Intelligent drilling deviation detection robot and drilling deviation correction method applied thereto

Abstract

The present invention discloses an intelligent drilling deviation detection robot and a drilling deviation correction method applied thereto. The robot body has a cylindrical shell smaller than the diameter of the drill hole. The end of the cylindrical shell is connected to a recovery cable. A number of sensor modules are provided at the head end. At least three groups of elastic mechanical support arms are symmetrically provided along the circumferential direction on the outer wall of the cylindrical shell. The elastic mechanical support arms are pressed against the hole wall when the robot body enters the hole and achieve rolling contact with the hole wall through support wheels. The robot body is provided with a control unit and a power supply in the shell that are communicatively connected to the sensor modules. The intelligent drilling deviation detection robot and the drilling deviation correction method applied thereto provided by the present invention can realize deviation monitoring during geological drilling, realize real-time automatic drilling deviation correction through intelligent analysis, effectively improve drilling accuracy and operating efficiency, significantly improve drilling construction quality, adapt to harsh drilling environments, and can be expanded to various geological drilling scenarios such as blasting holes and geological exploration holes.

Description

Technical Field

[0001] The invention discloses an intelligent drilling deviation detection robot and a drilling deviation correction method applied thereto, belonging to the technical field of underground drilling construction deviation correction detection. Background Art

[0002] Whether in underground mining drilling and blasting operations or in drilling and tunneling and blasting operations, it is necessary to compare the actual drilling angle with the designed angle to correct borehole deviation. Traditional borehole inclinometer detection methods require manual segmented measurements after drilling is stopped, which cannot achieve real-time monitoring of the drilling process, resulting in delayed deflection control, seriously affecting drilling accuracy and construction safety. In the existing technology, although some automated inclinometer equipment uses inertial navigation or fiber optic sensing technology, such as the Chinese patent application CN201410377948.9, which discloses an intelligent online deep-hole drilling vehicle for coal mines, it installs sensors and fiber optic Bragg gratings inside the drill pipe. This can monitor real-time online data (including rock hardness, gas concentration, geothermal temperature, water, drilling depth, etc.) and reflect geological structure in advance. However, the system is complex, has poor environmental adaptability, and lacks intelligent data analysis and real-time deviation correction functions, making it difficult to meet the needs of precise drilling in complex geological conditions. Summary of the Invention

[0003] The technical problem solved by the present invention is: in view of the fact that the existing underground drilling deviation monitoring technology is difficult to meet the drilling needs under increasingly complex geological conditions, an intelligent drilling deviation detection robot that can realize full-hole continuous monitoring, intelligent analysis and automatic correction and a drilling deviation correction method applied thereto are provided.

[0004] The present invention is implemented by the following technical solutions:

[0005] The present invention first discloses an intelligent drilling deflection detection robot, which is characterized in that: it includes a robot body 1, the robot body 1 has a cylindrical shell smaller than the diameter of the drill hole, the end of the cylindrical shell is connected to a recovery cable 6, a plurality of sensor modules 2 are arranged at the head end, and at least three groups of elastic mechanical support arms 11 are symmetrically arranged along the circumferential direction on the outer wall of the cylindrical shell. The elastic mechanical support arms 11 are pressed against the hole wall when the robot body enters the hole, and achieve rolling contact with the hole wall through support wheels 12. The robot body 1 is provided with a control unit 3 and a power supply 5 in the shell that are communicatively connected to the sensor module.

[0006] In the intelligent drilling deflection detection robot of the present invention, further, the sensor module 2 provided at the head end of the cylindrical shell includes:

[0007] MEMS inertial navigation module 21 that collects the robot's posture data in the hole,

[0008] 3D laser scanning probe 22 that scans the hole wall and collects 3D point cloud data of the hole wall,

[0009] Ultrasonic sensor 23 that collects distance data between the robot and the hole wall,

[0010] Visual sensor 24 for collecting hole wall environmental data,

[0011] A glass protective cover 25 is provided at the head end of all cylindrical shells to cover all sensor modules.

[0012] In the intelligent drilling deviation detection robot of the present invention, further, the elastic mechanical support arm 11 adopts a servo electric push rod to connect the support wheel 12 to achieve rolling contact with the hole wall. The support wheel 12 has a built-in pressure sensor and is feedback-connected to the servo electric push rod. The contact pressure feedback between the support wheel and the hole wall is used to adjust the tightening force of the elastic mechanical support arm on the hole wall.

[0013] In the intelligent drilling deflection detection robot of the present invention, further, the support wheel 12 is driven by a motor.

[0014] The present invention also discloses a drilling deviation correction method, comprising the following steps:

[0015] S1. The intelligent drilling deflection detection robot of the present invention is placed in the hole of the front section of the drill hole, and the robot body is driven to move along the wall of the front section of the hole. During the movement of the robot body, the MEMS inertial navigation module of the sensor module collects the posture data of the robot in the hole, the 3D laser scanning probe scans the hole wall to collect 3D point cloud data of the hole wall, the ultrasonic sensor collects the distance data between the robot and the hole wall, and the visual sensor collects the environmental data of the hole wall;

[0016] S2, denoising the collected robot posture data, hole wall 3D point cloud data, and environmental data, then fusing them using a particle filter algorithm, and building a 3D model of the drilling trajectory for the front section of the drilling in combination with the drilling coordinate system;

[0017] S3. Deflection data of the preceding drilling section is obtained using the three-dimensional model of the drilling trajectory. If the deflection angle in the deflection data is greater than a deflection threshold, the intelligent drilling deflection detection robot feeds the deflection data of the preceding drilling section back to the drilling rig control system to perform deviation correction for the subsequent drilling section.

[0018] S4. If the deflection angle in the front-end drilling deflection data obtained by the three-dimensional model of the drilling trajectory is less than or equal to the deflection threshold, the collected robot posture data, hole wall three-dimensional point cloud data and environmental data are input into the trained drilling deflection prediction model to predict the deflection trend of the rear-end drilling, and the predicted deflection data is output and fed back to the drilling rig control system of the rear-end drilling.

