A deep three-dimensional ground stress in-situ testing robot

Abstract

This invention relates to a deep three-dimensional geostress in-situ testing robot based on borehole indentation method and its in-situ collaborative identification method. The in-situ testing robot includes a bridle connection mechanism, an electronic control system section, an attitude and environment perception section, a pressure balance section, a hydraulic power system section, a push support section, and an indentation execution section. The above components are connected in series and all have a cylindrical sealed structure. Through integrated design, high-precision indentation execution mechanism, and data-physical dual-drive intelligent inversion, rapid in-situ collaborative identification of three-dimensional geostress and rock mechanical parameters in small-diameter boreholes can be achieved.

Description

Technical Field

[0001] This invention application relates to the field of in-situ testing technology for deep rock mass engineering, and in particular to a testing robot for in-situ testing of three-dimensional geostress and rock mass mechanical parameters in deep boreholes using the indentation method. Background Technology

[0002] With the deepening implementation of my country's "deep earth" strategy, deep rock engineering is playing an increasingly important role in resource development, energy reserves, and major infrastructure construction. Pumped storage power stations, deep mineral resource mining, nuclear waste geological disposal, and carbon sequestration projects are rapidly developing towards maximum burial depth, massive scale, and complex structures. For example, the main powerhouses of pumped storage power stations are generally buried at depths exceeding 300 meters, with some reaching 500-800 meters, spanning over 30 meters and exceeding 70 meters in height; deep metal mines have reached depths exceeding 2000 meters; and oil and gas reservoirs in the Tarim Basin are buried at depths exceeding 8000 meters. With increasing depth, the rock mass is generally in a "three-high" environment—high ground stress (often exceeding 30 MPa), high ground temperature (up to 125℃), and high osmotic pressure. This leads to a significant increase in the frequency and intensity of engineering disasters such as rock bursts, rock bursts, large deformations of surrounding rock, and well wall collapses, seriously threatening construction safety and long-term stability. Accurately obtaining the deep three-dimensional geostress state and in-situ mechanical parameters of rock mass has become a fundamental prerequisite for engineering safety design and disaster prevention.

[0003] Currently, mainstream geostress testing methods both domestically and internationally include hydraulic fracturing, traditional stress relief methods (such as the hollow inclusion method), local wall stress relief method (BWSRM), and inelastic strain recovery method. However, these methods have the following significant shortcomings when applied to small-diameter deep boreholes (φ90~130mm): 1. Poor environmental adaptability: Deep boreholes are generally filled with high-density drilling fluid, with pressures exceeding 50 MPa and temperatures exceeding 100°C. Furthermore, the borehole walls are often covered with rock cuttings and have irregular undulations. Traditional strain gauge bonding testing methods have extremely low success rates in this environment, and may even be impossible to implement.

[0004] 2. Reliance on intact rock cores: Most methods require obtaining high-quality rock cores for laboratory testing (such as elastic constants). After the rock core is taken out, the stress is completely released, and the measured parameters cannot represent the in-situ state. Moreover, it is often difficult to obtain cores from deep fractured strata.

[0005] 3. Excessive equipment diameter: Existing advanced equipment such as BWSRM has an outer diameter of more than 100mm, which makes it impossible to enter conventional geological exploration boreholes, thus limiting its widespread application.

[0006] 4. Limited parameter acquisition: A single test can usually only obtain in-situ stress or rock mass deformation parameters, lacking the ability for synchronous and collaborative identification, resulting in poor physical consistency of the data.

[0007] 5. Low testing efficiency: The single-hole testing cycle of the hydraulic fracturing method is 3 to 5 days, and the traditional stress relief method has a complicated procedure, which is difficult to meet the timeliness requirements of advanced geological prediction for rapid construction.

