An intelligent geological drilling machine system and control method based on multi-source perception and autonomous decision

Abstract

The application discloses an intelligent geological drilling machine system and a control method based on multi-source sensing and autonomous decision-making, wherein a centralized numerical control center synchronously collects drilling data through a plurality of types of sensor groups, adopts extended Kalman filtering to fuse fifteen-dimensional heterogeneous sensing data, automatically identifies stratum lithology through a dual-mode redundant architecture of random forest and BP neural network, generates optimized drilling parameters through three-level progressive decision-making, and realizes closed-loop tracking of an actuator through a dual-loop incremental PID. The dual-mode execution system comprises a motor frequency conversion control, a hydraulic proportional valve control, an auxiliary electromagnetic reversing valve control, an audible and visual alarm and a water quantity control subsystem, realizes intelligent and manual non-disturbance switching through a four-link mechanism, and adopts a three-index weighted fusion model to realize real-time evaluation of a stuck drill pipe risk. The application realizes automatic operation in the whole drilling process and stratum adaptive control, and overcomes defects of artificial experience dependence, insufficient stratum identification, extensive parameter regulation and control and high accident rate.

Description

Technical Field

[0001] This invention belongs to the field of geological drilling equipment technology, and in particular relates to an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making. Background Technology

[0002] Geological drilling is a fundamental technical means in mineral resource exploration, engineering geological investigation, and geological disaster prevention. During drilling operations, the selection of drilling parameters directly determines drilling efficiency, borehole quality, and operational safety. These parameters mainly include drill pressure (DPP), rotation speed, and flushing fluid flow rate. DPP determines the drill bit's ability to break up the rock mass at the bottom of the hole; rotation speed affects the drill bit's cutting efficiency; and flushing fluid flow rate serves multiple functions, including cooling the drill bit, carrying away rock cuttings, and protecting the borehole wall. The proper matching of these three parameters requires dynamic adjustment based on the lithology of the current drilling formation and the borehole depth. Soft soil formations require low DPP, moderate rotation speed, and high-flow-rate flushing fluid to prevent borehole collapse; hard rock formations require high DPP and high rotation speed to maintain drilling efficiency; and when drilling through fractured rock masses, strict control of DPP and rotation speed is necessary to avoid borehole instability and stuck drill bit accidents. However, the lithology of the formation is often complex and varied in the direction of drilling depth. Heterogeneous geological structures such as interbedded soft and hard layers, fault fracture zones, and weathering and alteration zones frequently occur, making the real-time adjustment of drilling parameters a complex decision-making task that is highly dependent on the experience of the operators.

[0003] In existing technologies, the operation of traditional geological drilling rigs mainly relies on operators to judge formation changes by observing superficial characteristics such as the color of the returned cuttings, sensing drilling vibrations, and listening to mechanical noises. Drilling parameters are then adjusted manually by operators based on personal experience, adjusting hydraulic valves and gear shifts. Some improved drilling rigs monitor specific drilling parameters by adding single-type sensors. For example, pressure sensors on the vertical shaft cylinder monitor feed pressure, speed sensors on the motor shaft monitor rotation speed, and current and voltage sensors indirectly estimate drilling torque. These modifications display the sensor data on an instrument panel or monitoring screen for operator reference. However, a closed-loop control link is not formed between the sensor data and the actuators. Operators still need to manually operate hydraulic handles, regulating valves, and control switches based on the displayed data to adjust drilling parameters. In recent years, measurement while drilling (MWD) technology and machine learning-based lithology identification methods have emerged in the oil drilling field. These technologies can invert formation lithology through MWD data or surface drilling parameters. However, these technologies are mainly applied to deep oil and gas drilling scenarios and rely on expensive MWD instruments and professional logging data, making them difficult to directly apply to shallow geological drilling projects.

[0004] The existing technologies described above have the following drawbacks: First, operators' reliance on experience to judge formation changes is subjective and subject to delays. Operator training is lengthy, and experience is difficult to quantify and pass on. In complex formations, the accuracy of manual judgment is limited, often resulting in reactive parameter adjustments only after formation changes have occurred and caused drilling anomalies, missing the optimal adjustment window. Second, existing sensor modification schemes have a single sensing dimension, typically monitoring only two or three parameters such as pressure and rotational speed. They lack the ability to simultaneously sense multi-dimensional physical quantities such as vibration, dip angle, and flow rate. There is a lack of data fusion and analysis mechanisms, making it impossible to extract formation response information from multi-dimensional features. Third, the data collected by sensors is not used to generate closed-loop control commands; the actuators still rely on manual adjustments, resulting in a data chain break between sensing and execution, failing to achieve automated decision-making from formation identification to parameter optimization. Furthermore, existing solutions lack automatic control mechanisms for flushing fluid flow and quantitative assessment methods for stuck drill risk, failing to provide early warning and proactive protection before abnormal conditions occur. Overall, existing technologies have not yet achieved a complete automated closed loop from multi-source sensing, formation identification, parameter decision-making to execution control; therefore, it is necessary to propose an intelligent geological drilling rig system and control method based on multi-source sensing and autonomous decision-making to solve the above problems. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide an intelligent geological drilling rig system based on multi-source perception and autonomous decision-making. It aims to overcome the shortcomings of existing geological drilling rigs, such as single perception dimension, broken data link, reliance on manual experience for stratum identification, extensive parameter control and lack of closed-loop control, which leads to low drilling efficiency and high accident rate. It features multi-source heterogeneous sensor fusion perception, dual-mode redundant automatic stratum identification, three-level progressive adaptive decision-making, dual-loop closed-loop execution control and multi-dimensional safety protection.

[0006] To achieve the above technical solution, the technical solution adopted by the present invention is as follows: An intelligent geological drilling rig system based on multi-source perception and autonomous decision-making includes: an integrated numerical control center, a geological drilling rig body, and a modular detachable tower. The integrated CNC center includes a multi-dimensional fusion sensing module, a real-time decision control module, and a dual-mode execution system; The multi-sensor fusion module collects drilling parameters, mechanical status and formation response data in real time through a multi-type sensor group deployed on the main body of the geological drilling rig. It also uses an extended Kalman filter algorithm to perform time synchronization and fusion state estimation on the heterogeneous data of the multi-type sensor group and outputs the fused drilling state vector. The real-time decision control module receives the drilling status vector and performs a three-level progressive decision-making process: formation identification layer, parameter matching layer, and safety verification layer. The formation identification layer inputs the drilling status vector into a random forest classifier to output the current formation lithology category and confidence level. The parameter matching layer uses the lithology category as an index to retrieve the target drill pressure value and target rotation speed value from the formation-drilling parameter lookup table and performs depth compensation in combination with the borehole depth. The safety verification layer verifies the compensated target parameters against the equipment rated range and formation safety boundary before outputting the final control target parameters. The dual-mode execution system includes a motor frequency conversion control system, a hydraulic proportional valve control system, an auxiliary electromagnetic directional valve control system, an audible and visual alarm system, and a water volume control system, supporting parallel input and seamless switching of intelligent control commands and manual operation commands. The modular and detachable tower is attached to the main body of the geological drilling rig. Two pulleys are installed at the top of the tower. The wire rope of the winch converts the winching force into the pulling force for raising and lowering the drill rod through the two pulleys, realizing the automatic raising and lowering of the drill rod.

[0007] Preferably, the multi-type sensor group includes a pressure sensor, a displacement sensor, a speed sensor, a six-axis sensor, a tilt sensor, an inlet water flow sensor, a return water flow sensor, and a temperature and humidity sensor. Three pressure sensors are installed on the multi-way hydraulic valve seat, the lifting port of the vertical shaft cylinder, and the pressing port of the vertical shaft cylinder, respectively, to monitor system pressure, lifting pressure, and lowering pressure. The displacement sensor is installed next to the vertical shaft cylinder to monitor the vertical shaft displacement. Two speed sensors are installed on the motor shaft and the vertical shaft, respectively, to monitor the motor speed and the vertical shaft speed. Two six-axis sensors are installed on the drilling rig body and the vertical shaft, respectively, to sense the rig body attitude, monitor rig body vibration, and changes in vertical shaft attitude. The tilt sensor is installed on the vertical shaft to monitor the drilling tilt angle. The inlet water flow sensor and the return water flow sensor are installed at the inlet water pipe and the return water pipe at the borehole opening, respectively, to monitor the circulating water volume. The temperature and humidity sensor is installed inside the control box to monitor the operating environment temperature and humidity.

[0008] Preferably, the data fusion process of the extended Kalman filter in the multi-dimensional fusion sensing module is as follows: using GPS timing signal as a unified time reference, UTC timestamps are added to the data of each sensor; a sliding time window with a width of 500ms and a step size of 100ms is set; the asynchronous data of each channel within the window are linearly interpolated to the center time of the window according to the timestamps to form a time-aligned multi-channel data frame; the system pressure, lifting pressure, lower pressure, vertical shaft displacement, motor speed, vertical shaft speed, three-axis acceleration of the machine body, three-axis angular velocity of the machine body, borehole inclination angle, water inflow rate, and return water flow rate are constructed as observation vectors and input into the extended Kalman filter; the filter process model adopts a constant acceleration model; and the output is the fused drilling state vector. The formation identification layer of the real-time decision control module calculates drilling resistance characteristic values ​​based on the drilling state vector within each fusion cycle. Vibration energy characteristic value Rotation speed fluctuation characteristic value Flow change characteristic value Cumulative characteristic value of mud consumption Five stratigraphic response feature values ​​are used to form a stratigraphic response feature vector, which is input into the random forest classifier. The random forest classifier contains 200 CART decision trees. The Gini coefficient is used as the splitting criterion. Each decision tree performs independent classification voting. The one with the highest vote rate is used as the stratigraphic lithology identification result, and the vote rate is used as the confidence level. In the parameter matching layer, when the confidence level is greater than or equal to 0.6, the formation-drilling parameter reference table is directly retrieved using the lithology identification result as the key to obtain the target drill pressure reference value and the target rotation speed reference value. When the confidence level is less than 0.6, the two candidate lithology categories with the highest probability values ​​output by the random forest and their probabilities are selected. The reference table parameters of the two categories are weighted and interpolated according to the probability ratio to generate a transitional reference value. The depth compensation formula is: ; in To retrieve or interpolate the drill pressure reference value, Mass per unit length of drill pipe It is the acceleration due to gravity. The current drilling depth is used as the reference depth. The safety verification layer compares the compensated target drilling pressure value with the maximum allowable pressure of the hydraulic system and the maximum allowable drilling pressure of the formation, and takes the minimum value. It also compares the target rotation speed value with the rated rotation speed of the frequency converter and takes the minimum value. Finally, it outputs the control target parameter pair.

