Dynamic adjustment method for mud cakes of different stratums of pile foundation

Abstract

The invention relates to the technical field of civil engineering, in particular to a dynamic adjustment method for mud cakes in different stratums of a pile foundation, which comprises the following steps: S1, designing a cast-in-place pile test device for simulating mountainous geology, analyzing the relationship between a filling coefficient and mud performance, and determining key indexes; according to the method, through an intelligent detection method fusing multi-frequency ultrasonic excitation and deep learning time-frequency feature enhancement, a high-robustness aperture-filling coefficient association database is constructed, accurate real-time monitoring of the state in the complex stratum hole is achieved, reliable data support is provided for dynamic adjustment, and the method is suitable for large-scale popularization and application. And by means of the reinforcement learning model and the digital twinborn platform with the filling coefficient deviation as the target, dynamic optimization of cooperative parameters of the slurry components and the drilling speed is achieved, stable control over the filling coefficient is ensured, and therefore the intelligence and precision level of slurry cake adjustment in pile foundation construction is improved, and the hole forming quality is effectively guaranteed.

Description

Technical Field

[0001] This invention relates to the field of civil engineering technology, specifically to a method for dynamically adjusting mud cake in different soil strata of pile foundations. Background Technology

[0002] Pile foundations are deep underground foundation structures, typically consisting of piles and pile caps. Their core function is to transfer the load of the superstructure or building to deeper, more resilient soil or rock layers, addressing issues such as insufficient bearing capacity and excessive settlement in shallow foundations, thus ensuring structural stability and safety. They are widely used in high-rise buildings, bridges, ports, and other engineering projects. In mountainous highway construction, pile foundations are a crucial foundation type for bridge engineering. Taking the Weixin-Yiliang Expressway in Yunnan Province as an example, this section traverses complex geological areas such as karst development zones and sand-clay cross-sections, presenting severe challenges for pile foundation construction: the geological conditions exhibit extremely high lateral heterogeneity, with prominent issues such as karst fissures and loose sand layers. Current technologies for pile foundation construction suffer from problems such as insufficient accuracy in borehole condition monitoring, lack of dynamic adaptability in mud adjustment, large fluctuations in filling coefficients, and unstable borehole quality due to variable geological conditions.

[0003] Based on this, the present invention provides a method for dynamically adjusting mud cake in different strata of pile foundations to solve the above-mentioned technical problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method for dynamic adjustment of mud cake in different strata of pile foundations. This invention integrates a multi-frequency ultrasonic excitation and a deep learning time-frequency feature enhancement intelligent detection method to construct a highly robust aperture-filling coefficient correlation database, realizing accurate real-time monitoring of the borehole state in complex strata, providing reliable data support for dynamic adjustment. Furthermore, by using a reinforcement learning model and digital twin platform targeting the filling coefficient deviation, dynamic optimization of the synergistic parameters of mud composition and drilling speed is achieved, ensuring stable control of the filling coefficient. This improves the intelligence and precision of mud cake adjustment in pile foundation construction and effectively guarantees the quality of borehole formation.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides a method for dynamically adjusting mud cake in different soil strata of pile foundations, comprising the following steps:

[0007] S1: Design a test device for cast-in-place piles to simulate mountainous geology, analyze the relationship between the filling coefficient and mud performance, and determine key indicators;

[0008] S2: Optimize the basic mud mix ratio through single-factor experiments, and study the differences in adaptability of dispersants and thickener modifiers to sand and clay layers, forming a formation-adaptive formula library;

[0009] S3: An intelligent detection method that integrates multi-frequency ultrasonic excitation and deep learning time-frequency feature enhancement is adopted. The acoustic impedance change is dynamically analyzed through a spatiotemporal attention network to construct a highly robust aperture-filling coefficient correlation database.

[0010] S4: Construct a reinforcement learning model with the goal of minimizing the filling coefficient deviation, and combine it with a digital twin platform to simulate the effects of different adjustment strategies in real time, and dynamically optimize the synergistic parameters of mud composition and drilling speed.

[0011] S5: Based on the actual engineering data, optimize the parameters of casing burial depth, drill bit selection, and rotary drilling technology to form a standardized construction plan for mountainous areas.

[0012] The S1 section describes the design of a cast-in-place pile test device to simulate mountainous geology. The relationship between the filling coefficient and mud performance is analyzed, and key indicators are determined. The specific steps are as follows:

[0013] S1.1 Design a modular test device that can simulate karst, sand, and clay mountain strata, integrating an adjustable pressure casing, a mud circulation system, and fiber optic strain sensors;

[0014] S1.2. With the borehole parameters fixed, the mud viscosity, density, and water loss were changed sequentially. The filling coefficient was measured by laser scanning to establish the original dataset.

[0015] S1.3. Using grey relational analysis and multiple regression modeling, viscosity, density, and water loss were determined as key indicators.

[0016] S1.4. Repeat the test at the critical value to verify the stability threshold of pile hole collapse rate <5%, and output a control manual including index range and API test standard.

[0017] In step S2, the basic mud mix ratio is optimized through single-factor experiments, and the adaptability differences of dispersants and thickener modifiers to sand and clay layers are studied to form a formation-adaptive formula library. The specific steps are as follows:

[0018] S2.1 Basic mud preparation: Using bentonite as the base material, add soda ash and CMC in the conventional proportion to prepare basic mud that meets the basic performance requirements;

[0019] S2.2 Single-factor experimental design: With other factors fixed, design multiple control experiments by changing the dosage of dispersant and thickener respectively;

[0020] S2.3 Formation Simulation Test: The mud was injected into the test device simulating sand and clay layers respectively to test the mud performance and hole formation quality under different modifier dosages;

[0021] S2.4 Adaptability Difference Analysis: Compare the effects of different modifiers on the mud wall protection effect and filling coefficient in sand and clay layers;

[0022] S2.5 Formulation Library Construction: Based on experimental data, the optimal combination and dosage range of modifiers are screened for different formations to form a formation-adaptive mud formulation library containing formulation parameters and applicable conditions.

