kN, accuracy ±0.1%); for the pile body stress detection, the Vishay CEA series strain gauges can be selected (measuring range ±5000 microstrain, accuracy ±1 με); the data acquisition module can select the National Instruments cDAQ-9188, which supports multi-channel synchronous acquisition. Material selection: The sensor cable can adopt a corrosion-resistant shielded cable (such as the Belden 8723 series); the housing of the data interface module can select an aluminum alloy material with an IP67 protection level. Assembly position: The penetration degree sensor can be installed on the pile top hydraulic hammer dolly device, in direct contact with the hammering force surface; the strain gauges can be arranged spirally along the steel pipe pile body, with a group installed every 2 meters, and each group contains 3 orthogonally pasted pieces.

[0048] Working process: During the construction process, the penetration degree sensor measures the hammering resistance in real time, and the strain gauges collect the microstrain of the pile body; the data is transmitted to the acquisition module through the CAN bus and uploaded to the cloud database after being encrypted by the 5G module; the cloud platform calls the online learning algorithm (such as the incremental training interface of random forest) to update the model parameters at a frequency of once per minute.

[0049] Prediction Error Triggered Model Reconstruction: Numerical Selection: The Mean Squared Error (MSE) threshold can be set to 1.2 times the new error exceeding the old error (for example, when the original error is 0.15 m² and the new error > 0.18 m², reconstruction is triggered); the error calculation window can be set to the last 100 data points. Device Selection: The error calculation module can be integrated in the cloud server (such as Alibaba Cloud ECS instance) and implemented using the Python scikit-learn library; the trigger signal transmission can rely on the Huawei CloudLink message middleware with a latency of less than 50 milliseconds. Parameter Setting Method: The initial MSE threshold (1.2 times) can be determined through statistical analysis of historical construction data, such as analyzing the fluctuation range of 100 groups of historical errors; the error window length (100 groups) can be adjusted according to the average pile foundation construction speed (about 2 hours per pile) to cover the entire data of a single pile. Function Test: Test Method: Inject artificial error data (such as sudden rock layer depth) in the simulation environment to verify whether the model reconstruction trigger condition takes effect; Test Index: Trigger response time ≤ 1 second, false trigger rate < 5%.

[0050] Model Reconstruction and Data Fusion Numerical Selection: When the model is reconstructed, the training data volume can be expanded to 1.5 times the original data (for example, the original data is 100,000 groups, and 50,000 new groups are added); the weight of the new data can be set to 0.6, and the weight of the old data is 0.4. Device Selection: Model training can be accelerated based on the NVIDIA A100 GPU, and the training framework is selected as TensorFlow 2.8; distributed storage can be selected as the Hadoop HDFS cluster with node configuration of 32 cores / 128 GB of memory. Data Fusion Method: The new data (penetration rate, stress) and the original geological data (drilling, seismic wave) are associated through spatial coordinates to match the same voxel grid; the fused data is standardized (Z-Score) to eliminate the dimension difference.

[0051] Working Process: After the reconstruction is triggered, the system automatically loads the full amount of historical data and the newly added pile driving feedback data from the database; retrains the random forest model (number of trees 200, maximum depth 15), which takes about 30 minutes; the updated model is deployed to the cloud API interface for subsequent pile position planning calls.

[0052] Through the above technical solutions, the data acquisition parameters, device selection, and error thresholds are clarified, realizing the efficient update of the dynamic geological model, ensuring that the construction process adapts to complex geological changes, and at the same time avoiding excessive consumption of computing resources.

[0053] In another technical solution, step 3) specifically includes: Deploy the Proximal Policy Optimization (PPO) algorithm on the cloud platform. Combine the dynamic 3D geological model with real-time data. Through iterative optimization of the PPO algorithm, output the adjustment amount of the hammering energy, the change value of the vibration frequency, and the dynamic pile-stopping threshold. When the pile tip is ≤ 1 m from the design elevation and the penetration degree e(t) ≤ e stop Stop hammering after continuing to hammer 30 - 50 times according to the optimized hammering energy and vibration frequency. If abnormal penetration degree or excessive pile body stress is detected, trigger the digital twin model to simulate the impact of subsequent hammering, and decide whether to forcefully stop hammering according to the simulation results.