[0008] Indentation, as a minimally invasive mechanical testing technique, has demonstrated unique advantages in rock mechanics in recent years—by recording the load-displacement curves during indentation, it can simultaneously infer the material's hardness, modulus, and residual stress. Successful application to deep borehole environments could potentially overcome the aforementioned bottlenecks. However, directly applying classical indentation theory (based on the assumption of semi-infinite space) to the borehole surface boundaries and high-stress environments will cause the theoretical model to fail. Therefore, there is an urgent need to develop a downhole testing robot specifically designed for small-diameter deep boreholes, integrating indentation testing principles, and possessing environmental adaptability and intelligent identification capabilities. Summary of the Invention

[0009] This invention addresses the technical problems existing in the prior art by providing a testing robot and its collaborative identification and usage method for in-situ testing of three-dimensional geostress and rock mechanics parameters in deep boreholes using the indentation method. This solves the problems of large diameter, poor environmental adaptability, single testing parameters, and low efficiency of existing deep borehole geostress testing equipment.

[0010] Specifically, a deep three-dimensional in-situ stress testing robot based on borehole indentation method is provided, comprising a bridle connection mechanism, an electronic control system section, an attitude and environment perception section, a pressure balancing section, a hydraulic power system section, a push-support section, and an indentation execution section, all connected in series and forming a cylindrical sealed structure; wherein... The push-support section includes an upper push-support group and a lower push-support group distributed along the robot axis. The upper push-support group consists of three first hydraulic push-support arms distributed at 120° intervals around the circumference. The lower push-support group consists of two second hydraulic push-support arms. The circumferential angle between the two second hydraulic push-support arms is 120°, and together with the orientation of the indentation probe of the indentation execution section, they form a triangular support. Each push-support arm is provided with a rigid support foot at its end. The indentation execution section includes a linear loading unit, an angle rotation unit, and an integrated indentation probe assembly. The linear loading unit includes a first servo motor, a precision ball screw, a screw nut, a wedge block, a linear guide rail, and a high-precision displacement sensor. The first servo motor drives the ball screw to rotate, and the screw nut drives the wedge block to move along the linear guide rail. The inclined surface of the wedge block slides in engagement with the base of the indentation probe assembly, converting the motion along the robot's axial direction into a radial linear pressing motion of the indentation probe. The angle rotation unit includes a second servo motor, a reducer, a drive gear, and a slewing bearing, driving the indentation probe assembly to rotate around the robot's axis. The integrated indentation probe assembly integrates a diamond or cemented carbide indenter, a miniature force sensor, and a displacement sensor, and is completely sealed. The electronic control system subsection includes a high-temperature resistant main control module, an ADC acquisition module, a Manchester encoding / decoding communication module, and a motor drive module, which are used to receive ground commands, acquire multiple sensor signals, and control various actuators. The attitude and environment perception section integrates a MEMS solid-state gyroscope, a three-axis accelerometer, a temperature sensor, and a high-pressure resistant high-definition camera to acquire robot attitude, drilling environment temperature, and borehole wall images in real time, and to help select intact rock surfaces as test points.

[0011] Furthermore, the hydraulic power system subsection includes a high-temperature resistant micro motor, a planetary reducer, a high-pressure piston pump, an electrically controlled relief valve, a solenoid directional valve, and an accumulator.

[0012] Furthermore, in the angle rotation unit, the second servo motor, in conjunction with the reducer and the drive gear, drives the slewing bearing, and the rotation angle accuracy is ≤±1°. At least three test orientations can be achieved within the same test section. The smaller circumferential angle between any two test orientations is 40°~120°. Preferably, the three test orientations are evenly distributed along the circumference and the adjacent angles are 120°.

[0013] Furthermore, in the attitude and environment perception section, the MEMS inertial measurement unit includes a MEMS gyroscope and a three-axis accelerometer, with an azimuth measurement error of <5° and a tilt measurement accuracy of ±1° for the three-axis accelerometer; the high-voltage resistant high-definition camera has a resolution of 1080P and is equipped with an LED illumination array.

[0014] Furthermore, the pressure balancing sub-section has a piston-type structure, with one side connected to the borehole annulus environment through a filter mesh, and the other side connected to the hydraulic oil circuit of the hydraulic power system sub-section, with a pressure balancing sensitivity ≤0.1MPa.

