A Collaborative Control Method and System for Loss of Magnetism and Overload of Magnetic Drill Based on Multi-Level Linkage Protection

By collecting the magnetic drill's adsorption force and load parameters in real time, dynamically calculating the torque threshold, and implementing four-level linkage protection, the problem of independent demagnetization and overload protection in traditional magnetic drills is solved, realizing linkage protection for safety control and improving the safety and service life of the equipment.

CN123121418APending Publication Date: 2026-11-17SHANGHAI CHENGXIANG ELECTROMECHANICAL EQUIPMENT CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611500002.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-24
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Traditional magnetic drills have two independent logic designs for demagnetization protection and overload protection. They cannot effectively detect and coordinate the problems of adsorption force attenuation and reaction force overturning torque coupling caused by overload, resulting in frequent safety accidents.

Method used

By collecting the actual adsorption force of the electromagnetic chuck and the load parameters of the spindle motor in real time, the maximum output torque threshold is dynamically calculated, and a four-level linkage protection is implemented. Combined with the magnetic flux sensing of the magnetic sensor, a three-dimensional safety strategy space is constructed for active exploration and safety terrain map-driven control, so as to realize the linkage protection of adsorption force and load.

Benefits of technology

It effectively prevents secondary demagnetization caused by overload, extends the service life of equipment, improves the pertinence and effectiveness of safety control, has the ability to actively perceive safety boundaries, and shortens the strategy convergence time of new equipment through group cognition aggregation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123121418A_ABST
    Figure CN123121418A_ABST
Patent Text Reader

Abstract

This invention provides a collaborative control method and system for demagnetization and overload of magnetic drilling rigs based on multi-level linkage protection, belonging to the field of intelligent safety control technology for magnetic drilling rigs. The method first ensures safe startup conditions through a self-checking interlock of the magnetic suction at startup. During operation, it collects the electromagnetic chuck's adsorption force parameters and the spindle motor load parameters in real time, dynamically calculates the maximum allowable output torque threshold based on the adsorption force, and executes four-level linkage protection. Based on this, a three-dimensional safety strategy space is constructed, spanned by the suction maintenance strategy axis, the load output strategy axis, and the contact state management strategy axis. A safety terrain map is actively explored through strategy offset disturbances, establishing a strategy drift mechanism driven by safety evaluation values. This invention transforms the passive response to demagnetization and overload into active exploration and strategy drift, solving the problem of recognizing safety boundaries under conditions where the true safety margin cannot be directly measured, and significantly improving the operational safety of magnetic drilling rigs under complex working conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of intelligent safety control technology for magnetic drills, and more specifically, to a collaborative control method and system for demagnetization overload of magnetic drills based on multi-level linkage protection. Background Technology

[0002] Magnetic drills rely on electromagnetic chucks to adhere to the surface of steel workpieces to complete drilling operations. The strength of this adhesion directly determines the safety of the equipment's operation. Traditional magnetic drill protection schemes typically design demagnetization protection and overload protection as two independent logics: demagnetization protection only detects whether the chuck's power supply voltage is below a threshold, and overload protection only monitors whether the motor current exceeds the rated value. However, during actual drilling, the machine vibration and axial tension caused by overload continuously act on the contact interface between the electromagnetic chuck and the workpiece, gradually deteriorating the contact state. This leads to irreversible attenuation of the actual effective adhesion force even under normal voltage conditions. Conversely, insufficient adhesion force generates overturning torque from drilling reaction force, further exacerbating load fluctuations. These two physical processes are bidirectionally coupled, but current technology lacks the means to perceive and coordinate this coupling. When the equipment encounters a sudden overload under weak adhesion conditions, it is highly susceptible to falling or overturning, causing serious safety accidents. Therefore, how to achieve coordinated perception and joint protection of demagnetization risk and overload state under the condition that the actual effective adsorption force cannot be directly and continuously measured is a technical problem that urgently needs to be solved in the field of magnetic drill safety control. Summary of the Invention

[0003] In view of the above problems, the present invention provides a collaborative control method for demagnetization overload of magnetic drill bit based on multi-level linkage protection, the method comprising: S1: During equipment operation, the actual adsorption force parameters of the electromagnetic chuck are collected in real time, and the load parameters of the spindle motor are also collected in real time. S2: Dynamically calculate the current maximum allowable output torque threshold based on the actual adsorption force parameters; S3: Compare the load parameters collected in real time with the maximum output torque threshold, and execute the linkage protection action according to the comparison result.

[0004] Furthermore, the actual adsorption force parameter is obtained by detecting the magnetic flux of the contact surface between the electromagnetic chuck and the workpiece using a magnetically sensitive Hall sensor.

[0005] Furthermore, the linked protection action includes the following four-level hierarchical control: Level 1: When the actual adsorption force parameter reaches the first preset threshold and the load parameter does not exceed the maximum output torque threshold, the current operating state is maintained; Level 2: When the actual adsorption force parameter is lower than the first preset threshold and the load parameter does not exceed the maximum output torque threshold, the maximum output torque threshold is lowered and a warning signal is issued. Level 3: When the load parameter exceeds the maximum output torque threshold and the actual adsorption force parameter reaches the first preset threshold, a flexible buffer load reduction is performed. Level 4: When the actual adsorption force parameter is lower than the second preset threshold and the load parameter exceeds the maximum output torque threshold, the main shaft power is cut off to maintain the adsorption force of the electromagnetic chuck and lock the moving parts of the whole machine. Wherein, the second preset threshold is lower than the first preset threshold.

[0006] Furthermore, the flexible buffer load reduction is achieved by at least one of the following methods: limiting the peak motor current, gradually reducing the PWM duty cycle, and pausing the feed motor.

[0007] Furthermore, it also includes a power-on self-test step. After the device is powered on, it detects the actual adsorption force parameter. Only when the actual adsorption force parameter reaches the preset safety threshold is the spindle motor unlocked to allow startup; when the actual adsorption force parameter does not reach the preset safety threshold, the spindle motor is locked and an alarm is issued.

[0008] Furthermore, in the second-level control, an active magnetization operation to enhance the excitation current of the electromagnetic chuck is also performed; after the fourth-level control is triggered, the system saves the fault record, and can only be unlocked and restarted after a manual physical reset operation is performed.

[0009] Furthermore, in step S3, the execution of the coordinated protection action based on the comparison result also includes the following control process driven by active exploration of the strategy space and a secure terrain map: A three-dimensional safety strategy space is constructed, which is spanned by a suction maintenance strategy axis, a load output strategy axis, and a contact state management strategy axis. The suction maintenance strategy axis corresponds to the adjustment method of the electromagnetic chuck excitation current, the load output strategy axis corresponds to the limiting method of the spindle motor output torque, and the contact state management strategy axis corresponds to the adjustment method of the drill bit feed speed. Each strategy axis is discretized into multiple selectable strategy levels, and any combination of three strategy levels constitutes a strategy space coordinate point. Within each control cycle, with the current strategy space coordinate point as the center, a strategy offset disturbance of a preset amplitude is applied along the three strategy axes of the three-dimensional safety strategy space to generate three candidate strategy offset directions for trial strategy coordinate points, and the trial strategy coordinate points are mapped to the corresponding trial excitation current adjustment amount, trial torque limit amount and trial feed speed adjustment amount respectively. Under the control parameters corresponding to the tentative strategy coordinate points, the response changes of the electromagnetic chuck adsorption force, the spindle motor load torque, and the drill bit feed thrust are collected respectively. The response deviation vector of each tentative strategy coordinate point relative to the current strategy coordinate point is calculated. The response deviation vector includes the adsorption force response deviation component, the load torque response deviation component, and the feed thrust response deviation component. The response deviation vector and its corresponding strategy coordinate point are stored in a safety topographic map database. The safety topographic map database uses the strategy coordinate point as an index to store the statistical characteristics of the historically accumulated response deviation vector under that coordinate point. The safety evaluation value of the strategy coordinate point is updated according to the magnitude direction and deviation duration of the response deviation vector. The safety evaluation value is inversely proportional to the magnitude of the response deviation vector and directly proportional to the stability of the deviation duration. When the safety evaluation value of the current strategy coordinate point is lower than the preset safety evaluation value threshold, the nearest neighbor strategy coordinate point with a safety evaluation value higher than the safety evaluation value threshold is retrieved from the safety terrain map database as the target strategy coordinate point. A strategy drift path is generated from the current strategy coordinate point to the target strategy coordinate point. The strategy drift path passes through one or more intermediate strategy coordinate points in the three-dimensional safety strategy space, and the change in strategy axis level between adjacent intermediate strategy coordinate points does not exceed the maximum change limit in a single step. The position of the strategy coordinate point in the three-dimensional security strategy space is gradually adjusted according to the strategy drift path. After each adjustment, the acquisition of the response deviation vector and the update of the security evaluation value are re-executed until the strategy coordinate point reaches the target strategy coordinate point, thus completing the strategy drift process. During the strategy drift process, if the magnitude of the response deviation vector corresponding to any strategy coordinate point exceeds the preset deviation magnitude alarm threshold, the strategy drift process is immediately terminated and the process reverts to the previous strategy coordinate point where the safety evaluation value is higher than the safety evaluation value threshold. At the same time, the strategy coordinate point is marked as a risk-free zone coordinate point and excluded from the subsequent search range. When the number of coordinate points marked as risk restricted areas in the safety topographic map database exceeds the preset threshold for the number of risk restricted areas, a cognitive reset operation is triggered. This resets all historical response deviation vectors in the safety topographic map database to zero and resets the current strategy coordinate point to the center origin position of the three-dimensional safety strategy space, restarting the exploration and accumulation process of the safety topographic map.