[0009] Preferably, the hydraulic proportional valve control system includes an electromagnetic proportional directional valve and an electromagnetic proportional relief valve; the control signal of the electromagnetic proportional directional valve is a voltage quantity of ±0 to 10 volts, which is converted into a corresponding analog voltage through a 485 communication module for stepless proportional adjustment of the vertical shaft feed speed and feed force; the control signal of the electromagnetic proportional relief valve is a standard current signal of 4 to 20 mA, which is converted into a current signal of 4 to 20 mA through a dedicated signal conversion control board for output, for precise setting and dynamic adjustment of system pressure; the 485 communication module supports dual-channel ±0 to 10 volt signal output, simultaneously controlling the electromagnetic proportional directional valve and the electromagnetic proportional relief valve.

[0010] Preferably, the auxiliary electromagnetic reversing valve control system includes front and rear control cylinders and a chuck cylinder, wherein the front and rear control cylinders are independently controlled by one electromagnetic reversing valve, and the chuck cylinder is independently controlled by another electromagnetic reversing valve. The motor frequency conversion control system supports 485 signal communication. The integrated CNC center sends speed command to the frequency converter and receives operating status feedback through the 485 communication interface. The water control system includes an electric regulating valve, an inlet flow sensor, a return flow sensor, and a water control unit. The water control unit dynamically generates the target flushing fluid flow rate according to the lithology of the formation and the borehole depth according to the zoning rules. It uses the real-time measurement value of the inlet flow sensor as feedback and drives the opening of the electric regulating valve through an incremental PID algorithm to make the actual flushing fluid flow rate track the target value. At the same time, it continuously monitors the leakage and water inrush conditions in the borehole through the difference between the inlet and outlet flow rates.

[0011] Preferably, the seamless handover includes four stages: handover triggering, state freezing, control quantity alignment, and shadow following. The switching trigger conditions include: the operator presses the physical switching button on the console, the system self-test detects that any 485 communication module has not returned a valid response frame for 3 consecutive communication cycles, or the formation identification confidence level output by the real-time decision control module is lower than 0.3 for 10 consecutive control cycles. The system starts the switching process when any of the conditions are met. When the state freeze occurs during switching startup, the dual-mode execution system immediately freezes all control variables for the current control cycle, including the voltage value of the electromagnetic proportional directional valve. Electromagnetic proportional relief valve current value Variable frequency drive speed setpoint , Front and rear control cylinder solenoid directional valve switch status chuck cylinder solenoid directional valve switch status and the opening value of the electric regulating valve A six-element control state snapshot is generated and stored in the shared memory area; Control alignment is as follows: When switching from intelligent mode to manual mode, the electrical zero positions of the joysticks, knobs, and buttons on the manual control panel are remapped to the frozen control state snapshot values ​​using a linear mapping function, so that the output corresponding to the current position of the control device is equal to the control quantity at the time of freezing. The mapping relationship is as follows: ; in For freezing control quantity, The sensitivity coefficient of the joystick is determined by the factory calibration. For the recalculated zero-position mapping point The current physical location corresponds to When switching back from manual mode to intelligent mode, the intelligent controller first reads the control values ​​corresponding to the current manual operation as the initial output values ​​of the incremental PID controller. Then, in each control cycle, it gradually transitions to the target value calculated by the intelligent algorithm; Shadow following means that after switching to manual mode, the intelligent controller continues to run in the background in shadow mode. Each control cycle calculates the theoretical target control quantity but does not output it to the actuator. When the deviation between the theoretical target control quantity and the actual manual control quantity exceeds 15% of the corresponding control channel range and lasts for more than 5 seconds, the HMI interface will pop up a prompt message to suggest that the operator confirm the current operation.

[0012] Preferably, a control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making includes the following steps: S1 synchronously collects data on inlet and outlet water flow, vertical shaft lifting pressure, vertical shaft lowering pressure, system pressure, displacement, rotation speed, machine body vibration, and drilling inclination angle using multiple types of sensor groups at a period of 500ms. S2. After performing sliding median filtering and time alignment preprocessing on the collected multi-source data, five formation response feature values ​​are calculated: drilling resistance feature value, vibration energy feature value, rotation speed fluctuation feature value, flow rate change feature value, and mud consumption accumulation feature value. The formation response feature vector is then input into a random forest classifier to invert the lithology and confidence level of the current drilling formation in real time. S3, based on the identified formation lithology and confidence level, retrieves or generates target drill pressure and target rotation speed values ​​from a pre-set drill pressure-rotation speed matching parameter library, and outputs the final control target parameters after depth compensation and safety boundary verification; during drilling, the parameters are adjusted according to the borehole depth. With critical depth The relationship dynamically adjusts the action mode of the vertical shaft hydraulic cylinder, and the critical depth. Determined by the target drill pressure and the weight per unit length of the drill pipe; S4 generates adaptive control commands, which drive the electromagnetic proportional relief valve and frequency converter respectively through the incremental PID controllers of the drilling pressure control loop and the speed control loop, so that the actual drilling parameters track the final control target parameters. S5 continuously monitors equipment operating status and calculates the stuck drill risk index. ,when When the set threshold is exceeded or an abnormal operating condition is detected, an audible and visual alarm is triggered and the machine is automatically shut down, or the machine is transferred to manual control mode according to the described non-disruptive switching process.

[0013] Preferably, the calculation formula for the five formation response characteristic values ​​in step S2 is as follows: Drilling resistance characteristic value: ; in For downward pressure, This represents the change in vertical axis displacement during the fusion cycle. The duration of the fusion cycle; vibrational energy characteristic value: ; in This represents the number of sampling points from the six-axis sensor within the fusion period. , , For the first Three-axis acceleration of the fuselage at each sampling point; characteristic value of rotational speed fluctuation. ,in The standard deviation of the vertical shaft speed within the fusion cycle. The average vertical shaft speed within this cycle; characteristic value of flow rate variation: ; in The number of flow sensor sampling points within the fusion period. and For the first Inflow and outflow rates at each sampling point; cumulative mud consumption characteristic value: ; in This refers to the sampling interval of the flow sensor.

[0014] Preferably, the critical depth mentioned in step S3 ,in For the target drilling pressure value, Mass per unit length of drill pipe m / s², The physical meaning is that the drill pipe's self-weight is exactly equal to the borehole depth when the target drill pressure is equal; when When the drilling is determined to be in the shallow hole drilling stage, the vertical shaft hydraulic cylinder applies downward pressure. ,in For the hydraulic system's floating back pressure, the cylinder actively presses down during this stage to compensate for the insufficient weight of the drill pipe; when This is considered a transition phase, where the downforce decreases linearly to zero with increasing depth; when... When the drilling is determined to be in the deep hole drilling stage, the vertical shaft cylinder switches to the lifting control mode, and the lifting force is increased. The hydraulic cylinder lifts the drill rod to partially offset its own weight, preventing the entire weight of the drill rod from pressing on the drill bit and causing excessive drilling pressure, thus maintaining the effective pressure of the drill bit on the bottom of the hole at a constant level. Within the allowable deviation range; When the lithology type of the strata changes, the control parameters change from the current value. Transition to the new target value ,in This indicates the current value of the control parameter. This indicates the new target values ​​for the control parameters, including target values ​​for drilling pressure, rotational speed, and flushing fluid flow rate; the transition process uses an S-shaped acceleration / deceleration curve within a preset transition time. Smooth transition within, the first The target parameter value for each control cycle is ,in The total number of transition cycles, To control the period, normalize the sigmoid function: ; Steepness parameters This segmentation definition ensures that the transition process starts precisely at... , terminated The rate of change is close to zero at both ends to avoid hydraulic shock and hole wall damage, while a rapid transition is made in the middle section to shorten the response time. Determined based on the hardness grade difference of the formation; the larger the grade difference... The longer.

[0015] Preferably, the quantitative calculation model for the stuck drill risk index in step S5 is as follows: ; ; ; ; in , , These are the weighting coefficients; This is a stress anomaly index. This is the sliding average of the downforce over the last 30 seconds. A positive and continuously increasing value indicates that the drill bit is encountering abnormal resistance. This is the torque anomaly index. This is the current effective value of the motor current fed back by the frequency converter. This is the moving average of the current over the last 30 seconds; For traffic anomaly index, when ,otherwise , An increase indicates that the circulation channel is blocked; when The system can detect the risk of stuck drill pipe in real time. The weighted fusion scheme of the three indicators makes the judgment of stuck drill pipe cover three independent physical dimensions: mechanical resistance, rotational resistance and circulation channel. When any one dimension is abnormal, the comprehensive index may not trigger the threshold. Only when multiple dimensions deteriorate at the same time will it be judged as stuck drill pipe, which greatly reduces the false alarm rate. The pressure indicator alone cannot distinguish between stuck drill pipe and normal hard rock. The torque indicator alone is not sensitive to slight sticking of drill pipe. The flow rate indicator alone may also fluctuate when the formation changes normally. The fusion of the three indicators makes up for the blind spots of the single indicator.

[0016] The beneficial effects of this invention are as follows: 1. This invention constructs a closed-loop control architecture through an integrated numerical control center. A multi-sensor fusion module uses extended Kalman filtering to perform time synchronization and fusion state estimation on 15-dimensional heterogeneous data from multiple sensor groups, outputting a fused drilling state vector. A real-time decision control module then executes three levels of progressive decision-making: formation identification, parameter matching, and safety verification, generating the final control target parameters. Finally, incremental PID controllers in the drill pressure control loop and speed control loop of the dual-mode execution system drive the electromagnetic proportional relief valve and frequency converter respectively, achieving fully automatic closed-loop adjustment of drilling parameters. This solution fundamentally eliminates reliance on operator experience and judgment, solving the shortcomings of existing technologies where drilling processes heavily rely on real-time manual adjustments, operator training cycles are long, and human judgment is subject to lag and subjectivity, leading to operational efficiency bottlenecks. It enables geological drilling rigs to independently complete drilling tasks in complex formations, improving operational consistency and reliability.