[0023] The S3 method employs an intelligent detection approach that integrates multi-frequency ultrasonic excitation with deep learning time-frequency feature enhancement. It dynamically analyzes acoustic impedance changes using a spatiotemporal attention network to construct a robust aperture-filling coefficient correlation database. The specific steps are as follows:

[0024] S3.1. A 20-200kHz multi-band ultrasonic array sensor is used to collect acoustic impedance signals in real time during the drilling process and simultaneously record the borehole diameter laser scanning calibration data.

[0025] S3.2 Perform wavelet transform time-frequency analysis on the original signal, and combine it with generative adversarial network to enhance feature extraction and eliminate signal noise caused by karst fissures;

[0026] S3.3 Construct a spatiotemporal attention network, using the acoustic impedance characteristic slope / intercept as input and outputting the dynamic change value of the aperture;

[0027] S3.4 Integrate ultrasonic data, inverted pore size and measured filling coefficient to establish a knowledge graph of pore size-filling coefficient relationship for multiple strata including sand, clay and karst.

[0028] In step S3.2, wavelet transform time-frequency analysis is performed on the original signal, and generative adversarial network is used to enhance feature extraction to eliminate signal noise caused by karst fissures. The specific steps are as follows:

[0029] S3.2.1 Perform 5-layer Daubechies wavelet packet decomposition on the original ultrasound signal to extract energy features of the 1-100kHz sub-band;

[0030] S3.2.2 Construct a generative adversarial network. The generator uses a U-Net structure to map noisy signals to clean signals, and the discriminator uses a 5-layer convolutional network to determine the authenticity of the signal.

[0031] S3.2.3 Optimize network parameters through adversarial training, with a loss function of L = 0.7L. re +0.3L a d v , where L re For wavelet coefficient reconstruction loss: W is the wavelet packet decomposition operator; L a d v To combat losses: Train until the signal-to-noise ratio is ≥25dB;

[0032] S3.2.4. Apply a weighting factor ω = 2.0 to the wavelet coefficients in the 20-50kHz karst fissure frequency band, and correct the loss function as L′ = L + ω × L re-k , where L re-k This is to compensate for the reconstruction loss in this frequency band and suppress gap artifacts.

[0033] In step S3.3, a spatiotemporal attention network is constructed, taking the acoustic impedance characteristic slope / intercept as input and outputting the dynamic change value of the aperture. The specific steps are as follows:

[0034] S3.3.1. Perform linear fitting on the acoustic impedance signal to obtain the slope k and intercept b. The fitting formula is as follows:

[0035] Z(t) = k·t + b + ∈

[0036] Where Z(t) is the acoustic impedance signal, ∈ is the fitting residual, and the least squares method is used to solve for ∑∈ 2 Minimum;

[0037] S3.3.2 Construct a spatiotemporal attention network, including a spatial attention module and a temporal attention module. The calculation formula for the spatial attention module is as follows:

[0038] M s =σ(W s ·[AvgPool(Z),MaxPool(Z)]+b s )

[0039] The formula for calculating the time attention module is:

[0040] M t =σ(W t ·[AvgPool(Z T MaxPool(Z) T )]+b t )

[0041] Where σ is the sigmoid activation function, W s b s W t b t Z is a trainable parameter. T This is the transpose of the acoustic impedance signal;

[0042] S3.3.3 Multiply the acoustic impedance characteristic slope k and intercept b by the spatiotemporal attention weights to obtain the weighted characteristic representation. The calculation formula is as follows:

[0043] Z att =M s ⊙M t ⊙Z

[0044] Where ⊙ represents element-wise multiplication;

[0045] S3.3.4. The weighted features are mapped to the dynamic aperture change value through a fully connected layer. The calculation formula for the fully connected layer is as follows:

[0046] ΔD=W fc ·Z att +b fc

[0047] Among them, W fc b fc Here are the parameters for the fully connected layer, and ΔD is the dynamic change value of the pore size.

[0048] In step S4, a reinforcement learning model is constructed with the goal of minimizing the filling coefficient deviation. This model is then combined with a digital twin platform to simulate the effects of different adjustment strategies in real time, dynamically optimizing the synergistic parameters between mud composition and drilling speed. The specific steps are as follows:

[0049] S4.1 Real-time access to ultrasonic borehole diameter data, mud performance sensor data, and drilling rig operating parameters to construct a dynamic dataset;

[0050] S4.2. Establish a coupled model of pile hole-mud-drilling tool based on the finite element method, and update the twin state through real-time data driving.

[0051] S4.3 Design a PPO algorithm with the filling coefficient deviation as the reward function, and pre-train it for 100,000 iterations in a digital twin environment;

[0052] S4.4 Deploy a lightweight inference model that receives real-time data every 5 seconds and outputs mud composition adjustment instructions and drilling speed suggestions;

[0053] S4.5 Compare the fluctuation range of the filling coefficient before and after adjustment. When the deviation is greater than 15%, the model retraining mechanism is triggered.

