A method for ground stress perception and inversion while drilling

Abstract

The application discloses a ground stress while drilling touch sensing and inversion method. The method comprises the following steps: preliminary estimated values of ground stress and rock mechanics parameters are respectively analyzed based on while drilling data through two parallel physical mechanism models; the preliminary estimated values, original data and known constants are input into a pre-trained residual deep learning network to learn system errors of the physical model and unmodeled dynamic characteristics, and high-precision residual correction values are output; and the preliminary estimated values and the residual correction are added to obtain final inversion results. The application realizes complementary advantages of physical interpretability and data-driven adaptability, solves problems of insufficient precision, data dependence and parameter inversion fragmentation of traditional methods, and can provide real-time, continuous and high-precision in-situ stratum information profile for underground engineering, and significantly improves construction safety and intelligent level.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of mining engineering, geotechnical engineering monitoring and intelligent sensing, and specifically relates to a method for sensing and inverting ground stress during drilling. Background Technology

[0002] With the increasing depletion of shallow mineral resources, mineral resource extraction and underground engineering construction are developing towards deeper deposits. In-situ geostress parameters of deep ore and rock are important indicators for preventing geological disasters in deep engineering projects, optimizing construction techniques, and improving mining efficiency.

[0003] Traditional methods for testing in-situ stress (such as stress relief methods and hydraulic fracturing methods) and rock mechanics parameters (such as indoor core tests) suffer from drawbacks such as being time-consuming and offline, highly destructive, and unable to reflect real-time field conditions. Monitoring While Drilling (MWD) technology provides a new approach for acquiring formation information in real time. Currently, MWD inversion methods are mainly divided into two categories: one is physical model-based methods, which invert formation parameters by establishing a physical and mechanical model of the drill bit-rock interaction; the other is purely data-driven methods, such as machine learning and deep learning models, which directly learn mapping relationships from data. However, the former, a single physical model, is often based on simplified and idealized assumptions (such as homogeneous, isotropic, and fixed failure modes of the rock mass). Its accuracy is limited under complex geological conditions, making it difficult to accurately characterize complex geological conditions and dynamic fracturing processes, leading to significant model errors. The latter, purely data-driven methods, heavily rely on high-quality, large-sample labeled data (i.e., simultaneously including MWD data and accurate true values ​​of stratigraphic parameters). Furthermore, the model is a "black box," and the prediction results lack physical interpretability. When data is scarce or the working conditions exceed the training set range, the generalization ability and reliability drop sharply. In addition, existing methods often independently invert geostress or rock mechanical parameters, ignoring the inherent physical relationship between the two, resulting in inconsistent inversion results.

[0004] Therefore, there is an urgent need in this field to develop a high-precision collaborative inversion method that can integrate prior physical knowledge to ensure the extrapolation capability and interpretability of the model, while using data intelligence to correct model errors and adapt to complex situations. This method has significant theoretical value and promising engineering applications. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method for sensing and inverting ground stress during drilling, which achieves the unity of physical interpretability and data adaptability through a hybrid intelligent paradigm of "physical mechanism as a first constraint and intelligent data residual correction".

[0006] This invention provides a method for sensing and inverting ground stress during drilling, specifically: first, drilling parameters are collected, and ground stress and rock mechanical parameters are initially inverted based on the drilling parameters through a physical mechanism model; then, a residual learning network is constructed and trained, and the trained residual learning network is used to intelligently correct the initial estimate, and finally, a high-precision inversion result is output.

[0007] The drilling parameters are acquired during the drilling process through a sensor array on the drill string, including time-series drilling mechanical parameters and motion parameters.

[0008] The physical mechanism model includes a first physical model and a second physical model; the first physical model is based on the theory of normal intrusion and tangential cutting of cutting tools, and obtains preliminary rock mechanical property parameters by analytical inversion of the input drilling data. The second physical model, based on the stress balance analysis of the cutter-compression fracture zone (RCZ)-surrounding rock system, the Mohr-Coulomb failure criterion, and stress coordinate transformation, constructs a closed analytical equation set regarding the principal components of geostress, and uses Cramer's rule for rapid solution to obtain preliminary geostress estimates. The inputs to the second physical model include drilling data and the outputs of the first physical model.