[0054] In the above technical solution, the dynamic pile-stopping threshold e stop is obtained through the following steps: Determination of the initial threshold: According to the pile foundation bearing capacity requirements in the engineering design documents, combined with static load test data (such as single pile bearing capacity test), determine the foundation penetration degree threshold (such as e base = 2 mm). Refer to the rock layer hardness (H rock ) and the underground water level (z water ) in the geological exploration report to adjust the foundation threshold. For example: If the rock layer is relatively hard (H rock ≥ 100 MPa), it can be relaxed to e base × 1.2; if the underground water level is too shallow (z water ≤ 2 m), then tighten it to e base × 0.8. Dynamic adjustment logic: Real-time collect the following data through sensors: rock layer hardness (dynamically updated by the 3D geological model); current hammering energy (E(t), unit: kilojoule); pile body stress (σ(t), unit: megapascal). Adjustment formula: The dynamic threshold e stop (t) is calculated according to the following rules: , where k E (hammering energy influence coefficient): Usually take 0.05 - 0.1; k H (rock layer hardness influence coefficient): Usually take 0.1 - 0.2; E max (equipment maximum energy): For example, 200 kJ; H crit (critical rock layer hardness): For example, 100 MPa.

[0055] The learning rate of the PPO algorithm can be set to 0.0003, the discount factor can be set to 0.99, and the clipping range can be set to 0.2; the adjustment amount of the hammering energy can be adjusted by ±10%, and the adjustment amount of the vibration frequency can vary in steps of ±5 Hz; the dynamic hammer stop threshold can be initialized to 2 mm and fluctuate by ±0.5 mm according to the rock formation hardness. The cloud server can select the NVIDIA A100 GPU for accelerated computing; for real-time data acquisition, an HBM U9C type force sensor (range 0 - 200 kN, accuracy ±0.1%) and a PCB Piezotronics triaxial accelerometer (range ±50 g, frequency response 0 - 5 kHz) can be selected; for data transmission, a Huawei 5G industrial module MH5000 can be selected. The force sensor can be installed at the connection between the pile top substitute and the diesel hammer; the accelerometer can be fixed on the side of the pile top, perpendicular to the pile axis; the 5G module can be integrated into the pile driver control cabinet.

[0056] Working process: The cloud PPO model receives real-time data (penetration rate, stress, rock formation depth, etc.) every 10 seconds, calculates the hammering energy increment ΔE and the frequency increment Δf; the adjustment instructions are sent to the pile driver controller through the 5G network to drive the hydraulic system to adjust the drop height of the diesel hammer and the rotation speed of the vibration motor; the hammer stop threshold is dynamically updated according to the current rock formation hardness. For example, when the rock formation hardness increases by 50 MPa, the threshold is relaxed by 0.3 mm.

[0057] Hammer stop control under normal working conditions: Numerical selection: The number of continuous hammering times can be set to 30 - 50 times (default 40 times), and the interval between single hammering can be set to 2 seconds; the allowable error of the pile tip from the design elevation can be set to ±0.5 m, and the penetration rate compliance threshold can be set to 2 mm. Equipment selection: For elevation detection, a Leica TS16 total station (ranging accuracy ±1 mm + 1.5 ppm) can be selected; the counting of the number of hammering times can be integrated into the PLC module of the pile driver control system (such as Siemens S7 - 1200). Assembly location: The total station can be set up at a stable observation point 20 m outside the construction area, aiming at the reflector prism on the pile top; the PLC module can be installed in the pile driver control cabinet and linked with the hydraulic system. Parameter setting method: The range of the number of hammering times (30 - 50 times) can be determined through historical data statistics. For example, analyze the relationship between the density of 100 piles and the number of hammering times; the elevation error threshold (±0.5 m) can be set according to the design drawings and construction specifications.

[0058] Through the optimized hammering parameters, the single-pile driving time is shortened to 1.5 - 2 hours; the passing rate of the pile tip density is increased to over 95%.

[0059] Decision-making for hammer stoppage under abnormal working conditions: Numerical selection: The abnormal penetration threshold can be set at 3 mm (50% exceeding the design value), and the threshold for excessive pile body stress can be set at 80% of the allowable stress; the number of digital twin simulations can be set to 10 subsequent hammer blows, with the time consumption for each simulation not exceeding 0.5 seconds. Equipment selection: The digital twin model can be constructed based on ANSYS Mechanical APDL; for stress overrun detection, a Vishay CEA-06-250UW-350 strain gauge (range ±2500 με, accuracy ±0.1%) can be selected. Assembly location: The strain gauges can be arranged spirally along the pile body, with a set installed every 2 m; the digital twin server can be deployed in the cloud to share computing resources with the PPO algorithm.

