Dynamic gripping control method and system in strong electromagnetic environment

Abstract

The application discloses a dynamic grabbing control method and system in a strong electromagnetic environment, a visual-inertial tightly coupled fusion positioning calculation device is used to calculate the pose of the device relative to the conductor, and an optimized target grabbing point is generated based on a recursive prediction model to predictively estimate the movement trend of the conductor; on this basis, a two-stage progressive approximation and compliant impedance centering control of "pre-approximation-fine adjustment" is performed to realize smooth alignment; in the grabbing stage, PID feedback and wind load / gravity feedforward are used to realize accurate closed loop of contact pressure, and hierarchical pressure is used to ensure that the conductor is not damaged; then, three types of heterogeneous information, i.e., pressure, attitude and vision, are fused to determine the grabbing state; after the grabbing is successful, active holding pressure and deflection compensation are dynamically applied according to the wind speed and attitude to ensure that the device can stably complete detection in a strong electromagnetic and strong wind environment. The scheme has high success rate, strong anti-interference, no damage to the conductor, and can rely on cloud-edge collaboration to realize parameter self-adaptive evolution, and has significant engineering value.

Description

Technical Field

[0001] This invention belongs to the field of live-line testing equipment and robot control technology for power transmission lines, specifically relating to a control method and system that enables a testing device to stably, non-destructively, and with high precision dynamically attach to a conductor in a strong electromagnetic environment. Background Technology

[0002] With the advancement of smart grid construction, transmission line crimping clamps, as core connecting components bearing current and mechanical stress, directly determine power supply reliability in their operational status. Internal defects in crimping clamps (such as incomplete crimping or broken strands) are often hidden faults, requiring diagnosis using non-destructive testing methods such as X-rays. However, traditional manual inspection methods during power outages suffer from significant losses, high-risk high-altitude operations, and low efficiency. In recent years, the technology of using drones to mount X-ray inspection devices for live-line work has gained increasing attention. However, the "dynamic attachment" process—the smooth transition of the device from the drone to the conductor and its reliable fixation—remains a key technological bottleneck restricting the automation and safety of the operation.

[0003] Existing dynamic attachment technology has the following obvious shortcomings:

[0004] Rigid gripping mechanisms damage conductors: Existing solutions mostly use mechanical claws or rigid clamps. The claw head material is too hard, which can easily cause scratches and indentations on the conductor surface during gripping (the damage depth can reach more than 0.1mm). Long-term operation can induce partial discharge and fatigue damage to the conductor, threatening line safety.

[0005] The sensing methods are limited and the gripping state estimation algorithm has poor robustness: Most solutions rely on a single sensor (such as vision or contact switch), which makes it difficult to make accurate and real-time judgments on key states such as "whether the gripping is stable, whether the gripping force is sufficient, and whether the posture is centered" under complex working conditions such as strong electromagnetic interference, wind-induced swaying, and changing lighting.

[0006] The docking process is undisturbed by the arm, but the success rate and stability are insufficient: the drone hovers with a jitter of ±5cm or even greater, and the existing control strategies are mostly open-loop or simple PID feedback, lacking dynamic prediction of the trend of the conductor movement and multi-level approximation strategies, resulting in low grabbing success rate, large impact during the transition process, and poor final positioning accuracy of the device on the conductor.

[0007] Lack of active holding mechanism after attachment: After attachment is completed, the device is still subjected to wind load and its own weight. The existing solution only relies on static friction to hold it. When the wind speed exceeds a certain threshold or is subjected to electromagnetic disturbance, it is very easy to slip or even fall off, causing equipment to fall.

[0008] Data and experience are not reused in a closed loop: the existing solution is an "isolated operation" for each attachment, lacking the accumulation and learning of historical data such as attachment force, attitude, and working conditions, and cannot provide intelligent decision support for the optimal attachment parameters under different wire diameters and working conditions.

[0009] In summary, there is an urgent need for a control method and system that can achieve stable, non-destructive, and high-precision dynamic attachment of the device to a conductor in a strong electromagnetic environment, in order to overcome the shortcomings of the existing technology. Summary of the Invention

[0010] The purpose of this section is to outline some aspects of embodiments of the present invention and to briefly describe some preferred embodiments. Simplifications or omissions may be made in this section, as well as in the abstract and title of this application, to avoid obscuring the purpose of these documents; however, such simplifications or omissions should not be construed as limiting the scope of the invention.