[0054] In step S4.2, a coupled model of pile hole-mud-drilling tool is established based on the finite element method, and the twin state is updated through real-time data. The specific steps are as follows:

[0055] S4.2.1: Establish a multiphysics coupling model of pile hole-mud-drilling tool, including:

[0056] ① The soil in the pile hole was modeled using the Mohr-Coulomb constitutive model, and the parameters were initialized using geological exploration data;

[0057] ② The Herschel-Bulkley rheological model was used for the mud fluid, and the viscosity parameters were connected to the mud sensor in real time;

[0058] ③The dynamics of the drilling tool is modeled using the Lagrange equation, integrating the torque-speed curve provided by the rotary drilling rig manufacturer;

[0059] S4.2.2: Set the adaptive meshing rule for the finite element mesh, with the mesh size near the hole wall ≤ 5 mm and in the far-field region ≤ 20 mm;

[0060] S4.2.3: Define the data-driven interface to receive ultrasonic hole diameter data, mud density meter readings, and drill pressure sensor data every 10 seconds, and update the boundary conditions of the digital twin;

[0061] S4.2.4: Use an explicit-implicit hybrid solver, set the time step to 0.1 second, and output the predicted hole wall stress distribution and mud flow field in real time.

[0062] In S4.3, design a PPO algorithm with the filling coefficient deviation as the reward function, and pre-train it for 100,000 iterations in the digital twin environment. The specific steps are as follows:

[0063] S4.3.1: Define the reasonable range of the filling coefficient as 1.05 - 1.15, set the target value C target = 1.10, and calculate the real-time deviation e = |C t - 1.10|;

[0064] S4.3.2: Construct a piecewise reward function:

[0065] ① When e ≤ 0.03, r = 10 - 333e;

[0066] ② When 0.03 < e ≤ 0.10, r = 5 - 50e;

[0067] ③ When e > 0.10, r = -20e;

[0068] Simultaneously introduce a hole wall stability penalty term p = 0.5 × the hole collapse risk index, and the final reward R = r - p;

[0069] S4.3.3: The policy network adopts a three-layer fully connected structure, with the output layer activated by Softmax. The action space includes:

[0070] ① The adjustment of the dispersant dosage Δw1 ∈ [-0.2%, +0.2%];

[0071] ② The adjustment of the viscosifier dosage Δw2 ∈ [-0.1%, +0.1%];

[0072] ③ The adjustment of the drilling speed Δv ∈ [-5 rpm, +5 rpm];

[0073] S4.3.4: The PPO algorithm sets the clipping parameter ∈1 = 0.2, the discount factor γ = 0.95, the GAE parameter λ = 0.9, and the objective function is:

[0074]

[0075] S4.3.5: Phased training: The first 30,000 iterations adopt a random exploration strategy, the middle 50,000 iterations gradually improve the strategy determinism, and the last 20,000 iterations introduce formation migration sample enhancement. The model parameters are saved every 500 iterations.

[0076] In step S5, by combining data from actual engineering projects, the parameters of casing burial depth, drill bit selection, and rotary drilling technology are optimized to form a standardized mountainous construction plan. The specific steps are as follows:

[0077] S5.1: Deploy IoT sensors at typical work sites along the Weiyi Expressway to record parameters such as casing vibration frequency, drill bit wear rate, and rotary drilling torque in real time;

[0078] S5.2: The contribution of casing burial depth, drill bit type, and rotation speed to the filling coefficient is analyzed using orthogonal experimental design.

[0079] S5.3: Decision tree model based on data mining: When encountering a sand layer and the groundwater level is <2m, the strategy of "deepening the casing to 2.5m + roller cone drill bit + reducing speed to 18rpm" is automatically triggered.

[0080] S5.4: Compile a "Technical Standard for Pile Foundation Construction in Mountainous Areas" that includes three types of strata and five abnormal working conditions, stipulating that the combination of "follow-up casing + impact drill bit" must be used in karst areas.

[0081] Compared with the prior art, the beneficial effects of the present invention are:

[0082] This invention utilizes an intelligent detection method that integrates multi-frequency ultrasonic excitation with deep learning time-frequency feature enhancement to construct a highly robust aperture-filling coefficient correlation database. This enables precise real-time monitoring of the borehole state in complex formations, providing reliable data support for dynamic adjustments. Furthermore, by employing a reinforcement learning model targeting the filling coefficient deviation and a digital twin platform, the invention achieves dynamic optimization of the synergistic parameters of mud composition and drilling speed, ensuring stable control of the filling coefficient. This enhances the intelligence and precision of mud cake adjustment in pile foundation construction, effectively guaranteeing borehole quality. Attached Figure Description

[0083] Figure 1 This is a flowchart of a method for dynamically adjusting mud cake in different strata of pile foundations according to the present invention.

[0084] Figure 2 This is a flowchart illustrating the construction process optimization method for dynamically adjusting mud cake in different soil strata of pile foundations according to the present invention. Detailed Implementation

[0085] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0086] Example:

[0087] like Figures 1-2 As shown, this embodiment provides a method for dynamic adjustment of mud cake in different strata of pile foundations, including the following steps: S1: Design a test device for cast-in-place piles simulating mountainous geology, analyze the relationship between the filling coefficient and mud performance, and determine key indicators; S2: Optimize the mud mix ratio of the foundation through single-factor experiments, and study the adaptability differences of dispersants and thickener modifiers to sand and clay layers, forming a stratum-adaptive formula library; S3: Adopt an intelligent detection method that integrates multi-frequency ultrasonic excitation and deep learning time-frequency feature enhancement, dynamically analyze acoustic impedance changes through a spatiotemporal attention network, and construct a highly robust aperture-filling coefficient correlation database; S4: Construct a reinforcement learning model with the goal of minimizing the filling coefficient deviation, and combine it with a digital twin platform to simulate the effects of different adjustment strategies in real time, dynamically optimizing the synergistic parameters of mud composition and drilling speed; S5: Combine physical engineering data to optimize the parameters of casing burial depth, drill bit selection, and rotary drilling technology, forming a standardized mountainous construction plan.