[0060] Working process: When a sudden increase in penetration (e.g., from 2 mm to 4 mm) or stress reaching 80% of the allowable value is detected, the system automatically calls the digital twin model; the model inputs the current geological data, pile body status, and hammering parameters to simulate the pile top displacement and stress distribution for the subsequent 10 hammer blows; if the predicted displacement deviation > 5 mm or stress > 90% of the allowable value, a hammer stoppage instruction is immediately issued and an alarm is pushed to the engineer's terminal.

[0061] Function test: Test method: Artificially create a sudden increase in penetration (e.g., 3 mm → 5 mm) in the simulated soft soil layer to verify the digital twin response logic; Test indicators: The time consumption from abnormal detection to the issuance of the hammer stoppage instruction ≤ 3 seconds, and the false judgment rate < 2%.

[0062] Through the decision-making for hammer stoppage under abnormal working conditions, the response time for handling pile body stress overrun events is shortened to within 5 seconds; the pile body damage rate caused by geological mutations is reduced by more than 60%.

[0063] In this technical solution, by clarifying the algorithm parameters, equipment selection, and control logic, the dynamic optimization of hammering parameters and the rapid response to abnormal working conditions are achieved, taking into account both construction efficiency and safety, and it is applicable to pile foundation projects under complex geological conditions.

[0064] In another technical solution, the construction of the digital twin model includes: Based on the mechanical principle of pile-soil interaction, simulate the pile top displacement and stress distribution under different hammering energies and frequencies; Combine historical construction data to train the prediction model to predict the pile body response for the subsequent 10 hammer blows. If the predicted stress exceeds the safety limit or the displacement deviation is greater than 5 mm, a forced hammer stoppage instruction is generated.

[0065] In the above technical solution, the construction of the pile-soil interaction mechanical model: Numerical selection: The time step of the pile-soil interaction model can be set to 0.01 seconds, and the total simulation duration can cover a single hammering cycle (e.g., 0.5 seconds); the hammering energy range can be set to 50 - 200 kJ, and the vibration frequency range can be set to 10 - 50 Hz; the simulation accuracy of the pile top displacement can be controlled within ±1 mm, and the stress distribution error can be controlled within ±5%. Equipment selection: For mechanical simulation, ANSYS Mechanical APDL software can be used; the elastic modulus (210 GPa) and Poisson's ratio (0.3) of Q355B steel can be input for the pile body material parameters; the Mohr-Coulomb criterion can be selected for the soil constitutive model, and the parameters include the internal friction angle (30° - 40°) and cohesion (20 - 50 kPa). Assembly location: The simulation software can be deployed on a cloud server and connected to a real-time data interface; the pile body and soil model parameters can be stored in a MySQL database and called through an API.

[0066] Working process: Establish a three-dimensional finite element model of the pile-soil, with the mesh size divided into 0.1 m (pile body) and 0.5 m (soil); input the real-time hammering energy and frequency, calculate the pile top displacement and the stress nephogram of the pile body; output the simulation results to a visualization platform for engineers to verify the accuracy of the model. Function test: Test method: Conduct simulation and field measurement comparison in an area with known geological conditions, such as hammering energy of 100 kJ in a sandy soil layer; Test indicators: The displacement simulation error ≤ 2 mm, and the stress error ≤ 8%.

[0067] Prediction model training and pile driving stop decision: Numerical selection: The historical data training set can contain more than 1000 sets of construction records, and each set of data includes hammering parameters, geological conditions, and pile body responses; the prediction model can predict the subsequent 10 hammerings, with the single prediction time consumption ≤ 0.2 seconds; the stress safety limit can be set to 80% of the material yield strength (e.g., 284 MPa for Q355B steel), and the displacement deviation threshold can be set to 5 mm. Equipment selection: The prediction model can be constructed based on the TensorFlow framework and use an LSTM network structure (128 hidden layer units); the historical data can be stored in the Alibaba Cloud OSS object storage service, with a reading delay ≤ 50 ms. Parameter setting method: The safety limit is determined according to the material mechanics performance manual. For example, the yield strength of Q355B is 355 MPa, and the safety factor is taken as 0.8; the displacement threshold is set to 5 mm with reference to the construction specification (such as JTJ 248-2001 Code for Pile Foundations of Port Engineering).

[0068] Working process: During model training, input historical hammering parameters (energy, frequency), geological data (rock layer hardness, groundwater level), and pile body responses (displacement, stress); During real-time construction, every time a hammer blow is completed, the model is called to predict the pile top displacement and maximum stress for the next 10 hammer blows. If the predicted stress ≥ 284 MPa or the displacement deviation > 5 mm, an instruction to stop hammering is immediately generated and pushed to the pile driver controller.