[0019] In the drilling deviation correction method of the present invention, further, in the step S1, the intelligent drilling deviation detection robot navigates according to a pre-set drilling path when moving in the hole, or uses a visual sensor to observe the hole wall environment of the front drilling in real time, and uses SLAM combined with the drilling coordinate system to locate the robot's position in real time and construct a navigation path for the robot. During the robot's navigation process, the ultrasonic sensor collects the distance data between the robot and the hole wall for anti-collision warning.

[0020] In the drilling deviation correction method of the present invention, further, in the step S2, the Kalman filter algorithm is used to denoise the collected robot posture data, hole wall three-dimensional point cloud data and environmental data, and the denoising process is as follows: the intelligent drilling deviation detection robot collects 30 seconds of robot posture data, hole wall three-dimensional point cloud data, distance data and environmental data in a stationary state in the hole to establish a zero bias reference; during the movement of the robot, FFT analysis is performed on the collected original data to identify the power frequency vibration components of the robot device noise and mechanical vibration frequency, and an anti-phase compensation amount with a phase opposite to the power frequency vibration and matching amplitude is injected into the state prediction equation; the state transfer matrix of the Kalman filter is adjusted and redundant check is started for the mutation data, and at the same time, the abnormal observation values ​​output by the Kalman filter are identified and eliminated through multi-sensor redundant check or chi-square detection, and the filtering results are phase compensated, and finally smoothed and outputted. Denoised robot posture data, hole wall three-dimensional point cloud data and environmental data.

[0021] In the drilling deviation correction method of the present invention, further, in step S2, the construction of the three-dimensional model of the drilling trajectory includes the following steps:

[0022] S21, initialize the particles in the drilling coordinate system through the robot posture data, and randomly generate a large number of particles according to the initial robot posture data collected by the MEMS inertial navigation module , is the particle of the i-th particle at time k, X represents the system state vector, including the position information of the robot posture data in the drilling coordinate system and the posture angle of the robot, N is the total number of particles, k represents the current moment, k-1 represents the previous moment, and each particle has a weight , initially all particles have equal weights, ;

[0023] S22. Based on the state transition equation of the system, the robot posture data collected by the MEMS inertial navigation module at the previous moment is used to predict the state of the particle at the next moment. The state transition equation is as follows:

[0024] ,

[0025] is the robot posture data represented by the particle at the last moment, where f( ) is the state transfer function, It is the predetermined drilling parameters input into the drilling rig at the last moment. is the process noise, which follows a Gaussian distribution , is the process noise covariance matrix, and according to the state transfer equation, each particle at the previous moment Make a prediction and get the particle size at the next moment The robot posture data represented;

[0026] S23, using the three-dimensional point cloud data of the hole wall collected by the three-dimensional laser scanning probe and the environmental data of the hole wall collected by the visual sensor as observation data to update the weight of the particle, first obtain the particle observation value through the following formula,

[0027] ,

[0028] in is the observed value of particle i at time k, h( ) is the observation function, v k is the observation noise, which follows a Gaussian distribution , is the observation noise covariance matrix;

[0029] Then, based on the predicted particle state and observation data, the weight update value of each particle is calculated by the following formula:

[0030] ),

[0031] in ) is in the particle state Lower observation value The probability density function of

[0032] Finally, the weights of all particles are normalized;

[0033] S24, resample the particles after weight update, and resample M particles from the current particle set according to the particle weight. Replace the original particle set, and all resampled particle weights are reset to ;

[0034] S25. During the robot's movement, the continuously collected robot posture data, hole wall three-dimensional point cloud data, and environmental data are iterated by repeating the above steps, and the particle states obtained in a series of drilling coordinate systems are fitted to establish a three-dimensional model of the drilling trajectory of the front drilling.

[0035] In the drilling deviation correction method of the present invention, further, in step S4, the drilling deviation prediction model is trained by the following steps:

[0036] S41. Using an intelligent drilling deviation robot to collect data from blastholes with different geological conditions and drilling parameters, sorting the data within the same blasthole according to the chronological order of segmented drilling, extracting robot posture data, hole wall 3D point cloud data, environmental data, and actual drilling deviation data from the sorted data to form a feature set for machine learning;

[0037] S42, dividing the feature set into a training set and a test set, inputting the robot posture data, hole wall three-dimensional point cloud data, and environmental data of the training set into a long short-term memory network model for training, adjusting the weights of the long short-term memory network model using a back propagation algorithm so that the predicted drilling deflection data output by the model is close to the actual drilling deflection data corresponding to the training set, and optimizing the model using mean square error as a loss function;

[0038] S43. Use the test set to evaluate the trained and optimized long short-term memory network model, calculate the error indicators between the drilling deviation data output by the model and the actual drilling deviation data corresponding to the test set through the root mean square error and mean absolute error, and complete the evaluation to obtain the drilling deviation prediction model.

[0039] In the drilling deviation correction method of the present invention, further, the drilling deviation prediction model is deployed in the control unit of the intelligent drilling deviation robot. During the actual drilling process, the posture data of the detection robot performed for correction in step S3, the three-dimensional point cloud data of the hole wall, the environmental data and the actual drilling deviation data are used as new feature sets to update and train the drilling deviation prediction model.

[0040] The present invention adopts the above technical solution to achieve the following beneficial effects:

[0041] (1) The intelligent drilling deflection detection robot of the present invention uses an elastic mechanical support arm + support wheel to enable the robot to move within the blasthole. The elastic mechanical support arm can adapt to blastholes of different diameters. The elastic mechanical support arm uses an electric push rod to adjust the pressure of the support wheel on the borehole wall. The pressure sensor built into the support wheel measures the pressure dynamically through contact, thereby adapting to the irregularities of the borehole wall during the robot's movement within the hole. The robot's compact design is suitable for drilling environments with a diameter of 100-250mm and can operate stably under harsh conditions such as dust and vibration.