[0017] 2. This invention utilizes a multi-sensor group, including a six-axis sensor, pressure sensor, displacement sensor, rotation speed sensor, flow sensor, and tilt sensor, to simultaneously collect data on drilling resistance, vibration energy, rotation speed fluctuations, and inflow / outflow differences from multiple physical dimensions. Based on a fused state vector, it calculates five formation response features: drilling resistance, vibration energy, rotation speed fluctuation, flow rate change, and mud consumption accumulation. A dual-mode redundant architecture, consisting of a random forest master classifier and a BP neural network auxiliary classifier, is employed for real-time inversion of formation lithology. When the results from the two classifiers are inconsistent, the confidence level is actively reduced, triggering a weighted interpolation strategy. This solution addresses the shortcomings of existing technologies where operators can only infer lithology based on superficial features such as cutter color and drilling rig vibration, and lack the ability to identify complex geological structures such as concealed fracture zones, thin interbedded layers, and abrupt changes in soft and hard rocks. It provides a reliable formation basis for subsequent adaptive parameter adjustments.

[0018] 3. This invention utilizes a real-time decision control module to dynamically generate combinations of drill pressure and rotation speed control parameters based on formation identification results and borehole depth. This drives a proportional directional valve, a proportional relief valve, and a frequency converter to achieve continuous stepless adjustment of feed pressure and rotation speed. Simultaneously, a water control unit dynamically adjusts the flushing fluid flow rate according to formation zoning rules to balance wall protection, cuttings removal, and cooling requirements. In terms of safety protection, the system uses a three-indicator weighted fusion stuck drill risk index to assess drilling safety in real time. When the risk exceeds a threshold or an abnormal condition is detected, an audible and visual alarm and automatic shutdown are triggered. Seamless switching between intelligent control and manual operation is achieved through four stages: switching trigger, state freezing, control quantity alignment, and shadow following, ensuring no sudden changes in control quantities during switching. This solution addresses the shortcomings of existing technologies that use fixed levels or simple threshold control for drilling parameters, cannot dynamically adjust them in real time according to formation mechanical characteristics, and frequently cause stuck drills, drill pipe breakage, and borehole collapse due to improper parameter adjustments. It significantly enhances the system's adaptability in heterogeneous and complex formations, reduces accident risks, and ensures operational safety. Attached Figure Description

[0019] Figure 1 This is a front view of the intelligent geological drilling rig system described in this embodiment of the invention; Figure 2 This is a side view of the intelligent geological drilling rig system described in an embodiment of the present invention; Figure 3 This is a top view of the intelligent geological drilling rig system described in this embodiment of the invention; Figure 4 This is a 45-degree side view of the intelligent geological drilling rig system described in this embodiment of the invention; Figure 5 This is a schematic diagram of the system architecture of the present invention; Figure 6 This is a schematic diagram of the method flow of the present invention; Figure 7 This is a schematic diagram of the curves showing the variation of five-dimensional formation response characteristic values ​​with drilling depth and the formation lithology classification results in an embodiment of the present invention; Figure 8 This is a schematic diagram comparing the weighted fusion assessment of three indicators for stuck drill risk with a single indicator method in an embodiment of the present invention; Figure 9 This is a schematic diagram comparing the smooth transition of the S-shaped curve and the step transition during formation switching in an embodiment of the present invention; In the diagram: 1. Integrated CNC center; 2. Geological drilling rig body; 3. Modular detachable tower. Detailed Implementation

[0020] Example 1: like Figures 1-4As shown, an intelligent geological drilling rig system based on multi-source perception and autonomous decision-making includes: an integrated numerical control center 1, a geological drilling rig body 2, and a modular detachable tower 3.

[0021] like Figure 5 As shown, the integrated CNC center further includes a multi-dimensional fusion sensing module, a real-time decision control module, and a dual-mode execution system; The multi-sensor fusion module collects drilling parameters, mechanical status and formation response data in real time through a multi-type sensor group deployed on the main body of the geological drilling rig. It also uses an extended Kalman filter algorithm to perform time synchronization and fusion state estimation on the heterogeneous data of the multi-type sensor group and outputs the fused drilling state vector. The real-time decision control module receives the drilling status vector and performs a three-level progressive decision-making process: formation identification layer, parameter matching layer, and safety verification layer. The formation identification layer inputs the drilling status vector into a random forest classifier to output the current formation lithology category and confidence level. The parameter matching layer uses the lithology category as an index to retrieve the target drill pressure value and target rotation speed value from the formation-drilling parameter lookup table and performs depth compensation in combination with the borehole depth. The safety verification layer verifies the compensated target parameters against the equipment rated range and formation safety boundary before outputting the final control target parameters. The dual-mode execution system includes a motor frequency conversion control system, a hydraulic proportional valve control system, an auxiliary electromagnetic directional valve control system, an audible and visual alarm system, and a water volume control system, supporting parallel input and seamless switching of intelligent control commands and manual operation commands. The modular and detachable tower is attached to the main body of the geological drilling rig. Two pulleys are installed at the top of the tower. The wire rope of the winch converts the winching force into the pulling force for raising and lowering the drill rod through the two pulleys, realizing the automatic raising and lowering of the drill rod.

[0022] Furthermore, before extending the Kalman filter, the multi-channel fusion sensing module first performs a sliding median filter on the raw data of each channel to remove impulse interference. The median filter window width is 5 sampling points. The specific data fusion process of the extended Kalman filter is as follows: using the GPS timing signal as a unified time reference, each frame of data collected by each sensor is stamped with a UTC timestamp. The sliding time window width is set to 500ms and the window sliding step size is 100ms. The asynchronous data of each channel within the window are uniformly aligned to the center time of the window according to the timestamp through linear interpolation, forming a time-aligned multi-channel data frame. The system pressure Increase pressure Down pressure Vertical axis displacement Motor speed Vertical shaft speed Three-axis acceleration of the fuselage , , Angular velocity of the fuselage on three axes , , Drilling angle Inlet flow rate and return water flow A total of fifteen dimensions of data are used to construct an observation vector. The input is fused and estimated using an extended Kalman filter. The filter process model adopts a constant acceleration model, and the state prediction equation is: ; The observation update equation is: ; in Here is the state transition matrix. To control the input matrix, For the observation matrix, Here is the Kalman gain matrix. For the first The observation vectors of each fusion cycle are used to output the fused drilling state vector. This serves as the unified input for subsequent stratigraphic identification and decision-making modules.

[0023] Furthermore, the signal flow relationship of the five subsystems in the dual-mode execution system is as follows: the final control target parameters output by the real-time decision control module are respectively sent to the hydraulic proportional valve control system and the motor frequency conversion control system, while the formation lithology type and borehole depth are sent to the water flow control system to generate the target flushing fluid flow rate; the auxiliary electromagnetic reversing valve control system receives the electromagnetic reversing valve switching command output by the real-time decision control module and operates independently of the drilling parameter adjustment loop; the audible and visual alarm system receives alarm signals from the real-time decision control module and the water level control system, including six types of alarm signals: stuck drill risk alarm, equipment overload alarm, sensor abnormality alarm, communication interruption alarm, borehole leakage alarm, and water inrush alarm, which are arbitrated according to priority and then drive the corresponding indicator lights and buzzers.

[0024] Preferably, the multi-type sensor group includes a pressure sensor, a displacement sensor, a speed sensor, a six-axis sensor, a tilt sensor, an inlet water flow sensor, a return water flow sensor, and a temperature and humidity sensor. Three pressure sensors are installed on the multi-way hydraulic valve seat, the lifting port of the vertical shaft cylinder, and the pressing port of the vertical shaft cylinder, respectively, to monitor system pressure, lifting pressure, and lowering pressure. The displacement sensor is installed next to the vertical shaft cylinder to monitor the vertical shaft displacement. Two speed sensors are installed on the motor shaft and the vertical shaft, respectively, to monitor the motor speed and the vertical shaft speed. Two six-axis sensors are installed on the drilling rig body and the vertical shaft, respectively, to sense the rig body attitude, monitor rig body vibration, and changes in vertical shaft attitude. The tilt sensor is installed on the vertical shaft to monitor the drilling tilt angle. The inlet water flow sensor and the return water flow sensor are installed at the inlet water pipe and the return water pipe at the borehole opening, respectively, to monitor the circulating water volume. The temperature and humidity sensor is installed inside the control box to monitor the operating environment temperature and humidity.

[0025] Preferably, the data fusion process of the extended Kalman filter in the multi-dimensional fusion sensing module is as follows: using GPS timing signal as a unified time reference, UTC timestamps are added to the data of each sensor; a sliding time window with a width of 500ms and a step size of 100ms is set; the asynchronous data of each channel within the window are linearly interpolated to the center time of the window according to the timestamps to form a time-aligned multi-channel data frame; the system pressure, lifting pressure, lower pressure, vertical shaft displacement, motor speed, vertical shaft speed, three-axis acceleration of the machine body, three-axis angular velocity of the machine body, borehole inclination angle, water inflow rate, and return water flow rate are constructed as observation vectors and input into the extended Kalman filter; the filter process model adopts a constant acceleration model; and the output is the fused drilling state vector. The formation identification layer of the real-time decision control module calculates drilling resistance characteristic values ​​based on the drilling state vector within each fusion cycle. Vibration energy characteristic value Rotation speed fluctuation characteristic value Flow change characteristic value Cumulative characteristic value of mud consumption Five stratigraphic response feature values ​​are used to form a stratigraphic response feature vector, which is input into the random forest classifier. The random forest classifier contains 200 CART decision trees. The Gini coefficient is used as the splitting criterion. Each decision tree performs independent classification voting. The one with the highest vote rate is used as the stratigraphic lithology identification result, and the vote rate is used as the confidence level. In the parameter matching layer, when the confidence level is greater than or equal to 0.6, the formation-drilling parameter reference table is directly retrieved using the lithology identification result as the key to obtain the target drill pressure reference value and the target rotation speed reference value. When the confidence level is less than 0.6, the two candidate lithology categories with the highest probability values ​​output by the random forest and their probabilities are selected. The reference table parameters of the two categories are weighted and interpolated according to the probability ratio to generate a transitional reference value. The depth compensation formula is: ; in To retrieve or interpolate the drill pressure reference value, Mass per unit length of drill pipe It is the acceleration due to gravity. The current drilling depth is used as the reference depth. The safety verification layer compares the compensated target drilling pressure value with the maximum allowable pressure of the hydraulic system and the maximum allowable drilling pressure of the formation, and takes the minimum value. It also compares the target rotation speed value with the rated rotation speed of the frequency converter and takes the minimum value. Finally, it outputs the control target parameter pair.