[0009] The first physical model is represented by the following formula: ;in, It is the cohesive force within the rock; T is the drilling propulsion force for the cutting teeth, and T is the drilling torque. Where is the radius of the drill bit's geometric rotation center; N is the rotational speed; V is the drilling speed. The internal friction angle between the rock compression and fracture zone and the intact rock. The internal friction angle of the rock. denoted as the coefficient of friction between the drill bit and the rock contact surface, m as the number of drill bit cutting teeth, and S as the comprehensive cutting parameters;

[0010] Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

[0011] The second physical model is represented by the following formula:

[0012] ; ;in, The cutting tooth penetration angle; For intact rock breaking angle; Indicating the direction of ground stress; The internal friction angle of the rock; The cutting tooth force received by the rock in the compression and fracturing zone; It is the cohesive force within the rock; This represents the theoretical maximum principal stress. d represents the theoretical minimum principal stress; d is the width of the cutting tooth. The maximum principal stress transfer factor, The minimum principal stress transfer factor is calculated using the following formula:

[0013] ; .

[0014] The residual deep learning network described above uses a Temporal Convolutional Network (TCN) as its core. The input layer of this network is a multi-channel layer, with channels including those for preliminary inversion of multi-dimensional parameter temporal sequences. Time series of drilling data parameters Known constant timing ;in, It is the angle of inclination; It is the azimuth angle; These are the dimensions of the cutting teeth.

[0015] The residual deep learning network uses mean squared error loss as the main loss function, and then introduces a physical information regularization term to obtain the final total loss function. Express it using the following formula: ;in, This is the mean square error loss; For hyperparameters; For physical information regularization terms;

[0016] The physical information regularization term The residual is obtained by feeding the output of the residual deep learning network into the physical mechanism model to calculate the residual.

[0017] The residual deep learning network is trained offline using publicly available historical datasets; at the same time, new measured times are obtained as fine-tuning samples to perform online incremental learning on the residual deep learning network.

[0018] The present invention also provides a system for realizing the in-situ stress sensing and inversion method during drilling, including an intelligent sensing drill bit subsystem, an edge computing and acquisition terminal, a cloud-based intelligent inversion server, and an interactive and visualization platform;

[0019] The intelligent sensing drilling tool subsystem includes a specially designed cutting tooth (integrated with a triaxial force sensor), a drill pipe (integrated with a speed-torque coincidence sensor), a propulsion mechanism (integrated with a pressure and displacement wire sensor), and an attitude measurement unit (IMU). It collects drilling data by drilling the target area and uploads the data to the edge computing and acquisition terminal.

[0020] The edge computing and acquisition terminal is responsible for synchronously acquiring, preprocessing, and caching the data uploaded by the intelligent sensing drilling tool subsystem, running the first and second physical models to achieve preliminary inversion, and uploading the results to the cloud intelligent inversion server;

[0021] The cloud-based intelligent inversion server deploys a pre-trained residual deep learning network, receives data packets uploaded from edge computing and acquisition terminals, performs high-precision residual prediction and result fusion calculation, and uploads the results to the interactive and visualization platform; it also manages model update tasks.

[0022] The interactive and visualization platform is used to receive the output of the cloud-based intelligent inversion server and to visualize, store, analyze, and provide early warnings of the final inversion results.

[0023] This invention discloses a method for sensing and inverting ground stress during drilling. Through a hybrid intelligent paradigm of "physical mechanism as a first constraint and intelligent data residual correction", it achieves the goal of unifying physical interpretability and data adaptability. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating the overall workflow of the method of this invention;

[0025] Figure 2 This is a schematic diagram of the first physical model construction principle of the method of the present invention; wherein, (a) is a schematic diagram of the force analysis of the cutting tooth intruding into the broken rock; (b) is a schematic diagram of the force analysis of the cutting tooth; (c) is a schematic diagram of the force analysis of the broken rock; and (d) is a schematic diagram of the force analysis of the intact rock.