[0088] In this embodiment, it should also be noted that the design of the grouting pile test device simulating mountainous geology in S1, the analysis of the relationship between the filling coefficient and mud performance, and the determination of key indicators are as follows: S1.1, Design a modular test device that can simulate karst, sand, and clay mountainous strata, integrating an adjustable pressure casing, a mud circulation system, and fiber optic strain sensors; S1.2, Fix the drilling parameters, and sequentially change the mud viscosity, density, and water loss, and measure the filling coefficient by laser scanning to establish an original dataset; S1.3, Use grey relational analysis and multiple regression modeling to determine viscosity, density, and water loss as key indicators; S1.4, Conduct repeated tests at critical values ​​to verify the stability threshold of pile hole collapse rate <5%, and output a control manual containing the indicator range and API testing standards.

[0089] Furthermore, it should be noted that the main body of the device in S1.1 is a transparent plexiglass tank with a length × width × height of 2m × 1m × 3m. It adopts a modular design to realize multi-stratum simulation: ① Karst module: with internal cavities of 5-20cm in diameter (filled with gypsum to simulate soluble rock mass) and a fracture network of 0.5-2mm; ② Sand layer module: filled with graded sand (particle size 0.5-2mm, relative density 70%); ③ Clay layer module: using silty clay with a liquid limit of 35% and a plastic limit of 18%, compacted in layers (compaction degree 90%). The integrated components include: ① Adjustable pressure casing: 50cm in diameter, with internal pressure adjusted via a pneumatic valve (0.1-0.5MPa) to simulate soil and water pressure at different burial depths; ② Mud circulation system: including a mixing tank (500L capacity), a centrifugal pump (flow rate 10-30L / min), and a return pipe to achieve positive mud circulation; ③ Fiber optic strain sensors: one sensor every 30° along the borehole wall (range -2000-2000με) to monitor borehole wall deformation in real time. Stability verification and manual output in S1.4: For critical values ​​(viscosity 22s, density 1.2g / cm³)... 3 Ten repeated tests were conducted (with a water loss of 15 mL / 30 min). The collapse rate of the pile hole was consistently <5%. The reasonable range for the following indicators was determined: viscosity: 20-25 s; density: 1.15-1.25 g / cm³. 3 Water loss: 12-18 mL / 30 min. Output the "Drilling Performance Control Manual," specifying that the testing methods for the indicators comply with API RP 13B-1 standards.

[0090] In this embodiment, it should also be noted that in S2, the basic mud mix ratio is optimized through single-factor experiments, and the adaptability differences of dispersants and thickener modifiers to sand and clay layers are studied to form a formation-adaptive formula library. The specific steps are as follows: S2.1, Basic mud preparation: Using bentonite as the base material, soda ash and CMC are added in conventional proportions to prepare basic mud that meets the basic performance requirements; S2.2, Single-factor experiment design: With other factors fixed, the dosage of dispersant and thickener is changed to design multiple groups. Control experiment; S2.3, Formation simulation test: Mud was injected into test devices simulating sand and clay layers respectively to test mud performance and hole formation quality under different modifier dosages; S2.4, Adaptability difference analysis: The influence of different modifiers on mud wall protection effect and filling coefficient in sand and clay layers was compared; S2.5, Formula library construction: Based on experimental data, the optimal combination of modifiers and dosage range were screened for different formations to form a formation-adaptive mud formula library containing formula parameters and applicable conditions.

[0091] Furthermore, it should be noted that the basic mud preparation in S2.1 is as follows: the formula is: bentonite (8%) + soda ash (0.5%) + CMC (0.2%) + water (91.3%). After stirring for 30 minutes, the viscosity was tested to be 22s and the density to be 1.18g / cm³.3 The water loss was 15 mL / 30 min, meeting the basic indicators determined in S1. Single-factor experimental design in S2.2: Dispersant: FCLS (ferrochromium lignin sulfonate) was selected, with a dosage gradient of 0.1%-0.5% (based on bentonite mass); Thickening agent: PHP (partially hydrolyzed polyacrylamide) was selected, with a dosage gradient of 0.05%-0.3%; other parameters were fixed, and three parallel tests were set up for each dosage group. Formation simulation test in S2.3: Sand layer test: Mud was injected into the sand layer module, and the permeability coefficient of the mud wall was measured using a permeameter (target < 1 × 10⁻⁶). -7 (cm / s), observe the collapse of the borehole through in-hole camera; clay layer test: inject clay layer module, measure mud cake thickness (target 2-5mm) and borehole wall diameter reduction rate (target <3%).