[0026] Furthermore, the specific design of the extended Kalman filter is as follows: State vector The process model is 15-dimensional and adopts a constant acceleration model, which assumes that the second derivative of the drilling state vector is driven by Gaussian white noise. The discretized state transition matrix is... It is a 15×15 block diagonal matrix, where the displacement The corresponding sub-block is: ; System pressure Increase pressure Down pressure Inlet flow rate and return water flow The corresponding sub-blocks use a random walk model. Motor speed and vertical shaft speed The corresponding sub-blocks use a first-order Markov model; triaxial acceleration , , and angular velocity , , The corresponding sub-blocks adopt a first-order Markov model; borehole inclination angle A constant-value model is used because the changes are slow during drilling. Observation matrix This is a 15×15 identity matrix, representing the state variables directly observed by each sensor. The process noise covariance matrix is ​​also shown. A modular differentiated design is adopted, with displacement and speed variables corresponding to... The diagonal elements of the sub-blocks are taken as 0.05 times the variance of the corresponding sensor measurements; the sub-blocks corresponding to pressure and flow rate variables are taken as 0.2 times the variance of the corresponding sensor measurements; and the sub-blocks corresponding to vibration acceleration and angular velocity variables are taken as 0.5 times the variance of the corresponding sensor measurements. These scaling factors can be fine-tuned according to the actual drilling conditions during the on-site commissioning phase. (Observation noise covariance matrix) The diagonal elements are taken as the measurement variance values ​​calibrated at the factory for each sensor.

[0027] Furthermore, among the five formation response characteristic values: drilling resistance characteristic value: ; in For vertical shaft feed rate, This represents the change in vertical axis displacement within the last 500ms. ms, The unit is N·s / mm, which physically reflects the specific work required for the drill bit to overcome formation resistance; vibration energy characteristic value: ; in This represents the number of sampling points for the six-axis sensor within a 500ms window. , , For the first The fuselage triaxial acceleration values ​​at each sampling point The unit is m / s², which physically reflects the total mechanical vibration energy generated by the interaction between the drill bit and the formation; the characteristic value of rotational speed fluctuation. ,in The standard deviation of the vertical shaft rotation speed within the window. This represents the average vertical shaft rotation speed within this window. Dimensionless, physically reflecting the degree of disturbance of drill bit rotation caused by formation heterogeneity; characteristic value of flow rate variation: ; in This represents the number of sampling points for the flow sensor within the window. and For the first The inflow and outflow rates at each sampling point The unit is m³ / h, which physically reflects the degree of absorption or seepage of flushing fluid by formation pores and fractures; cumulative mud consumption characteristic value: ; in The sampling interval for the flow sensor. The unit is m³, which physically reflects the net leakage of fluid within the pore over a period of time. The above five characteristic values ​​are combined to form a formation response characteristic vector. , as input to the stratigraphic identification model.

[0028] Furthermore, the maximum depth of each CART decision tree in the random forest classifier is 15, the minimum number of leaf node samples is 5, and the calibration drilling data with known lithology labels is used as the training set during the training phase. The formation-drilling parameter reference table uses formation lithology as the primary key. Each record includes four fields: target drill pressure reference value, target rotation speed reference value, recommended flushing fluid flow rate, and maximum permissible drilling speed. Specific parameter values ​​are as follows: soft soil corresponds to a target drill pressure reference value of 500 to 1000 N and a target rotation speed reference value of 60 to 120 rpm; soft rock corresponds to a target drill pressure reference value of 1000 to 2000 N and a target rotation speed reference value of 120 to 200 rpm; hard rock corresponds to a target drill pressure reference value of 2000 to 3500 N and a target rotation speed reference value of 200 to 350 rpm; hard rock corresponds to a target drill pressure reference value of 3500 to 5000 N and a target rotation speed reference value of 350 to 500 rpm; and fractured rock mass corresponds to a target drill pressure reference value of 800 to 1500 N and a target rotation speed reference value of 80 to 150 rpm. The reference table is established through field calibration tests during the system initialization phase and stored in the non-volatile memory of the integrated CNC center.

[0029] Furthermore, in the safety verification layer, the maximum permissible pressure of the hydraulic system is 90% of the rated working pressure of the hydraulic system, determined by the parameters on the equipment nameplate; the maximum permissible drilling pressure of the formation is set according to the lithology of the formation: 1500N for soft soil, 3000N for soft rock, 5000N for hard rock, 7000N for hard rock, and 2000N for fractured rock. The above values ​​are recommended upper limits based on conventional hole diameters of 75mm to 110mm. During the field calibration stage, these values ​​can be adjusted according to the actual hole diameter and drill bit type and updated to the system configuration parameter table.

[0030] Furthermore, the three-level progressive decision chain adopts a fusion architecture of random forest coarse classification and confidence gating. Random forest, based on decision tree ensemble, has natural robustness and anti-overfitting ability for high-dimensional features, making it suitable for processing noisy drilling data and providing highly robust multi-class classification and natural confidence. Confidence gating triggers weighted interpolation instead of forced decision when the classification result is unreliable, avoiding the defects of low-confidence misclassification leading to incorrect parameter output when using random forest alone, or the inability to cope with stratigraphic transition zones when using pure table lookup alone.

[0031] Preferably, the hydraulic proportional valve control system includes an electromagnetic proportional directional valve and an electromagnetic proportional relief valve; the control signal of the electromagnetic proportional directional valve is a voltage quantity of ±0 to 10 volts, which is converted into a corresponding analog voltage through a 485 communication module for stepless proportional adjustment of the vertical shaft feed speed and feed force; the control signal of the electromagnetic proportional relief valve is a standard current signal of 4 to 20 mA, which is converted into a current signal of 4 to 20 mA through a dedicated signal conversion control board for output, for precise setting and dynamic adjustment of system pressure; the 485 communication module supports dual-channel ±0 to 10 volt signal output, simultaneously controlling the electromagnetic proportional directional valve and the electromagnetic proportional relief valve.

[0032] Preferably, the auxiliary electromagnetic reversing valve control system includes front and rear control cylinders and a chuck cylinder, wherein the front and rear control cylinders are independently controlled by one electromagnetic reversing valve, and the chuck cylinder is independently controlled by another electromagnetic reversing valve. The motor frequency conversion control system supports 485 signal communication. The integrated CNC center sends speed command to the frequency converter and receives operating status feedback through the 485 communication interface. The water control system includes an electric regulating valve, an inlet flow sensor, a return flow sensor, and a water control unit. The water control unit dynamically generates the target flushing fluid flow rate according to the lithology of the formation and the borehole depth according to the zoning rules. It uses the real-time measurement value of the inlet flow sensor as feedback and drives the opening of the electric regulating valve through an incremental PID algorithm to make the actual flushing fluid flow rate track the target value. At the same time, it continuously monitors the leakage and water inrush conditions in the borehole through the difference between the inlet and outlet flow rates.

[0033] Furthermore, in the water volume control system, the target flushing fluid flow rate... The partitioning rules are as follows: When the strata are soft soil To enhance wall protection and debris-carrying capacity; when the strata lithology is fractured rock mass When the strata are composed of soft or hard rock When the strata are hard rocks This is to reduce unnecessary erosion of the orifice walls; the reference flushing fluid flow rate... According to the borehole diameter and drill pipe outer diameter According to the circular return speed m / s calculation The water volume control unit adopts the same incremental PID closed-loop control as the drilling pressure control loop, in order to... The set value is the real-time measurement value of the inlet water flow sensor. For feedback, deviation The control quantity, after being limited, drives the opening of the electric regulating valve, so that... track PID parameters , , The reference ranges are [0.2, 1.0], [0.02, 0.1], and [0.05, 0.2], respectively, with a control cycle of 500ms. Specific values ​​are tuned using the step response method during on-site commissioning and then stored in the configuration parameter table. The leakage criterion within the orifice is the difference between the inlet and outlet water flow rates. continuously greater than for 30 seconds 20%, the criterion for determining water inrush is continuously greater than for 30 seconds Both operating conditions, at 120% of their rated capacity, report corresponding alarm signals to the real-time decision control module.

[0034] Preferably, the seamless handover includes four stages: handover triggering, state freezing, control quantity alignment, and shadow following. The switching trigger conditions include: the operator presses the physical switching button on the console, the system self-test detects that any 485 communication module has not returned a valid response frame for 3 consecutive communication cycles, or the formation identification confidence level output by the real-time decision control module is lower than 0.3 for 10 consecutive control cycles. The system starts the switching process when any of the conditions are met. When the state freeze occurs during switching startup, the dual-mode execution system immediately freezes all control variables for the current control cycle, including the voltage value of the electromagnetic proportional directional valve. Electromagnetic proportional relief valve current value Variable frequency drive speed setpoint , Front and rear control cylinder solenoid directional valve switch status chuck cylinder solenoid directional valve switch status and the opening value of the electric regulating valve A six-element control state snapshot is generated and stored in the shared memory area; Control alignment is as follows: When switching from intelligent mode to manual mode, the electrical zero positions of the joysticks, knobs, and buttons on the manual control panel are remapped to the frozen control state snapshot values ​​using a linear mapping function, so that the output corresponding to the current position of the control device is equal to the control quantity at the time of freezing. The mapping relationship is as follows: ; in For freezing control quantity, The sensitivity coefficient of the joystick is determined by the factory calibration. For the recalculated zero-position mapping point The current physical location corresponds to When switching back from manual mode to intelligent mode, the intelligent controller first reads the control values ​​corresponding to the current manual operation as the initial output values ​​of the incremental PID controller. Then, in each control cycle, it gradually transitions to the target value calculated by the intelligent algorithm; Shadow following means that after switching to manual mode, the intelligent controller continues to run in the background in shadow mode. Each control cycle calculates the theoretical target control quantity but does not output it to the actuator. When the deviation between the theoretical target control quantity and the actual manual control quantity exceeds 15% of the corresponding control channel range and lasts for more than 5 seconds, the HMI interface will pop up a prompt message to suggest that the operator confirm the current operation.