[0011] The technical problem to be solved by the present invention is to overcome the shortcomings of the prior art, such as damage to the attached wire, poor robustness of the attached state perception, large impact during docking, lack of active holding mechanism after attachment, and inability to adaptively adjust parameters according to working conditions, and to provide a dynamic attachment control method and system in a strong electromagnetic environment.

[0012] To solve the above technical problems, the present invention provides the following technical solution: a dynamic gripping control method under strong electromagnetic environment, comprising the following steps: before gripping, acquiring relative pose information between the detection device and the target conductor, motion trend information of the target conductor under environmental load, and suspension attitude information of the detection device; processing the relative pose information through multi-source sensing fusion, and performing trend prediction and environmental load correction processing on the motion trend information; weighting and fusing the current relative pose estimation and the corrected trend prediction results to generate the target gripping point and approximation trajectory; driving the detection device to perform graded progressive approximation according to the target gripping point: first, pre-approximation converging to the first region around the target gripping point, and then iterative correction based on the real-time updated relative pose in each control cycle until convergence to the second region around the target gripping point with higher accuracy than the first region; and during the approximation process, performing compliant alignment of the attitude of the detection device through impedance adjustment, so that the gripping reference plane of the detection device and the axis of the target conductor maintain a set alignment relationship; when the detection device enters the second region... After the gripping mechanism contacts the target conductor, the contact pressure applied by the gripping mechanism is adjusted in a closed loop based on the real-time gripping force collected as feedback. Feedforward compensation is performed for environmental load and gravity components, and an upper limit constraint is set for the single pressure increment. Pressure is only increased when the actual contact pressure change at adjacent times meets the set safety conditions, allowing the gripping mechanism to perform graded, non-destructive, and compliant pressing of the target conductor. When the contact pressure reaches the set safety conditions, the gripping force information, device attitude information, and visual fit information are normalized and weighted to obtain a gripping fit evaluation quantity. If the evaluation quantity continuously meets the stable gripping criterion within a set time, the gripping is considered successful; otherwise, the gripping parameters are adjusted until the criterion is met or an abandonment decision is triggered. After successful gripping, based on environmental load information and device attitude information, an active adjustable holding pressure related to external disturbances is applied to the detection device, and incremental compensation is performed for the deflection trend of the detection device caused by external disturbances. This allows the detection device to maintain gripping under dynamic environmental loads until the detection operation is completed and a safe release is performed.

[0013] Specifically: Multi-source sensing fusion involves tightly coupling and filtering the high-frequency output of inertial sensing with the pose observation of visual sensing. When any sensing channel experiences measurement degradation in a strong electromagnetic environment, the availability of relative pose information is maintained through another sensing channel.

[0014] Specifically, trend prediction involves weighted extrapolation of the relative pose sequence of the target traverse over a historical period, and correction of the weighted extrapolation results based on the deviation between the environmental load and the set benchmark.

[0015] Specifically: impedance adjustment dynamically shapes the attitude of the detection device around the target conductor's directional axis into a set inertia-damping-stiffness characteristic, so that the detection device gradually adapts and smoothly returns to normal under the action of external disturbance torque.

[0016] Specifically: the closed-loop regulation of the contact pressure is a proportional-integral-derivative feedback regulation; the feedforward compensation includes the projection compensation of gravity along the gripping direction under the current attitude of the detection device, as well as the compensation for real-time environmental loads.

[0017] Specifically: the normalized weighted summation is to normalize the gripping force information, device attitude information and visual fit information to a unified dimension, and then sum them according to the set weight coefficients; where the weight coefficients are adjusted according to the credibility of each piece of information under the current working conditions.

[0018] Specifically: the active adjustable holding pressure is dynamically adjusted based on the deviation between the real-time environmental load and the set threshold, so that the holding pressure increases with the increase of external disturbance and decreases with the decrease of external disturbance, so as to suppress slippage while avoiding long-term overpressure; the incremental compensation is generated based on the attitude deflection amount accumulated by the detection device within the set observation period.

[0019] Specifically, it also includes feasibility assessment of the grabbing process: comprehensively analyzing multi-source sensing data, trend prediction results, approximation trajectory and control parameters during the grabbing process. When the assessment shows that the feasibility of grabbing under the current working conditions is lower than the set threshold, a decision to abandon the grabbing is generated and the detection device is controlled to exit the grabbing process and return to the standby state.