[0015] In addition, this application proposes a method for using any of the above-mentioned in-situ testing robots, characterized by comprising the following steps: Step S1: Lower the robot to the target depth inside the borehole via cable, obtain the depth through the ground winch depth measurement system, acquire azimuth, dip, temperature and borehole wall video images using the attitude and environment perception sub, and identify and screen test points with intact rock mass and flat borehole wall through the ground control system or downhole edge computing module. Step S2: The ground sends a command, and the hydraulic power system short section drives the three arms of the upper push support group of the push support short section to extend synchronously. By comparing the values ​​of the displacement sensors of each arm, the extension amount of each arm is automatically adjusted so that the robot axis coincides with the drilling axis, realizing active centering; then the upper and lower push support groups are simultaneously supported on the hole wall to provide the reaction force support required for testing. Step S3: In the anchored state, the short section angle rotation unit of the indentation execution section drives the indentation probe assembly to rotate to the first test orientation. The linear loading unit drives the indenter to press into the borehole wall rock at a constant rate in displacement control mode, and load-displacement curves are collected simultaneously. After unloading, the indenter angle rotation unit drives the indenter to rotate to the second and third test angles, which are 45° apart. The indentation test is repeated to obtain the load-displacement curves at different angles in the first test orientation. After completing the indentation test in the first test orientation, the short section angle rotation unit drives the indentation probe assembly to rotate to the second and third test orientations, which are 120° apart. The indentation test is repeated. Step S4: The acquired load-displacement curves are uploaded to the ground via cable and input into the pre-trained data-physics dual-drive intelligent inversion model. This model uses the elastoplastic indentation theory model as a physical constraint embedded in the neural network to collaboratively invert the three-dimensional geostress components and in-situ mechanical parameters of the rock mass at the test point. Step S5: Retract the push arm, move to the next depth measuring point, repeat steps S2 to S4, and after completing the full hole multi-point test, retract all actuators and remove the robot from the drilling hole.

[0016] Furthermore, in step S1, the borehole wall image transmitted in real time by a high-definition camera is used to automatically identify defects such as joints, fissures, and fracture zones using an image recognition algorithm, and to automatically evaluate the rock mass quality at the test point; flat rock surfaces in the borehole wall image where the area ratio of joints, fissures, and fracture zones is lower than a preset threshold are selected as test points.

[0017] Furthermore, in step S2, when it is necessary to change the test orientation within the same depth section, the upper push-support group is kept taut, and only the lower push-support group needs to be retracted. The angle rotation unit drives the indentation probe assembly to rotate around the robot axis to the target orientation, and then the lower push-support group is taut again. There is no need to move the entire robot.

[0018] Furthermore, the data-physical dual-driven intelligent inversion model in step S4 is a multi-task learning neural network. Its input consists of load-displacement curves and their derived features (slope of the loading segment, unloading stiffness, yield point load, plastic work, and maximum indentation depth) at at least three orientations of the same depth section. The output includes the rock mass elastic modulus, strength parameters, and three-dimensional geostress components. The model loss function includes a physical regularization term derived from the indentation theory model, which is used to constrain the mechanical self-consistency of the prediction results.

[0019] This invention, through integrated design, a high-precision indentation actuator, and data-physical dual-drive intelligent inversion, enables rapid in-situ collaborative identification of three-dimensional geostress and rock mechanical parameters within small-diameter boreholes. Compared with existing technologies, this invention offers the following significant advantages: 1. Small diameter and wide adaptability: The robot has a maximum outer diameter of only 75mm, which can enter conventional geological exploration boreholes (φ90~130mm), filling the gap that existing equipment such as BWSRM cannot enter small diameter boreholes.

[0020] 2. Extreme environment tolerance: Withstands pressure of 50MPa and temperature of -10 to 125℃. With a fully sealed design and a dual-stage sealing structure, it can work stably for a long time in the open hole section of deep borehole filled with drilling fluid.

[0021] 3. In-situ integrated collaborative identification: No core sampling is required. Three-dimensional geostress (error <10%) and in-situ mechanical parameters of rock mass (elastic modulus, hardness) are obtained simultaneously during the same physical loading process, ensuring the physical self-consistency of the data.

[0022] 4. High precision and high resolution: Displacement control accuracy ±0.005mm, displacement resolution 1μm, force resolution ≤1N, capable of obtaining fine elastoplastic load-displacement curves, providing high-quality data for inversion.

[0023] 5. High testing efficiency: The fully automated testing time for a single borehole with multiple measuring points at the kilometer level can be controlled within 2 hours, which is much faster than the water pressure fracturing method (3 to 5 days).