[0035] like Figure 6 As shown, a control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making includes the following steps: S1 synchronously collects data on inlet and outlet water flow, vertical shaft lifting pressure, vertical shaft lowering pressure, system pressure, displacement, rotation speed, machine body vibration, and drilling inclination angle using multiple types of sensor groups at a period of 500ms. S2. After performing sliding median filtering and time alignment preprocessing on the collected multi-source data, five formation response feature values ​​are calculated: drilling resistance feature value, vibration energy feature value, rotation speed fluctuation feature value, flow rate change feature value, and mud consumption accumulation feature value. The formation response feature vector is then input into a random forest classifier to invert the lithology and confidence level of the current drilling formation in real time. S3, based on the identified formation lithology and confidence level, retrieves or generates target drill pressure and target rotation speed values ​​from a pre-set drill pressure-rotation speed matching parameter library, and outputs the final control target parameters after depth compensation and safety boundary verification; during drilling, the parameters are adjusted according to the borehole depth. With critical depth The relationship dynamically adjusts the action mode of the vertical shaft hydraulic cylinder, and the critical depth. Determined by the target drill pressure and the weight per unit length of the drill pipe; S4 generates adaptive control commands, which drive the electromagnetic proportional relief valve and frequency converter respectively through the incremental PID controllers of the drilling pressure control loop and the speed control loop, so that the actual drilling parameters track the final control target parameters. S5 continuously monitors equipment operating status and calculates the stuck drill risk index. ,when When the set threshold is exceeded or an abnormal operating condition is detected, an audible and visual alarm is triggered and the machine is automatically shut down, or the machine is transferred to manual control mode according to the described non-disruptive switching process.

[0036] Furthermore, the specific data collected in step S1 includes: influent flow rate. Return water flow rate Vertical shaft lifting pressure Vertical shaft downforce System pressure Vertical axis displacement Motor speed Vertical shaft speed Three-axis acceleration of the fuselage , , Angular velocity of the fuselage on three axes , , and drilling inclination angle There are a total of fifteen data channels.

[0037] Further, the data preprocessing in step S2 is as follows: first, the raw data of each sensor channel is subjected to sliding median filtering to remove pulse interference, and the filtering window width is 5 sampling points; then, the data of each channel is aligned to a unified 500ms fusion period according to the GPS timestamp. For channels with a sampling rate higher than 2Hz, the arithmetic mean of all sampling points within the period is taken as the representative value of the period. For channels with a sampling rate equal to or lower than 2Hz, the nearest neighbor interpolation method is used to determine the representative value of the period.

[0038] Furthermore, the specific control algorithms for the drilling pressure control loop and the rotational speed control loop in step S4 are dual-loop incremental PID closed-loop control. The drilling pressure control loop uses the target drilling pressure output in step S3. The actual drilling pressure is calculated from the real-time feedback of the vertical shaft hydraulic cylinder working pressure by the pressure sensor, which is the set value. Calculate drilling pressure deviation as a process variable. The incremental PID control law is: ; The control quantity is updated as follows: ; After being limited by the limiter, the signal is linearly mapped to a 4-20mA current signal and output to the electromagnetic proportional relief valve to regulate the system pressure and thus change the drilling pressure. , , The PID parameters for the drilling pressure loop are calibrated using the Ziegler-Nichols step response method during the system's on-site commissioning phase and then stored in the configuration parameter table. The speed control loop uses the target speed output in step S3. The set value is the vertical shaft speed fed back in real time by the speed sensor. Calculate the speed deviation as a process variable: ; The speed control quantity is calculated using an incremental PID control law of the same type as the drilling pressure circuit. , After being limited, the signal is sent to the frequency converter via the 485 communication interface in Modbus RTU protocol frame format. The frequency converter then adjusts the motor speed accordingly. , , The PID parameters for the rotational speed loop are shown below, taking the typical configuration of the XY-2 drilling rig as an example. , , Both control loops are based on ms is for synchronous operation of control cycles. At the beginning of each control cycle, the sensor feedback value is read through the 485 communication interface and analog signal acquisition channel. Then, PID calculation is performed. Finally, the control quantity is output to the corresponding actuator through the communication interface or analog signal output channel to ensure that the dynamic response of the actual drilling parameters has no overshoot and the steady-state error does not exceed 5% of the target value.

[0039] Furthermore, the specific execution process of the seamless switch to manual control mode in step S5 is as follows: At the moment the switch is triggered, the system freezes all control quantities output in the current control cycle; then, the frozen control quantity values ​​are displayed to the operator through the HMI interface as the current operation reference value; at the same time, the system automatically maps the electrical zero position of the physical control device on the manual control console to the frozen control quantity value, so that the control position felt by the operator when taking over control is consistent with the current equipment status; during the transition period after the switch is completed, the intelligent controller continues to run in the background in shadow mode and calculate the target control quantity, but does not output it to the actuator, and is only used to compare and monitor with the manual operation quantity. When the deviation between the two exceeds the safety range, an alarm prompt is issued again.

[0040] Furthermore, the specific execution method of the automatic shutdown described in step S5 is as follows: The integrated CNC center simultaneously sends an emergency stop command to the frequency converter, sets the frequency converter enable signal to a low level, and sends a disconnect command to the hydraulic pump station motor contactor, causing the drilling rig's rotation and feed actions to stop simultaneously; after shutdown, the system keeps the sensor power supply circuit and 485 communication bus working, continuously collecting and recording equipment status data for operators to review when troubleshooting the cause of the fault; after confirming that the fault has been eliminated, the operator clicks the reset button through the HMI interface, and the system executes the startup self-test process, sequentially checking the communication status of each sensor, the frequency converter's readiness status, and whether the hydraulic system pressure is within the normal range. Only after all self-tests pass can the drilling operation be restarted.

[0041] Preferably, the calculation formula for the five formation response characteristic values ​​in step S2 is as follows: Drilling resistance characteristic value: ; in For downward pressure, This represents the change in vertical axis displacement during the fusion cycle. The duration of the fusion cycle; vibrational energy characteristic value: ; in This represents the number of sampling points from the six-axis sensor within the fusion period. , , For the first Three-axis acceleration of the fuselage at each sampling point; characteristic value of rotational speed fluctuation. ,in The standard deviation of the vertical shaft speed within the fusion cycle. The average vertical shaft speed within this cycle; characteristic value of flow rate variation: ; in The number of flow sensor sampling points within the fusion period. and For the first Inflow and outflow rates at each sampling point; cumulative mud consumption characteristic value: ; in This refers to the sampling interval of the flow sensor.

[0042] Furthermore, the stratigraphic lithology is classified into five categories: soft soil, soft rock, hard rock, hard rock, and fractured rock mass. During inference, the random forest classifier runs a BP neural network auxiliary classifier in parallel for redundancy verification. The BP neural network has a 5-12-5 three-layer structure. The input layer has 5 nodes corresponding to five stratigraphic response feature values, the hidden layer has 12 nodes using the Sigmoid activation function, and the output layer has 5 nodes that are normalized by Softmax to output the probability values ​​of the five lithological categories. When the identification results of the random forest main classifier and the BP neural network auxiliary classifier are consistent, the result of the main classifier is used as the final lithological identification result, and the overall confidence score is the average of the two confidence scores. When the two identification results are inconsistent, it indicates that the current stratigraphic transition zone or lithological ambiguity zone is in the range, and the overall confidence score is forcibly set to a value below 0.6, triggering the weighted interpolation strategy of the parameter matching layer.

[0043] Furthermore, the five stratigraphic response feature values ​​are first subjected to Z-score standardization before being input into the random forest classifier and BP neural network. The processing method is as follows: ,in For the first Original values ​​of each feature. and These are the mean and standard deviation of the feature on the training dataset, respectively. The standardized feature values ​​have a mean of 0 and a standard deviation of 1. and The computation and storage are performed during the offline training phase of the model, and the same method is used during the online inference phase. and The real-time feature values ​​are standardized before being input into the classifier.

[0044] Furthermore, the random forest main classifier and the BP neural network auxiliary classifier constitute a dual-mode redundancy architecture. The two classifiers are based on completely different mathematical principles. The random forest, based on decision tree ensemble, has natural robustness and anti-overfitting ability for high-dimensional features, making it suitable for processing noisy drilling data. The BP neural network has a strong ability to model the nonlinear coupling relationship between features and can capture subtle stratigraphic transition patterns that the random forest may miss. Using the random forest alone may miss gradual stratigraphic transitions, while using the BP neural network alone may be oversensitive to sensor outliers. Dual-mode redundancy allows the criteria of the two to cross-check each other. When classification divergence occurs in the ambiguous area of ​​stratigraphic boundaries, confidence downgrading and weighted interpolation are triggered, avoiding forced misclassification in the boundary area by a single classifier.

[0045] Furthermore, the BP neural network-assisted classifier is trained offline using a cross-entropy loss function and an Adam optimizer. The training samples are derived from historical drilling data with calibrated lithology at the construction site. The initial learning rate during network training is 0.001, the batch size is 64, and the training epochs are 200. Training is terminated early when the validation set loss does not decrease for 10 consecutive epochs. After training, the network weight parameters are permanently stored in the non-volatile memory of the integrated CNC center. During the online inference phase, the weights are directly loaded for forward computation, and the inference time does not exceed 5ms.

[0046] Furthermore, the multidimensional characteristic index mentioned in step S2 also includes: the rate of change of drilling resistance. Defined as: ; in This represents the characteristic value of drilling resistance at the current moment. The second represents the relative rate of change of drilling resistance within a 5-second window, measured in seconds (s). - ¹; Rate of change of vibrational energy Defined as: ; in The characteristic value of vibration energy at the current moment; rotational speed stability index Defined as: ; in The standard deviation of the vertical shaft rotation speed within the sliding window over the most recent 10 seconds. The average vertical shaft rotation speed within this window; inlet and outlet water flow rate variation index. Defined as: ; when ,otherwise The above four derived indicators are related to the current drilling depth. Together they form a five-dimensional extended feature vector. , and formation response feature vector They are also included in the input feature space for stratigraphic identification.