[0010] Furthermore, the update and policy drift process of the secure terrain map database also includes a two-loop learning mechanism driven by prediction bias: In the secure terrain map database, the security evaluation value of each strategy coordinate point is generated by a weighted fusion of short-term sliding window evaluation value and long-term decay integral evaluation value. The short-term sliding window evaluation value is calculated based on the reciprocal of the mean amplitude of the response deviation vector collected during the most recent N visits to the strategy coordinate point. The long-term decay integral evaluation value is calculated based on the reciprocal of the mean amplitude obtained by exponentially weighting and accumulating the response deviation vectors in all historical access records of the strategy coordinate point according to the time decay weight. The time decay weight decays exponentially with the increase of the recording time, and the decay time constant is preset to a fixed multiple of the system operating cycle. After completing the strategy drift in each control cycle, the ratio of the adsorption force response deviation component to the load torque response deviation component in the response deviation vector corresponding to the current strategy coordinate point is calculated to generate the coupling deviation direction angle corresponding to that strategy coordinate point. The coupling deviation direction angle represents the dynamic phase relationship between the adsorption force and the load torque at the current strategy coordinate point. The variance of the historical coupling deviation direction angle sequence of the same strategy coordinate point is calculated. When the variance is lower than the preset direction angle stability threshold, the strategy coordinate point is marked as a low-disturbance strategy coordinate point, and the safety evaluation value of the low-disturbance strategy coordinate point is multiplied by a preset stabilization reward coefficient. When the variance is higher than the preset direction angle fluctuation threshold, the strategy coordinate point is marked as a high-disturbance strategy coordinate point, and the safety evaluation value of the high-disturbance strategy coordinate point is multiplied by a preset stabilization penalty coefficient. When the total number of policy coordinate points in the security topographic map database reaches a preset saturation capacity threshold, a topographic map compression operation is triggered: the product of the access frequency and security evaluation value of each policy coordinate point is calculated as the retention priority score of the policy coordinate point, and policy coordinate points are deleted in order of retention priority scores from low to high until the total number of policy coordinate points drops to a preset compressed capacity threshold. The statistical characteristics of the response deviation vector corresponding to the deleted policy coordinate points are merged into the long-term decay integral evaluation value of the nearest neighboring policy coordinate points with retention priority scores higher than the preset retention threshold, so as to achieve lossless transfer of the cognitive information carried by the deleted coordinate points. In the generation of the strategy drift path, a prediction bias-driven path planning method is adopted: when planning an intermediate path from the current strategy coordinate point to the target strategy coordinate point, for each candidate intermediate strategy coordinate point, based on the direction of the historical response bias vector of the intermediate strategy coordinate point, the predicted response bias direction generated when continuing to shift along the intermediate strategy coordinate point towards the target direction is predicted. If the angle between the predicted response bias direction and the direction of the current response bias vector is less than a preset direction consistency threshold, the path cost of the intermediate strategy coordinate point is set to a first cost value; if the angle between the predicted response bias direction and the direction of the current response bias vector is greater than or equal to the direction consistency threshold, the path cost of the intermediate strategy coordinate point is set to a second cost value greater than the first cost value, so that the strategy drift path preferentially passes through the safe terrain map area consistent with the current system response characteristics; the path costs of each intermediate strategy coordinate point are accumulated, and the sequence of intermediate strategy coordinate points with the smallest accumulated path cost is selected as the strategy drift path. During the execution of the strategy drift path, the difference between the actual response deviation vector and the predicted response deviation vector at each intermediate strategy coordinate point is recorded in real time. The difference between the predicted deviation vector and the predicted response deviation vector is integrated and accumulated along the time axis. When the accumulated value exceeds the preset prediction deviation accumulation threshold, it is determined that the safety evaluation value of the strategy coordinate point in the current safety terrain map database has deviated from the real system response characteristics, triggering the cognitive reset operation and clearing the accumulated prediction deviation value to zero.

[0011] Furthermore, the cognitive reset operation also includes cognitive imbalance grading processing and predictive anchoring mechanisms: The cumulative prediction deviation value that triggers the cognitive reset operation is classified into levels: when the cumulative prediction deviation value exceeds the first cumulative prediction deviation threshold but does not exceed the second cumulative prediction deviation threshold, it is determined to be a level one cognitive imbalance, and a local cognitive reset operation is performed. The local cognitive reset operation only clears the historical response deviation vectors corresponding to the top K policy coordinate points with the largest difference in prediction deviation in the safety terrain map database, while the historical response deviation vectors of the remaining policy coordinate points remain unchanged. The value of K is determined by the product of the degree to which the cumulative prediction deviation value exceeds the first cumulative prediction deviation threshold and a preset degree coefficient. When the cumulative prediction deviation value exceeds the second cumulative prediction deviation threshold, it is determined to be a level two cognitive imbalance, and a global cognitive reset operation is performed. The global cognitive reset operation clears the historical response deviation vectors of all policy coordinate points in the safety terrain map database to zero and resets the current policy coordinate point to the center origin position of the three-dimensional safety policy space. After the global cognitive reset operation is completed, the accelerated exploration mode is started: the number of strategy levels on each strategy axis in the three-dimensional safety strategy space is increased to twice that in the normal exploration mode to refine the exploration granularity. At the same time, the maximum single-step change limit of strategy drift is expanded to twice that in the normal exploration mode to accelerate the coverage of the strategy space. The duration of the accelerated exploration mode is proportional to the degree to which the cumulative prediction deviation exceeds the second cumulative prediction deviation threshold. After the duration is reached, it automatically switches back to the normal exploration mode. During the continuous process of the accelerated exploration mode, the amplitude change trend of the newly acquired response deviation vector is monitored in real time. When the amplitude of the response deviation vector shows a monotonically decreasing trend within several consecutive control cycles and the final amplitude is lower than the preset steady-state amplitude threshold, it is determined that the accelerated exploration has reached the convergence state. The accelerated exploration mode is terminated in advance and switched back to the normal exploration mode. At the same time, all newly acquired response deviation vectors in the accelerated exploration mode are written into the safe topographic map database with a weight coefficient higher than that in the normal exploration mode, so that the cognitive information obtained by the new exploration can quickly take the lead. When the cognitive reset operation is triggered more than a preset frequent reset threshold within a preset time window, the system is determined to be in a state of continuous cognitive imbalance, and an ultimate safety lock signal is generated. The ultimate safety lock signal triggers the following operations: the three-dimensional safety strategy space is forcibly converged to an extremely conservative strategy coordinate point composed of the maximum level of the suction maintenance strategy axis, the minimum level of the load output strategy axis, and the minimum level of the contact state management strategy axis, and is maintained at the extremely conservative strategy coordinate point until the operator performs a manual reset operation. At the same time, all strategy coordinate points that are not marked as risk restricted area coordinate points are batch marked as temporary risk restricted area coordinate points, and the safety evaluation value of the temporary risk restricted area coordinate points is uniformly reduced to a preset temporary lock evaluation value. The temporary lock evaluation value is lower than the safety evaluation value threshold, so that the system cannot leave the extremely conservative strategy coordinate point by strategy drift before the manual reset is completed.

[0012] Furthermore, the calculation and utilization of the coupling deviation direction angle also includes a causal failure path discrimination and differentiated protection mechanism based on the temporal evolution of the coupling deviation direction angle: The coupling deviation direction angle sequence of the current strategy coordinate point and its neighboring strategy coordinate points is continuously recorded within multiple control cycles. A first-order difference operation is performed on the coupling deviation direction angle sequence to obtain the coupling deviation direction angle change rate sequence. When the coupling deviation direction angle change rate sequence shows a positive increasing trend within a consecutive preset number of control cycles, and the change in the load torque response deviation component precedes the change in the adsorption force response deviation component, the current system is determined to be in an overload-dominated failure path. The overload-dominated failure path is characterized by load impact occurring before adsorption force attenuation, with load impact being the cause and adsorption force attenuation being the result. When the coupling deviation direction angle change rate sequence shows a negative increasing trend within a consecutive preset number of control cycles, and the change in the adsorption force response deviation component precedes the change in the load torque response deviation component, the current system is determined to be in a demagnetization-dominated failure path. The demagnetization-dominated failure path is characterized by adsorption force attenuation occurring before load impact, with adsorption force attenuation being the cause and load impact being the result. Differential protection strategies are implemented based on the determined failure path type: If the failure path is determined to be overload-dominant, strategy drift on the load output strategy axis is prioritized. First-priority drift is performed on the load output strategy axis in the direction of reducing the maximum allowable output torque, while strategy drift on the suction maintenance strategy axis and the contact state management strategy axis are performed in parallel as auxiliary drifts. The step size of the auxiliary drift does not exceed a preset proportion of the first-priority drift step size, so as to prioritize cutting off the load impact source and blocking the causal transmission path from overload to demagnetization. If the failure path is determined to be demagnetization-dominant, strategy drift on the suction maintenance strategy axis is prioritized. First-priority drift is performed on the suction maintenance strategy axis in the direction of increasing the excitation current, while preparatory load reduction drift is performed on the load output strategy axis. The step size of the preparatory load reduction drift is a preset proportion of the first-priority drift step size, so as to reserve a safety margin while replenishing magnetization and prevent sudden overload from occurring before the replenishment of magnetization is fully effective. During the strategy drift process, the sign change of the coupling deviation direction angle change rate sequence is continuously monitored. When the sign change rate exceeds the preset path turning threshold, it is determined that the failed path type has reversed. The current strategy drift direction is immediately stopped, and the drift is reversed to the direction of the starting strategy coordinate point to perform a backtracking drift. After backtracking to the starting strategy coordinate point, the starting strategy coordinate point is marked as the failed path reversal boundary point. A repulsion field is established with the failed path reversal boundary point as the origin. The repulsion field increases the path cost of strategy coordinate points that pass through the neighborhood of the failed path reversal boundary point in the subsequent strategy drift path planning. The increase in path cost is inversely proportional to the distance of the strategy coordinate point from the failed path reversal boundary point.

[0013] Furthermore, the construction and updating of the security terrain map database also includes a multi-machine collaborative group cognition aggregation and a general security prior knowledge transfer mechanism: When two or more magnetic drilling rigs are within the same network coverage area or connected to the same cloud-based cognitive aggregation server via a remote communication interface, each magnetic drilling rig uploads the coordinate values ​​of its local safety topographic map database, including policy coordinate points, safety evaluation values, risk exclusion zone coordinate point markers, and coupling deviation direction angle statistical characteristics, to the cloud-based cognitive aggregation server via an encrypted communication protocol. After receiving the safety topographic map data uploaded by each device, the cloud-based cognitive aggregation server performs cross-device coordinate alignment processing according to the coordinate values ​​of the policy coordinate points. For policy coordinate points with the same coordinate values ​​or coordinate value differences less than a preset coordinate tolerance threshold, the corresponding safety evaluation values ​​are weighted and aggregated to generate a group aggregated safety evaluation value. The weight coefficient of the weighted aggregation is positively correlated with the total number of local policy drifts of the device that uploaded the policy coordinate point. The cloud-based cognitive aggregation server identifies, among all aligned policy coordinate points, policy coordinate points whose group aggregation security evaluation value is higher than a preset group excellence threshold and which have not been marked as risk-prone areas by any device. These points are marked as general security prior policy coordinate points, and all general security prior policy coordinate points and their corresponding group aggregation security evaluation values ​​are encapsulated into a general security prior knowledge package. When a newly connected magnetic drilling device starts up for the first time, the device downloads the general security prior knowledge package from the cloud-based cognitive aggregation server. The general security prior policy coordinate points and their group aggregation security evaluation values ​​in the package are used as the initial data for the device's security topographic map database. This enables the newly connected device to possess group-verified security cognitive capabilities from the startup phase, shortening the policy exploration convergence time after startup. After the newly connected magnetic drilling rig completes local policy drift and accumulates a local response deviation vector, it compares the local security evaluation value corresponding to the coordinate point of the general security prior policy in the local security terrain map database with the group aggregated security evaluation value. If the absolute value of the deviation between the local security evaluation value and the group aggregated security evaluation value is less than a preset group-individual deviation tolerance threshold, the weight coefficient of the policy coordinate point is increased in subsequent local security evaluation value updates to enhance the stability of the policy corresponding to the policy coordinate point. If the absolute value of the deviation is greater than or equal to the group-individual deviation tolerance threshold, it is determined that the policy combination corresponding to the policy coordinate point is affected by the device's unique characteristics. Due to physical constraints or operating conditions, the policy coordinate point is marked as a device-specific policy coordinate point, and its weight coefficient is reduced in subsequent local safety evaluation value updates. Simultaneously, when uploading the next safety terrain map data to the cloud-based cognitive aggregation server, the marking information of the device-specific policy coordinate point and its corresponding local operating condition parameters are also uploaded. This allows the cloud-based cognitive aggregation server to further divide the general safety prior knowledge package into operating condition subclasses according to the operating condition parameters during subsequent updates, generating multiple operating condition-specific general safety prior knowledge sub-packages corresponding to different operating condition subclasses. These sub-packages are then selectively downloaded by new magnetic drilling equipment connected under different operating conditions based on operating condition matching. The cloud-based cognitive aggregation server adaptively adjusts the update cycle of the general security prior knowledge package based on the update frequency of the security terrain map data uploaded by each device: when the amount of data uploaded per unit time exceeds a preset group activity threshold, the update cycle is shortened and the weight coefficient of recently uploaded data in the weighted aggregation is increased; when the amount of data uploaded per unit time is lower than the group activity threshold, the update cycle is extended and the weight coefficient of recently uploaded data in the weighted aggregation is decreased, so that the evolution speed of the general security prior knowledge package and the update speed of group cognition are dynamically matched.