[0042] (2) The intelligent drilling deviation detection robot of the present invention uses a multi-sensor module to achieve full-hole continuous monitoring of drilling deviation. The multi-sensor module integrates a variety of advanced sensors, including a high-precision MEMS inertial navigation module, a three-dimensional laser scanning probe, a visual sensor, and an ultrasonic sensor. Among them, the visual sensor is set on the top of the robot and is equipped with a rotating pan-tilt platform, which can achieve 180° horizontal rotation. The measurement accuracy of the high-precision MEMS inertial navigation module is 0.01° (static) and 0.1° (dynamic). The scanning frequency of the three-dimensional laser scanning probe is 10Hz, and the point cloud density reaches 500 points / cm². The visual sensor is combined with SLAM technology to achieve positioning and path reconstruction in the borehole, with a positioning error of <1cm / m. The ultrasonic sensor is used to monitor the distance between the robot and the hole wall in real time, with an anti-collision warning accuracy of ±2mm. These sensors work together to obtain information in the borehole from different angles, providing data support for accurate detection. The detection efficiency is more than 5 times higher than that of traditional inclinometers.

[0043] (3) The drilling deviation correction method of the present invention uses the robot posture data, hole wall three-dimensional point cloud data, and environmental data collected by the intelligent drilling deviation detection robot. First, the sensor noise is eliminated by the improved Kalman filter algorithm to improve the data accuracy in the complex hole environment; secondly, the particle filter algorithm is used for data fusion, so that different types of data can be effectively integrated, and a three-dimensional model of the drilling trajectory is established to express the drilling deviation data. The deviation angle error is <0.1°, which is convenient for operators to quickly and intuitively understand the drilling deviation situation, establish a "detection-analysis-feedback-correction" closed-loop system, support rapid decision-making in the later stage of drilling correction, realize real-time adaptive correction in the drilling process, effectively improve drilling accuracy, and reduce the cost increase and safety hazards caused by drilling deviation.

[0044] (4) The drilling deviation correction method of the present invention also predicts the drilling deviation trend of small deviation data through the drilling deviation prediction model. When the deviation data expressed by the three-dimensional model of the drilling trajectory does not need to be corrected, the deviation trend of the subsequent drilling is predicted in advance. The long short-term memory network (LSTM) is selected as the prediction model, which can effectively process long sequence data and avoid the gradient vanishing problem. It is suitable for modeling the drilling deviation trend. The correction decision of the subsequent drilling is adjusted in advance according to the prediction results. The drilling deviation trend is predicted based on the machine learning algorithm and is directly linked with the drilling rig control system. Once the deviation is detected, the system can respond quickly and realize real-time correction, reducing the possibility of large drilling deviation and further improving the drilling accuracy.

[0045] To sum up, the intelligent drilling deviation detection robot and the drilling correction method applied thereto provided by the present invention can realize the deflection monitoring in the geological drilling process, realize the real-time automatic correction of drilling through intelligent analysis, effectively improve the drilling accuracy and operation efficiency, significantly improve the drilling construction quality, adapt to the harsh drilling environment, and can be expanded to various geological drilling scenarios such as blasting holes and geological exploration holes.

[0046] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 Schematic diagram of the overall structure of the intelligent drilling deviation detection robot in the embodiment.

[0048] Figure 2 It is a front view of the intelligent drilling deviation detection robot in the embodiment.

[0049] Figure 3 Schematic diagram of the distribution of sensor modules at the head end of the intelligent drilling deviation detection robot in the embodiment.

[0050] Figure 4 Schematic diagram of the internal structure of the intelligent drilling deviation detection robot in the embodiment.

[0051] Figure 5 It is a flow chart of the drilling deviation correction method performed by the intelligent drilling deviation detection robot in the embodiment.

[0052] Figure 6 Schematic diagram of training the borehole deflection prediction model in the embodiment.

[0053] Numbers in the figure: 1. Robot body; 2. Sensor module; 3. Control unit; 4. Winch interface; 5. Power supply; 6. Recovery cable; 11. Elastic mechanical support arm; 12. Support wheel; 21. MEMS inertial navigation module; 22. 3D laser scanning probe; 23. Ultrasonic sensor; 24. Visual sensor; 25. Glass protective cover; 26. Rotating gimbal. DETAILED DESCRIPTION

[0054] Example

[0055] See also Figure 1-4The intelligent drilling deflection detection robot shown in the figure is a specific embodiment of the present invention, comprising a robot body 1, a sensor module 2, a control unit 3, a winch interface 4, a power supply 5, and a recovery cable 6. The robot body 1 has a cylindrical shell smaller than the diameter of the drill hole. The end of the cylindrical shell is connected to the recovery cable 6 via a winch interface 4. Several sensor modules 2 are installed at the front end of the cylindrical shell of the robot body 1 to collect robot posture data and in-hole data of the robot. At least three sets of elastic mechanical support arms 11 are symmetrically arranged along the outer wall of the cylindrical shell of the robot body 1 along the circumference. The elastic mechanical support arms 11 are equipped with support wheels 12. When the robot body enters the hole, the elastic mechanical support arms 11 are pressed against the hole wall and achieve rolling contact with the hole wall through the support wheels 12. The shell of the robot body 1 is provided with a control unit 3 and a power supply 5 that are communicatively connected to the sensor modules. The several sensor modules installed on the robot can transmit data through the control unit using wireless signals such as 5G / Wi-Fi6, or use the recovery cable 6 connected to the end of the robot body shell to transmit wired signals through signal lines.