[0024] 6. Intelligent anti-interference: Through high-definition vision and AI algorithms, joints and fracture zones are identified and avoided, eliminating interference from unfavorable geological bodies at the physical level, which greatly improves the success rate of testing and the representativeness of data.

[0025] 7. Intrinsic safety guarantee: Equipped with a pressure balancing sub and an automatic accumulator recovery system in case of power failure to prevent high-pressure damage and well jamming accidents, reducing the risk of equipment loss. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall structure of the robot according to an embodiment of the present invention. Figure 2 This is a partial cross-sectional view of the indentation execution section in the robot according to an embodiment of this invention. Figure 3 This is a schematic diagram illustrating the actual workflow of the robot used in an embodiment of this invention. Detailed Implementation

[0027] The principles and features of this invention are described below with reference to the accompanying drawings. The examples given are only for explaining this invention and are not intended to limit the scope of this invention.

[0028] This invention application achieves in-situ, collaborative identification of three-dimensional geostress and rock mechanics parameters in small-diameter boreholes through integrated design, high-precision indentation actuator, and data-physical dual-drive intelligent inversion.

[0029] The following is in conjunction with the appendix Figure 1 The schematic diagram of the overall structure of the robot used in the embodiment provides a detailed description of each component of the robot of this invention.

[0030] A deep three-dimensional in-situ stress testing robot based on the borehole indentation method is generally slender and cylindrical, composed of multiple functional sections connected in series via threaded joints, with a maximum outer diameter ≤75mm. Its structure, from top to bottom, includes: a bridle connection mechanism, an electronic control system section, an attitude and environment perception section, a pressure balancing section, a hydraulic power system section, a push-support section, and an indentation execution section. The specific structure and function of each section are described in detail below.

[0031] The bridle connection mechanism, located at the top of the robot, is used to connect the armored logging cable. It integrates a tension bar and shear pins, providing mechanical load-bearing, electrical connection, and safe release in case of jamming.

[0032] The electronic control system features a dual-layer composite sealing structure: an outer titanium alloy insulation cylinder coated with silicon nitride, and an inner ceramic composite sealing cylinder with a vacuum interlayer effectively isolating it from external high temperatures. Internally, it houses high-temperature resistant electronic components, including an FPGA main control chip, an ADC acquisition module, a Manchester encoding / decoding communication module, and a motor drive module. This system is responsible for: receiving ground control commands, real-time acquisition of multi-channel sensor signals such as displacement, force, temperature, pressure, and attitude, closed-loop control of the servo motor and hydraulic valve group, and high-speed data exchange with the ground via a logging cable.

[0033] The hydraulic power system subsection integrates the following functional units: Main drive motor: Provides power to the entire hydraulic system, converting electrical energy into mechanical energy.

[0034] Planetary reducer: Connects the motor and the hydraulic pump, and improves drive efficiency by reducing speed and increasing torque to match the motor output characteristics with the pump's working requirements.

[0035] High-pressure piston pump: converts mechanical energy into hydraulic energy, providing stable and controllable pressure oil for pushing and indenting actions.

[0036] Electrically controlled relief valve: driven by a stepper motor, it realizes continuous stepless adjustment of system pressure to adapt to the different requirements of pushing force and indentation force for different rock types and different pore sizes.

[0037] Solenoid directional valve: controls the flow direction of hydraulic oil, thereby precisely controlling the extension and retraction of each push arm.

[0038] Accumulator: As a safety redundancy device, it automatically releases the stored pressure oil in abnormal operating conditions such as system power failure, driving the push arm to retract quickly and preventing the equipment from getting stuck in the well.

[0039] Through the coordinated work of the aforementioned functional units, this short section provides high-pressure hydraulic power to the entire robot, enabling the upper and lower push-and-hold groups to generate sufficient support reaction force, ensuring the stability of the robot body and preventing displacement during the indentation test.

[0040] The push-support section comprises an upper push-support assembly and a lower push-support assembly, spaced axially apart. The upper push-support assembly consists of three first hydraulic push-support arms spaced 120° apart circumferentially, each independently controlled and equipped with a built-in displacement sensor (accuracy ±0.1mm). It is used to actively push the robot to the borehole center and achieve initial anchoring. The lower push-support assembly consists of two second hydraulic push-support arms spaced 120° apart circumferentially (60° offset from the upper assembly), and is fixedly connected to the indentation execution section. It provides direct reaction force support during indentation testing.