[0092] In this embodiment, it should also be noted that the intelligent detection method in S3, which integrates multi-frequency ultrasonic excitation with deep learning time-frequency feature enhancement, dynamically analyzes acoustic impedance changes through a spatiotemporal attention network to construct a highly robust aperture-filling coefficient correlation database. The specific steps are as follows: S3.1, Using a 20-200kHz multi-band ultrasonic array sensor, acoustic impedance signals are collected in real time during drilling, and aperture laser scanning calibration data is recorded simultaneously; S3.2, Wavelet transform time-frequency analysis is performed on the original signal, combined with generative adversarial network to enhance feature extraction and eliminate signal noise caused by karst fissures; The specific steps are as follows: S3.2.1, The original ultrasonic signal is decomposed into 5-layer Daubechies wavelet packets to extract energy features of the 1-100kHz sub-band; S3.2.2, A generative adversarial network is constructed. The generator uses a U-Net structure to map the noisy signal to a clean signal, and the discriminator uses a 5-layer convolutional network to judge the authenticity of the signal; S3.2.3, The network parameters are optimized through adversarial training, with a loss function of L = 0.7L. re +0.3L a d v , where L re For wavelet coefficient reconstruction loss: W is the wavelet packet decomposition operator; L a d v To combat losses: Training until the signal-to-noise ratio is ≥25dB; S3.2.4, apply a weighting factor ω=2.0 to the wavelet coefficients in the 20-50kHz karst fissure frequency band, and correct the loss function to L′=L+ω×L re-k , where L re-k To address the reconstruction loss in this frequency band and suppress gap artifacts, S3.3, construct a spatiotemporal attention network, using the acoustic impedance characteristic slope / intercept as input and outputting the dynamic change value of the aperture; the specific steps are as follows: S3.3.1, perform linear fitting on the acoustic impedance signal to obtain the slope k and intercept b, with the fitting formula as follows:

[0093] Z(t) = k·t + b + ∈

[0094] Where Z(t) is the acoustic impedance signal, ∈ is the fitting residual, and the least squares method is used to solve for ∑∈ 2 Minimum;

[0095] S3.3.2 Construct a spatiotemporal attention network, including a spatial attention module and a temporal attention module. The calculation formula for the spatial attention module is as follows:

[0096] M s =σ(W s ·[AvgPool(Z),MaxPool(Z)]+b s )

[0097] The formula for calculating the time attention module is:

[0098] M t =σ(W t ·[AvgPool(Z T MaxPool(Z) T )]+b t )

[0099] Where σ is the sigmoid activation function, W s b s W t b t Z is a trainable parameter. T This is the transpose of the acoustic impedance signal; S3.3.3, multiply the acoustic impedance characteristic slope k and intercept b by the spatiotemporal attention weights to obtain the weighted characteristic representation, calculated using the following formula:

[0100] Z att =M s ⊙M t ⊙Z

[0101] Where ⊙ represents element-wise multiplication; S3.3.4, the weighted features are mapped to the dynamic aperture change value through a fully connected layer. The calculation formula for the fully connected layer is:

[0102] ΔD=W fc ·Z att +b fc

[0103] Among them, W fc b fc Here, ΔD represents the dynamic variation of the pore size, and S3.4 integrates ultrasonic data, inverted pore size, and measured filling coefficient to establish a knowledge graph of the pore size-filling coefficient relationship for multiple strata, including sand, clay, and karst.

[0104] Furthermore, it should be noted that the multi-frequency ultrasonic array sensor employs a 64-channel array with a center frequency of 20-200kHz and a sampling rate of 1MHz, arranged circumferentially along the drill rod (one group every 10cm). The generator uses a U-Net structure containing four encoder / decoder blocks (3×3 convolutional kernels) with a learning rate of 0.0002; the discriminator is a 5-layer convolutional network (the last layer is a 1×1 convolutional layer that outputs probabilities).

[0105] In this embodiment, it should also be noted that in S4, a reinforcement learning model is constructed with the goal of minimizing the filling coefficient deviation. This model, combined with a digital twin platform, simulates the effects of different adjustment strategies in real time, dynamically optimizing the synergistic parameters of mud composition and drilling speed. The specific steps are as follows: S4.1: Real-time access to ultrasonic borehole diameter data, mud performance sensor data, and drilling rig operating parameters to construct a dynamic dataset; S4.2: Establishing a pile hole-mud-drilling tool coupling model based on the finite element method, updating the twin state through real-time data-driven updates; the specific steps are as follows: S4.2.1: Establishing a pile hole-mud-drilling tool multiphysics coupling model, including: ① The pile hole soil adopts the Mohr-Coulomb constitutive model, and the parameters are obtained through geological... Exploration data initialization; ② The Herschel-Bulkley rheological model is used for mud fluid, and the viscosity parameters are connected to the mud sensor in real time; ③ Drilling tool dynamics are modeled using the Lagrange equation, and the torque-speed curve provided by the rotary drilling rig manufacturer is integrated; S4.2.2: Set the adaptive mesh generation rules for finite element meshes, with mesh size ≤5mm near the borehole wall and ≤20mm in the far field region; S4.2.3: Define the data-driven interface, which receives ultrasonic borehole diameter data, mud density meter readings, and drill pressure sensor data every 10 seconds, and updates the twin boundary conditions; S4.2.4: Use an explicit-implicit hybrid solver, with a calculation step size set to 0.1 seconds, and output the borehole wall stress distribution and mud flow field prediction in real time. S4.3: Design the PPO algorithm with the filling coefficient deviation as the reward function, and pre-train it for 100,000 iterations in the digital twin environment; the specific steps are as follows: S4.3.1: Define the reasonable range of the filling coefficient as 1.05-1.15, and set the target value C. target =1.10, calculate the real-time deviation e = |C t-1.10|; S4.3.2: Construct a piecewise reward function: ① When e ≤ 0.03, r = 10 - 333e; ② When 0.03 < e ≤ 0.10, r = 5 - 50e; ③ When e > 0.10, r = -20e; Synchronously introduce a hole wall stability penalty term p = 0.5 × the hole collapse risk index, and the final reward R = r - p; S4.3.3: The policy network adopts a 3-layer fully connected structure, and the output layer is activated by Softmax. The action space includes: ① The adjustment of the dispersant dosage Δw1 ∈ [-0.2%, +0.2%]; ② The adjustment of the viscosifier dosage Δw2 ∈ [-0.1%, +0.1%]; ③ The adjustment of the drilling speed Δv ∈ [-5 rpm, +5 rpm]; S4.3.4: The PPO algorithm sets the clipping parameter ∈1 = 0.2, the discount factor γ = 0.95, the GAE parameter λ = 0.9, and the objective function is:

[0106]

[0107] S4.3.5: Training in stages: In the first 30,000 iterations, a random exploration strategy is adopted. In the middle 50,000 iterations, the policy certainty is gradually improved. In the last 20,000 iterations, formation migration samples are introduced for enhancement, and the model parameters are saved every 500 iterations. S4.4: Deploy a lightweight inference model, receive real-time data every 5 seconds and output instructions for adjusting the mud components and drilling speed suggestions; S4.5: Compare the fluctuation range of the fullness coefficient before and after adjustment. When the deviation continuously > 15%, trigger the model retraining mechanism.

[0108] Furthermore, it should be noted that for the lightweight model: It is compressed by TensorRT, the inference time < 0.1 s, and the adjustment instructions are output every 5 seconds; For the retraining mechanism: When the deviation > 15% continuously for 3 times, trigger incremental training (add 20,000 real-time data).

[0109] In this embodiment, it should also be noted that in S5, combined with the entity project data, optimize the parameters of the casing burial depth, bit selection, and rotary drilling process to form a standardized mountain construction plan. The specific steps are as follows: S5.1: Deploy Internet of Things sensors at typical work points on the Weiyi Expressway to record the parameters of the casing vibration frequency, bit wear rate, and rotary drilling torque in real time; S5.2: Use the orthogonal test method to analyze the contribution of the casing burial depth, bit type, and rotation speed to the fullness coefficient; S5.3: Establish a decision tree model based on data mining: When encountering a sand layer and the groundwater level < 2 m, automatically trigger the strategy of "deepening the casing to 2.5 m + roller cone bit + reducing the speed to 18 rpm"; S5.4: Compile the "Mountain Pile Foundation Construction Technology Standard" including 3 types of formations and 5 abnormal working conditions, and stipulate that the "follow-up casing + impact bit" combination must be used in the karst area.

[0110] Furthermore, it should be noted that the following sensors were deployed at the Weiyi Expressway construction site: casing vibration sensor (range 0-100Hz), drill bit wear sensor (accuracy ±0.1mm), and torque sensor (0-5000N·m).

[0111] In this embodiment, a method for dynamically adjusting mud cake in different geological strata of pile foundations is described as follows: First, a test platform simulating complex mountainous strata is constructed to clarify key mud performance indicators. This involves designing a modular test device including modules for karst, sand, and clay layers, integrating an adjustable pressure casing, a mud circulation system, and sensors. By changing mud viscosity, density, and water loss, combined with laser scanning, a raw dataset of filling coefficients is established. Then, key indicators are determined through grey relational analysis and multiple regression. Critical value verification is used to form a control manual, providing basic parameters and experimental basis for subsequent research. Based on this, the mud formulation is optimized to adapt to different geological strata. First, a basic mud is prepared, and then the dosage of dispersant and thickener is changed through single-factor experiments. Mud performance and hole formation quality are tested in simulated sand and clay layers, and the adaptability differences of modifiers are analyzed. Finally, a library of suitable formulations for different geological strata is constructed to ensure that the mud exerts a good wall-protecting effect in specific geological strata. Subsequently, an intelligent detection system was constructed to monitor the borehole status in real time. A multi-frequency ultrasonic array sensor was used to collect acoustic impedance signals, and laser scanning calibration data was recorded simultaneously. Wavelet packet decomposition and generative adversarial network were used to denoise the original signals to eliminate karst fissure noise. The acoustic impedance characteristics were analyzed through a spatiotemporal attention network to invert the dynamic changes in borehole diameter. Finally, the data was integrated to establish a knowledge graph of borehole diameter-filling coefficient correlation in multiple formations, providing data support for dynamic adjustment. Next, dynamic optimization of mud composition and drilling speed was achieved. Various monitoring data were accessed in real time to construct a dynamic dataset. A digital twin model of pile hole-mud-drilling tool coupling was established based on the finite element method. The model was pre-trained in the twin environment using the PPO reinforcement learning algorithm. A lightweight model was deployed to output adjustment commands in real time. When the filling coefficient deviation continued to exceed the limit, retraining was triggered to ensure parameter co-optimization to maintain the stability of the filling coefficient. Finally, a standardized construction plan was formed by combining the actual engineering data. Sensors were deployed at actual work sites to collect parameters. The impact of process parameters on the filling coefficient was analyzed through orthogonal experiments. A decision tree model was established and construction process standards covering multiple strata and abnormal working conditions were compiled to achieve standardization and efficiency in pile foundation construction in mountainous areas, and to ensure the quality of hole formation and construction safety.