[0056] like Figure 2 and Figure 3 As shown, the intelligent drilling deflection detection robot of this embodiment utilizes a wheeled walking mechanism within the borehole. The outer diameter of the cylindrical shell of the robot body 1 is generally no larger than 80% of the borehole diameter, leaving space for the installation of the elastic mechanical support arms 11 and support wheels 12 that allow the robot to move smoothly within the borehole. The wheeled walking mechanism of this embodiment is equipped with three sets of symmetrically distributed elastic mechanical support arms 11. The elastic mechanical support arms 11 are mounted on the outside of the shell of the robot body 1 via elastic hinges. They can be opened at different angles to accommodate blastholes of varying diameters, and can be adjusted through swinging to maintain stable contact with the borehole wall even when the wall is convex or concave. The elastic mechanical support arm 11 uses a micro servo electric push rod to connect the support wheel 12 to the hole wall to achieve rolling contact. The support wheel 12 has a built-in pressure sensor and is connected to the servo electric push rod for feedback. The feedback of the contact pressure between the support wheel 12 and the hole wall is used to adjust the support force of the elastic mechanical support arm 11 on the hole wall. During the robot's movement in the hole, the support force of the mechanical support arm 11 on the hole wall provides friction for the support wheel to move, thereby adapting to the irregularities of the hole wall. In addition, the pressure sensor on the support wheel 12 can also be used for contact measurement to monitor the contact pressure distribution between the support wheel 12 and the hole wall in real time. By comparing the pressure data differences of the support wheels symmetrically distributed around the periphery of the robot body and combining them with the geometric model of the robot body, the local deformation and deflection direction of the drilling hole are calculated. If the pressure of the support wheel on one side increases significantly, such as the pressure on the right side is greater than 10% on the left side, it indicates that the hole wall is offset to that side. The displacement measured by the push rod telescopic encoder of the elastic mechanical support arm 11 is integrated with the pressure data of the pressure sensor to obtain the trend of the change in the drilling diameter.

[0057] The support wheels 12 are driven by a motor, which is powered by a power supply 5 within the robot body. For horizontal drilling and upward vertical drilling, an electric push rod first presses the support wheels 12 against the borehole wall. The motor then drives the support wheels, allowing the robot to self-propelled within the borehole. For downward vertical drilling, the robot is hoisted, with a recovery cable interface at the top of the borehole. A winch mechanism allows manual retraction of the recovery cable 6, allowing the robot to propel itself through the borehole using its own gravity. The support wheels 12 are covered with a high-friction rubber material to further enhance stability and friction during movement within the borehole.

[0058] Specific as Figure 3 As shown, the sensor module 2 set at the head end of the intelligent drilling deviation detection robot in this embodiment includes a MEMS inertial navigation module 21, a three-dimensional laser scanning probe 22, an ultrasonic sensor 23, and a visual sensor 24. Among them, the visual sensor 24 adopts a binocular visual sensor, which is installed at the head end of the robot body through a rotating pan-tilt platform 26 and can achieve 180° horizontal rotation. The MEMS inertial navigation module 21 collects the robot's posture data within the hole, including the robot's posture angle within the hole, and records changes in the robot's posture, such as tilt and rotation, as it moves within the hole. The measurement accuracy is 0.01° static and 0.1° dynamic. The 3D laser scanning probe 22 scans the hole wall and constructs 3D point cloud data of the hole wall. The scanning frequency is 10Hz, and the point cloud density reaches 500 points / cm². Through this 3D point cloud data of the hole wall, information such as the shape, size, and surface features of the hole wall can be accurately obtained. The ultrasonic sensor 23 collects distance data between the robot and the hole wall, monitoring the distance between the robot and the hole wall in real time. The anti-collision warning accuracy is ±2mm. Once the distance approaches the set safety distance, an anti-collision warning is issued to ensure the safe movement of the robot within the borehole. The visual sensor 24 collects hole wall environmental data. With the help of the rotating pan-tilt head 26, it can rotate 180° horizontally to obtain a full-scale observation of the environment within the borehole. Combined with SLAM technology, the robot can achieve in-hole positioning and path reconstruction, with a positioning error of <1cm / m. Multiple sensor modules work together to acquire borehole information from different angles, providing data support for accurate detection of borehole deflection. The sensor modules in this embodiment all use commercially available, mature sensors, such as the TDK ICM-42688-P MEMS inertial navigation module, the Faro Focus Core 130 3D laser scanning probe, the Pepperl+Fuchs UC2000-30GM-IUR2-V15 ultrasonic sensor, and the Intel RealSense D455 binocular vision sensor.

[0059] The robot body 1 of this embodiment is provided with a glass protective cover 25 at the head end of the cylindrical shell to cover all sensor modules, thereby reducing the influence of the environment in the hole on the sensors.

[0060] like Figure 4 As shown, this embodiment incorporates a high-energy-density lithium battery within the robot housing 1, powering five pairs of sensors and the support wheel motors. The battery life is ≥6 hours, enabling extended operation. The robot housing 1 boasts an IP68 protection rating, effectively withstanding the harsh environment of underground mines. A drill linkage interface allows for both automated control and manual operation via a retractable cable, increasing operational flexibility.

[0061] The "front drilling section" referred to in this embodiment is the drilling section that has been completed during the drilling process, and the "back drilling section" is the next drilling section to be drilled. The back drilling section is corrected immediately by detecting the deviation data of the completed front drilling section.

[0062] This embodiment also discloses a drilling deviation correction method using the above-mentioned intelligent drilling deviation detection robot, such as Figure 5 As shown, the specific steps include:

[0063] S1. The intelligent drilling deviation detection robot described above in this embodiment is placed into the hole mouth of the front section of the drill hole, and the robot body is driven to move along the front section of the hole wall. During the movement of the robot, the MEMS inertial navigation module of the sensor module collects the posture data of the robot in the hole, the three-dimensional laser scanning probe scans the hole wall to collect the three-dimensional point cloud data of the hole wall, the ultrasonic sensor collects the distance data between the robot and the hole wall, and the visual sensor collects the environmental data of the hole wall.

[0064] When preparing to perform drill hole deflection detection, the intelligent drill hole deflection detection robot is first transported to the drilling site. The robot's elastic mechanical support arms 11 and support wheels 12 begin working in tandem. Pressure sensors monitor the tension in real time and dynamically adjust the tension based on the actual hole wall conditions. If the hole wall is smooth, the tension is reduced appropriately. If the hole wall is rough or has bumps or depressions, the tension is automatically increased to ensure the robot maintains a stable, consistent grip and adapts to any irregularities.