[0014] In addition, the present invention also provides a multi-level linkage protection-based magnetic drill demagnetization overload collaborative control system, which is used to store and execute the above-mentioned methods.

[0015] Based on the above, the magnetic drill demagnetization overload collaborative control method and system based on multi-level linkage protection provided by the present invention has achieved the following significant technical effects compared with the prior art.

[0016] First, it breaks through the traditional architecture where demagnetization protection and overload protection are independent. By establishing a dynamic threshold model linking adsorption force and allowable output torque, it achieves dynamic matching control that fully utilizes processing capacity when adsorption force is sufficient and automatically narrows the load limit when adsorption force decreases. When a sudden overload impact generates vibration and tension, the system prioritizes flexible buffering and load reduction instead of hard shutdown, cutting off the causal chain of overload vibration transmission to the adsorption interface from the physical source, effectively preventing secondary demagnetization caused by overload.

[0017] Secondly, it fundamentally overcomes the technical deficiency of traditional voltage detection methods, which cannot identify the actual decline in adsorption force. The magnetic Hall sensor directly senses the magnetic flux density at the contact surface between the suction cup and the workpiece, effectively detecting the actual decrease in adsorption force caused by factors such as contact surface corrosion, unevenness of the plate surface, and coil aging, providing a high-fidelity data foundation for linkage control. The three-level layered protection logic enables the system to initiate load reduction, magnetization replenishment, and early warning operations at the stage of slight demagnetization, intervening before the problem worsens, avoiding the mechanical shock and damage to transmission components caused by direct hard power cut-off, and significantly extending the service life of the entire machine.

[0018] Third, through proactive exploration of the strategy space and a control mechanism driven by a safety topography map, the system acquires the ability to proactively perceive its own safety boundaries. During drilling, controlled strategy disturbance exploration is continuously executed, transforming the predictive deviations that traditional methods treat as control errors into positive feedback signals driving the evolution of safety cognition, establishing a dual-loop learning architecture that includes risk exclusion zone markers. The inner loop corrects the safety evaluation values ​​of each strategy coordinate point, while the outer loop triggers a graded cognitive reset when cognitive imbalance occurs, enabling the system to maintain continuous awareness of safety boundaries even when the true safety margin cannot be directly measured. The causal failure path discrimination mechanism further distinguishes between two fundamentally different failure modes: overload-dominated and demagnetization-dominated, executing differentiated priority drift strategies for each, significantly improving the targeting and effectiveness of control measures.

[0019] Fourth, the multi-machine collaborative swarm intelligence aggregation mechanism enables the security topography knowledge of a single device to be aggregated in the cloud and migrated across devices. Newly connected devices do not need to start from scratch to obtain initial security knowledge that has been verified by the swarm, significantly shortening the policy convergence time. The communication interface unit adaptively adjusts the knowledge package update cycle, so that the evolution speed of swarm intelligence is dynamically matched with the activity level of the swarm, realizing the leap from single-machine intelligence to swarm intelligence.

[0020] In summary, this invention systematically solves the technical challenges of demagnetization and overload coupling protection in the field of magnetic drill safety control from four aspects: perception accuracy, control strategy, cognitive architecture, and group collaboration. It has outstanding substantive features and significant progress. Attached Figure Description

[0021] Figure 1This is a flowchart illustrating the method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the system provided by the present invention.

[0022] Figure 3 This diagram illustrates the comparison between predicted and measured values ​​of the response deviation amplitude during the drift process of the strategy of this invention. The horizontal axis represents the control cycle number, and the vertical axis represents the amplitude of the response deviation vector, which is the normalized percentage value of the sum of the absolute values ​​of the adsorption force response deviation component, the load torque response deviation component, and the feed thrust response deviation component. The predicted values ​​are generated from a safety topographic map database based on historical statistical characteristics, while the measured values ​​are the actual values ​​collected by the sensors.

[0023] Figure 4 This diagram illustrates a comparison of the control effects of the conventional and present solutions when subjected to hard particle impacts; the vertical axis represents the load torque and the adsorption force retention rate. In the conventional solution, the torque surges dramatically under hard particle impact, triggering a hard shutdown, while vibration causes a significant drop in adsorption force. The present solution achieves flexible buffering through strategy drift, effectively suppressing torque fluctuations and maintaining stable adsorption force. Detailed Implementation

[0024] like Figures 1-4 As shown.

[0025] This technical solution provides a collaborative control method and system for demagnetization overload of magnetic drill bit based on multi-level linkage protection. The technical details are explained in detail below using a specific application scenario.

[0026] The magnetic drill bit involved in this embodiment is a high-pressure brushless magnetic drill bit. Its electromagnetic chuck has a rated suction force of 30 kN, its spindle motor has a rated power of 2200 watts, and its maximum speed is 2500 revolutions per minute. The entire control system of the magnetic drill bit is centered on a main control MCU, which uses an STM32F407 series microcontroller with a main frequency of 168 MHz and a floating-point unit to meet the computational requirements of the real-time control algorithm. The main control MCU connects to the signal conditioning circuits of various sensors through an analog-to-digital converter interface, controls the excitation current of the electromagnetic chuck through a pulse width modulation output interface, drives the brushless spindle motor through a six-step commutation or vector control method, and controls the audible and visual warning unit and the fault reset and locking unit through a general-purpose input / output interface. Data communication within the control system uses a controller local area network bus with a communication rate of 500 kilobits per second, ensuring real-time exchange of control commands and feedback data between units.

[0027] After the equipment is powered on, it first performs a self-test process.

[0028] Upon power-up, the main control MCU immediately initiates a magnetic self-test program, without performing any motor drive operations. The self-test program reads the current actual adsorption force parameters of the electromagnetic chuck through the magnetic detection unit. The magnetic detection unit employs a magnetic Hall sensor, which is mounted on the surface of the electromagnetic chuck's magnetic poles, maintaining a 0.5 mm gap from the workpiece contact surface. The magnetic Hall sensor directly senses the magnetic flux density between the magnetic poles and the workpiece contact surface. This magnetic flux density signal is amplified and filtered by a signal conditioning circuit before being sent to the analog-to-digital conversion channel of the main control MCU. The main control MCU internally stores a calibration curve between magnetic flux density and effective adsorption force. This calibration curve, obtained through experimental calibration, covers the full range from zero to rated adsorption force, achieving a measurement accuracy of 2% of the rated adsorption force.

[0029] The main control MCU compares the detected actual adsorption force parameter with a preset safety threshold. The preset safety threshold is set to 80% of the rated suction force, i.e., 24KN. If the actual adsorption force parameter reaches or exceeds 24KN, the main control MCU determines that the adsorption conditions meet the safety requirements, unlocks the spindle motor, and allows the operator to start drilling operations. If the actual adsorption force parameter is lower than 24KN, the main control MCU determines that the adsorption is insufficient, issues an audible and visual alarm signal, and simultaneously locks the spindle motor drive circuit, preventing the spindle motor from starting under any operating signal. The audible and visual alarm is triggered by a constantly lit red indicator light and an intermittent buzzer sound, continuously reminding the operator to check the power supply line of the electromagnetic chuck, the surface condition of the workpiece, and the adhesion between the chuck and the workpiece. After the self-check process is completed, if the adsorption conditions are not met, the system remains locked until the adsorption force parameter is detected to meet the standard before it can be unlocked.

[0030] After the self-test is passed and the operator issues the start command, the equipment enters the drilling operation stage.

[0031] During the drilling operation, the main control MCU performs real-time synchronous acquisition of dual signals. The acquisition operation is repeated at a fixed control cycle, set to ten milliseconds. At the beginning of each control cycle, the main control MCU simultaneously triggers the magnetic attraction detection unit and the load acquisition unit to sample data. The magnetic attraction detection unit acquires the actual attraction force parameters at the current moment in the manner described above. The load acquisition unit includes two sub-units: a motor current sampling unit and a Hall speed acquisition unit. The motor current sampling unit acquires the three-phase current signal through a shunt resistor connected in series in the motor power supply line. The shunt resistor has a resistance of one milliohm. The signal is sent to the main control MCU after differential amplification and analog-to-digital conversion. The Hall speed acquisition unit receives the output signals from three Hall sensors built into the spindle motor and calculates the current speed by measuring the time interval between adjacent Hall pulses. Based on the acquired motor current and speed values, combined with motor parameters and drive efficiency curves, the main control MCU calculates and outputs the current output torque value of the spindle motor in real time.

[0032] After the main control MCU completes the acquisition and conversion of the above two parameters in each control cycle, it enters the dynamic threshold calculation stage.

[0033] The core basis for dynamic threshold calculation is the actual adsorption force parameter. The main control MCU internally stores an adsorption force-allowable torque mapping table, implemented as a piecewise linear function. The input to the mapping table is the actual adsorption force parameter, and the output is the currently allowed maximum output torque threshold. The higher the actual adsorption force parameter, the larger the allowable torque threshold output by the mapping table. The specific numerical relationships of the mapping table are set as follows: when the actual adsorption force parameter is 100% of the rated suction force, the allowable torque threshold is set to 90% of the motor's rated torque; when the actual adsorption force parameter is 80% of the rated suction force, the allowable torque threshold is set to 70% of the motor's rated torque; when the actual adsorption force parameter is 60% of the rated suction force, the allowable torque threshold is set to 40% of the motor's rated torque. For adsorption force values ​​between the above calibration points, a linear interpolation method is used to calculate the corresponding allowable torque threshold. This mapping relationship ensures that the equipment can fully utilize its processing capacity when the adsorption force is sufficient, and that the equipment's load capacity automatically decreases when the adsorption force decreases, achieving a dynamic match between load capacity and adsorption safety margin.

[0034] After obtaining the current maximum allowable output torque threshold, the main control MCU will compare the load parameters collected in real time with the threshold and execute the four-level linkage protection logic based on the comparison result.

[0035] The first level corresponds to normal operating conditions. When the actual adsorption force parameter reaches or exceeds the first preset threshold and the current output torque value does not exceed the maximum allowable output torque threshold, the main control MCU determines that the system is in a safe operating state. The first preset threshold is set to 70% of the rated suction force. In this state, the main control MCU does not take any limiting measures, the electromagnetic chuck maintains the current excitation current, the spindle motor runs freely according to the speed and load given by the operator, and the feed mechanism operates normally. However, the main control MCU continues to perform the aforementioned data acquisition and comparison operations to provide a data basis for the decision-making in the next cycle.