[0026] Figure 3 This is a schematic diagram illustrating the construction principle of the second physical model of the method of the present invention;

[0027] Figure 4 This is a force analysis diagram of the contact interface of the core drill bit in the second physical model of the method of the present invention;

[0028] Figure 5 This is a schematic diagram of the residual temporal convolutional network of the method of the present invention;

[0029] Figure 6This is a schematic diagram of the system hardware and inverted geostress profile in an embodiment of the present invention, wherein the markings are as follows: 1, drill rod; 2, rotational speed-torque composite sensor; 3, wire displacement sensor; 4, stress sensor; 5, cutting tooth stress sensor. Detailed Implementation

[0030] This invention provides a method for sensing and inverting in-situ stress during drilling, the flowchart of which is shown below. Figure 1 As shown, the specific steps are as follows: First, drilling parameters are collected, and the in-situ stress and rock mechanics parameters are initially inverted based on the drilling parameters using a physical mechanism model; then, a residual learning network is constructed and trained, and the trained residual learning network is used to intelligently correct the initial estimates, finally outputting high-precision inversion results.

[0031] The drilling parameters are acquired during the drilling process through a sensor array on the drill string, including time-series drilling mechanical parameters and motion parameters.

[0032] The physical mechanism model includes a first physical model and a second physical model; the first physical model is based on the theory of normal intrusion and tangential cutting of cutting tools, and obtains preliminary rock mechanical property parameters by analytical inversion of the input drilling data. Its principle diagram is as follows Figure 2 As shown; where, The cutting teeth are subjected to a tangential force from mechanical transmission; The cutting tooth is subjected to the normal force of mechanical transmission; The resultant force at the center of the cutting tooth particle; This is the decomposition of the resultant force of the cutting teeth along the x-axis tangentially; This is the decomposition of the resultant force of the cutting teeth along the y-axis normal. This refers to the cutting tooth penetration angle (an inherent characteristic of the cutting tooth). This represents the normal reaction force of the rock fracture zone (RCZ) on the contact surface of the cutting teeth; This refers to the tangential frictional force between the rock fracture zone (RCZ) and the contact surface of the cutting teeth; The normal supporting force of the intact mother stone on the bottom end of the cutting tooth; The tangential frictional resistance of the intact parent rock to the bottom of the cutting tooth intrusion; The normal supporting force of the parent rock on the fractured zone (RCZ); The tangential frictional resistance between the lower parent rock and the fractured zone (RCZ); This represents the normal thrust at the interface between the intact parent rock and the fractured zone (RCZ). η represents the frictional resistance at the interface between the intact parent rock and the fractured rock zone (RCZ); h represents the unit intrusion depth of the cutting tooth. For rock breaking angle; The normal stress on the rock failure surface; This represents the tangential stress at the rock fracture surface. The cutting width is equal to the drill bit's outer diameter minus its inner diameter;

[0033] Figure 2 This invention presents a force analysis of the cutting and rock-breaking process using cutting tools, based on classical rock mechanics principles. During rock breaking, the core drilling rig uses pressure applied by the propulsion cylinder to keep the cutting tools in close contact with the rock, and uses torque to drive the drill bit's rotation. The rock is broken based on the coupling effect of propulsion pressure and rotational torque. To simplify the analysis, this invention decomposes the cutting tool penetration and rock-breaking process into a cutting tool penetration process and a rotational cutting process, and performs force analysis on each separately to analyze the physical relationship between rock properties and mechanical parameters. The cutting tool penetration process is similar to the penetration of a rigid indenter; during the rotary cutting process, the cutting tool's front face compresses the intact rock in front, and the rock in front, under the contact pressure of the cutting tool, will undergo shear failure along the failure surface, forming a rock failure zone (RBZ). Finally, force balance equations are established by separately analyzing the cutting tool, the broken rock, and the intact rock to obtain the mapping relationship between rock properties and the forces acting on the cutting tool.

[0034] The second physical model is based on the stress balance analysis of the cutter-compression fracture zone (RCZ)-surrounding rock system, the Mohr-Coulomb failure criterion, and stress coordinate transformation. It constructs a closed analytical equation set about the principal components of geostress and uses Cramer's rule for fast solution to obtain preliminary geostress estimates. Its principle diagram is as follows Figure 3 and Figure 4 As shown; the inputs to the second physical model include drilling data and the output of the first physical model.