[0065] When navigating within a borehole, the intelligent drilling deviation detection robot navigates according to a pre-set drilling path. If no pre-set path is available, a visual sensor can be used to observe the borehole wall environment in real time. SLAM, combined with the borehole coordinate system, allows the robot to locate its position in real time and construct a navigation path. While the physical space within the borehole in the pre-drilled section is a single tubular structure, the borehole wall may contain localized collapse, accumulated rock debris, or tool residue. Therefore, a "multi-path" approach is required to adapt to the complex environment within the borehole. Different path modes, such as tilted travel, spiral scanning, and vibration propulsion, can also be preset, allowing the robot to adapt to complex environments using different modes. SLAM is a simultaneous positioning and mapping technology that uses its position and map or surrounding environment information to locate itself during movement. Based on this positioning, an incremental map is constructed, enabling autonomous positioning and navigation. SLAM technology is a well-known positioning and mapping technology. This embodiment aims to utilize well-known SLAM technology to enable the intelligent drilling deviation detection robot to navigate within the pre-drilled hole. The specific SLAM data processing process is not described in detail.

[0066] During downward drilling, the robot can be lowered to the hole opening of the preceding drilled section via a winch device using a retrieving cable 6, using the drilling gantry. During the lowering process, the robot's posture must be correct to avoid collisions with the hole wall. Once the robot reaches the hole opening, it adjusts its initial posture, activates autonomous navigation, and begins moving into the hole.

[0067] The robot travels at a speed of 0.5-1.0 m / s. Every 0.5 m, it pauses, and the sensor module 2 at its front end begins operating. The MEMS inertial navigation module 21 monitors the robot's attitude angles in real time, recording any changes in the robot's tilt, rotation, and other posture data within the borehole.

[0068] The 3D laser scanning probe 22 scans the hole wall of the previous drilled hole and constructs a 3D point cloud of the hole wall. The 3D point cloud data can accurately obtain information such as the shape, size, and surface features of the hole wall. The visual sensor 24 uses the rotating pan-tilt head 26 to rotate 180° horizontally to conduct all-round observation of the environmental data within the borehole, and combines SLAM technology to achieve positioning and path reconstruction within the borehole. The ultrasonic sensor 23 monitors the distance data between the robot and the hole wall in real time. During the robot's navigation process, the ultrasonic sensor collects the distance data between the robot and the hole wall to issue an anti-collision warning. Once the distance between the robot and the hole wall approaches the set safety distance, an anti-collision warning will be issued to ensure the safe operation of the robot in the borehole.

[0069] S2. De-noise the collected robot posture data, hole wall 3D point cloud data, and environmental data, and then fuse them through the particle filter algorithm. Combined with the drilling coordinate system, a 3D model of the drilling trajectory of the front drilling is established.

[0070] In step S2, the Kalman filter algorithm is first used to denoise the collected robot posture data, hole wall three-dimensional point cloud data, and environmental data to eliminate sensor noise. By establishing a composite noise model that includes device noise, mechanical vibration, etc., and integrating a real-time compensation mechanism, an adaptive observation weight adjustment function is introduced. When a data mutation is detected, the outlier weight is automatically reduced. At the same time, curvature-noise correlation compensation is specially designed for the curved section of the drilling hole.

[0071] The specific denoising process is as follows:

[0072] The intelligent drilling deviation detection robot collects 30 seconds of robot posture data, hole wall 3D point cloud data, distance data, and environmental data while stationary in the hole to establish a zero-bias benchmark. During the robot's movement, it performs FFT analysis on the collected raw data to identify the power frequency vibration components of the robot device noise and mechanical vibration, and injects an inverted compensation into the state prediction equation. The inverted compensation is generated by analyzing the frequency and amplitude of the power frequency vibration component of the noise, with a phase opposite and amplitude matching, to offset the impact of noise on the system. The Kalman filter state equation is: , where is the robot posture data at time k, is the Kalman filter state transfer matrix, describing the system dynamics model, The influence of the control input, in this embodiment, refers to the drilling rig parameters, is the process noise, This is the reverse compensation designed for the noise of the power frequency vibration component. The noise model is extracted from the zero-bias benchmark data, assuming that the noise component is: ; The compensation amount is: , at discrete time step k, the compensation amount The Kalman filter state equation is injected by lookup table or real-time calculation. The compensated state prediction is closer to the true value and reduces the deviation caused by noise. The covariance prediction based on the Kalman filter state equation is updated as ,in It needs to be adjusted according to the compensation effect (such as reducing the covariance term corresponding to high-frequency noise). It represents the error covariance matrix when predicting the state at time k when the state estimate at time k−1 is known, reflecting the uncertainty of the predicted state. is the state transfer matrix, which is used to transfer the state error at time k−1 to time k. is the posterior error covariance matrix of the state estimate at time k−1, which represents the uncertainty of the state estimate at the previous moment and is calculated by the update step of the Kalman filter. is the process noise covariance matrix, which describes the uncertainty of the system dynamic model, that is, the unmodeled factors in the state transition process, such as the noise introduced by the drilling environment vibration and the robot mechanical error in this embodiment.

[0073] The Kalman filter's state transition matrix is ​​adjusted based on the current borehole curvature, and redundancy checking is initiated for mutation data (e.g., single-point jumps > 3σ). Mutation data is identified by comparison with pre-established zero-bias baseline data. If the real-time data deviates from the zero-bias baseline by more than three standard deviations, it is considered mutation data and is downgraded or eliminated to ensure accurate drilling trajectory modeling. Anomalous observations in the Kalman filter output are identified and eliminated through multi-sensor redundancy checking or chi-square testing. Phase compensation is performed on the filtered results, and finally, a third-order Butterworth filter is used for smoothing and outputting denoised robot posture data, 3D point cloud data of the hole wall, and environmental data.

[0074] Taking the chi-square test to identify abnormal data as an example, the chi-square test ( Test) Eliminate abnormal observations of mutation data: Calculate residual observations Predicting the state with Kalman filter The differences: ,in is the observation matrix, is the state prediction value. Calculate the chi-square statistic: , the residual covariance matrix S is calculated as: , where is the prediction covariance, is the observation noise covariance. Abnormality judgment: If > 阈值 (such as 95% confidence level corresponds to the threshold), then Determine the abnormal value and remove it. Set the delay time of phase compensation ,The de-noised robot posture data, hole wall 3D point cloud data and environmental data are smoothed by discretization processing (bilinear transformation) using a third-order Butterworth low-pass filter (cut-off frequency fc , e.g., 10 Hz).