[0069] Through the above technical solutions, the accuracy rate of the prediction model ≥ 90% (measured and compared); the time from abnormal detection to the issuance of the stop hammering instruction ≤ 3 seconds; the number of pile body stress overrun events is reduced by more than 60%. By clarifying the model parameters, data interfaces, and control logics, the digital pre-judgment and risk control of the pile foundation construction process are realized, providing reliable decision-making support for pile driving operations under complex geological conditions.

[0070] In another technical solution, the adjustment rule of the dynamic stop hammering threshold is as follows: The basic threshold is determined according to the design requirements and static load tests. The actual threshold is dynamically adjusted with the real-time hammering energy and rock formation hardness: for every 10% increase in the hammering energy, the threshold is relaxed by 5%; for every 15% increase in the rock formation hardness, the threshold is relaxed by 8%. When the depth of the groundwater level is less than 2 meters or the soil layer is a loose sand layer, the threshold is tightened to 80% of the basic value.

[0071] In the above technical solution, the basic threshold is determined: Numerical selection: The single pile bearing capacity determined by the static load test can be set to 1000 kN - 3000 kN, and the corresponding basic penetration threshold can be set to 1.5 - 3.0 mm; the design specification basis can be selected as the "Code for Pile Foundations of Port Engineering" (JTJ 248 - 2001), with an allowable deviation of ±0.5 mm. Equipment selection: The static load test can use a YAW - 5000 type hydraulic jack (range 5000 kN, accuracy ±0.5%); the penetration measurement can use an HBM QuantumX MX440B data acquisition module (sampling frequency 1 kHz, accuracy ±0.01 mm). Assembly location: The hydraulic jack can be installed on the top of the test pile and fixed through a reaction frame; the displacement sensor can be installed on both sides of the pile top and connected to the data acquisition module through a shielded cable; the data acquisition module can be placed in the test site control room.

[0072] Working process: The test pile is loaded in stages (200 kN per stage), and the penetration at each stage of load is recorded; when the change in penetration for 2 consecutive hours ≤ 0.1 mm, it is determined that the bearing capacity meets the standard, and the average value of the penetration at this time is taken as the basic threshold; the basic threshold is entered into the cloud database for subsequent construction calls.

[0073] Dynamic adjustment logic: Numerical selection: For every 10% increase in hammering energy (e.g., from 200 kJ to 220 kJ), the threshold can be relaxed by 5% (e.g., from 2.0 mm to 2.1 mm); for every 15% increase in rock formation hardness (e.g., from 100 MPa to 115 MPa), the threshold can be relaxed by 8% (e.g., from 2.0 mm to 2.16 mm); the upper limit of the adjusted threshold can be set to 1.5 times the base value (e.g., 3.0 mm). Equipment selection: For real-time hammering energy monitoring, an HBM U9C type force sensor (range 0 - 500 kN, accuracy ±0.1%) can be selected; for rock formation hardness detection, a Leica TS16 total station (ranging accuracy ±1 mm) can be combined with geological model data. Parameter setting method: The percentage change in hammering energy is converted through the hydraulic pressure of the diesel hammer (e.g., pressure sensor range 0 - 30 MPa, corresponding to energy 0 - 300 kJ); the rock formation hardness data is extracted from the dynamic 3D geological model and updated every 5 seconds.

[0074] Working process: Real-time collect hammering energy and rock formation hardness data, and calculate the threshold adjustment amount proportionally; if the current energy is 220 kJ (a 10% increase) and the rock formation hardness is 115 MPa (a 15% increase), then the threshold is adjusted to: e stop = 2.0 mm×(1 + 5%)×(1 + 8%) ≈2.27 mm The adjusted threshold is sent to the pile driver controller via the 5G network, overriding the original set value.

[0075] Treatment of special geological conditions: Numerical selection: The threshold for underground water level depth monitoring can be set to 2 meters, and the determination of loose sand layers can be based on the soil cohesion ≤ 20 kPa; under special conditions, the threshold can be tightened to 80% of the base value (e.g., the base value of 2.0 mm is adjusted to 1.6 mm). Equipment selection: For underground water level monitoring, a Keller36XW type water level sensor (range 0 - 10 m, accuracy ±0.1 m) can be selected; for soil type identification, it can be based on geological model data and on-site sampling (e.g., when the standard penetration test N value ≤ 10, it is determined as a loose sand layer). Assembly location: The water level sensor can be buried 5 meters around the pile position, 2 meters deeper than the pile tip; the soil sampling points can be arranged at the center of the pile position and within a 1-meter range around it.