[0020] Specifically: During the gripping and holding process, the deviation between the actual value and the set value of the holding parameters of the detection device is monitored in real time, and correction is made when the deviation exceeds the allowable range; after the detection operation is completed, the contact pressure is gradually unloaded, the attitude centering constraint is released, and the detection device is released from the target guide in the reverse order of gripping and pressing.

[0021] To solve the above-mentioned technical problems, the present invention also provides the following technical solution: a dynamic gripping control system under strong electromagnetic environment, comprising: an environmental sensing unit, used to acquire relative pose information between the detection device and the target conductor, motion trend information of the target conductor under environmental load, and suspension attitude information of the detection device, and output the fused relative pose estimate and the target gripping point; a docking control unit, used to drive the detection device to perform graded progressive approximation based on the target gripping point, and to perform compliant centering of the device attitude through impedance adjustment; and a compliant gripping unit, used to perform closed-loop adjustment and feedforward compensation of the contact pressure with real-time gripping force as feedback, and to perform graded pressure adjustment and feedforward compensation of the contact pressure. The system constrains the single pressure increment and pressure change conditions during the gripping process; the state determination unit normalizes and weights the gripping force information, device attitude information, and visual fit information to synthesize a gripping fit evaluation quantity, and determines whether the device has reached a stable gripping state based on the time window criterion; the active holding unit applies an active adjustable holding pressure and deflection increment compensation to the gripped detection device based on environmental load and device attitude information to maintain gripping until the detection operation is completed and the device is safely released; the edge computing unit, located locally on the detection device, performs gripping feasibility assessment and local processing of multi-source data, and collaborates with the cloud to update the gripping strategy.

[0022] This invention provides a dynamic gripping control method and system for strong electromagnetic environments, which has the following beneficial effects:

[0023] The combination of graded pressure application, compliant resistance, and incremental contact force limitation enables soft contact and gentle pressing on the conductor surface, and the scratch depth can be controlled to below 0.01mm.

[0024] Visual-inertial tight coupling + advanced motion prediction + dual-ring recursive approximation, docking deviation ≤3mm, gripping success rate ≥95%;

[0025] Multi-source fusion judgment remains reliable even when a single sensor degrades (failure rate ≤3%), and the pre-selection abandonment logic avoids blind attachment;

[0026] Wind speed correlation maintains pressure + deflection compensation, and remains stable under level 12 wind load;

[0027] Edge-cloud collaboration enables continuous optimization of parameters and models, resulting in strong generalization capabilities. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0029] Figure 1The method flowchart provided by the present invention. Detailed Implementation

[0030] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0031] See Figure 1 This invention provides a dynamic gripping control method under strong electromagnetic environment, comprising the following specific steps:

[0032] Step S1: Pre-attachment environmental perception and state prediction:

[0033] (1) This solution requires three types of sensors, as shown in Table 1 below:

[0034] Table 1: Sensor Statistics Table

[0035]

[0036] Among them, wind load is the original disturbance that causes conductor swaying and device shaking, so the wind speed term is introduced into the prediction model as an exogenous input.

[0037] (2) SA1 visual-inertial tightly coupled positioning:

[0038] Visual measurements and inertial navigation predictions are fed into the same state estimation filter, allowing them to constrain and jointly correct each other at every moment. This ensures that when one measurement degrades under strong electromagnetic interference, the other can still maintain a usable estimate. The process includes the following steps:

[0039] 1. The state vector of the device relative to the traverse coordinate system is:

[0040]

[0041] in:

[0042] : The longitudinal (along the direction of the conductor) displacement of the device's center of mass relative to the conductor's attachment point, in meters;

[0043] : The lateral displacement (perpendicular to the direction of the conductor and in the horizontal plane) of the device's center of mass relative to the conductor, in meters;

[0044] : The vertical displacement of the device's center of mass relative to the conductor (i.e., the direction of "pressure" applied during the gripping and approximation), in meters;

[0045] The roll angle of the device about the axis of the conductor;

[0046] The pitch angle of the device;

[0047] : The yaw angle of the device;

[0048] The superscript T indicates matrix transpose.