[0036] The second level corresponds to a mild loss of magnetization warning condition. When the actual adsorption force parameter is lower than the first preset threshold but the current output torque value does not exceed the maximum allowable output torque threshold, the main control MCU determines that the system has experienced a decrease in adsorption force but has not yet experienced overload. At this time, the main control MCU performs two operations. The first operation is to lower the maximum allowable output torque threshold. The reduction is proportional to the difference between the actual adsorption force parameter and the first preset threshold; the larger the difference, the larger the reduction, further narrowing the upper limit of the motor's load in subsequent control cycles. The second operation is to issue a yellow warning signal, with the yellow light of the audible and visual warning unit flashing to alert the operator that the current adsorption state has deviated from the optimal range. At the same time, the main control MCU initiates an active magnetization operation, increasing the pulse width modulation duty cycle of the electromagnetic chuck's excitation current to enhance the magnetic attraction strength. The magnitude of the magnetization operation is limited by the rated current limit of the electromagnetic chuck coil. The main control MCU monitors the coil current feedback value to ensure that the current does not exceed 110% of the rated value during the magnetization process, preventing the coil from overheating and being damaged. The magnetization operation is performed in a slow, incremental manner, increasing the duty cycle by one percentage point every five control cycles until the adsorption force parameter rises back above the first preset threshold or the duty cycle reaches its upper limit.

[0037] The third level corresponds to the overload linkage protection condition. When the output torque value exceeds the maximum allowable output torque threshold, but the actual adsorption force parameter still reaches or exceeds the first preset threshold, the main control MCU determines that the system has experienced a sudden load impact, but the adsorption foundation remains firm. At this time, the main control MCU does not take a hard shutdown approach, but instead performs a flexible buffer load reduction operation. The flexible buffer load reduction is achieved through three coordinated methods. The first method is to limit the peak motor current. The main control MCU clamps the upper limit of the reference current of the current loop to the current value corresponding to the current threshold of the current allowable torque, preventing the current from rising further. The second method is to gradually reduce the pulse width modulation duty cycle. The main control MCU reduces the motor drive voltage at a rate of 5% every two control cycles, so that the output torque decreases smoothly. The third method is to pause the feed motor. By stopping the stepper motor drive pulse of the feed mechanism, the drill bit stops advancing deeper into the workpiece, thereby blocking the source of further load increase. The combined use of the above three methods allows the overload energy to be released gradually, avoiding the mechanical impact caused by a sudden power outage.

[0038] Level 4 corresponds to a dual-hazard lockout condition. When both conditions are met simultaneously—the actual adsorption force parameter being lower than the second preset threshold and the output torque value exceeding the maximum allowable output torque threshold—the main control MCU determines the system to be in an extremely dangerous state. The second preset threshold is set to 50% of the rated adsorption force. In this situation, the adsorption force is severely insufficient and overload occurs simultaneously, and the equipment may fall or overturn at any time. The main control MCU immediately executes the emergency stop procedure. The first action of the emergency stop procedure is to block the pulse width modulation output of the spindle motor, causing the motor to stop rotating in the shortest possible time. After completing the first action, the main control MCU delays for 50 milliseconds. After the spindle motor has completely stopped, it does not cut off the power supply to the electromagnetic chuck, but instead maintains (or increases to a safe holding current) the current excitation current, only blocking the drive pulses of the spindle motor and the feed motor, so that the entire machine enters a fault stop while maintaining the adsorption lockout state. The purpose of the delayed power cut-off is to prevent the back electromotive force generated at the moment the motor stops from interfering with the chuck control circuit, and also to prevent the spindle from still rotating under load and generating additional impact at the moment the chuck is powered off. After the power outage is complete, the main control MCU writes the fault lock flag to the non-volatile memory and locks all drive units of the entire machine, preventing the device from restarting under any operating signal. The audible and visual warning unit issues a red alarm signal, the red indicator light flashes rapidly, and the buzzer sounds continuously.

[0039] In the four-level protection logic, there is a progressive transition relationship from level one to level four. If the adsorption force decreases in level one, it transitions to level two. If magnetization is ineffective in level two and the adsorption force continues to decrease while the load increases, it transitions to level four. If a sudden overload occurs in level one, it directly transitions to level three. If an overload impact causes the adsorption force to drop below a second preset threshold in level three, it transitions to level four. This progressive relationship allows the system to intervene through the first three levels of protection in most abnormal situations, triggering a hard lockout at level four only in extreme cases.

[0040] It is important to note the active magnetization operation in the second-level control. The active magnetization operation starts immediately after a slight loss of magnetization warning is triggered, but its execution is gradual rather than abrupt. Within each control cycle, the main control MCU calculates the difference between the current adsorption force parameter and the first preset threshold, determining the magnetization step size based on the magnitude of the difference. When the difference is large, the magnetization step size is appropriately increased, but the increment per step does not exceed two percent of the rated excitation current. When the difference is small, the magnetization step size is correspondingly decreased to avoid overshoot that could cause a sudden increase in adsorption force and lead to mechanical shock. The magnetization operation continues until the adsorption force parameter stabilizes above the first preset threshold again. At this point, the main control MCU slowly reduces the excitation current to the level before magnetization, maintaining the system in a low-energy-consumption state.

[0041] The fault lock triggered by the fourth-level control requires manual reset to be released. After troubleshooting and eliminating the cause of the fault, the operator needs to press and hold the reset button on the equipment panel for three seconds. Upon detecting the long press of the reset button, the main control MCU reads the information from the fault record memory, confirms that the fault has been recorded, clears the fault lock flag, and unlocks the entire machine. After unlocking, the equipment will not start automatically; the operator needs to re-execute the power-on self-test procedure and reissue the start command. This design ensures that a complete safety check process must be performed after each fault lock, preventing secondary accidents caused by incomplete fault resolution. The information stored in the fault record memory includes the fault trigger time, fault type identifier, adsorption force parameter value and load parameter value at the time of fault triggering, as well as snapshots of key operating parameters from the last ten control cycles before the fault occurred. This information can be read through the debugging interface during later maintenance for fault cause tracing and equipment status assessment.

[0042] The entire control process executes continuously in control cycles. After each 10-millisecond control cycle, the main MCU saves the decision result and execution status of the current cycle in a circular buffer of its internal random access memory, and then proceeds to the next cycle for data acquisition and processing. The circular buffer stores data from the most recent 1,000 cycles, covering a 10-second runtime window. When the system detects any abnormal event, the data in this buffer is locked and transferred to non-volatile memory, forming a snapshot of the fault scene. This mechanism provides complete time-series data support for post-fault analysis and helps identify the root cause of intermittent anomalies.

[0043] Building upon the aforementioned four-level hierarchical linkage protection logic, this embodiment also introduces a control mechanism driven by proactive strategy space exploration and security terrain map. This mechanism runs continuously in the background of the main control MCU, executing in parallel with the aforementioned main loop of acquisition-comparison-execution, without interfering with the real-time response of the main loop.

[0044] To facilitate understanding of this mechanism, a specific application scenario is provided below. This scenario follows the equipment configuration and operating parameters of the aforementioned embodiment, specifically drilling a hole in a 40mm thick Q345 structural steel plate. The hole diameter is 20mm, the target depth is 35mm, and the process sequence is to first use a center drill for positioning, then switch to a 20mm drill bit for drilling. After the operator starts drilling, the spindle motor operates at 800 rpm, the electromagnetic chuck excitation current is maintained at 90% of its rated value, and the feed rate is set to 0.15mm per rpm. When the drilling reaches approximately two-thirds of its depth, due to the presence of localized hard points within the workpiece, the load torque experiences slight fluctuations, accompanied by minor vibrations. Under these conditions, the equipment's control system not only performs the routine monitoring under the aforementioned first-level normal operating conditions but also simultaneously initiates a background strategy space exploration process to proactively assess the actual margin of the current safety boundary.

[0045] The main control MCU internally constructs a three-dimensional safety strategy space. The three dimensions of this space correspond to the suction maintenance strategy axis, the load output strategy axis, and the contact status management strategy axis, respectively. The suction maintenance strategy axis corresponds to the adjustment method of the electromagnetic chuck's excitation current, and this axis is discretized into five strategy levels, corresponding to excitation currents of 60%, 70%, 80%, 90%, and 100% of the rated value. The load output strategy axis corresponds to the limiting method of the spindle motor's output torque, and this axis is discretized into five strategy levels, corresponding to the maximum allowable output torque of 40%, 55%, 70%, 80%, and 90% of the rated torque. The contact status management strategy axis corresponds to the adjustment method of the drill bit feed rate, and this axis is discretized into five strategy levels, corresponding to feed rates of 50%, 75%, 100%, 125%, and 150% of the set rate. Any combination of three strategy levels constitutes a coordinate point in the strategy space, and this coordinate point completely describes a set of control strategy combinations. The three policy axes are divided into 5x5x5 grids, forming a policy space grid with a total of 125 discrete coordinate points.

[0046] Under normal operating conditions, the current strategy coordinate point is located at the intersection of the third gear of the three strategy axes, that is, the suction maintenance strategy axis corresponds to the 80% gear of the rated excitation current, the load output strategy axis corresponds to the 70% gear of the rated torque, and the contact state management strategy axis corresponds to the 100% gear of the set feed rate.

[0047] Within each 10-millisecond control cycle, in addition to performing the aforementioned data acquisition and comparison operations, the main control MCU also performs a strategy space exploration operation in the background. Specifically, using the current strategy coordinate point as the center, it applies a strategy level offset disturbance along each of the three strategy axes, generating three tentative strategy coordinate points. The first tentative coordinate point is offset one level positively along the suction maintenance strategy axis, i.e., the excitation current is increased to 90% of the rated value, while the other two strategy axes remain unchanged. The second tentative coordinate point is offset one level negatively along the load output strategy axis, i.e., the maximum output torque is reduced to 55% of the rated torque, while the other two strategy axes remain unchanged. The third tentative coordinate point is offset one level positively along the contact state management strategy axis, i.e., the feed speed is increased to 125% of the set rate, while the other two strategy axes remain unchanged.

[0048] The main control MCU maps each tentative strategy coordinate point to a corresponding tentative excitation current adjustment, tentative torque limit, and tentative feed speed adjustment, and sends these parameters to the electromagnetic chuck drive unit, brushless spindle drive unit, and feed drive unit. The control parameters for each tentative strategy coordinate point operate continuously for two control cycles, or twenty milliseconds. Within these twenty milliseconds, the main control MCU collects the response changes of the electromagnetic chuck's adsorption force, the spindle motor's load torque, and the drill bit's feed thrust through the sensor acquisition layer.

[0049] After data acquisition, the main control MCU calculates the response deviation vector for each trial strategy coordinate point relative to the current strategy coordinate point. This response deviation vector contains three components: the adsorption force response deviation component, the load torque response deviation component, and the feed thrust response deviation component. All three components are expressed as relative deviations from their respective rated values, with units of percentage, thus ensuring a unified dimension for the three components. For example, if, after shifting one step forward along the positive direction of the suction maintenance strategy axis, the measured adsorption force increases by 2% relative to the baseline adsorption force value at the current strategy coordinate point, then the adsorption force response deviation component at that trial coordinate point is 2%. Similarly, if the load torque response deviation component is -1% and the feed thrust response deviation component is +0.5%, then the response deviation vector for that trial coordinate point is (+2%, -1%, +0.5%).

[0050] The main control MCU stores the response deviation vector and its corresponding policy coordinates into a safety topographic map database. The safety topographic map database uses a multi-dimensional grid index structure, with the three tier numbers of the policy coordinates as combined keys, to store the statistical characteristics of the historically accumulated response deviation vectors at that coordinate point. Specifically, the statistical characteristics include the mean vector, covariance matrix, access count, and most recent access timestamp of all response deviation vectors at that coordinate point. The safety topographic map database is stored in the main control MCU's off-chip serial flash memory, with a capacity of 2 megabytes, capable of storing data for up to 65,536 policy coordinates.