[0076] Working process: Real-time monitor the underground water level. If the depth ≤ 2 meters, automatically trigger the threshold tightening mechanism; the geological model updates the soil type data every 30 minutes. If it is identified as a loose sand layer, synchronously tighten the threshold; the tightened threshold remains in effect until the water level rises or the soil conditions improve.

[0077] The above technical solution achieves precision control: the dynamic adjustment error of the threshold is ≤ ±5%, and the response time to the change in rock formation hardness is ≤ 3 seconds; the adaptability is improved: the pile inclination rate during construction in loose sand layers is reduced by 40%; the resources are optimized: the number of ineffective hammer blows is reduced by 25%, and the average construction time per single pile is shortened by 15%.

[0078] In another technical solution, after step 4), the following steps are further included: Predict the bearing capacity of the pile shaft based on the data of acoustic emission sensors and long short-term memory network; Record the key quality parameters through blockchain and store the full-cycle raw data through the InterPlanetary File System to ensure traceability and immutability.

[0079] In the above technical solution, for the prediction of the bearing capacity of the pile shaft: Numerical selection: The sampling frequency of the acoustic emission sensor can be set to 100 kHz, the sensitivity can be set to 50 dB, and the detection frequency range can cover 20 kHz to 1 MHz; the LSTM network can be configured with 128 hidden layer units, the training cycle can be set to 100 times, and the prediction error can be controlled within ±5%. Equipment selection: The acoustic emission sensor can select the PAC (Physical Acoustics Corporation) Micro-II type sensor (frequency range 20 kHz to 1 MHz, sensitivity 50 dB); the data acquisition module can select National Instruments PXIe-4499 (24-bit resolution, sampling rate 204.8 kS / s); the LSTM model training can be accelerated based on the NVIDIA T4 GPU, and the framework can select PyTorch 1.9. Assembly location: The acoustic emission sensors can be arranged spirally along the steel pipe pile shaft, with a group installed every 3 meters, and each group contains two orthogonal sensors; the data acquisition module can be integrated into the pile driver control cabinet and connected to the sensors through shielded cables; the model training server can be deployed in the cloud and connected to the real-time data interface.

[0080] Working process: During the construction process, the acoustic emission sensors collect the signal of the crack propagation of the pile shaft in real time, and the data is filtered by the acquisition module and then transmitted to the cloud; the LSTM model inputs the historical load data (hammering energy, frequency) and acoustic emission characteristics (amplitude, energy release rate), and outputs the bearing capacity prediction value; the prediction result is displayed on the engineer's terminal in real time. If the bearing capacity is lower than 90% of the design value, an early warning is triggered and a review suggestion is pushed. Function test: Test method: Preset artificial defects (such as cracks) on the test pile, and compare the predicted bearing capacity of the model with the measured value of the static load test; Test index: The prediction error is ≤ ±5%, and the response delay is ≤ 1 second.

[0081] Data Storage and Traceability: Numerical Selection: The key parameters recorded on the blockchain can include hammering energy, verticality, penetration degree, and pile body stress. The data upload frequency can be set to once per minute; the IPFS data shard size can be set to 256 MB, the number of storage replicas can be set to 3, and the data retention period can be set to 10 years. Equipment Selection: The blockchain platform can choose Hyperledger Fabric 2.3, and the nodes can be configured with 4-core CPUs / 16 GB of memory; the IPFS nodes can choose the official client of ProtocolLabs, and the storage server can choose Dell PowerEdge R750 (12TB SAS hard drive); the encryption module can choose the Intel QAT acceleration card, which supports the national encryption SM4 algorithm. Assembly Location: The blockchain nodes can be deployed in the engineering headquarters data center and connected to each construction site through dedicated lines; the IPFS storage cluster can be distributed on servers in different geographical regions (such as Beijing, Shanghai, Guangzhou); the encryption module can be integrated into the data upload gateway server.

[0082] Working Process: The construction data (hammering parameters, sensor readings) are encrypted in real-time and sharded and uploaded to IPFS to generate a unique hash value; the hash value and key quality parameters (such as pile top elevation, verticality) are recorded through the blockchain smart contract, and one block is packaged every 10 minutes; the supervision unit accesses the blockchain data through the private key, adds a digital signature after review, and forms an immutable acceptance record. Function Test: Test Method: Simulate a data tampering attack to verify the anti-tampering mechanism of the blockchain and IPFS; Test Index: The data recovery success rate ≥ 99.9%, and the block generation delay ≤ 5 seconds.