[0041] During operation, the upper push-support group's three arms extend first, achieving active centering by comparing displacements; then, the lower push-support group's two arms extend, forming a triangular stable structure with the indentation probe. When it is necessary to change the test direction, simply retract the lower push-support group and slightly release one arm of the upper push-support group. The rotating unit of the indentation actuator section then completes the orientation adjustment and tightens it again.

[0042] The indentation execution section is the core test unit in this embodiment of the invention; see appendix. Figure 2 It consists of a linear loading unit, an angle rotation unit, and an integrated indentation probe assembly.

[0043] Linear loading unit: Includes a first servo motor (with encoder), precision ball screw, screw nut, wedge block, linear guide rail, and high-precision LVDT displacement sensor. The first servo motor drives the ball screw to rotate, the nut drives the wedge block to move linearly, and the wedge block pushes the probe seat through the inclined surface, converting the horizontal motion into radial vertical pressing motion. The LVDT sensor directly measures the radial displacement of the probe relative to the housing.

[0044] Angle rotation unit: includes a second servo motor, a reducer, a drive gear, and a slewing bearing. The second servo motor drives the drive gear through the reducer, which in turn rotates the slewing bearing, which is fixed to the mounting frame, thereby achieving circumferential rotation of the entire probe assembly. The rotation angle is precisely controlled by a motor encoder with an accuracy of ≤±1°.

[0045] Integrated indentation probe assembly: Enclosed entirely in a pressure-resistant housing, it extends from a side window of the short section. Internally, it integrates an indenter, a miniature strain gauge force sensor, and an LVDT core connecting rod. A Glyd ring + dust seal is used between the probe and the guide sleeve, allowing direct operation in drilling fluid environments.

[0046] Attitude and environment perception section; integrates the following sensors: MEMS solid-state gyroscope; measuring the azimuth angle of equipment; Triaxial accelerometer; measuring the tilt angle of the equipment to ensure the indentation probe is pressed in vertically; PT1000 platinum resistance temperature sensor; monitors downhole temperature with a resolution of 0.1℃ and an accuracy of ±1℃; High-pressure resistant, high-definition camera; equipped with a high-brightness LED array and a sapphire lens window, capable of withstanding 50MPa pressure. The camera acquires real-time images of the borehole wall, which are then transmitted to the ground via cable. Combined with image recognition algorithms, it automatically identifies defects such as joints, fissures, and fracture zones, assisting in selecting intact rock surfaces as test sites.

[0047] The pressure-balancing sub-section employs a piston-type structure, with the piston dividing the chamber into an ambient pressure chamber and a hydraulic oil compensation chamber. The ambient pressure chamber is connected to the borehole annulus through a multi-stage filter screen at the end of the outer casing; the hydraulic oil compensation chamber is connected to the oil tank of the hydraulic power system. When the downhole pressure increases, the piston moves automatically to balance the internal and external pressures in real time (balancing sensitivity ≤ 0.1 MPa), thereby protecting the internal electronic components and hydraulic elements from overpressure damage.

[0048] The workflow for the collaborative identification and application of the deep three-dimensional in-situ stress testing robot based on the borehole indentation method, as described in this invention application, is attached. Figure 3 This includes the following steps: Step 1, Well Deployment and Intelligent Site Selection: The robot is lowered to the target depth, and the attitude and environmental perception subsection acquires real-time data. The ground control system selects complete and flat test sites based on the AI ​​recognition results of the borehole wall image.

[0049] Step 2, Anchoring and Centering: The hydraulic system drives the three arms of the upper push-support group to extend synchronously and automatically center; then it drives the two arms of the lower push-support group to tighten, forming a high-rigidity support.