[0075] Mutation data refers to sudden changes in data that deviate significantly from historical baselines within a short period of time, such as sudden changes in attitude angle or pressure values. Anomaly detection values, on the other hand, refer to data with significant deviations from the Kalman filter's predicted output, such as outliers caused by sensor noise or environmental interference. This embodiment uses a complementary approach to eliminate mutation data and anomaly detection values ​​in the denoising process, ensuring that the robot can output accurate deflection detection data even in harsh environments such as dust and vibration, facilitating the construction of high-precision 3D models of drilling trajectories.

[0076] Next, a particle filter algorithm is used for data fusion, unifying the multi-source data—including robot posture data, 3D point cloud data of the hole wall, distance data, and environmental data—into a drilling coordinate system. After data fusion, a 3D model of the drilling trajectory is constructed. This 3D model intuitively displays the actual drilling trajectory as deflection data, including information such as deflection angle and azimuth, providing accurate data support for subsequent deviation correction operations.

[0077] The construction of the three-dimensional model of the drilling trajectory specifically includes the following steps:

[0078] S21. Initialize the particles in the drilling coordinate system using the robot posture data. Before starting data fusion, initialize the particles in the drilling coordinate system based on the prior knowledge of the drilling system. Assume that the system state vector is ),in Represents the position information of the robot in the drilling coordinate system, Represents the robot's attitude angles (yaw angle, pitch angle, roll angle). Based on the initial robot attitude data collected by the MEMS inertial navigation module, including the robot's initial attitude and position estimation, a large number of particles are randomly generated. , X represents the system state vector, including the position information of the robot posture data in the drilling coordinate system and the robot posture angle, N is the total number of particles, k represents the current moment, (k-1) represents the previous moment. Each particle has a weight , initially all particles have equal weights, .

[0079] S22. Predict the state of the particle at the next moment based on the system's state transition equation. The MEMS inertial navigation module can measure the robot's acceleration and angular velocity, and the robot's position and posture changes can be approximated through integration operations. Specifically, the robot's posture data collected by the MEMS inertial navigation module at the previous moment is used to predict the state of the particle at the next moment. The state transition equation is as follows:

[0080] ,

[0081] is the robot posture data at the previous moment, where f() is the state transfer function, It is the predetermined drilling parameters input into the drilling rig at the last moment. is the process noise, which follows a Gaussian distribution , is the process noise covariance matrix, and the state of each particle at the previous moment is calculated according to the state transfer equation. Make a prediction and get the state of the particle at the next moment .

[0082] S23, using the three-dimensional point cloud data of the hole wall collected by the three-dimensional laser scanning probe and the environmental data of the hole wall collected by the visual sensor as observation data to update the weight of the particle, first obtain the particle observation value through the following formula,

[0083] ,

[0084] in is the observed value of particle i at time k, h() is the observation function, is the observation noise, which follows a Gaussian distribution , is the observation noise covariance matrix;

[0085] Then, based on the predicted particle state and observation data, the weight update value of each particle is calculated by the following formula:

[0086] ),

[0087] in ) is in the particle state Lower observation value The probability density function of can be calculated based on the Gaussian distribution of the observation noise.

[0088] Finally, the weights of all particles are normalized.

[0089] S24. After the weight update, the weights of some particles will become very small, while the weights of a few particles will be relatively large. In order to avoid the problem of particle degradation, the particles after weight update are resampled. According to the weight of the particles, M particles are resampled from the current particle set. The particles with larger weights have higher probability of being selected. Replace the original particle set, and all resampled particle weights are reset to .

[0090] S25. During the movement of the robot, the continuously collected robot posture data, hole wall three-dimensional point cloud data and environmental data are iterated by repeating the above steps. That is, as the robot moves, the prediction, update and resampling steps are continuously repeated, and the particle states obtained in a series of drilling coordinate systems are fitted to establish a three-dimensional model of the drilling trajectory of the front drilling. The least squares method can be used for model fitting.

[0091] S3. Obtain the deflection data of the front section of drilling through the three-dimensional model of the drilling trajectory, compare the drilling trajectory in the three-dimensional model with the set drilling trajectory to obtain the deflection angle and deflection direction of the front section of drilling, and determine that the drilling deflection threshold that needs to be corrected is 0.5°. If the deflection angle in the deflection data is greater than 0.5°, the intelligent drilling deflection detection robot will feed back the deflection data of the front section of drilling to the drilling rig control system to execute deviation correction for the rear section of drilling.

[0092] The transmission of deflection data is preferably done via 5G / Wi-Fi 6 wireless transmission. The control unit inside the robot body 1 has a built-in wireless communication module with a response delay of less than 100ms, supporting direct linkage with the drill rig control system. Once deflection data requiring correction is detected, it can react quickly and send adjustment instructions to the ground operating system or drill rig to achieve real-time correction. If the mine uses a segmented drilling process, such as pausing every 3-5m of drilling to lower the inspection robot, the robot will enter the front section of the drill hole during the pause to detect drill hole deflection data, and provide real-time feedback to the drill rig control system to adjust subsequent drilling parameters (such as drilling pressure, rotation speed, and drilling direction), achieving "drilling and measurement integrated" dynamic correction. If the drill rig supports measurement while drilling (MWD), the robot data can be integrated with the MWD system to achieve real-time correction without stopping drilling.

[0093] S4. If the three-dimensional model of the drilling trajectory obtains a deflection angle of ≤0.5° in the deflection data of the front section of the drilling, the collected robot posture data, the three-dimensional point cloud data of the hole wall and the environmental data are input into the trained drilling deflection prediction model, the deflection data of the tendency of the back section of the drilling is predicted, and the predicted deflection data is output and fed back to the drilling rig control system of the back section of the drilling. The deflection threshold of 0.5° can still be used to determine whether correction is required for the back section of the drilling.