[0035] Figure 4 For the stress analysis of the contact interface of the coring drill bit, it is known that due to the stress relief effect of coring, the rock mass in front of the coring drill bit will be under plane stress, and there exists a maximum principal stress. and minimum principal stress The biaxial force acting on the cutting tooth assumes an angle between the direction of the maximum principal stress and the specified system coordinate system. Based on this, a force model for the drilling cutting tooth is established, and the relationship between the force on the cutting tooth and the maximum and minimum principal stresses is analyzed according to the force equilibrium equation. All markings in the figure represent force symbols. The inner diameter of the drill bit; The angle between the direction of the maximum principal stress in the ground and the reference coordinate system; The normal compressive stress of the cutting teeth on the fractured zone (RCZ) of the rock; The tangential stress of the cutting tooth on the rock fracture zone (RCZ); The normal stress of the rock fracture zone (RCZ) on the intact rock in front of the drill bit outer wall; The tangential stress in the rock fracture zone (RCZ) on the intact rock in front of the drill bit outer wall; The normal stress of the rock fracture zone (RCZ) on the intact rock in front of the drill bit inner wall; The tangential stress in the rock fracture zone (RCZ) on the intact rock in front of the drill bit inner wall; The resultant force of radial stress on the rock stress element along the inner wall of the drill bit, representing the maximum and minimum principal stresses; The resultant force of radial stress on the rock stress element along the inner wall of the drill bit, representing the maximum and minimum principal stresses; Let be the rotation angle of a certain analytical single-tooth section relative to the reference coordinate system.

[0036] The first physical model is represented by the following formula:

[0037] ;in, It is the cohesive force within the rock; T is the drilling propulsion force for the cutting teeth, and T is the drilling torque. Where is the radius of the drill bit's geometric rotation center; N is the rotational speed; V is the drilling speed. The internal friction angle between the rock compression and fracture zone and the intact rock. The internal friction angle of the rock. denoted as the coefficient of friction between the drill bit and the rock contact surface, m as the number of drill bit cutting teeth, and S as the comprehensive cutting parameters;

[0038] Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

[0039] The second physical model is represented by the following formula:

[0040] ; ;in, The cutting tooth penetration angle; For intact rock breaking angle; Indicating the direction of ground stress; The internal friction angle of the rock; The cutting tooth force received by the rock in the compression and fracturing zone; It is the cohesive force within the rock; This represents the theoretical maximum principal stress. d represents the theoretical minimum principal stress; d is the width of the cutting tooth. The maximum principal stress transfer factor, The minimum principal stress transfer factor is calculated using the following formula:

[0041] ; .

[0042] The residual deep learning network described above uses a Temporal Convolutional Network (TCN) as its core, and its network structure diagram is shown below. Figure 5 As shown; the input layer of the residual deep learning network is a multi-channel input layer, with channels including preliminary inversion of multi-dimensional parameter timing. Time series of drilling data parameters Known constant timing ;in, It is the angle of inclination; It is the azimuth angle; These are the dimensions of the cutting teeth.

[0043] The residual deep learning network uses mean squared error loss as the main loss function, and then introduces a physical information regularization term to obtain the final total loss function. Express it using the following formula: ;in, This is the mean square error loss; For hyperparameters; For physical information regularization terms;

[0044] The physical information regularization term The residual is obtained by feeding the output of the residual deep learning network into the physical mechanism model to calculate the residual.

[0045] The residual deep learning network is trained offline using publicly available historical datasets; at the same time, new measured times are obtained as fine-tuning samples to perform online incremental learning on the residual deep learning network.

[0046] The present invention also provides a system for realizing the in-situ stress sensing and inversion method during drilling, including an intelligent sensing drill bit subsystem, an edge computing and acquisition terminal, a cloud-based intelligent inversion server, and an interactive and visualization platform;

[0047] The intelligent sensing drilling tool subsystem includes a specially designed cutting tooth (integrated with a triaxial force sensor), a drill pipe (integrated with a speed-torque coincidence sensor), a propulsion mechanism (integrated with a pressure and displacement wire sensor), and an attitude measurement unit (IMU). It collects drilling data by drilling the target area and uploads the data to the edge computing and acquisition terminal.

[0048] The edge computing and acquisition terminal is responsible for synchronously acquiring, preprocessing, and caching the data uploaded by the intelligent sensing drilling tool subsystem, running the first and second physical models to achieve preliminary inversion, and uploading the results to the cloud intelligent inversion server;

[0049] The cloud-based intelligent inversion server deploys a pre-trained residual deep learning network, receives data packets uploaded from edge computing and acquisition terminals, performs high-precision residual prediction and result fusion calculation, and uploads the results to the interactive and visualization platform; it also manages model update tasks.