[0050] Step 3, Multi-directional Indentation Test: In the anchored state, the short section angle rotation unit of the indentation execution section drives the indentation probe assembly to rotate to the first test orientation. The linear loading unit drives the indenter to press into the borehole wall rock at a constant rate in displacement control mode, and load-displacement curves are collected simultaneously. After unloading, the indenter angle rotation unit drives the indenter to rotate to the second and third test angles, which are 45° apart, and repeats the indentation test to obtain load-displacement curves at different angles in the first test orientation. After completing the indentation test in the first test orientation, the short section angle rotation unit drives the indentation probe assembly to rotate to the second and third test orientations, which are 120° apart, and repeats the indentation test. Finally, displacement-load curves at 9 locations are collected at the same depth. Step 4, Intelligent Inversion: Upload the acquired load-displacement curves to the ground and input a data-physical dual-driven multi-task learning neural network. This network uses the indentation theory model (a three-dimensional elastoplastic solution considering the borehole surface boundary and the pre-loading effect of geostress) as a physical regularization term to constrain the mechanical self-consistency of the prediction results; its input consists of load-displacement curves and their derived features (slope of the loading segment, unloading stiffness, yield point load, plastic work, and maximum indentation depth) at at least three orientations of the same depth section; the output includes parameters such as σ1, σ2, σ3, rock mass elastic modulus E, and hardness H. Step 5, Retrieval: Move to the next depth measuring point and repeat steps 2 to 4 to complete the full-hole multi-point test. Upon completion of the test, the system is powered off, the accumulator drives the push arm to automatically retract, retracting all actuators and removing the robot from the borehole.

[0051] The foregoing has provided a detailed description of one embodiment of this invention, but the description is merely a preferred embodiment and should not be construed as limiting the scope of this invention. All equivalent variations and modifications made within the scope of this invention should still fall within the patent coverage of this invention.

Claims

1. A robot for in-situ testing of deep three-dimensional geostress based on borehole indentation method, characterized in that, It includes a bridle connection mechanism, an electronic control system section, an attitude and environment sensing section, a pressure balancing section, a hydraulic power system section, a push-support section, and an indentation execution section, all connected in series and each having a cylindrical sealing structure; among them, The push-support section includes an upper push-support group and a lower push-support group distributed along the robot axis. The upper push-support group consists of three first hydraulic push-support arms distributed at 120° intervals around the circumference. The lower push-support group consists of two second hydraulic push-support arms. The circumferential angle between the two second hydraulic push-support arms is 120°, and together with the orientation of the indentation probe of the indentation execution section, they form a triangular support. Each push-support arm is provided with a rigid support foot at its end. The indentation execution section includes a linear loading unit, an angle rotation unit, and an integrated indentation probe assembly. The linear loading unit includes a first servo motor, a precision ball screw, a screw nut, a wedge block, a linear guide rail, and a high-precision displacement sensor. The first servo motor drives the ball screw to rotate, and the screw nut drives the wedge block to move along the linear guide rail. The inclined surface of the wedge block slides in engagement with the base of the indentation probe assembly, converting the motion along the robot's axial direction into a radial linear pressing motion of the indentation probe. The angle rotation unit includes a second servo motor, a reducer, a drive gear, and a slewing bearing, driving the indentation probe assembly to rotate around the robot's axis. The integrated indentation probe assembly integrates a diamond or cemented carbide indenter, a micro force sensor, and a displacement sensor, and is completely sealed. The electronic control system subsection includes a high-temperature resistant main control module, an ADC acquisition module, a Manchester encoding / decoding communication module, and a motor drive module, which are used to receive ground commands, acquire multiple sensor signals, and control various actuators. The attitude and environment perception section integrates a MEMS solid-state gyroscope, a three-axis accelerometer, a temperature sensor, and a high-pressure resistant high-definition camera to acquire robot attitude, drilling environment temperature, and borehole wall images in real time, and to help select intact rock surfaces as test points.

2. The deep three-dimensional in-situ stress testing robot based on the borehole indentation method according to claim 1, characterized in that, The hydraulic power system subsection includes a high-temperature resistant micro motor, a planetary reducer, a high-pressure piston pump, an electrically controlled relief valve, a solenoid directional valve, and an accumulator.

3. The deep three-dimensional in-situ stress testing robot based on the borehole indentation method according to claim 1, characterized in that, In the angle rotation unit, the second servo motor, in conjunction with the reducer and the drive gear, drives the slewing bearing. The rotation angle accuracy is ≤±1°. At least three test orientations can be achieved within the same test section. The smaller circumferential angle between any two test orientations is 40°~120°. Preferably, the three test orientations are evenly distributed along the circumference and the adjacent angles are 120°.