[0051] The main control MCU updates the safety evaluation value of the strategy coordinate point based on the magnitude direction and duration of the response deviation vector. The safety evaluation value is a dimensionless scalar, ranging from 0 to 1, with higher values ​​indicating greater safety margins for the strategy coordinate point. The specific calculation method for the safety evaluation value is as follows: First, take the absolute values ​​of the three components of the response deviation vector and sum them to obtain the total deviation magnitude; the reciprocal of the total deviation magnitude is used as the basic safety score. Then, calculate the sliding window standard deviation of each component of the response deviation vector over the most recent 10 visits, and the reciprocal of the standard deviation is used as the stability score. The safety evaluation value equals the product of the basic safety score and the stability score, and is then normalized to fall within the 0-1 range. According to this calculation method, the smaller the magnitude of the response deviation vector, the higher the safety evaluation value; the higher the stability of the deviation duration, the higher the safety evaluation value. Conversely, if the total magnitude of the response deviation vector exceeds 15% in a trial exploration, or if the components show a divergent trend in consecutive visits, the safety evaluation value of the strategy coordinate point will decrease accordingly.

[0052] When the main control MCU detects that the security evaluation value of the current policy coordinate point is lower than the preset security evaluation value threshold, it initiates the policy drift process. The security evaluation value threshold is set to 0.35. This threshold is determined based on the statistical distribution of security evaluation values ​​under normal system operation, taking the value of the mean security evaluation value under normal operation shifted downwards by two standard deviations. When the security evaluation value is lower than this threshold, it indicates that the security margin of the current policy combination has significantly decreased, and it is necessary to actively migrate to a safer policy region.

[0053] After the strategy drift process begins, the main control MCU retrieves the nearest neighboring strategy coordinate point with a security evaluation value higher than 0.35 from the security terrain map database as the target strategy coordinate point. The distance in the strategy space uses normalized Euclidean distance; that is, the gear numbers of the three strategy axes are first normalized to the interval between 0 and 1, and then the three-dimensional Euclidean distance is calculated. The search range is a spherical neighborhood centered on the current strategy coordinate point with a normalized distance not greater than 0.4. Upon a successful search, the main control MCU generates a strategy drift path from the current strategy coordinate point to the target strategy coordinate point. This path is a polygonal path formed by gradually moving along the grid edges in the three-dimensional grid space. The change in the strategy axis gear between adjacent intermediate strategy coordinate points does not exceed one gear; that is, it moves only one gear on one strategy axis at a time. The main control MCU gradually adjusts the position of the strategy coordinate point according to this strategy drift path. After each adjustment, the acquisition of the response deviation vector and the update of the security evaluation value are re-executed until the strategy coordinate point reaches the target strategy coordinate point.

[0054] During strategy drift, if the sum of the magnitudes of the response deviation vectors exceeds a preset deviation magnitude alarm threshold after any adjustment step, the strategy drift process is immediately terminated. The deviation magnitude alarm threshold is set at 20%, meaning an alarm is triggered when the sum of the absolute values ​​of the adsorption force response deviation component, the load torque response deviation component, and the feed thrust response deviation component exceeds 20%. This threshold is determined based on the system's mechanical structure and the ultimate load-bearing capacity of the electromagnetic chuck. Exceeding this value indicates that the strategy combination has approached or exceeded the equipment's safe operating boundary. At this point, the main control MCU reverts the current strategy coordinate point to the previous strategy coordinate point with a safety evaluation value higher than 0.35, and simultaneously marks the strategy coordinate point that triggered the alarm as a risk-free zone coordinate point. This risk-free zone coordinate point is excluded from the search range in subsequent strategy drift searches to prevent the system from re-entering this unsafe area.

[0055] When the number of coordinate points marked as risk exclusion zones in the safety topographic map database exceeds a preset threshold, a cognitive reset operation is triggered. The risk exclusion zone threshold is set to 12, meaning that when 12 unsafe policy coordinate points are identified, it indicates that a large portion of the cognitive information in the current safety topographic map database has become invalid. The cognitive reset operation clears all historical response deviation vectors in the safety topographic map database to zero and resets the current policy coordinate points to the central origin of the three-dimensional safety policy space, i.e., the intersection of the third gear of the three policy axes, restarting the exploration and accumulation process of the safety topographic map. This mechanism ensures that the system's cognitive ability does not degrade due to the accumulation of too much invalid information.

[0056] Building upon the ongoing active exploration and safe topographic map-driven control mechanism in the strategy space, this embodiment further introduces a dual-loop learning mechanism driven by prediction bias, and a predictive anchoring mechanism for graded handling of cognitive imbalance. These two mechanisms, superimposed on the aforementioned safe topographic map database update and strategy drift processes, enable the system to continuously learn from historical exploration experience and dynamically adjust its cognitive boundaries.

[0057] Following the aforementioned application scenario, as drilling continues through workpieces containing localized hard points, the safety topography data accumulated by the system gradually becomes richer. The main control MCU maintains two sets of independent evaluation values ​​for each strategy coordinate point in the safety topography database. The first set of evaluation values ​​is a short-term sliding window evaluation value, calculated based on the reciprocal of the average amplitude of the response deviation vector collected during the most recent 100 visits to the strategy coordinate point. The window length of 100 visits is set based on a system operating cycle of 10 milliseconds, with 100 visits corresponding to a 1-second operating time. This timescale is sufficient to capture real-time fluctuations in the equipment's status during continuous drilling, while avoiding dulling the response to sudden changes due to an excessively long window. The second set of evaluation values ​​is a long-term decay integral evaluation value, calculated based on the reciprocal of the average amplitude obtained by exponentially weighting and accumulating the response deviation vectors from all historical access records of the strategy coordinate point according to time decay weights. The time decay weight decays exponentially with the increase of recording time, and the decay time constant is set to 1000 times the system operating cycle, corresponding to a 10-second operating time. The two sets of evaluation values ​​are weighted and merged with weight coefficients of 0.3 and 0.7 to generate a comprehensive security evaluation value for the coordinate point of the strategy. The short-term evaluation value has a lower weight and the long-term evaluation value has a higher weight. This setting ensures that the security evaluation value remains sensitive to sudden situations while mainly reflecting the inherent security characteristics of the strategy area that have been verified over a long period of time.

[0058] After completing the strategy drift in each control cycle, the main control MCU calculates the ratio of the adsorption force response deviation component to the load torque response deviation component in the response deviation vector corresponding to the current strategy coordinate point, generating the coupling deviation direction angle corresponding to that strategy coordinate point. This angle uses the adsorption force response deviation component as the reference axis and takes the arctangent ratio of the load torque response deviation component to the adsorption force response deviation component, with a value ranging from -90 degrees to 90 degrees. A positive angle indicates that the load torque response deviation direction leads the adsorption force response deviation direction, while a negative angle indicates lag. This angle numerically quantifies the dynamic phase relationship between adsorption force and load torque, and is an important basis for determining the causal order of the two physical quantities on the time axis.

[0059] The main control MCU calculates the variance of the historical coupling deviation direction angle sequence for the same strategy coordinate point. When the variance is lower than the preset direction angle stability threshold, it indicates that the dynamic phase relationship between the adsorption force and the load torque at that strategy coordinate point remains stable, without significant oscillations or fluctuations. The direction angle stability threshold is set to 15 degrees. In this case, the main control MCU marks the strategy coordinate point as a low-disturbance strategy coordinate point and multiplies its safety evaluation value by a 1.2-fold stabilization reward coefficient. When the variance is higher than the preset direction angle fluctuation threshold, it indicates that the phase relationship between the two physical quantities at that strategy coordinate point is unstable. The direction angle fluctuation threshold is set to 45 degrees. In this case, the main control MCU marks the strategy coordinate point as a high-disturbance strategy coordinate point and multiplies its safety evaluation value by a 0.7-fold destabilization penalty coefficient. This correction mechanism allows the system to focus not only on the magnitude of the response deviation but also on the phase stability within the response deviation during the evaluation of strategy safety, further improving the accuracy of the safety evaluation value in depicting the operating state.

[0060] The secure topographic map database continuously accumulates new strategy coordinate point data during operation. When the total number of strategy coordinate points in the database reaches a preset saturation capacity threshold of 2000, a topographic map compression operation is triggered. 2000 strategy coordinate points correspond to more than half of the entire 5x5x5 strategy space grid; reaching this value indicates that most strategy areas have been explored and covered. After the compression operation is initiated, the main control MCU calculates the product of the access frequency and the security evaluation value of each strategy coordinate point, using this product as the retention priority score for that strategy coordinate point. Access frequency reflects the system's familiarity with the strategy area, while the security evaluation value reflects the area's security; their product comprehensively measures the retention value of the coordinate point. The main control MCU gradually deletes strategy coordinate points in ascending order of retention priority scores until the total number drops to 1000. The statistical characteristics of the response deviation vector corresponding to the deleted strategy coordinate points are merged into the long-term decay integral evaluation value of the nearest neighboring strategy coordinate points with retention priority scores higher than 0.5, thereby migrating the cognitive information carried by the deleted coordinate points to nearby valid coordinate points, achieving lossless compression of cognitive information.

[0061] In the generation of the strategy drift path, the main control MCU also employs a prediction bias-driven path planning method instead of simple distance-first planning. When planning an intermediate path from the current strategy coordinate point to the target strategy coordinate point, for each candidate intermediate strategy coordinate point, the main control MCU predicts the possible direction of response deviation when continuing to shift towards the target direction along that intermediate strategy coordinate point, based on the direction of the historical response deviation vector of that intermediate strategy coordinate point. If the angle between the predicted response deviation direction and the direction of the current response deviation vector is less than the 45-degree direction consistency threshold, the path cost of that intermediate strategy coordinate point is set to 1. If the angle is greater than or equal to 45 degrees, the path cost of that intermediate strategy coordinate point is set to 3. The first cost value is much lower than the second cost value, allowing the path planning algorithm to prioritize safe terrain map areas consistent with the current system response characteristics, avoiding traversing unknown areas with inconsistent response characteristics. The main control MCU accumulates the path costs of all candidate intermediate strategy coordinate points and selects the sequence of intermediate strategy coordinate points with the smallest accumulated path cost as the final executed strategy drift path.

[0062] During the actual execution of the strategy drift path, the main control MCU records in real time the difference between the actual response deviation vector and the predicted response deviation vector at each intermediate strategy coordinate point. This prediction deviation difference characterizes the degree of deviation between the historical perception stored in the safety topography database and the current actual system response. The main control MCU integrates and accumulates this prediction deviation difference along the time axis. When the accumulated value exceeds a preset prediction deviation accumulation threshold of 0.3, it is determined that the safety evaluation value of the strategy coordinate point in the current safety topography database has deviated from the actual system response characteristics. This threshold is set based on the fact that the root mean square of the prediction deviation difference under normal operating conditions is approximately 0.05, and 0.3 is equivalent to 6 times the root mean square value. Exceeding this value indicates that the system state has undergone a non-negligible change. At this time, the main control MCU triggers a perception reset operation and clears the accumulated prediction deviation value to zero.