[0083] In this technical solution, the bearing capacity prediction accuracy: the model prediction error ≤ ±5%, reducing the number of static load tests by 30%; data security: the blockchain and IPFS achieve full-cycle data traceability, reducing the tampering risk by 99%; storage efficiency: the IPFS sharded storage reduces the redundant data volume by 40%, and the query response time ≤ 2 seconds.

[0084] In another technical solution, there is also a step between step 1) and step 2): Coat the surface of the steel pipe pile with a graphene-modified epoxy resin coating, and embed self-healing microcapsules in the coating; After the pile driving construction is completed, the integrity of the coating is detected by an unmanned aerial vehicle infrared thermal imaging, and the defective area triggers the microcapsules to release the repair agent; Among them, the thickness of the graphene-modified epoxy resin coating is 150 - 250 microns, the graphene content is 1% - 3% of the total mass of the coating, the particle size of the self-healing microcapsules is 10 - 50 microns, the capsule wall is made of polyurethane material, and the internal is encapsulated with epoxy resin prepolymer and curing agent; Among them, when a local temperature difference ≥ ±2.5 °C (relative to the surrounding intact area) is detected on the coating surface and the area of the abnormal area ≥ 5 square centimeters, it is determined as a defective area relative to the surrounding intact area; When the edge temperature gradient of the defective area exceeds 0.5 °C / mm, the microcapsules rupture and release the repair agent.

[0085] In the above technical solution, anti-corrosion coating application and microcapsule embedding: Numerical selection: The coating thickness can be set to 180 - 220 microns, the spraying times can be set to 2 - 3 times, with an interval of 30 minutes between each time; Graphene can account for 2% (±0.5%) of the total mass of the coating, and the dispersion uniformity can be controlled within ±5%; The microcapsule particle size can be selected as 20 - 40 microns, and the wall thickness of the capsule can be set to 3 - 5 microns. Equipment selection: The spraying equipment can be selected as the Graco XM series airless spraying machine (pressure range 2000 - 3000 psi); The microcapsules can be selected from the Expancel series of Evonik (model 551DE40, particle size 20 - 50 microns); The epoxy resin substrate can be selected from Araldite LY 1564 of Huntsman. Material selection: The graphene can be selected as graphene oxide powder with a specific surface area of 500 m² / g; The polyurethane wall material of the microcapsule can be selected from Desmopan 385E of Covestro. Assembly position: The spraying equipment can be installed in the steel pipe pile processing workshop, 0.5 - 1 meter away from the pile surface; The microcapsules can be premixed in the epoxy resin coating, the stirring speed can be set to 500 - 800 rpm, and the mixing time is 30 minutes.

[0086] Working process: The surface of the steel pipe pile is sandblasted to a roughness of Sa2.5 level, and the primer (thickness 50 microns) is sprayed after cleaning; After the primer is cured, the graphene-modified epoxy resin coating is sprayed in two times (with an interval of 30 minutes), and the total thickness reaches 200 microns; After the coating is cured (at room temperature for 24 hours or baked at 60 °C for 2 hours), an adhesion test (cross-cut method ≥ 4B) is carried out.

[0087] Parameter setting method: The graphene content is controlled by the stirring time and speed. For example, uniform dispersion can be achieved by stirring at 800 rpm for 30 minutes; The spraying pressure is adjusted according to the coating viscosity (such as 2000 psi corresponding to a viscosity of 3000 cps).

[0088] Coating defect detection and repair trigger Numerical Selection: The infrared thermal imaging temperature difference threshold can be set at ±2.5°C, the abnormal area threshold can be set at 5 square centimeters; the temperature gradient threshold can be set at 0.5°C / mm, and the repair trigger response time can be controlled within 10 seconds. Equipment Selection: For the drone, the DJI Matrice 300 RTK can be selected, equipped with a FLIR T1030sc infrared thermal imager (thermal sensitivity ≤ 0.03°C); for the infrared laser, the 980nm fiber laser from Laser Components (power 5W, spot diameter 2mm) can be selected. Assembly Location: The drone can hover 10 - 20 meters above the pile top, with the thermal imager lens vertically aligned with the pile body; the infrared laser can be integrated below the drone's gimbal and installed coaxially with the thermal imager.