4. The deep three-dimensional in-situ stress testing robot based on the borehole indentation method according to claim 1, characterized in that, In the attitude and environment perception section, the MEMS inertial measurement unit includes a MEMS gyroscope and a three-axis accelerometer, with an azimuth measurement error of <5° and a tilt measurement accuracy of ±1° for the three-axis accelerometer; the high-voltage resistant high-definition camera has a resolution of 1080P and is equipped with an LED illumination array.

5. The deep three-dimensional in-situ stress testing robot based on the borehole indentation method according to claim 1, characterized in that, The pressure balancing sub-section has a piston-type structure. One side of it is connected to the borehole annulus environment through a filter mesh, and the other side is connected to the hydraulic oil circuit of the hydraulic power system sub-section. The pressure balancing sensitivity is ≤0.1MPa.

6. A method for in-situ testing using the robot according to any one of claims 1 to 5, characterized in that, Includes the following steps: Step S1: Lower the robot to the target depth inside the borehole via cable, obtain the depth through the ground winch depth measurement system, acquire azimuth, dip, temperature and borehole wall video images using the attitude and environment perception sub, and identify and screen test points with intact rock mass and flat borehole wall through the ground control system or downhole edge computing module. Step S2: The ground sends a command, and the hydraulic power system short section drives the three arms of the upper push support group of the push support short section to extend synchronously. By comparing the values ​​of the displacement sensors of each arm, the extension amount of each arm is automatically adjusted so that the robot axis coincides with the drilling axis, realizing active centering; then the upper and lower push support groups are simultaneously supported on the hole wall to provide the reaction force support required for testing. Step S3: In the anchored state, the short section angle rotation unit of the indentation execution section drives the indentation probe assembly to rotate to the first test position, and the linear loading unit drives the indenter to press into the borehole wall rock at a constant rate in the displacement control mode, and the load-displacement curve is collected simultaneously. After unloading, the indenter angle rotation unit drives the indenter to rotate to the second and third test angles, which are 45° apart, and repeats the indentation test to obtain the load-displacement curves at different angles in the first test orientation; after completing the indentation test in the first test orientation, the short section angle rotation unit drives the indentation probe assembly to rotate to the second and third test orientations, which are 120° apart, and repeats the indentation test. Step S4: The acquired load-displacement curves are uploaded to the ground via cable and input into the pre-trained data-physics dual-drive intelligent inversion model. This model uses the elastoplastic indentation theory model as a physical constraint embedded in the neural network to collaboratively invert the three-dimensional geostress components and in-situ mechanical parameters of the rock mass at the test point. Step S5: Retract the push arm, move to the next depth measuring point, repeat steps S2 to S4, and after completing the full hole multi-point test, retract all actuators and remove the robot from the drilling hole.

7. The method according to claim 6, characterized in that, In step S1, the borehole wall image transmitted in real time by a high-definition camera is used to automatically identify defects such as joints, fissures, and fracture zones using an image recognition algorithm, and to automatically evaluate the rock mass quality at the test point; flat rock surfaces with a proportion of joint, fissure, and fracture zone area lower than a preset threshold in the borehole wall image are selected as test points.

8. The method according to claim 6, characterized in that, In step S2, when it is necessary to change the test orientation within the same depth section, keep the upper push-support group taut, and only need to retract the lower push-support group. The angle rotation unit drives the indentation probe assembly to rotate around the robot axis to the target orientation, and then taut the lower push-support group again. There is no need to move the entire robot.

9. The method according to claim 6, characterized in that, The data-physics dual-driven intelligent inversion model in step S4 is a multi-task learning neural network. Its input consists of load-displacement curves and their derived features (slope of the loading segment, unloading stiffness, yield point load, plastic work, and maximum indentation depth) at at least three orientations of the same depth section. The output includes the rock mass elastic modulus, strength parameters, and three-dimensional geostress components. The model loss function includes a physical regularization term derived from the indentation theory model, which is used to constrain the mechanical self-consistency of the prediction results.

Images