[0063] In response to the triggering of the cognitive reset operation, the main control MCU executes a tiered processing strategy. When the cumulative prediction deviation exceeds 0.3 but does not exceed 0.6, it is judged as a Level 1 cognitive imbalance, and a local cognitive reset operation is performed. The 0.6 is set based on the fact that this value is 12 times the average cumulative prediction deviation under normal operating conditions, indicating that the deviation has significantly deviated from the normal range but has not yet reached the level of global failure. The local cognitive reset operation only clears the historical response deviation vectors corresponding to the top K strategy coordinate points with the largest difference in prediction deviation from the safety topographic map database, while the historical response deviation vectors of the remaining strategy coordinate points remain unchanged. The value of K is determined by multiplying the degree of exceeding 0.3 by a preset degree coefficient of 50, with the degree of exceeding expressed as a percentage. For example, when the degree of exceeding 0.3 is 20%, the value of K is 20 multiplied by 50, which is 10. When the cumulative prediction deviation exceeds 0.6, it is judged as a Level 2 cognitive imbalance, and a global cognitive reset operation is performed. The global cognitive reset operation clears all historical response deviation vectors of all policy coordinate points in the safety terrain map database to zero and resets the current policy coordinate point to the center origin of the three-dimensional safety policy space, that is, all three policy axes are restored to the third level.

[0064] After the global cognitive reset operation is completed, the main control MCU automatically activates the accelerated exploration mode to quickly reconstruct the safe terrain map. In accelerated exploration mode, the number of strategy levels on each strategy axis in the 3D safe strategy space increases from 5 in normal exploration mode to 10, doubling the exploration granularity and enabling more precise detection of local terrain features in the strategy space. Simultaneously, the maximum single-step change limit for strategy drift is expanded from one level in normal exploration mode to two levels, accelerating the coverage of the strategy space. The duration of accelerated exploration mode is proportional to the degree to which the cumulative prediction deviation exceeds 0.6, specifically, 5 seconds of duration for every percentage point of deviation. Once the duration is reached, the system automatically switches back to normal exploration mode. It should be noted that the 10-level refined coordinates in this acceleration mode are only used as temporary exploration granularity for local strategy drift. Before the device triggers the operation of uploading safe topographic map data to the cloud, the system forcibly determines whether the current strategy axis division is the standard 5-level scheme. If it is still in the 10-level acceleration state, the upload will be automatically delayed until the acceleration mode ends and it returns to 5 levels, so as to ensure that the uploaded data is based on a unified reference coordinate system.

[0065] During the accelerated exploration mode, the main control MCU monitors the amplitude trend of newly acquired response deviation vectors in real time. When the amplitude of the response deviation vector shows a monotonically decreasing trend for three consecutive control cycles, and the final amplitude is lower than 5% of the preset steady-state amplitude threshold, it is determined that the accelerated exploration has reached convergence, the accelerated exploration mode is terminated early, and the system switches back to normal exploration mode. The 5% steady-state amplitude threshold is set based on the fact that this value corresponds to 5% of the sensor's range. Below this value, it indicates that the system state has stabilized, and further increasing the exploration density will not yield significant benefits. When switching back to normal exploration mode, the main control MCU writes all newly acquired response deviation vectors in the accelerated exploration mode into the safe topographic map database with a weighting factor twice that of the normal mode, so that the cognitive information obtained from the new exploration quickly takes over the dominant position.

[0066] When the cognitive reset operation is triggered more than 5 times within a 1-hour time window, the main control MCU determines that the system is in a state of continuous cognitive imbalance and generates a final safety lock signal. The threshold for the number of frequent resets is set based on the fact that the average trigger frequency of cognitive reset under normal operating conditions is about once every 10 hours. Five times in 1 hour is about 10 times the normal operating frequency, indicating that the system state has seriously deviated from the normal range. After the final safety lock signal is triggered, the main control MCU forcibly converges the three-dimensional safety strategy space to the extremely conservative strategy coordinate point. This extremely conservative strategy coordinate point is composed of the maximum setting of the suction maintenance strategy axis (100% of the rated excitation current), the minimum setting of the load output strategy axis (40% of the rated torque), and the minimum setting of the contact state management strategy axis (50% of the set feed rate). The system continues to execute at this extremely conservative strategy coordinate point until the operator performs a manual reset operation. At the same time, the main control MCU batch marks all strategy coordinate points that have not been marked as risk restricted area coordinate points as temporary risk restricted area coordinate points, and uniformly reduces the safety evaluation value of these temporary risk restricted area coordinate points to a temporary lock evaluation value of 0.1. The temporary lockout evaluation value is far below the safety evaluation value threshold of 0.35, which prevents the system from leaving the extremely conservative policy coordinate point through policy drift before the manual reset is completed, thereby ensuring that the device always operates in the safest state under the extreme condition of continuous cognitive imbalance.

[0067] Building upon the dual-loop learning mechanism and the hierarchical cognitive reset mechanism, this embodiment further introduces a causal failure path discrimination and differentiated protection mechanism based on the temporal evolution of coupling deviation direction angle. This mechanism enables the system not only to perceive changes in safety margin but also to identify the inherent causal direction of the change, thereby adopting targeted differentiated protection strategies.

[0068] Following the aforementioned drilling scenario, as the drill enters the hard-point region inside the workpiece, the fluctuation amplitude of the load torque gradually increases. The main control MCU continuously records the coupling deviation direction angle sequence of the current strategy coordinate point and its neighboring strategy coordinate points over multiple control cycles. This coupling deviation direction angle uses the adsorption force response deviation component as the reference axis, taking the arctangent ratio of the load torque response deviation component to the adsorption force response deviation component, with a value ranging from -90 degrees to +90 degrees. A positive angle indicates that the load torque response deviation direction leads the adsorption force response deviation direction, while a negative angle indicates lag. The main control MCU performs a first-order difference operation on this sequence to obtain the coupling deviation direction angle change rate sequence. This change rate sequence reflects the changing trend and rate of change of the dynamic phase relationship between the adsorption force and the load torque.

[0069] The main control MCU analyzes the sign and trend of the rate of change sequence to determine the type of failure path. When the rate of change sequence of the coupling deviation direction angle shows a positive increasing trend for five consecutive control cycles, and the change of the load torque response deviation component leads the adsorption force response deviation component on the time axis, the main control MCU determines that the current system is in an overload-dominated failure path. The determination of whether the load torque response deviation component leads the adsorption force response deviation component is based on the time delay estimation method of the cross-correlation function: the cross-correlation coefficient between the load torque response deviation component sequence and the adsorption force response deviation component sequence is calculated. When the cross-correlation coefficient reaches its maximum value in the positive time offset direction, it is determined that the change of the load torque response deviation component leads the adsorption force response deviation component. In the overload-dominated failure path, the load impact is the cause, and the adsorption force decay is the result. The physical process is that the drill bit encounters a hard point, causing a sudden increase in load. The impact and vibration are transmitted through the mechanical structure to the contact surface between the electromagnetic chuck and the workpiece, gradually destroying the micro-contact state of the contact surface, thus leading to a decrease in adsorption force.

[0070] Conversely, when the rate of change of the coupling deviation direction angle shows a negative growth trend for five consecutive control cycles, and the change of the adsorption force response deviation component leads the change of the load torque response deviation component, the main control MCU determines that the current system is in a demagnetization-dominated failure path. In the demagnetization-dominated failure path, adsorption force attenuation is the cause, and load impact is the result. The physical process is that the actual effective adsorption force of the workpiece contact surface gradually decreases due to vibration or poor adhesion. When the adsorption margin is insufficient, the overturning torque generated by the drilling reaction force applies a load to the adsorption surface that exceeds its bearing capacity, which manifests as a passive increase in load torque.

[0071] Based on the determined failure path type, the main control MCU executes a differentiated priority drift strategy. If the failure path is determined to be overload-dominated, the main control MCU prioritizes strategy drift on the load output strategy axis, performing first-priority drift along the direction of reducing the maximum allowable output torque. Specifically, the maximum allowable output torque is gradually reduced from the current level, one level per control cycle, until the torque limit value drops to 40% of the rated torque. Simultaneously, strategy drifts on the suction maintenance strategy axis and the contact state management strategy axis are executed in parallel as auxiliary drifts. The step size of the auxiliary drift does not exceed 50% of the first-priority drift step size. The purpose of this differentiated strategy is to prioritize cutting off the source of load impact, blocking the causal transmission path from overload to demagnetization, and preventing continuous load impact from further deteriorating the adsorption state of the contact surface.

[0072] If the failure path is determined to be dominated by demagnetization, the main control MCU prioritizes strategy drift on the attraction-maintaining strategy axis, performing first-priority drift along the direction of increasing excitation current. Specifically, the excitation current is gradually increased from its current value, one level at a time per control cycle, until the excitation current reaches 100% of its rated value. Simultaneously, the main control MCU performs preparatory load reduction drift on the load output strategy axis, gradually reducing the maximum allowable output torque from its current level. The step size of the preparatory load reduction drift is set to 50% of the first-priority drift step size, i.e., one level down every two control cycles. The purpose of this differentiated strategy is to reserve a safety margin while supplementing magnetization, preventing sudden overloads before magnetization is fully effective, and achieving coordinated magnetization and load reduction.

[0073] During strategy drift execution, the master control MCU continuously monitors the sign change of the coupling deviation direction angle change rate sequence. When the sign of this change rate sequence is opposite to the sign before drift initiation for three consecutive control cycles, the master control MCU determines that the failure path type has reversed. A reversal determination is triggered when the sign change rate exceeds a preset path reversal threshold, which is set to flip the sign of the change rate once per control cycle. At this point, the master control MCU immediately stops the current strategy drift direction, reverses the drift direction, and performs a backdrift, causing the strategy coordinate point to return to the starting strategy coordinate point along the original path. After the backdrift is completed, the master control MCU marks the starting strategy coordinate point as the failure path reversal boundary point. The master control MCU establishes a repulsion field with this reversal boundary point as the origin. The mathematical form of this repulsion field is a potential function, the value of which is inversely proportional to the normalized Euclidean distance from the strategy coordinate point to the reversal boundary point. In subsequent policy drift path planning, the path cost of policy coordinates passing through the neighborhood of the reversal boundary needs to be increased. The increase is inversely proportional to the distance from the policy coordinate to the reversal boundary; the closer the distance, the greater the increase. This repulsion field mechanism enables the policy drift path to automatically avoid sensitive areas where failed path type reversals have occurred in subsequent planning, preventing the system from repeatedly entering these areas and causing control oscillations.

[0074] Building upon the aforementioned single-machine strategy space exploration and dual-loop learning mechanism, this embodiment further introduces a multi-machine collaborative group cognition aggregation and general security prior knowledge transfer mechanism. This mechanism aggregates the security terrain map knowledge accumulated by a single device during the exploration process through the cloud, enabling cross-device sharing and allowing newly connected devices to obtain initial security cognition validated by the group from the startup phase.

[0075] When two or more magnetic drilling rigs connect to the same cloud-based cognitive aggregation server via a wireless network, each rig uploads the coordinates, safety assessment values, risk exclusion zone coordinates, and coupling deviation direction angle statistics of its local safety topographic map database to the cloud-based cognitive aggregation server using an encrypted communication protocol. The upload operation is automatically triggered after each rig completes a full drilling operation. The uploaded data volume is approximately 50 bytes per strategic coordinate point, totaling approximately 6.25 kilobytes for 125 strategic coordinate points. Transmission time in a wireless network environment is less than 1 second.