[0112] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0113] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for dynamically adjusting mud cake in different strata of pile foundations, characterized in that, Includes the following steps: S1: Design a test device for cast-in-place piles to simulate mountainous geology, analyze the relationship between the filling coefficient and mud performance, and determine key indicators; S2: Optimize the basic mud mix ratio through single-factor experiments, and study the differences in adaptability of dispersants and thickener modifiers to sand and clay layers, forming a formation-adaptive formula library; S3: An intelligent detection method that integrates multi-frequency ultrasonic excitation and deep learning time-frequency feature enhancement is adopted. The acoustic impedance change is dynamically analyzed through a spatiotemporal attention network to construct a highly robust aperture-filling coefficient correlation database. S4: Construct a reinforcement learning model with the goal of minimizing the filling coefficient deviation, and combine it with a digital twin platform to simulate the effects of different adjustment strategies in real time, and dynamically optimize the synergistic parameters of mud composition and drilling speed. S5: Based on the actual engineering data, optimize the parameters of casing burial depth, drill bit selection, and rotary drilling technology to form a standardized construction plan for mountainous areas.

2. The method for dynamically adjusting mud cake in different strata of pile foundations according to claim 1, characterized in that, The S1 section describes the design of a test device for cast-in-place piles simulating mountainous geology. It analyzes the relationship between the filling coefficient and mud performance, and determines key indicators. The specific steps are as follows: S1.1 Design a modular test device that can simulate karst, sand, and clay mountain strata, integrating an adjustable pressure casing, a mud circulation system, and fiber optic strain sensors; S1.

2. With the borehole parameters fixed, the mud viscosity, density, and water loss were changed sequentially. The filling coefficient was measured by laser scanning to establish the original dataset. S1.

3. Using grey relational analysis and multiple regression modeling, viscosity, density, and water loss were determined as key indicators. S1.

4. Repeat the test at the critical value to verify the stability threshold of pile hole collapse rate <5%, and output a control manual including index range and API test standard.

3. The method for dynamically adjusting mud cake in different strata of pile foundations according to claim 1, characterized in that, In step S2, the basic mud mix ratio is optimized through single-factor experiments, and the adaptability differences of dispersants and thickener modifiers to sand and clay layers are studied to form a formation-adaptive formula library. The specific steps are as follows: S2.1 Basic mud preparation: Using bentonite as the base material, add soda ash and CMC in the conventional proportion to prepare basic mud that meets the basic performance requirements; S2.2 Single-factor experimental design: With other factors fixed, design multiple control experiments by changing the dosage of dispersant and thickener respectively; S2.3 Formation Simulation Test: The mud was injected into the test device simulating sand and clay layers respectively to test the mud performance and hole formation quality under different modifier dosages; S2.4 Adaptability Difference Analysis: Compare the effects of different modifiers on the mud wall protection effect and filling coefficient in sand and clay layers; S2.5 Formulation Library Construction: Based on experimental data, the optimal combination and dosage range of modifiers are screened for different formations to form a formation-adaptive mud formulation library containing formulation parameters and applicable conditions.

4. The method for dynamically adjusting mud cake in different strata of pile foundations according to claim 1, characterized in that, The S3 method employs an intelligent detection approach that integrates multi-frequency ultrasonic excitation with deep learning time-frequency feature enhancement. It dynamically analyzes acoustic impedance changes using a spatiotemporal attention network to construct a robust aperture-filling coefficient correlation database. The specific steps are as follows: S3.

1. A 20-200kHz multi-band ultrasonic array sensor is used to collect acoustic impedance signals in real time during the drilling process and simultaneously record the borehole diameter laser scanning calibration data. S3.2 Perform wavelet transform time-frequency analysis on the original signal, and combine it with generative adversarial network to enhance feature extraction and eliminate signal noise caused by karst fissures; S3.3 Construct a spatiotemporal attention network, using the acoustic impedance characteristic slope / intercept as input and outputting the dynamic change value of the aperture; S3.4 Integrate ultrasonic data, inverted pore size and measured filling coefficient to establish a knowledge graph of pore size-filling coefficient relationship for multiple strata including sand, clay and karst.

5. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 4, characterized in that, In step S3.2, wavelet transform time-frequency analysis is performed on the original signal, and generative adversarial network is used to enhance feature extraction to eliminate signal noise caused by karst fissures. The specific steps are as follows: S3.2.1 Perform 5-layer Daubechies wavelet packet decomposition on the original ultrasound signal to extract energy features of the 1-100kHz sub-band; S3.2.2 Construct a generative adversarial network. The generator uses a U-Net structure to map noisy signals to clean signals, and the discriminator uses a 5-layer convolutional network to determine the authenticity of the signal. S3.2.3 Optimize network parameters through adversarial training, with a loss function of L = 0.7L. re +0.3L a d v , where L re For wavelet coefficient reconstruction loss: W is the wavelet packet decomposition operator; L a d v To combat losses: Train until the signal-to-noise ratio is ≥25dB; S3.2.

4. Apply a weighting factor ω = 2.0 to the wavelet coefficients in the 20-50kHz karst fissure frequency band, and correct the loss function as L′ = L + ω × L re-k , where L re-k This is to compensate for the reconstruction loss in this frequency band and suppress gap artifacts.

6. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 4, characterized in that, In step S3.3, a spatiotemporal attention network is constructed, taking the acoustic impedance characteristic slope / intercept as input and outputting the dynamic change value of the aperture. The specific steps are as follows: S3.3.