[0047] Preferably, the critical depth mentioned in step S3 ,in For the target drilling pressure value, Mass per unit length of drill pipe m / s², The physical meaning is that the drill pipe's self-weight is exactly equal to the borehole depth when the target drill pressure is equal; when When the drilling is determined to be in the shallow hole drilling stage, the vertical shaft hydraulic cylinder applies downward pressure. ,in For the hydraulic system's floating back pressure, the cylinder actively presses down during this stage to compensate for the insufficient weight of the drill pipe; when This is considered a transition phase, where the downforce decreases linearly to zero with increasing depth; when... When the drilling is determined to be in the deep hole drilling stage, the vertical shaft cylinder switches to the lifting control mode, and the lifting force is increased. The hydraulic cylinder lifts the drill rod to partially offset its own weight, preventing the entire weight of the drill rod from pressing on the drill bit and causing excessive drilling pressure, thus maintaining the effective pressure of the drill bit on the bottom of the hole at a constant level. Within the allowable deviation range; When the lithology type of the strata changes, the control parameters change from the current value. Transition to the new target value ,in This indicates the current value of the control parameter. This indicates the new target values ​​for the control parameters, including target values ​​for drilling pressure, rotational speed, and flushing fluid flow rate; the transition process uses an S-shaped acceleration / deceleration curve within a preset transition time. Smooth transition within, the first The target parameter value for each control cycle is ,in The total number of transition cycles, To control the period, normalize the sigmoid function: ; Steepness parameters This segmentation definition ensures that the transition process starts precisely at... , terminated The rate of change is close to zero at both ends to avoid hydraulic shock and hole wall damage, while a rapid transition is made in the middle section to shorten the response time. Determined based on the hardness grade difference of the formation; the larger the grade difference... The longer.

[0048] Furthermore, a preset transition time is set. The specific determination rules are as follows: the five types of lithology are graded from 1 to 4 according to their hardness, with soft soil being grade 1, soft rock being grade 2, hard rock being grade 3, fractured rock mass being grade 3, and hard rock being grade 4, and the hardness grade difference is calculated. ,when hour Seconds, when hour Seconds, when hour Seconds. Control cycle. The S-curve smooth transition is applied to three control parameter channels: target drilling pressure, target rotation speed, and target flushing fluid flow rate. All three channels share the same... and Function synchronization transition ensures that the actions of each actuator are coordinated and consistent.

[0049] Furthermore, The floating back pressure of the hydraulic system is determined by running the vertical shaft cylinder under no-load during each startup self-test. The residual pressure in the lower chamber of the vertical shaft cylinder under its own weight is measured, and the average value of 10 consecutive measurements is taken as the back pressure. Store in the system configuration parameter table.

[0050] Preferably, the quantitative calculation model for the stuck drill risk index in step S5 is as follows: ; ; ; ; in , , These are the weighting coefficients; This is a stress anomaly index. This is the sliding average of the downforce over the last 30 seconds. A positive and continuously increasing value indicates that the drill bit is encountering abnormal resistance. This is the torque anomaly index. This is the current effective value of the motor current fed back by the frequency converter. This is the moving average of the current over the last 30 seconds; For traffic anomaly index, when ,otherwise , An increase indicates that the circulation channel is blocked; when The system promptly identifies the risk of stuck drill pipe. This weighted fusion scheme of three indicators covers three independent physical dimensions for determining stuck drill pipe: mechanical resistance, rotational resistance, and circulation channel. An anomaly in any single dimension may not trigger the threshold of the comprehensive index; only when multiple dimensions deteriorate simultaneously is it considered stuck drill pipe, significantly reducing the false alarm rate. Using pressure indicators alone cannot distinguish between stuck drill pipe and normal encounter with hard rock; using torque indicators alone is not sensitive to slight drill pipe jamming; and using flow rate indicators alone may fluctuate during normal formation changes. The fusion of the three indicators compensates for the blind spots of single indicators.

[0051] Furthermore, the abnormal operating conditions include equipment overload operation, drilling parameters exceeding the safe range, abnormal sensor data, and communication interruption. When an abnormal operating condition is triggered, the system automatically records the event data and triggers a maintenance reminder. The event data includes the event type code, trigger timestamp, a snapshot of all sensor readings and actuator control quantities at the trigger time, and continuous historical data within 10 seconds before the trigger. This data is written to non-volatile memory in a structured format and alerts the operator through HMI interface pop-ups and flashing warning lights in the audible and visual alarm system.

[0052] Furthermore, the specific criteria for identifying the abnormal operating condition are: the equipment is operating under overload, and the motor current fed back by the frequency converter... The current exceeds the motor's rated current for 5 consecutive seconds. 1.2 times; Drilling parameters exceed the safe range, any pressure sensor reading exceeds 90% of its calibrated range upper limit, or the vertical shaft speed exceeds 110% of the rated speed; Sensor data is abnormal, any sensor channel returns a constant data value within 5 consecutive sampling cycles with the variation not exceeding the noise level, or the data value exceeds the calibrated range of the sensor; Communication is interrupted, the integrated CNC center polls each 485 communication node at 100ms intervals and sends query frames, if any node does not return a valid response frame for 3 consecutive query cycles, the node is determined to be in communication interruption.

[0053] Furthermore, the structured format of the event data is JSON, and the recorded content includes: a unique event sequence number consisting of a 32-bit Unix timestamp plus a 3-digit incremental sequence number; an event type code defined by an enumeration value, where 0x01 indicates equipment overload, 0x02 indicates parameter exceeding safe range, 0x03 indicates sensor data anomaly, 0x04 indicates communication interruption, 0x05 indicates drill jam risk alarm, 0x06 indicates hole leakage alarm, and 0x07 indicates water inrush alarm; an event trigger timestamp accurate to milliseconds; a snapshot of the instantaneous readings of all 15 sensor channels at the event trigger moment; a snapshot of the control output of the 6 actuator channels at the event trigger moment; and a continuous historical record of all sensor data and control quantities within 10 seconds prior to the event, in 100 frames at 100ms intervals. The event log file is named "event type code_trigger timestamp.json" and stored in the event log directory of the integrated CNC center's built-in solid-state storage. The file system uses FAT32 format to support export to external storage devices via USB interface. The system pre-defines an event-maintenance countermeasure mapping table, using event type codes as keys. Each record includes an event name, impact level (divided into three levels: alert, warning, and critical), a description of the maintenance measures, and suggested operation steps. Upon event triggering, a modal maintenance reminder window pops up in the main display area of ​​the HMI interface, showing the event name, occurrence time, current equipment status summary, and detailed steps for maintenance. Simultaneously, the yellow warning light in the audible and visual alarm system flashes at a frequency of 1Hz. After the operator clicks the confirmation button on the HMI interface, the system records the timestamp of the confirmation operation and the operator's employee number. The yellow warning light stops flashing, the maintenance reminder window closes, and the event data file is archived to the confirmed event subdirectory. If the operator does not perform the confirmation operation within 5 minutes, the system automatically increases the flashing frequency of the yellow warning light to 2Hz and adds a short beeping sound.

[0054] Example 2: This embodiment uses a geological exploration project in a mining area as the application scenario. The borehole is designed to a depth of 50m and a diameter of 91mm. The strata from top to bottom are: 0 to 8m alluvial soft soil layer, 8 to 18m weathered soft sandstone layer, 18 to 30m hard dense limestone layer, 30 to 38m fractured rock fault fracture zone, and 38 to 50m hard granite gneiss layer.

[0055] The experimental platform was built on an XY-2 hydraulic feed geological drilling rig, equipped with an integrated CNC center, and deployed 15 sensor channels across 8 categories: 3 QDW90A-COL pressure sensors installed on the multi-way hydraulic valve seat, the vertical shaft cylinder lifting port, and the vertical shaft cylinder pressing port, respectively; 1 KPM displacement sensor installed beside the vertical shaft cylinder; 2 DIN2N speed sensors installed on the motor shaft and the vertical shaft; 2 six-axis sensors installed on the drilling rig body and the vertical shaft, respectively; 1 tilt sensor installed on the vertical shaft; 2 DN20 electromagnetic flow sensors installed on the inlet pipe and the orifice return pipe; and 1 B-TH-RS30 temperature and humidity sensor installed inside the control box. Sensor data were fused with a 500ms fusion period, and the integrated CNC center ran the EKF fusion algorithm and a three-level decision chain algorithm. 50mm outer diameter wireline coring drill rods were used, with a unit length mass... The reference flushing fluid flow rate is 6.04 kg / m. The flow rate is 3.0 m³ / h, and the control cycle is... It takes 100ms.

[0056] Key parameters are set as follows: the random forest classifier contains 200 CART decision trees, with a maximum depth of 15 and a minimum number of leaf node samples of 5; the BP neural network auxiliary classifier has a 5-12-5 structure, an initial learning rate of 0.001, a batch size of 64, and 200 training epochs; the weights in the drill jam risk index are... , , The threshold is set to 0.7; the steepness parameter of the S-curve. The formation-drilling parameter comparison table is set according to the values ​​described in claim 3; critical depth Calculate in real time according to the formula described in claim 9; take the PID parameters of the drilling pressure circuit. , , PID parameters for the speed loop , , PID parameters for water flow loop , , .

[0057] The existing methods used for comparative testing are: the traditional scheme of drilling control using a single pressure monitoring method plus manual speed and pressure adjustment, hereinafter referred to as the traditional manual scheme; and the semi-automatic scheme of adding pressure and speed dual sensors on the basis of the traditional scheme and displaying the data on the instrument panel for operators' reference, hereinafter referred to as the dual-parameter monitoring scheme.

[0058] Neither of the two comparative schemes included EKF fusion, automatic formation identification, PID closed-loop control, or automatic assessment of stuck drill risk. Operators relied on experience to judge the formation and manually adjust the hydraulic handle and gear shift. The comparative experiments were conducted at adjacent boreholes approximately 15m apart in the same borehole, with identical geological conditions.

[0059] Complete Implementation Process: After the drilling rig is in place, the integrated CNC center is started and a self-test procedure is executed. This procedure sequentially checks the communication status of 15 sensor channels, the inverter's readiness status, and the hydraulic system pressure within the normal range. Drilling begins after the self-test passes. During drilling, the integrated CNC center synchronously acquires data from the 15 channels at 500ms intervals. This data is then fused using EKF and used to invert the formation lithology in real time. and confidence level Dynamically generate target drilling pressure and target speed The system employs a dual-loop PID controller to drive the actuator in a closed-loop tracking manner. The system synchronously calculates the stuck drill risk index. At a depth of 34m When the value reaches 0.82, exceeding the threshold of 0.7, an audible and visual alarm is triggered. After confirmation by the operator, the control is taken over through a four-stage seamless switching mechanism. After the issue is resolved, the system switches back to intelligent mode and continues drilling to a final hole depth of 50m. Specific data are shown in Tables 1-3 below.