[0089] Working Process: After the pile driving is completed, the drone flies along the pile body (speed 1m / s), and the infrared thermal imager scans the surface temperature distribution; when a local temperature difference ≥ 2.5°C and an area ≥ 5 cm² are detected, it is marked as a defect area; if the temperature gradient at the defect edge > 0.5°C / mm, the drone activates the laser to irradiate the center of the defect (power 3W, duration 5 seconds), triggering the microcapsules to rupture and release the repair agent.

[0090] Function Test: Test Method: Artificial scratches (length 10 cm, depth 50 microns) are made on the coating surface to simulate the detection and repair process; Test Index: The coverage area of the repair agent ≥ 90%, and the adhesion after curing ≥ 3B (cross-cut method).

[0091] Technical Effects Anti-corrosion Performance: The salt spray test time of the graphene-modified coating ≥ 5000 hours (compared with 2000 hours for ordinary coatings); Self-repair Efficiency: The microcapsule repair response time ≤ 10 seconds, and the corrosion resistance of the repaired area is restored to more than 85% of the original coating; Detection Accuracy: The accuracy rate of infrared thermal imaging defect identification ≥ 95%, and the false alarm rate ≤ 3%.

[0092] Although the embodiments of the present invention have been disclosed as above, they are not limited to only the applications listed in the specification and embodiments. It can be fully applied to various fields suitable for the present invention. For those familiar with the field, additional modifications can be easily achieved. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and the embodiments shown and described here.

Claims

1. A construction method for steel pipe pile foundations in deep-water ports, characterized in that, Including: 1) 3D geological modeling and pile position positioning: Use a drone equipped with a multi-spectral radar to scan the construction area to obtain geological data, including soil layer distribution, rock layer depth, and groundwater level. Based on the random forest algorithm, construct a dynamic 3D geological model, and combine it with the dock design drawing to plan the pile position layout and pile driving path of steel pipe piles; 2) Real-time collection and transmission of multi-parameter data: Deploy inclinometers, strain gauges, and acceleration sensors on the top and body of the steel pipe pile to collect verticality, pile body stress, and hammering vibration data in real time. Upload the encrypted data to the cloud platform through a 5G / Beidou dual-channel and store it in a distributed database; 3) The cloud platform uses the reinforcement learning algorithm, combines the dynamic 3D geological model with real-time data, and dynamically calculates the optimal hammering energy, vibration frequency, and pile stopping criteria; 4) Pile driving: Monitor the pile body posture through lidar and inertial measurement unit, and use the PID control algorithm to drive the hydraulic system of the walking pile driver to control the verticality error within ≤5mm / m for pile driving.

2. The construction method of steel pipe pile foundation for deep-water port according to claim 1, characterized in that In step 1), constructing the dynamic 3D geological model based on the random forest algorithm specifically includes: S1. Use a manned multi-spectral radar to scan the construction area to obtain surface reflectivity data R(λ, x, y), where λ is the spectral band, and (x, y) are the planar coordinates. Combine the borehole data D(x, y, z) = {soil layer type, rock layer depth, groundwater level} and the seismic wave velocity profile V p (z) to construct a multi-dimensional data set; S2. Model training: Extract spectral feature F spectral = ∑ λ w λ R(λ, x, y), topographic feature F topo = ∇H(x, y), and spatial correlation feature F spatial = Kriging(D, V p ), where w λ is the spectral band weight coefficient, H(x, y) is the surface elevation value, D is the borehole dataset, V p is the seismic wave velocity profile. Input the feature vector F = [F spectral , F topo , F spatial into the random forest model, with the rock formation depth z rock and the groundwater depth z water as the target variables, and train the model by minimizing the mean squared error (MSE): Among them, is the model prediction value, z i is the measured value, N is the number of samples; S3. Dynamic 3D geological model construction: The construction area is discretized into a 3D voxel grid G(x, y, z), where the size of each voxel is 1 m × 1 m × 0.5 m. For the center point (x c , y c , z c ) of each voxel, the trained random forest model is used to predict geological parameters, generating a 3D attribute field G(x, y, z) = {soil layer type, z rock , z water}. Spatial optimization is performed on the sparse area through Kriging interpolation. The interpolation formula is: Among them, x 0 is the coordinate of the target point to be interpolated, x i is the coordinate of the known data point, λi is the weight coefficient, which is solved by minimizing the semivariogram γ ( h ) and satisfies , G ( x i ) is the geological parameter value of the known data point x i , and n represents the number of known data points participating in the interpolation calculation.