[0076] After receiving security terrain map data uploaded by various devices, the cloud-based cognitive aggregation server performs cross-device coordinate alignment processing. Under normal operating conditions, all devices use the same policy axis discretization scheme, meaning all three policy axes are uniformly divided into 5 levels. Therefore, the coordinate values ​​of policy coordinate points are naturally aligned, requiring no scaling transformation. As a data fault tolerance mechanism under extreme conditions, if the data received by the cloud aggregation server contains non-level 5 refined coordinates (such as level 10), the server automatically performs linear downsampling mapping (rounding down the level value to the corresponding level 5 interval), converting the non-standard data to the standard level 5 coordinate space before subsequent weighted aggregation calculations. For policy coordinate points with the same coordinate values, the cloud-based cognitive aggregation server weights and aggregates their corresponding security evaluation values ​​to generate a group aggregated security evaluation value. The weight coefficient of the weighted aggregation is positively correlated with the total number of local policy drifts uploaded by the device. Devices with more total policy drifts have higher reliability in their accumulated security evaluation values ​​and occupy a larger weight in the aggregation. The specific weighted aggregation calculation formula is that the group evaluation value is equal to the sum of the products of the safety evaluation value of each device and the total number of policy drifts of that device, divided by the sum of the total number of policy drifts of each device.

[0077] The cloud-based cognitive aggregation server identifies policy coordinates among all aligned policy coordinates that have a group aggregation security evaluation value higher than a preset group excellence threshold and are not marked as risk-free zones by any device. The group excellence threshold is set to 0.7. The cloud-based cognitive aggregation server marks these eligible policy coordinates as general security prior policy coordinates and encapsulates all general security prior policy coordinates and their corresponding group aggregation security evaluation values ​​into a general security prior knowledge package. This knowledge package is a structured data file containing four fields: version number, policy coordinate point list, corresponding group aggregation security evaluation value list, and operational condition label.

[0078] When a new magnetic drilling rig is first started and connected to the same cloud-based cognitive aggregation server, the rig downloads a general safety prior knowledge package from the server and loads the general safety prior policy coordinates and their collective aggregated safety evaluation values ​​from the knowledge package as initial data for its local safety topographic map database. This loading operation enables the new rig to possess collectively validated safety cognitive capabilities from the startup phase, obtaining an effective initial safety topographic map without starting from scratch in policy space exploration. The policy exploration convergence time is reduced from approximately 50 drilling operations to approximately 5 drilling operations.

[0079] After the new device completes local policy drift and accumulates the local response deviation vector, the main control MCU compares the local security evaluation value corresponding to the coordinate point of the general security prior policy in the local security terrain map database with the group aggregated security evaluation value. If the absolute value of the deviation between the local security evaluation value and the group aggregated security evaluation value is less than the preset group individual deviation tolerance threshold of 0.15, it is determined that the policy combination corresponding to the policy coordinate point reproduces the security characteristics verified by the group on the local device. The main control MCU increases the weight coefficient of the policy coordinate point to 1.2 times in subsequent local security evaluation value updates to enhance the stability of the policy. If the absolute value of the deviation is greater than or equal to 0.15, it is determined that the policy combination corresponding to the policy coordinate point is affected by the physical constraints or operating conditions specific to the local device. For example, the aging degree of the suction cup coil of the device is different or there are differences in the roughness of the workpiece contact surface. The main control MCU marks the policy coordinate point as a device-specific policy coordinate point and reduces the weight coefficient of the coordinate point to 0.8 times in subsequent local security evaluation value updates. When uploading the next safety terrain map data to the cloud-based cognitive aggregation server, the device also uploads the marker information of its device-specific strategy coordinates and the corresponding local operating condition parameters. These operating condition parameters include four fields: current operating head parameter, workpiece material type, borehole diameter, and current load rate. Upon receiving this data, the cloud-based cognitive aggregation server, during subsequent updates to the general safety prior knowledge package, divides the knowledge package into operating condition subclasses based on these parameters, generating multiple operating condition-specific knowledge sub-packages corresponding to different subclasses. Newly connected devices can selectively download the corresponding sub-packages based on the matching degree of their current operating conditions when downloading the knowledge package, further improving the targeting and effectiveness of knowledge transfer.

[0080] The cloud-based cognitive aggregation server adaptively adjusts the update cycle of the general security prior knowledge package based on the update frequency of the security terrain map data uploaded by each device. When the amount of data uploaded per unit time exceeds the group activity threshold of 100 uploads per hour, the server shortens the update cycle from the default 24 hours to 6 hours and increases the weight coefficient of recently uploaded data to 0.6 in weighted aggregation. When the amount of data uploaded is less than 10 uploads per hour, the server extends the update cycle to 72 hours and decreases the weight coefficient of recently uploaded data to 0.2. This adaptive adjustment mechanism ensures that the evolution speed of the general security prior knowledge package dynamically matches the update speed of group cognition. When group cognition is active, the knowledge package iterates rapidly; when group cognition is sparse, the knowledge package remains stable, avoiding knowledge fluctuations caused by excessive updates.

[0081] After describing the foregoing method embodiments, this section further provides system embodiments for implementing the above control method.

[0082] The multi-level linkage protection-based demagnetization overload collaborative control system for magnetic drills provided in this embodiment is applied to a magnetic drill rig. The magnetic drill's electromagnetic chuck has a rated suction force of 30KN, the spindle motor is a three-phase brushless DC motor with a rated power of 2200 watts, and the feed mechanism is driven by a stepper motor. The entire control system uses a main control MCU as its computing core. The main control MCU is a microcontroller based on the ARM Cortex-M7 core, operating at a clock frequency of 400 MHz, and has a built-in floating-point unit and digital signal processing instruction set, capable of meeting the real-time calculation requirements of complex algorithms such as strategy space exploration, security topographic map database retrieval, and dual-loop learning. The main control MCU integrates multiple analog-to-digital conversion channels, pulse width modulation output channels, general-purpose input / output interfaces, and a controller area network bus controller. An embedded real-time operating system runs on the main control MCU, scheduling all task modules with a control cycle of ten milliseconds, including sensor data acquisition tasks, strategy space exploration tasks, security evaluation value update tasks, strategy drift planning tasks, and communication management tasks.

[0083] The control system is divided into the physical execution layer, sensor acquisition layer, three-dimensional safety strategy space management unit, safety terrain map database, strategy drift controller, dual-loop learning unit, cognitive reset unit, causal failure path discrimination unit, extreme safety locking unit, communication interface unit and central processing unit according to function.

[0084] The physical execution layer comprises an electromagnetic chuck drive unit, a brushless spindle drive unit, and a feed drive unit. The electromagnetic chuck drive unit receives pulse-width modulation (PWM) signals from the main control MCU and adjusts the excitation current in the electromagnetic chuck coil through a power amplifier circuit. This unit includes a current feedback loop, sending the actual coil current value back to the main control MCU via an analog-to-digital converter (ADC) channel to achieve closed-loop regulation of the excitation current. The brushless spindle drive unit receives six PWM signals from the main control MCU, driving a three-phase full-bridge inverter circuit to control the spindle motor's speed and output torque. This unit includes a three-phase current sampling circuit, amplifying and filtering each phase's current signal before sending it to the MCU's ADC channel for current loop calculation in the field-oriented control algorithm. The feed drive unit receives stepper motor pulse signals and direction signals from the main control MCU and controls the feed mechanism's speed and direction through a stepper motor driver.

[0085] The sensor acquisition layer includes a magnetic attraction detection unit, a motor load detection unit, and a feed thrust detection unit. The magnetic attraction detection unit uses a magnetic Hall sensor mounted on the surface of the magnetic pole of the electromagnetic chuck, maintaining a 0.5 mm gap from the workpiece contact surface. It directly senses the magnetic flux density between the magnetic pole and the workpiece contact surface. After signal conditioning and filtering, the signal is sent to the analog-to-digital conversion channel of the main control MCU. The main control MCU internally stores a calibration curve between magnetic flux density and effective attraction force, converting the analog-to-digital conversion value into actual attraction force parameters. The motor load detection unit includes a current sampling circuit and a Hall speed acquisition circuit. The current sampling circuit acquires three-phase current signals through a shunt resistor connected in series in the motor power supply line. The shunt resistor has a resistance of one milliohm. The signal is differentially amplified and sent to the analog-to-digital conversion channel. The Hall speed acquisition circuit receives output signals from three Hall sensors built into the spindle motor and calculates the current speed by measuring the time interval between adjacent Hall pulses. Based on the acquired current and speed values, combined with motor parameters and drive efficiency curves, the main control MCU calculates and outputs the current torque value in real time. The feed thrust detection unit uses a strain gauge thrust sensor installed between the feed mechanism and the drill bit. The thrust signal is processed by a bridge amplifier circuit and then sent to the analog-to-digital conversion channel.

[0086] The main control MCU internally operates a three-dimensional safety policy space management unit. This unit maintains a 5x5x5 three-dimensional array structure in memory, where each array element corresponds to a policy space coordinate point, storing the coordinate value, current safety evaluation value, access count, and most recent access timestamp of that coordinate point. Within each control cycle, the three-dimensional safety policy space management unit generates tentative policy coordinate points along the three policy axes, centered on the current policy space coordinate point. These tentative policy coordinate points are then mapped to corresponding tentative excitation current adjustment, tentative torque limit, and tentative feed speed adjustment, and sent to the physical execution layer via pulse width modulation and pulse output channels.

[0087] The safety terrain map database is stored in the off-chip serial flash memory of the main control MCU, with a capacity of 2 megabytes, capable of storing data for up to 65,536 policy coordinate points. The database uses a hash index built with the three-level numbers of the policy coordinate points as combined keys. Each data entry includes the safety evaluation value for that coordinate point, the mean vector of the response deviation vector, the covariance matrix of the response deviation vector, the access count, the most recent access timestamp, and a risk exclusion zone marker. The main control MCU accesses the database via the serial peripheral interface bus, with the time to read or write a single data entry not exceeding one hundred microseconds.

[0088] The strategy drift controller operates as an independent task within the main control MCU. When the safety evaluation value of the current strategy coordinate point is lower than a preset threshold of 0.35, the strategy drift controller retrieves the target strategy coordinate point from the safety terrain map database, plans the strategy drift path, and gradually adjusts the position of the strategy coordinate point according to the path. After each adjustment, the strategy drift controller triggers the sensor acquisition layer to re-acquire the response deviation vector and update the safety evaluation value in the safety terrain map database. During the strategy drift process, if the magnitude of the response deviation vector exceeds 20% of the deviation magnitude alarm threshold, the strategy drift controller immediately terminates the drift process, reverts to the previous safe coordinate point, and marks the coordinate point that triggered the alarm as a risk-free zone coordinate point.

[0089] The dual-loop learning unit is implemented in the main control MCU as a collection of multiple calculation modules. The short-term evaluation value calculation module maintains the most recent 100 access records for each strategy coordinate point, calculates the sliding window mean of the response deviation vector magnitude, and takes its reciprocal. The long-term evaluation value calculation module maintains the exponentially weighted cumulative value of all historical access records for each strategy coordinate point, with the decay time constant set to 1000 times the system operating cycle. The coupling deviation direction angle calculation module extracts the adsorption force response deviation component and the load torque response deviation component from the response deviation vector, calculates their arctangent ratio, and accumulates the historical direction angle sequence of the same coordinate point for variance calculation.

[0090] The cognitive reset unit monitors the number of risk restricted area coordinates and the cumulative prediction deviation in the main control MCU. When the number of risk restricted area coordinates exceeds 12, or the cumulative prediction deviation exceeds 0.3, the cognitive reset unit triggers a cognitive reset operation, clearing all or part of the historical response deviation vectors in the safety terrain map database to zero, and resetting the current policy coordinates to the center origin position of the three-dimensional safety policy space.