[0060] Table 1: Operational characteristic data of the method of the present invention during the drilling process;

[0061] Table 2: Comparison of drilling performance data of the three methods on the same formation profile;

[0062] Table 3: Comparison of performance in early warning of stuck drill bit risk;

[0063] As can be seen from the data in Table 1, the method of the present invention achieved automatic identification of five types of strata throughout the entire 50m drilling process, with a confidence level of [missing information]. The stability strata remain between 0.88 and 0.95, and at a depth of 18m in the stratigraphic transition zone... The value was reduced to 0.52, and a weighted interpolation strategy was automatically triggered to generate transitional parameters, avoiding abrupt parameter changes. Actual drilling pressure With target drilling pressure The average deviation was only 1.2%, and the actual rotational speed deviation was also within a controllable range. The flushing fluid flow rate... The flow rate is automatically adjusted according to the stratigraphic zoning rules. In soft soil layers, the flow rate is increased to 4.5 m³ / h to enhance borehole wall protection; in hard and soft rock layers, the flow rate remains at the baseline value of 3.0 m³ / h; and in hard rock layers, the flow rate is reduced to 2.4 m³ / h to reduce borehole wall erosion. The flow rate remains stable throughout the entire borehole. In the 34m fractured rock section… When the value reaches 0.82, exceeding the 0.7 threshold, the system issues an early warning 13 seconds before a stuck drill accident occurs, providing operators with an ample window to react.

[0064] As shown in Table 2, the average mechanical drilling speed of this method is 2.8 m / h, which is higher than the 2.3 m / h of the dual-parameter monitoring scheme and significantly higher than the 1.7 m / h of the traditional manual scheme. Regarding drilling pressure control accuracy, the average drilling pressure deviation of this method throughout the entire process is only 1.2%, while the dual-parameter monitoring scheme has a deviation of 11.9%, and the traditional manual scheme is as high as 23.3%. In the fractured rock section from 30 to 38 m, the traditional manual scheme triggered equipment protection shutdown at 30 m due to severely excessive drilling pressure, and at 34 m, it failed to identify the early signs of stuck drill pipe, ultimately resulting in a stuck drill pipe accident and the borehole being scrapped. Although the dual-parameter monitoring scheme completed drilling in this section, the drilling speed was too low and it did not provide a stuck drill pipe warning.

[0065] As can be seen from Table 3, this method... The weighted average of the three indicators showed zero false alarms, while the single traffic indicator showed zero false alarms. False alarms up to 4 times, single stress indicator Two false alarms occurred, and the fusion of the three indicators significantly suppressed the false alarm rate while maintaining alarm sensitivity. In addition, the data in Table 1 shows that the smooth transition of the S-curve at the 18m stratum transition zone resulted in a parameter adjustment time of only 2.8 seconds with no overshoot, demonstrating the contribution of the parameter smooth switching mechanism to equipment protection and borehole wall stability.

[0066] In summary, this invention, through the synergistic effect of EKF multi-source fusion sensing, RF+BP dual-mode redundant formation identification, three-level progressive decision-making, dual-loop PID closed-loop control, three-index weighted stuck drill risk assessment, and S-curve smooth parameter transition, significantly outperforms existing technologies in terms of drilling efficiency, parameter control accuracy, stuck drill risk early warning capability, and adaptability to complex formations.

[0067] Figure 7 The data presents five formation response characteristic values ​​and drilling resistance characteristic values ​​within the drilling depth range of 0 to 50 m. Vibration energy characteristic value Rotation speed fluctuation characteristic value Flow change characteristic value Cumulative characteristic value of mud consumption The curves show continuous changes with depth, with different background colors indicating the five lithological zones of the strata. The figure shows that current technologies typically monitor only a single parameter such as pressure or rotational speed. At the interface between soft and hard rock, at depths of 8m and 18m, a rough judgment can be made based on changes in a single parameter. However, in fractured rock sections from 30 to 38m, relying solely on a single parameter becomes insufficient. The value might be misjudged as a soft rock return due to frequent sudden drops in peak values, while this invention simultaneously integrates... and The information cross-validated the existence of the fractured rock mass. Furthermore, The continuous upward trend in the fractured rock mass section reflects the cumulative effect of flushing fluid leakage along the fractures, an information completely missing in existing single-parameter monitoring schemes; the five characteristic values ​​exhibit differentiated combination patterns in different strata, with soft soil layers showing low... ,Low ,middle and high Hard rock strata exhibit high... ,middle ,Low and low The fractured rock mass is characterized by violent fluctuations, high ,high and high The combination of these parameters provides sufficient inter-class discriminability for the subsequent random forest and BP neural network dual-mode redundant classifier, making the stratum identification accuracy significantly better than existing methods that rely on single or dual parameters.

[0068] Figure 8 The graph shows the change of the stuck drill risk index over time in the fractured rock section with a drilling depth of 30 to 38 meters. The subplot above compares the three-index weighted fusion model used in this invention. Compared with using the pressure anomaly index alone Torque anomaly index Or traffic anomaly index Differences in identifying stuck drill bits using a single indicator. The complex anomaly injected at a depth of 34m lasted approximately 20 seconds, as shown in the figure: [using only...] The pressure exceeded the 0.7 threshold at 15 seconds and triggered an alarm, but this alarm was indistinguishable from the normal pressure increase when encountering hard rock. The same alarm occurred at 32m. A brief spike above 0.7 without actual drill jamming constitutes a false alarm; used alone. The response was slow, taking 18 seconds to exceed the 0.7 threshold; used alone. The flow sensor malfunctions due to noise at multiple times. This invention utilizes a weighted fusion of three indicators. The alarm consistently exceeded the 0.7 threshold at 13 seconds without any false alarms prior to that, 2 seconds earlier than the single best indicator, and no false alarms occurred in the normal hard rock section at 32m. The following sub-figure shows... The three-dimensional evolution surface with drilling depth and time clearly shows that the risk index forms an isolated peak at the 34m stuck drill event, while remaining low at other depths, which verifies the spatial positioning and anti-interference capabilities of the weighted fusion model for stuck drill risk.

[0069] Figure 9 This paper compares two transition methods for target drill pressure (DP) values ​​when the lithology changes from soft rock to hard rock. The subfigure compares the S-curve smooth transition used in this invention with the step parameter switching commonly used in existing technologies. The horizontal axis represents the control cycle number, and the vertical axis represents the target DP value. As can be seen from the figure, the step switching abruptly jumps the target DP from 1200N to 2500N at the moment of reaching a depth of 18m. This generates a pressure shock in the hydraulic system, which may cause instantaneous overpressure of the drill bit on the bottom of the hole, enough to cause hole wall rupture or drill bit damage in the vulnerable section of the hole wall. In contrast, the S-curve transition, within a preset transition time... Within seconds, the parameter smoothly transitions from the current value of 1200N to the new target value of 2500N. The initial transition segment shows a gradual change to test the actual response of the new formation; the middle segment rapidly approaches the target value to shorten the parameter arrival time; and the final segment decelerates again to avoid overshoot. The subplot below shows the adaptive adjustment of the S-curve under different formation hardness levels. When the level difference is 1, Seconds, when the grade difference is 2 levels Seconds, when the grade difference is 3 levels The transition slope tends to level off as the gradient increases, reflecting an adaptive match between the parameter transition rate and the degree of formation change. Existing technologies, such as fixed-gear switching or simple slope transitions, lack this mechanism for adaptively adjusting the transition time based on gradient differences.

Claims

1. An intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making, characterized in that, include: Integrated CNC center, geological drilling rig body and modular detachable tower; The integrated CNC center includes a multi-dimensional fusion sensing module, a real-time decision control module, and a dual-mode execution system; The multi-sensor fusion module collects drilling parameters, mechanical status and formation response data in real time through a multi-type sensor group deployed on the main body of the geological drilling rig. It also uses an extended Kalman filter algorithm to perform time synchronization and fusion state estimation on the heterogeneous data of the multi-type sensor group and outputs the fused drilling state vector. The real-time decision control module receives the drilling status vector and performs a three-level progressive decision-making process: formation identification layer, parameter matching layer, and safety verification layer. The formation identification layer inputs the drilling status vector into a random forest classifier to output the current formation lithology category and confidence level. The parameter matching layer uses the lithology category as an index to retrieve the target drill pressure value and target rotation speed value from the formation-drilling parameter lookup table and performs depth compensation in combination with the borehole depth. The safety verification layer verifies the compensated target parameters against the equipment rated range and formation safety boundary before outputting the final control target parameters. The dual-mode execution system includes a motor frequency conversion control system, a hydraulic proportional valve control system, an auxiliary electromagnetic directional valve control system, an audible and visual alarm system, and a water volume control system, supporting parallel input and seamless switching of intelligent control commands and manual operation commands. The modular and detachable tower is attached to the main body of the geological drilling rig. Two pulleys are installed at the top of the tower. The wire rope of the winch converts the winching force into the pulling force for raising and lowering the drill rod through the two pulleys, realizing the automatic raising and lowering of the drill rod.

2. The intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 1, characterized in that, The multi-type sensor group includes pressure sensors, displacement sensors, speed sensors, six-axis sensors, tilt sensors, inlet water flow sensors, return water flow sensors, and temperature and humidity sensors. Three pressure sensors are installed on the multi-way hydraulic valve seat, the lifting port of the vertical shaft cylinder, and the pressing port of the vertical shaft cylinder, respectively, to monitor system pressure, lifting pressure, and lowering pressure. The displacement sensor is installed next to the vertical shaft cylinder to monitor the vertical shaft displacement. Two speed sensors are installed on the motor shaft and the vertical shaft, respectively, to monitor the motor speed and the vertical shaft speed. Two six-axis sensors are installed on the drilling rig body and the vertical shaft, respectively, to sense the rig body attitude, monitor rig body vibration, and changes in vertical shaft attitude. The tilt sensor is installed on the vertical shaft to monitor the drilling tilt angle. The inlet water flow sensor and the return water flow sensor are installed at the inlet water pipe and the return water pipe at the borehole opening, respectively, to monitor the circulating water volume. The temperature and humidity sensors are installed inside the control box to monitor the operating environment temperature and humidity.