1. Perform linear fitting on the acoustic impedance signal to obtain the slope k and intercept b. The fitting formula is as follows: Z(t) = k·t + b + ∈ Where Z(t) is the acoustic impedance signal, ∈ is the fitting residual, and the least squares method is used to solve for ∑∈ 2 Minimum; S3.3.2 Construct a spatiotemporal attention network, including a spatial attention module and a temporal attention module. The calculation formula for the spatial attention module is as follows: M s =σ(W s [AvgPool(Z),MaxPool(Z)]+b s ) The formula for calculating the time attention module is: M t =σ(W t ·[AvgPool(Z T ),MaxPool(Z T )]+b t ) Where σ is the sigmoid activation function, W s b s W t b t Z is a trainable parameter. T This is the transpose of the acoustic impedance signal; S3.3.3 Multiply the acoustic impedance characteristic slope k and intercept b by the spatiotemporal attention weights to obtain the weighted characteristic representation. The calculation formula is as follows: From att =M s ⊙M t ⊙Z Where ⊙ represents element-wise multiplication; S3.3.

4. The weighted features are mapped to the dynamic aperture change value through a fully connected layer. The calculation formula for the fully connected layer is as follows: ΔD=W fc ·WITH att +b fc Among them, W fc b fc Here are the parameters for the fully connected layer, and ΔD is the dynamic change value of the pore size.

7. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 1, characterized in that, In step S4, a reinforcement learning model is constructed with the goal of minimizing the filling coefficient deviation. This model is then combined with a digital twin platform to simulate the effects of different adjustment strategies in real time, dynamically optimizing the synergistic parameters between mud composition and drilling speed. The specific steps are as follows: S4.1 Real-time access to ultrasonic borehole diameter data, mud performance sensor data, and drilling rig operating parameters to construct a dynamic dataset; S4.

2. Establish a coupled model of pile hole-mud-drilling tool based on the finite element method, and update the twin state through real-time data driving. S4.3 Design a PPO algorithm with the filling coefficient deviation as the reward function, and pre-train it for 100,000 iterations in a digital twin environment; S4.4 Deploy a lightweight inference model that receives real-time data every 5 seconds and outputs mud composition adjustment instructions and drilling speed suggestions; S4.5 Compare the fluctuation range of the filling coefficient before and after adjustment. When the deviation is greater than 15%, the model retraining mechanism is triggered.

8. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 7, characterized in that, In step S4.2, a coupled model of pile hole-mud-drilling tool is established based on the finite element method, and the twin state is updated through real-time data. The specific steps are as follows: S4.2.1: Establish a multiphysics coupling model of pile hole-mud-drilling tool, including: ① The soil in the pile hole was modeled using the Mohr-Coulomb constitutive model, and the parameters were initialized using geological exploration data; ②The mud fluid adopts the Herschel-Bulkley rheological model, and the viscosity parameters are connected to the mud sensor in real time; ③The dynamics of the drill string is modeled using the Lagrange equation, integrating the torque-speed curve provided by the rotary drilling rig manufacturer; S4.2.2: Set the finite element mesh adaptive division rule, the mesh size near the hole wall ≤ 5mm, and the far field area ≤ 20mm; S4.2.3: Define the data-driven interface, receive ultrasonic hole diameter data, mud density meter readings, and drill pressure sensor data every 10 seconds, and update the boundary conditions of the digital twin; S4.2.4: Use an explicit-implicit hybrid solver, set the calculation step to 0.1 second, and output the hole wall stress distribution and mud flow field prediction in real time.

9. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 7, characterized in that, In S4.3, a PPO algorithm with the filling coefficient deviation as the reward function is designed and pre-trained for 100,000 iterations in the digital twin environment. The specific steps are as follows: S4.3.1: Define the reasonable range for the fullness coefficient as 1.05-1.15, and set the target value C. target =1.10, calculate the real-time deviation e = |C t -1.10|; S4.3.2: Construct a piecewise reward function: ①When e ≤ 0.03, r = 10 - 333e; ②When 0.03 < e ≤ 0.10, r = 5 - 50e; ③When e > 0.10, r = -20e; Simultaneously introduce the hole wall stability penalty term p = 0.5 × the hole collapse risk index, and the final reward R = r - p; S4.3.3: The policy network adopts a 3-layer fully connected structure, and the output layer is activated by Softmax. The action space includes: ①Adjustment of the dispersant dosage Δw1 ∈ [-0.2%, +0.2%]; ②Adjustment of the thickener dosage Δw2 ∈ [-0.1%, +0.1%]; ③Adjustment of the drilling speed Δv ∈ [-5 rpm, +5 rpm]; S4.3.4: The PPO algorithm sets the clipping parameter ∈1 = 0.2, the discount factor γ = 0.95, the GAE parameter λ = 0.9, and the objective function is: S4.3.5: Training in stages: In the first 30,000 iterations, a random exploration strategy is adopted. In the middle 50,000 iterations, the policy certainty is gradually improved. In the last 20,000 iterations, formation migration samples are introduced for enhancement, and the model parameters are saved every 500 iterations.

10. A method for dynamically adjusting mud cake in different strata of pile foundations according to claim 1, characterized in that, In S5, combined with the entity project data, optimize the parameters of the casing burial depth, bit selection, and rotary drilling process to form a standardized mountain construction plan. The specific steps are as follows: S5.1: Deploy Internet of Things sensors at typical work points on the Weiyi Expressway to record the parameters of the casing vibration frequency, bit wear rate, and rotary drilling torque in real time; S5.2: Use the orthogonal test method to analyze the contribution of the casing burial depth, bit type, and rotation speed to the filling coefficient; S5.3: Establish a decision tree model based on data mining: When encountering a sand layer and the groundwater level < 2m, automatically trigger the strategy of "deepening the casing to 2.5m + cone bit + reducing the speed to 18 rpm"; S5.4: Compile the "Mountain Pile Foundation Construction Technology Standard" including 3 types of formations and 5 abnormal working conditions, and stipulate that the "follow-up casing + impact bit" combination must be used in the karst area.

Images