3. The construction method of steel pipe pile foundation for deep-water port according to claim 2, characterized in that, In step 1), planning the pile position layout and pile driving path of steel pipe piles specifically includes: Define a risk weight function based on a dynamic three-dimensional geological model W ( x , y ) = α ⋅ z water + β ⋅Softmax(1 / z rock ), where α , β is the geological risk coefficient; Constrained by the wharf design drawing, the pile position coordinates are optimized using the gradient descent method( x p , y p ), and the objective function is: + γ ⋅ Distance ( x p , y p , obstacle area) Among them, P is the total number of planned pile foundations, γ is the obstacle avoidance penalty factor, and Distance is the Euclidean distance between the pile position coordinates ( x p , y p ) and the obstacle area.

4. The construction method of steel pipe pile foundation in deep-water port according to claim 2, characterized in that, Step 1) also includes the update of the dynamic 3D geological model, including: During the construction process, access the pile driving feedback data in real time: penetration degree, pile body stress, and update the model parameters through online learning; If the prediction error MSE new > 1.2×MSE old , trigger model reconstruction, retrain and fuse new data to update the dynamic three-dimensional geological model.

5. The construction method of steel pipe pile foundation in deep-water port according to claim 1, characterized in that, Step 3) specifically includes: Deploy the Proximal Policy Optimization (PPO) algorithm on the cloud platform, combine the dynamic 3D geological model with real-time data, and through the iterative optimization of the PPO algorithm, output the adjustment amount of hammering energy, the change value of vibration frequency, and the dynamic pile stopping threshold; When the distance between the pile tip and the design elevation ≤ 1 m and the penetration e(t) ≤ e stop continue to hammer 30 - 50 times and then stop hammering according to the optimized hammering energy and vibration frequency; If abnormal penetration degree or excessive pile body stress is detected, trigger the digital twin model to simulate the subsequent hammering impact, and decide whether to force the pile to stop according to the simulation results.

6. The construction method of steel pipe pile foundation for deep-water port according to claim 5, characterized in that The construction of the digital twin model includes: Based on the pile-soil interaction mechanics principle, simulate the pile top displacement and stress distribution under different hammering energies and frequencies; Combine historical construction data to train a prediction model to predict the pile body response of the subsequent 10 hammerings. If the predicted stress exceeds the safety limit or the displacement deviation is greater than 5 mm, generate a forced pile stopping instruction.

7. The construction method of steel pipe pile foundation for deep-water port according to claim 6, characterized in that, The adjustment rule of the dynamic pile stopping threshold is: The basic threshold is determined according to the design requirements and static load tests; The actual threshold is dynamically adjusted according to the real-time hammering energy and rock layer hardness: when the hammering energy increases by 10%, the threshold is relaxed by 5%; When the rock layer hardness increases by 15%, the threshold is relaxed by 8%; When the groundwater level depth is less than 2 meters or the soil layer is a loose sand layer, the threshold is tightened to 80% of the basic value.

8. The construction method of steel pipe pile foundation for deep-water port according to claim 7, characterized in that, After step 4), it also includes the steps: Predict the pile body bearing capacity based on the acoustic emission sensor data and long short-term memory network; Record the key quality parameters through the blockchain and store the full-cycle original data through the InterPlanetary File System to ensure traceability and immutability.

9. The construction method of steel pipe pile foundation for deep-water port according to claim 8, characterized in that, Between step 1) and step 2), it also includes the step: Coat the surface of the steel pipe pile with a graphene-modified epoxy resin coating, and embed self-healing microcapsules in the coating; After the pile driving construction is completed, detect the coating integrity through drone infrared thermal imaging, and the defective area triggers the microcapsules to release the repair agent.

10. The construction method of steel pipe pile foundation for deep-water port according to claim 9, characterized in that, The thickness of the graphene-modified epoxy resin coating is 150 to 250 microns, where the graphene content is 1% to 3% of the total mass of the coating. The particle size of the self-healing microcapsules is 10 to 50 microns, the capsule wall is made of polyurethane material, and the interior encapsulates epoxy resin prepolymer and curing agent; Among them, when the detected local temperature difference on the coating surface is ≥ ±2.5 °C (relative to the surrounding intact area), and the area of the abnormal area is ≥ 5 square centimeters, it is determined as a defect area relative to the surrounding intact area; When the edge temperature gradient of the defect area exceeds 0.5 °C / mm, the microcapsules are triggered to rupture and release the repair agent.

Images