[0094] like Figure 6 As shown in the figure, the drilling deviation prediction model is specifically based on a machine learning algorithm. It uses the data of the previous drilling period to predict the drilling deviation trend of the subsequent drilling period and automatically recommends the drilling pressure / speed adjustment plan for the drilling rig in the subsequent drilling period based on the prediction results. The training is specifically carried out through the following steps:

[0095] S41. Utilize the multi-sensor module of the intelligent drilling deflection robot to collect data in blastholes with different geological conditions and drilling parameters. This data covers information on different geological conditions, drilling depths, and construction stages. Organize the collected data, remove outliers and erroneous data, and sort the robot's multi-sensor time series data and in-hole geological parameters in the same blasthole according to the chronological order of segmented drilling. Extract the robot's posture data, hole wall 3D point cloud data, environmental data, and actual drilling deflection data from the sorted data to form a feature set for machine learning.

[0096] S42. Extract key features from the organized data. For robot posture data, extract features such as acceleration, angular velocity, and posture angle; for hole wall 3D point cloud data, extract hole wall shape features and distance information; and for environmental data, extract texture features within the borehole and positional changes of target objects. Given that borehole deflection prediction requires processing time series data and capturing complex patterns within the data, a long short-term memory (LSTM) network is selected as the prediction model. The feature set is divided into training and test sets, typically with a ratio of 80% to 20%. The robot posture data, hole wall 3D point cloud data, and environmental data from the training set are fed into the LSTM network for training. Appropriate hyperparameters, such as the learning rate and the number of hidden layer neurons, are set. The backpropagation algorithm is used to adjust the LSTM model weights so that the predicted borehole deflection data output by the model approximates the actual borehole deflection data corresponding to the training set. The model is optimized using the mean squared error (MSE) as the loss function until the loss function converges to a small value, indicating that the model has learned the patterns in the data. The training process of the long short-term memory network model is a mature machine learning model, and this embodiment does not elaborate on its specific training process.

[0097] S43. Use the test set to evaluate the long short-term memory network model after training and optimization, and use the root mean square error (RMSE), mean absolute error (MAE), and determination coefficient (R 2 Calculate the error metrics between the predicted borehole deviation data output by the model and the actual borehole deviation data corresponding to the test set. Cross-validate these metrics to determine whether the model's predictive performance meets expectations. If the model performance is poor, adjust the model structure or hyperparameters and retrain and evaluate. Upon completion of the evaluation, the borehole deviation prediction model is obtained.

[0098] The qualified model is deployed to the control unit of the intelligent drilling deviation detection robot through software. The control unit has a built-in storage module that stores the software program for executing the drilling deviation prediction model. The drilling deviation prediction program is called and executed by the processing module. During the actual drilling process, the posture data of the detection robot that performs the correction in step S3, the three-dimensional point cloud data of the hole wall, the environmental data, and the actual drilling deviation data are used as a new feature set. After feature extraction, they are input into the control unit. The deviation trend of the subsequent drilling is predicted by calling the execution drilling deviation prediction model. At the same time, as new drilling data is continuously generated, the model is regularly updated and trained with new data to adapt to different geological conditions and changes in the construction environment to ensure the prediction accuracy of the model. After receiving the correction adjustment, the drilling rig control system immediately adjusts the drilling pressure and speed of the drilling rig to achieve real-time correction, ensure that the drilling can be carried out according to the predetermined trajectory, and improve the drilling accuracy.

[0099] This innovative combination of intelligent sensing and real-time control promises to overcome the industry challenge of controlling borehole deflection in underground mines, providing key technical equipment for intelligent mine construction. In practice, different underground mines may have varying geological conditions and drilling requirements. However, the intelligent borehole deflection detection robot and real-time deflection correction method presented in this invention offer excellent adaptability, allowing for parameter adjustment and optimization based on specific circumstances to achieve optimal detection and correction results.

[0100] In this document, the directions or positional relationships indicated by terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside", "outside", "vertical", and "horizontal" are based on the directions or positional relationships shown in the accompanying drawings and are only for the clarity of the technical solution and the convenience of description, and therefore should not be understood as limiting the present invention.

[0101] As used herein, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion of elements other than the listed elements and may also include additional elements not specifically listed.