3. The intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 1, characterized in that, The data fusion process of the extended Kalman filter in the multi-dimensional fusion sensing module is as follows: Using GPS timing signals as a unified time reference, UTC timestamps are added to the data from each sensor. A sliding time window is set, and the asynchronous data from each channel within the window are linearly interpolated to the center time of the window according to the timestamps, forming a time-aligned multi-channel data frame. The system pressure, lifting pressure, lower pressure, vertical shaft displacement, motor speed, vertical shaft speed, three-axis acceleration of the drilling rig, three-axis angular velocity of the drilling rig, borehole inclination angle, inlet water flow rate, and return water flow rate are constructed as observation vectors and input into the extended Kalman filter. The filter process model adopts a constant acceleration model, and the output is a fused drilling state vector. The formation identification layer of the real-time decision control module calculates drilling resistance characteristic values ​​based on the drilling state vector within each fusion cycle. Vibration energy characteristic value Rotation speed fluctuation characteristic value Flow change characteristic value Cumulative characteristic value of mud consumption Five stratigraphic response feature values ​​are used to form a stratigraphic response feature vector, which is input into the random forest classifier. The random forest classifier contains multiple CART decision trees. The Gini coefficient is used as the splitting criterion. Each decision tree performs independent classification voting. The one with the highest vote rate is used as the stratigraphic lithology identification result, and the vote rate is used as the confidence level. In the parameter matching layer, when the confidence level is greater than or equal to the set threshold, the formation-drilling parameter reference table is directly retrieved using the lithology identification result as the key to obtain the target drill pressure reference value and the target rotation speed reference value. When the confidence level is less than the set threshold, the two candidate lithology categories with the highest probability values ​​output by the random forest and their probabilities are selected. The reference table parameters of the two categories are weighted and interpolated according to the probability ratio to generate a transitional reference value. The depth compensation formula is: ； in To retrieve or interpolate the drill pressure reference value, Mass per unit length of drill pipe It is the acceleration due to gravity. The current drilling depth is used as the reference depth. The safety verification layer compares the compensated target drilling pressure value with the maximum allowable pressure of the hydraulic system and the maximum allowable drilling pressure of the formation, and takes the minimum value. It also compares the target rotation speed value with the rated rotation speed of the frequency converter and takes the minimum value. Finally, it outputs the control target parameter pair.

4. The intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 1, characterized in that, The hydraulic proportional valve control system includes an electromagnetic proportional directional valve and an electromagnetic proportional relief valve. The electromagnetic proportional directional valve converts the main control signal into a corresponding analog voltage via a 485 communication module, which is used to control the stepless proportional adjustment of the vertical shaft feed speed and feed force. The electromagnetic proportional relief valve converts the voltage signal into a current signal via a dedicated signal conversion control board and outputs it to achieve precise setting and dynamic adjustment of the system pressure. The 485 communication module supports dual-channel signal output and simultaneously controls the electromagnetic proportional directional valve and the electromagnetic proportional relief valve.

5. The intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 1, characterized in that, The auxiliary electromagnetic reversing valve control system includes front and rear control cylinders and a chuck cylinder. The front and rear control cylinders are independently controlled by one electromagnetic reversing valve, and the chuck cylinder is independently controlled by another electromagnetic reversing valve. The motor frequency conversion control system supports 485 signal communication. The integrated CNC center sends speed command to the frequency converter and receives operating status feedback through the 485 communication interface. The water control system includes an electric regulating valve, an inlet flow sensor, a return flow sensor, and a water control unit. The water control unit dynamically generates the target flushing fluid flow rate according to the lithology of the formation and the borehole depth according to the zoning rules. It uses the real-time measurement value of the inlet flow sensor as feedback and drives the opening of the electric regulating valve through an incremental PID algorithm to make the actual flushing fluid flow rate track the target value. At the same time, it continuously monitors the leakage and water inrush conditions in the borehole through the difference between the inlet and outlet flow rates.

6. The intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 1, characterized in that, The seamless handover includes four stages: handover triggering, state freezing, control quantity alignment, and shadow following. The switching trigger conditions include: the operator presses the physical switching button on the console, the system self-test detects that any 485 communication module has not returned a valid response frame for a fixed number of consecutive communication cycles, or the formation identification confidence level output by the real-time decision control module is lower than the set value for a fixed number of consecutive control cycles. The system starts the switching process when any of the conditions are met. When the state freeze occurs during switching startup, the dual-mode execution system immediately freezes all control variables for the current control cycle, including the voltage value of the electromagnetic proportional directional valve. Electromagnetic proportional relief valve current value Variable frequency drive speed setpoint , Front and rear control cylinder solenoid directional valve switch status chuck cylinder solenoid directional valve switch status and the opening value of the electric regulating valve A six-element control state snapshot is generated and stored in the shared memory area; Control alignment is as follows: When switching from intelligent mode to manual mode, the electrical zero positions of the joysticks, knobs, and buttons on the manual control panel are remapped to the frozen control state snapshot values ​​through a linear mapping function, so that the output corresponding to the current position of the control device is equal to the control quantity at the time of freezing. The mapping relationship is as follows: ； in To control the amount of freezing, The sensitivity coefficient of the joystick is determined by the factory calibration. For the recalculated zero-position mapping point The current physical location corresponds to When switching back from manual mode to intelligent mode, the intelligent controller first reads the control values ​​corresponding to the current manual operation as the initial output values ​​of the incremental PID controller. Then, in each control cycle, it gradually transitions to the target value calculated by the intelligent algorithm; Shadow following means that after switching to manual mode, the intelligent controller continues to run in the background in shadow mode. Each control cycle calculates the theoretical target control quantity but does not output it to the actuator. When the deviation between the theoretical target control quantity and the actual manual control quantity exceeds the set percentage of the corresponding control channel range and continues for more than the set time, the HMI interface will pop up a prompt message to suggest that the operator confirm the current operation.

7. A control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to any one of claims 1-6, characterized in that, Includes the following steps: S1 uses a multi-type sensor group to synchronously collect data on inlet and outlet water flow, vertical shaft lifting pressure, vertical shaft lowering pressure, system pressure, displacement, rotation speed, machine body vibration, and drilling inclination angle at fixed intervals. S2. After performing sliding median filtering and time alignment preprocessing on the collected multi-source data, five formation response feature values ​​are calculated: drilling resistance feature value, vibration energy feature value, rotation speed fluctuation feature value, flow rate change feature value, and mud consumption accumulation feature value. The formation response feature vector is then input into a random forest classifier to invert the lithology and confidence level of the current drilling formation in real time. S3, based on the identified formation lithology and confidence level, retrieves or generates target drill pressure and target rotation speed values ​​from a pre-set drill pressure-rotation speed matching parameter library, and outputs the final control target parameters after depth compensation and safety boundary verification; during drilling, the parameters are adjusted according to the borehole depth. With critical depth The relationship dynamically adjusts the action mode of the vertical shaft hydraulic cylinder, and the critical depth. Determined by the target drill pressure and the weight per unit length of the drill pipe; S4 generates adaptive control commands, which drive the electromagnetic proportional relief valve and frequency converter respectively through the incremental PID controllers of the drilling pressure control loop and the speed control loop, so that the actual drilling parameters track the final control target parameters. S5 continuously monitors equipment operating status and calculates the stuck drill risk index. ,when When the set threshold is exceeded or an abnormal operating condition is detected, an audible and visual alarm is triggered and the machine is automatically shut down, or the machine is transferred to manual control mode according to the described non-disruptive switching process.

8. The control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 7, characterized in that, The calculation formulas for the five formation response characteristic values ​​mentioned in step S2 are as follows: Drilling resistance characteristic value: ； in For downward pressure, This represents the change in vertical axis displacement during the fusion cycle. The duration of the fusion cycle; vibrational energy characteristic value: ； in This represents the number of sampling points from the six-axis sensor within the fusion period. , , For the first Three-axis acceleration of the fuselage at each sampling point; characteristic value of rotational speed fluctuation. ,in The standard deviation of the vertical shaft speed within the fusion cycle. The average vertical shaft speed within this cycle; characteristic value of flow rate variation: ； in The number of flow sensor sampling points within the fusion period. and For the first Inflow and outflow rates at each sampling point; cumulative mud consumption characteristic value: ； in This refers to the sampling interval of the flow sensor.

9. The control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 7, characterized in that, The critical depth mentioned in step S3 ,in For the target drilling pressure value, Mass per unit length of drill pipe The physical meaning is that the drill pipe's self-weight is exactly equal to the borehole depth when the target drill pressure is equal; when When the drilling is determined to be in the shallow hole drilling stage, the vertical shaft hydraulic cylinder applies downward pressure: ； in For the hydraulic system's floating back pressure, the cylinder actively presses down during this stage to compensate for the insufficient weight of the drill pipe; when This is considered a transition phase, where the downforce decreases linearly to zero with increasing depth; when... When the drilling is determined to be in the deep hole drilling stage, the vertical shaft cylinder switches to the lifting control mode, with the lifting force as follows: ； The hydraulic cylinder lifts the drill rod to partially offset its own weight, preventing the entire weight of the drill rod from pressing on the drill bit and causing excessive drilling pressure, thus maintaining the effective pressure of the drill bit on the bottom of the hole at a constant level. Within the allowable deviation range; When the lithology type of the strata changes, the control parameters change from the current value. Transition to the new target value ,in This indicates the current value of the control parameter. This indicates the new target values ​​for the control parameters, including target values ​​for drilling pressure, rotational speed, and flushing fluid flow rate; the transition process uses an S-shaped acceleration / deceleration curve within a preset transition time. Smooth transition within, the first The target parameter value for each control cycle is ,in The total number of transition cycles, To control the period, normalize the sigmoid function: ； As a steepness parameter, this segmentation definition ensures that the transition process starts precisely at... , terminated The rate of change at both ends is close to zero to avoid hydraulic shock and hole wall damage, while a rapid transition in the middle section is used to shorten the response time. Determined based on the hardness grade difference of the formation; the larger the grade difference... The longer.

10. The control method for an intelligent geological drilling rig system based on multi-source sensing and autonomous decision-making according to claim 7, characterized in that, The quantitative calculation model for the stuck drill risk index in step S5 is as follows: ； ； ； ； in , , These are the weighting coefficients; This is a stress anomaly index. This is the sliding average of the downforce over the last 30 seconds. A positive and continuously increasing value indicates that the drill bit is encountering abnormal resistance. This is the torque anomaly index. This is the current effective value of the motor current fed back by the frequency converter. This is the moving average of the current over the last 30 seconds; For traffic anomaly index, when ,otherwise , An increase indicates that the circulation channel is blocked; when The risk of the drill getting stuck was immediately determined.

Images