[0050] The interactive and visualization platform is used to receive the output of the cloud-based intelligent inversion server and to visualize, store, analyze, and provide early warnings of the final inversion results.

[0051] The method of the present invention will be further described below with reference to an embodiment:

[0052] The implementation of this invention is a systematic project that requires deployment, debugging and operation in accordance with strict construction procedures to ensure the reliability and accuracy of the intelligent inversion system on site.

[0053] Figure 6 The diagram illustrates the specific construction process and the final visualization. The main image shows the drilling process, displaying the drilling sensors. The lower left corner shows the stress sensor of the coring bit. The lower right corner visualizes the drilling parameter collection process, including curves showing the changes in various mechanical data collection processes. The upper right corner shows the inverse geostress information derived from the drilling depth, including the magnitude and orientation of the maximum and minimum principal stresses. The diagram below shows the size and orientation of the stress ellipse at the corresponding cross section during the drilling progress.

[0054] The specific construction process can be divided into the following four stages:

[0055] S1. Construction Preparation and System Deployment

[0056] On-site investigation and scheme design: Collect geological data and drilling design for the work area, and clarify the monitoring objectives; based on the borehole diameter, drilling rig model and rock mass hardness, customize or select suitable intelligent sensing drilling tool subsystems (including sensing cutters, composite sensor arrays, etc.).

[0057] Hardware installation and integration: Install the special sensing cutting teeth onto the drill bit, fix the speed-torque composite sensor at the designated position on the drill rod, and install the wire displacement sensor on the propulsion mechanism; calibrate all sensors for force, torque, and displacement in the laboratory or on-site to establish accurate signal-physical quantity conversion relationships;

[0058] Software system initialization: Start the data acquisition and preprocessing software, the calculation programs for the first and second physical mechanism models, and the residual TCN model.

[0059] S2, System Integration and Parameter Calibration

[0060] No-load and simulation test: Start all equipment and test the integrity of the complete link of sensor data acquisition, transmission and reception without drilling. Use historical data or simulated data stream to test the operation of physical model, cloud residual network, result fusion and visualization system.

[0061] Trial drilling and field calibration: Trial boreholes are drilled in sections of known lithological zones. The system operates in real time, collecting drilling data and outputting preliminary inversion results. Simultaneously, core samples are taken from the trial borehole sections for laboratory rock mechanics tests (uniaxial compression, etc.) and in-situ stress measurements (hydraulic fracturing, etc.). The "true values" obtained in the laboratory are compared and analyzed with the inversion results of the system for that section. Based on the deviations, the empirical coefficients in the physical model (friction coefficient μRZ, parameters in the comprehensive cutting parameters S) and the input feature weights of the residual network are calibrated and fine-tuned in the field to adapt the system to the geological conditions and drilling tool conditions.

[0062] Model warm-up and confidence level establishment: Using the calibrated data and model, the system is run in the initial formal drilling section, and the inversion results are used as the initial benchmark; through initial core sampling verification, the initial confidence level of the system in this work area is evaluated, providing a reliable basis for subsequent full-hole continuous inversion.

[0063] S3, Formal Drilling and Real-Time Intelligent Inversion

[0064] Standardized drilling operation: Operators carry out drilling construction according to conventional procedures, while the intelligent system runs automatically in the background, keeping the preset drilling parameters relatively stable within a reasonable range and avoiding drastic fluctuations to ensure the quality of the inverted data sequence.

[0065] Fully automated inversion execution: After real-time sensor data is preprocessed by the model, it is input into the first and second physical models in parallel, and preliminary rock mechanical property parameters are output within seconds. and preliminary geostress estimates The preliminary estimates and the original data packets are then input into a pre-trained TCN residual network, which outputs correction values ​​for each parameter. The cloud-based system integrates the correction values ​​with the preliminary estimates to obtain a high-precision final result, which is then sent to the visualization platform in real time.

[0066] Real-time monitoring and decision support: Based on the visualization platform, the "hole depth-mechanical parameters" profile and the direction of in-situ stress can be monitored in real time to guide the optimization of the process on site.

[0067] S4, Continuous Model Optimization

[0068] Incremental learning and updating of the model: When drilling at a fixed distance or encountering special rock formations, core samples are taken for verification, and new "measured-inverted" data pairs are added to the sample library; when a certain number of new samples are accumulated, the incremental learning process of the residual TCN network is started in the cloud, and the network empirical parameters are fine-tuned with new data, so that the model continues to evolve and becomes more and more adapted to the specific geological conditions of this project.