[0102] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A drilling deviation correction method, characterized in that: The steps include: S1. An intelligent drilling deflection detection robot is placed into the hole of the front section of the drilling hole. The intelligent drilling deflection detection robot includes a robot body (1). The robot body (1) has a cylindrical shell smaller than the diameter of the drilling hole. The end of the cylindrical shell is connected to a recovery cable (6). A plurality of sensor modules (2) are arranged at the head end. The sensor modules (2) include a MEMS inertial navigation module (21), a three-dimensional laser scanning probe (22), an ultrasonic sensor (23) and a visual sensor (24). The robot body (1) is provided with a control unit (3) and a power supply (5) in the shell for communicating with the sensor module. The robot body is driven to move along the wall of the front section of the hole. During the movement of the robot body, the MEMS inertial navigation module of the sensor module collects the posture data of the robot in the hole, the three-dimensional laser scanning probe scans the hole wall to collect the three-dimensional point cloud data of the hole wall, the ultrasonic sensor collects the distance data between the robot and the hole wall, and the visual sensor collects the environmental data of the hole wall. S2. De-noise the collected robot posture data, hole wall 3D point cloud data, and environmental data, and then fuse them using a particle filter algorithm. Combined with the drilling coordinate system, a 3D model of the drilling trajectory of the front section of the drilling is established. The construction of the 3D drilling trajectory model includes the following steps: S21, initialize the particles in the drilling coordinate system through the robot posture data, and randomly generate a large number of particles according to the initial robot posture data collected by the MEMS inertial navigation module , is the particle of the i-th particle at time k, X represents the system state vector, including the position information of the robot posture data in the drilling coordinate system and the posture angle of the robot, N is the total number of particles, k represents the current moment, k-1 represents the previous moment, and each particle has a weight , initially all particles have equal weights, ; S22. Based on the state transition equation of the system, the robot posture data collected by the MEMS inertial navigation module at the previous moment is used to predict the state of the particle at the next moment. The state transition equation is as follows: ， is the robot posture data represented by the particle at the last moment, where f( ) is the state transfer function, It is the predetermined drilling parameters input into the drilling rig at the last moment. is the process noise, which follows a Gaussian distribution , is the process noise covariance matrix, and according to the state transfer equation, each particle at the previous moment Make a prediction and get the particle size at the next moment The robot posture data represented; S23, using the three-dimensional point cloud data of the hole wall collected by the three-dimensional laser scanning probe and the environmental data of the hole wall collected by the visual sensor as observation data to update the weight of the particle, first obtain the particle observation value through the following formula, ， in is the observed value of particle i at time k, h( ) is the observation function, is the observation noise, which follows a Gaussian distribution , is the observation noise covariance matrix; Then, based on the predicted particle state and observation data, the weight update value of each particle is calculated by the following formula: )， in ) is in the particle state Lower observation value The probability density function of Finally, the weights of all particles are normalized; S24, resample the particles after weight update, and resample M particles from the current particle set according to the particle weight. Replace the original particle set, and all resampled particle weights are reset to ; S25, during the robot's movement, the robot posture data, the hole wall three-dimensional point cloud data, and the environmental data continuously collected are iterated by repeating the above steps, and the particle states obtained in a series of drilling coordinate systems are fitted to establish a three-dimensional model of the drilling trajectory of the front drilling section; S3. Deflection data of the preceding drilling section is obtained using the three-dimensional model of the drilling trajectory. If the deflection angle in the deflection data is greater than a deflection threshold, the intelligent drilling deflection detection robot feeds the deflection data of the preceding drilling section back to the drilling rig control system to perform deviation correction for the subsequent drilling section. S4. If the deflection angle in the front-end drilling deflection data obtained by the three-dimensional model of the drilling trajectory is less than or equal to the deflection threshold, the collected robot posture data, hole wall three-dimensional point cloud data and environmental data are input into the trained drilling deflection prediction model to predict the deflection trend of the rear-end drilling, and the predicted deflection data is output and fed back to the drilling rig control system of the rear-end drilling.

2. A drilling deviation correction method according to claim 1, characterized in that: At least three groups of elastic mechanical support arms (11) are symmetrically arranged along the circumferential direction on the outer side wall of the cylindrical shell of the intelligent drilling deflection detection robot. When the robot body enters the hole, the elastic mechanical support arms (11) are pressed against the hole wall and achieve rolling contact with the hole wall through the support wheels (12).

3. A drilling deviation correction method according to claim 2, characterized in that: The elastic mechanical support arm (11) uses a servo electric push rod to connect the support wheel (12) to achieve rolling contact with the hole wall. The support wheel (12) has a built-in pressure sensor and is feedback-connected to the servo electric push rod. The contact pressure feedback between the support wheel and the hole wall is used to adjust the support force of the elastic mechanical support arm on the hole wall.

4. A drilling deviation correction method according to claim 3, characterized in that: The supporting wheel (12) is driven by a motor.

5. The drilling deviation correction method according to claim 1, characterized in that: A glass protective cover (25) is provided at the head end of the cylindrical shell of the intelligent drilling deflection detection robot to cover all sensor modules.

6. A drilling deviation correction method according to claim 1, characterized in that: In step S1, when the intelligent drilling deviation detection robot moves in the hole, it navigates according to a pre-set drilling path, or uses a visual sensor to observe the hole wall environment of the previous drilling in real time, and uses SLAM combined with the drilling coordinate system to locate the robot's position in real time and construct a navigation path for the robot. During the robot's navigation process, the ultrasonic sensor collects the distance data between the robot and the hole wall for anti-collision warning.

7. The drilling deviation correction method according to claim 1, characterized in that: In step S2, a Kalman filter algorithm is used to denoise the collected robot posture data, hole wall three-dimensional point cloud data, and environmental data. The denoising process is as follows: the intelligent drilling deviation detection robot collects 30 seconds of robot posture data, hole wall three-dimensional point cloud data, distance data, and environmental data in a stationary state in the hole to establish a zero bias reference; during the movement of the robot, an FFT analysis is performed on the collected original data to identify the power frequency vibration components of the robot device noise and mechanical vibration frequency, and an anti-phase compensation amount with a phase opposite to and an amplitude matching the power frequency vibration components is injected into the state prediction equation; the state transfer matrix of the Kalman filter is adjusted and redundant check is started for the mutation data, and at the same time, abnormal observation values ​​output by the Kalman filter are identified and eliminated through multi-sensor redundant check or chi-square detection, and phase compensation is performed on the filtering results. Finally, the denoised robot posture data, hole wall three-dimensional point cloud data, and environmental data are smoothed and output.

8. The drilling deviation correction method according to claim 1, characterized in that: In step S4, the drilling deviation prediction model is trained through the following steps: S41. Using an intelligent drilling deviation robot to collect data from blastholes with different geological conditions and drilling parameters, sorting the data within the same blasthole according to the chronological order of segmented drilling, extracting robot posture data, hole wall 3D point cloud data, environmental data, and actual drilling deviation data from the sorted data to form a feature set for machine learning; S42, dividing the feature set into a training set and a test set, inputting the robot posture data, hole wall three-dimensional point cloud data, and environmental data of the training set into a long short-term memory network model for training, adjusting the weights of the long short-term memory network model using a back propagation algorithm so that the predicted drilling deflection data output by the model is close to the actual drilling deflection data corresponding to the training set, and optimizing the model using mean square error as a loss function; S43. Use the test set to evaluate the trained and optimized long short-term memory network model, calculate the error indicators between the drilling deviation data output by the model and the actual drilling deviation data corresponding to the test set through the root mean square error and mean absolute error, and complete the evaluation to obtain the drilling deviation prediction model.

9. A drilling deviation correction method according to claim 8, characterized in that: The drilling deviation prediction model is deployed in the control unit of the intelligent drilling deviation robot. During the actual drilling process, the posture data of the detection robot that performs correction in step S3, the three-dimensional point cloud data of the hole wall, the environmental data, and the actual drilling deviation data are used as new feature sets to update and train the drilling deviation prediction model.

Images