[0091] The causal failure path discrimination unit continuously acquires the coupling deviation direction angle sequence in the main control MCU and performs first-order difference calculation to obtain the rate of change sequence. When the rate of change sequence shows a positive increasing trend for five consecutive control cycles and the change of the load torque response deviation component leads the change of the adsorption force response deviation component, it is judged as an overload-dominated failure path, and a command is sent to the strategy drift controller to prioritize drifting along the load output strategy axis. When the rate of change sequence shows a negative increasing trend for five consecutive control cycles and the change of the adsorption force response deviation component leads the change of the load torque response deviation component, it is judged as a demagnetization-dominated failure path, and a command is sent to the strategy drift controller to prioritize drifting along the adsorption force maintenance strategy axis.

[0092] The extreme security lock-up unit monitors the trigger frequency of cognitive reset operations in the main control MCU. When the cognitive reset operation is triggered more than 5 times within 1 hour, the extreme security lock-up unit generates an ultimate security lock-up signal, forcibly converging the three-dimensional security policy space to the extremely conservative policy coordinate points, and batching all policy coordinate points that were not marked as risk restricted area coordinate points as temporary risk restricted area coordinate points, and uniformly reducing their security evaluation value to 0.1.

[0093] The communication interface unit includes a wireless module supporting dual-mode Wi-Fi and Ethernet communication, which connects to the main control MCU via a universal asynchronous transceiver interface. This unit communicates with the cloud-based cognitive aggregation server using the MQTT protocol, uploading local security topographic map data and receiving general security prior knowledge packages. The upload operation is automatically triggered after each complete drilling operation, with approximately 50 bytes of data uploaded per policy coordinate point.

[0094] The central processing unit (CPU), also known as the main control MCU, connects to all the aforementioned units via an internal bus and is responsible for global data processing and logic control. Data exchange between units is accomplished through a shared data area in the CPU's internal random access memory. Timing coordination between units is uniformly managed by the task scheduler of the embedded real-time operating system, ensuring that tasks such as sensor data acquisition, policy space exploration, security evaluation value updates, and policy drift planning are executed strictly synchronously with a control cycle of ten milliseconds. The hardware architecture of the entire system realizes a complete closed-loop control link from sensor signal acquisition to actuator driving. All core algorithms are calculated internally by the main control MCU, without relying on an external host computer, ensuring the independence and real-time response capability of the control system.

[0095] The safety control process described above during drilling is continuously executed in a 10-millisecond cycle. Within each cycle, the sensor acquisition layer completes data sampling, the 3D safety strategy space management unit completes strategy exploration and safety evaluation value updates, the strategy drift controller completes safety topographic map retrieval and path planning, the dual-loop learning unit completes the fusion update of short-term and long-term evaluation values, and the causal failure path discrimination unit completes real-time determination of failure path types. Each functional module executes sequentially according to the task scheduling sequence of the embedded real-time operating system, completing all calculations and outputting execution instructions within one control cycle to ensure the real-time nature and determinism of control decisions. Once drilling is complete, the operator shuts down the equipment power, and the main control MCU writes all data from the current safety topographic map database to non-volatile storage, serving as the initial cognitive basis for the next startup. Simultaneously, the communication interface unit automatically triggers a data synchronization operation with the cloud-based cognitive aggregation server, uploading the newly accumulated safety topographic map data from this drilling process to the cloud for use in group cognitive aggregation. Thus, the system completes a full closed loop from real-time safety control in a single drilling operation to cross-device knowledge accumulation and transfer.

Claims

1. A multi-level linkage protection-based magnetic seat drill loss-of-magnetic field overload cooperative control method, characterized in that, The method includes: S1: During equipment operation, the actual adsorption force parameters of the electromagnetic chuck are collected in real time, and the load parameters of the spindle motor are also collected in real time. S2: Dynamically calculate the current maximum allowable output torque threshold based on the actual adsorption force parameters; S3: Compare the load parameters collected in real time with the maximum output torque threshold, and execute the linkage protection action according to the comparison result.

2. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 1, characterized in that: The actual adsorption force parameters are obtained by detecting the magnetic flux of the contact surface between the electromagnetic chuck and the workpiece using a magnetic Hall sensor.

3. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 1, characterized in that: The coordinated protection action includes the following four-level hierarchical control: Level 1: When the actual adsorption force parameter reaches the first preset threshold and the load parameter does not exceed the maximum output torque threshold, the current operating state is maintained; Level 2: When the actual adsorption force parameter is lower than the first preset threshold and the load parameter does not exceed the maximum output torque threshold, the maximum output torque threshold is lowered and a warning signal is issued. Level 3: When the load parameter exceeds the maximum output torque threshold and the actual adsorption force parameter reaches the first preset threshold, a flexible buffer load reduction is performed. Level 4: When the actual adsorption force parameter is lower than the second preset threshold and the load parameter exceeds the maximum output torque threshold, the main shaft power is cut off to maintain the adsorption force of the electromagnetic chuck and lock the moving parts of the whole machine. Wherein, the second preset threshold is lower than the first preset threshold.

4. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 3, characterized in that: The flexible buffer load reduction is achieved by limiting the peak motor current, gradually reducing the PWM duty cycle, and pausing the feed motor, at least one of the following methods:

5. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 1, characterized in that: It also includes a power-on self-test step. After the device is powered on, it detects the actual adsorption force parameter. Only when the actual adsorption force parameter reaches the preset safety threshold is the spindle motor unlocked to allow start-up. When the actual adsorption force parameter does not reach the preset safety threshold, the spindle motor is locked and an alarm is issued.

6. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 3, characterized in that: In the second-level control, an active magnetization operation to enhance the excitation current of the electromagnetic chuck is also performed; after the fourth-level control is triggered, the system saves the fault record, and can only be unlocked and restarted after a manual physical reset operation is performed.

7. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 1, characterized in that: In step S3, the execution of the linkage protection action based on the comparison result further includes a control process driven by active exploration of the strategy space and a security terrain map: A three-dimensional safety strategy space is constructed, consisting of three strategy axes: suction maintenance, load output, and contact status management. Each strategy axis has multiple gears, which are combined to form strategy coordinate points. Within each control cycle, a trial offset is applied near the current coordinate point to generate trial coordinate points and convert them into corresponding control parameter adjustment amounts; The system response is acquired under trial control parameters, and the response deviation vector of each trial point relative to the current point is calculated. The response deviation vector is stored in the safety topographic map database, and the safety evaluation value of each coordinate point is updated according to the deviation magnitude and duration. When the current point's security evaluation value is lower than the threshold, search the database for nearby points with high security evaluation values ​​as target points, plan a strategy drift path that passes through intermediate points, and gradually adjust the coordinate points according to the path. After each step, re-collect the response and update the evaluation value. If the deviation exceeds the limit at any point during the drift, the drift will be terminated and the drift will return to the previous safe point. This point will be marked as a risk-free zone and eliminated. When the number of risky restricted areas exceeds the limit, a cognitive reset is triggered, the database is cleared, the current point is reset to the origin, and exploration restarts.

8. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 7, characterized in that: The update and policy drift process of the secure topographic map database also includes a two-loop learning mechanism driven by prediction bias: The safety evaluation value of each coordinate point is obtained by weighted fusion of short-term and long-term historical response statistics; After each control cycle, the coupling direction angle is calculated based on the phase relationship between the response deviation components, and rewards or penalties are applied to the evaluation value based on the stability of the direction angle. When the database capacity is saturated, the retention priority is determined according to the access frequency and evaluation value of each point. Low priority points are deleted and their historical information is merged into the neighboring retention points. When planning the strategy drift path, the deviation direction of continuing forward is predicted based on the historical response direction of the candidate intermediate point. If the predicted direction is consistent with the current actual direction, the path cost of that point is reduced; otherwise, the cost is increased, and the path with the minimum total cost is selected. During path execution, the actual deviation is compared with the predicted deviation in real time. If the accumulated deviation exceeds the limit, a cognitive reset is triggered.

9. The multi-stage linkage protection based magnetic seat drill loss-of-magnetic field overload cooperative control method according to claim 8, characterized in that: The cognitive reset operation also includes cognitive imbalance grading and predictive anchoring mechanisms: Cognitive reset is divided into two levels based on the degree of cumulative prediction bias: a partial reset is performed when there is a slight imbalance, which only clears the historical data of the coordinate points with the largest bias in the database; a global reset is performed when there is a severe imbalance, which clears all historical data and resets the current point to the origin. After a global reset, an accelerated exploration mode is activated, increasing the number of strategy levels and the maximum change per step to speed up spatial coverage. The acceleration duration is related to the degree of imbalance, and once the target is reached, the normal mode is restored. If the response deviation continues to decrease and tends to stabilize during the acceleration period, the acceleration can be terminated early and the new data can be written to the database with a higher weight. If a reset is triggered frequently within a specified time window, it is determined to be a continuous imbalance and an ultimate lock signal is generated: the system is forced to converge to an extremely conservative coordinate point composed of the maximum suction, minimum load and minimum feed, and the point is locked until it is manually reset. At the same time, all points not marked as risk-prone areas are temporarily downgraded to low safety ratings, so that the system cannot drift away from the conservative point.

10. A multi-level linkage protection-based magnetic seat drill loss-of-magnetic field overload cooperative control system, characterized in that, include: The physical execution layer includes an electromagnetic chuck drive unit, a brushless spindle drive unit, and a feed drive unit, which are used to adjust the excitation current, limit the spindle output torque, and adjust the feed speed, respectively. The sensor acquisition layer includes a magnetic attraction detection unit, a motor load detection unit, and a feed thrust detection unit, which are used to acquire the attraction force, load torque, and feed thrust, respectively. The three-dimensional safety strategy space management unit is used to construct a strategy space consisting of three strategy axes: suction maintenance, load output, and contact state management. Each strategy axis has multiple gears, which are combined to form strategy coordinate points. In each control cycle, a tentative offset is applied to generate tentative control parameters and send them to the physical execution layer. A safety topographic map database is used to store the statistical characteristics of historical response deviation vectors and their corresponding safety evaluation values, indexed by policy coordinate points. The strategy drift controller is used to search for nearby high safety evaluation value points in the database as target points when the safety evaluation value of the current coordinate point is lower than the threshold. It plans a drift path that passes through the intermediate points and adjusts the coordinate points step by step according to the path. If the deviation of a point exceeds the limit during the drift process, the drift is terminated, the retreat is performed, and a risk-prone area is marked. The dual-loop learning unit is used to update the security evaluation value based on the statistical weighting of short-term and long-term historical responses, and to correct the evaluation value according to the phase relationship between the response deviation components. The cognitive reset unit is used to trigger a cognitive reset when the number of risk-prone areas exceeds the limit or the cumulative prediction deviation exceeds the limit, clearing historical data in the database and resetting the current point to the origin. The causal failure path discrimination unit is used to determine the current failure path type based on the temporal evolution trend of the response deviation component, and send differentiated drift instructions to the strategy drift controller accordingly. An extreme security locking unit is used to lock the system to an extremely conservative policy coordinate point when frequent resets are triggered, until manual reset is performed; The communication interface unit is used to communicate with the cloud server, upload local data, and receive general security prior knowledge as initial data for the database; and, The central processing unit (CPU) is used to perform data processing and logic control.