[0069] Data archiving and result generation: The system automatically stores all raw drilling data, intermediate inversion results, and final result data. The trained and optimized model is archived to form a "knowledge model" applicable to the mining area. When conducting subsequent projects in the same mining area, the system can be directly invoked or quickly initialized based on the existing "knowledge model," enabling knowledge accumulation and transfer, and significantly improving the efficiency and accuracy of subsequent project implementation.

Claims

1. A method for sensing and inverting in-situ stress during drilling, characterized in that, include: First, drilling parameters are collected, and then the in-situ stress and rock mechanics parameters are initially inverted based on the drilling parameters using a physical mechanism model. Then, a residual learning network is constructed and trained. The trained residual learning network is used to intelligently correct the preliminary estimate, and finally, a high-precision inversion result is output. The physical mechanism model includes a first physical model and a second physical model; the first physical model is based on the theory of normal intrusion and tangential cutting of cutting tools, and obtains preliminary rock mechanical property parameters by analytical inversion of the input drilling data. ,in, It is the cohesive force within the rock; The internal friction angle of the rock is given. The second physical model is based on the force balance analysis of the cutter-compression fracture zone-surrounding rock system, the Mohr-Coulomb failure criterion, and stress coordinate transformation. A closed analytical equation system of principal components of geostress is constructed, and Cramer's rule is used for rapid solution to obtain preliminary geostress estimates. ,in, This represents the theoretical maximum principal stress. The theoretical minimum principal stress is the value; the inputs to the second physical model include drilling data and the outputs of the first physical model.

2. The method for sensing and inverting in-situ stress during drilling as described in claim 1, characterized in that, The drilling parameters are acquired during the drilling process through a sensor array on the drill string, including time-series drilling mechanical parameters and motion parameters.

3. The method for sensing and inverting in-situ stress during drilling as described in claim 2, characterized in that, The first physical model is represented by the following formula: ;in, T is the drilling propulsion force for the cutting teeth, and T is the drilling torque. Where is the radius of the drill bit's geometric rotation center; N is the rotational speed; V is the drilling speed. The internal friction angle between the rock compression and fracture zone and the intact rock. denoted as the coefficient of friction between the drill bit and the rock contact surface, m as the number of drill bit cutting teeth, and S as the comprehensive cutting parameters; Combined cutting parameters S and tooth entry angle Coefficient of friction of rock contact surface and the internal friction angle between the rock compression fracture zone and the intact rock. Regarding the specific calculation formula: .

4. The method for sensing and inverting in-situ stress during drilling as described in claim 3, characterized in that, The second physical model is represented by the following formula: ; ;in, The cutting tooth penetration angle; For intact rock breaking angle; Indicating the direction of ground stress; d represents the cutting tooth force received by the rock compression and fracturing zone; d is the cutting tooth width. The maximum principal stress transfer factor, The minimum principal stress transfer factor is calculated using the following formula: ； 。 5. The method for sensing and inverting in-situ stress during drilling as described in claim 1, characterized in that, The residual learning network described above uses a temporal convolutional network as its core. The input layer of the residual learning network is a multi-channel input layer, with channels including those for preliminary inversion of multi-dimensional parameter temporal sequences. Time series of drilling data parameters Known constant timing ;in, It is the angle of inclination; It is the azimuth angle; Here are the cutting tooth dimensions; T is the drilling torque; N is the rotational speed; V is the drilling rate; and F is the thrust. The internal friction angle between the rock compression and fracture zone and the intact rock; The coefficient of friction between the drill bit and the rock contact surface.

6. The method for sensing and inverting in-situ stress during drilling as described in claim 1, characterized in that, The residual learning network uses mean squared error loss as the main loss function, and then introduces a physical information regularization term to obtain the final total loss function. Express it using the following formula: ;in, This is the mean square error loss; For hyperparameters; This refers to the physical information regularization term; the physical information regularization term The residual is obtained by inputting the output of the residual learning network into the physical mechanism model.

7. The method for sensing and inverting in-situ stress during drilling as described in claim 1, characterized in that, The residual learning network is trained offline using publicly available historical datasets; at the same time, new measured times are obtained as fine-tuning samples to perform online incremental learning on the residual learning network.

Images