[0049] 2. The angular velocity ω measured by the IMU is ω = [ω x ω y ω z ] T and acceleration a=[a x a y a z ] T The state vector is recursively shifted forward by one sampling period dt according to the kinematic relationship:

[0050]

[0051] Where: the subscript k represents the kth sampling time; f(*) is the prior prediction of the state at the next moment; f(*) is the system's equation of motion, which includes the cubic integral of position (the position is obtained by the two integrals of acceleration) and the angular velocity integral of attitude.

[0052] 3. The visual algorithm detects the conductors in the image and solves for the pose constraints, thus constructing the measurement equations:

[0053]

[0054] Where: z k V is the visual measurement vector (containing relative position and attitude observations); H is the measurement matrix, representing the linear mapping from state to measurement; v k For measuring noise, its covariance is used to determine the reliability of visual measurements.

[0055] 4. Prior prediction The system combines visual measurements with weighted least squares (Kalman filtering), automatically allocating weights based on the noise covariance of the two types of information. When strong electromagnetic interference increases the variance of the visual measurements, the system automatically reduces the visual weights and increases the inertial navigation weights, thereby suppressing jitter in state estimation. The final result... This is the approximation state estimation.

[0056] (3) SA2 based on recursive prediction model for conductor motion advance prediction:

[0057] Under wind load, the conductor will swing, and its displacement has short-term quasi-periodicity—that is, on a time scale of hundreds of milliseconds, its motion trend can be approximately extrapolated from the sampled values ​​of multiple past periods.

[0058] Let the current time be t. k Take the historical displacement sequence relative to that moment. , where d k-j The displacement of the conductor in the "direction of interest" (the lateral y-component of the visual fusion output) at the j-th sampling period in the past is N, which is the number of historical sampling points selected and determined by the conductor's swing fundamental frequency (electromagnetic vibration caused by power frequency 50Hz or wind-induced low-frequency oscillation).

[0059] Therefore, the recursive prediction formula is:

[0060]

[0061] in:

[0062] d k+1 : Predicted displacement of the conductor at the next moment, in meters;

[0063] d k-j : The historical displacement value of the j-th sampling period before the current time, in meters;

[0064] N: The number of historical displacement samples used in the prediction (positive integer), reflecting the length of the time window used for the prediction;

[0065] α: Memory decay weighting factor, 0 < α < 1, used to assign greater weight to "historical data that is closer to the present", reflecting the short-term continuity of the movement; the closer α is to 1, the more the prediction depends on the recent overall trend.

[0066] Incorporate wind speed into forecast corrections. Definition

[0067]

[0068] in:

[0069] Predicted displacement after wind load correction, in meters;

[0070] K w Wind correction factor, in m·s / m, represents the displacement response of the calibration conductor under unit wind speed variation.

[0071] v wind Current measured wind speed, in m / s;

[0072] v0: Reference wind speed, taken as the threshold wind speed when the conductor does not swing significantly, in m / s; this item is approximately zero when the wind speed is lower than v0.

[0073] This prediction allows the system to "predict where the conductor will move" instead of "only chasing after seeing the conductor move," thus providing an advanced target for the double-ring approach, reducing lag and improving docking success rate.

[0074] (4) SA3 state fusion and attachment point generation:

[0075] Estimate the current approximation state x k With future predicted displacement Confidence-weighted fusion is used to generate the target attachment point coordinates:

[0076]

[0077] in:

[0078] : The three-dimensional position vector of the target attachment point, in meters;

[0079] : The three-dimensional position components of the device relative to the conductor at the current moment, i.e. (x, y, z), in meters;

[0080] Corrected predicted displacement scalar, in meters;

[0081] n: Unit direction vector in the direction of conductor swing (take the horizontal y-axis unit vector).

[0082] β: Prediction confidence weight coefficient, 0≤β≤1. β is larger when the prediction model error is small and smaller when the error is large, in order to balance the relative weight of "based on the current state" and "based on prediction".

[0083] The target attachment point after fusion becomes the reference point for subsequent approximation control instructions.

[0084] Step S2: Dynamic docking approximation control:

[0085] (1) To resolve the contradiction between "drone hovering jitter (±5cm level)" and "the device needs to be aligned with an accuracy of ±3mm", a two-stage strategy of "fast first, then accurate" is adopted:

[0086] First-stage pre-approach: The control device moves towards the guide wire at a relatively high speed along the approach trajectory, with the goal of sending the device into a pre-approach sphere with a radius R1 (e.g., 5cm) centered on the target's gripping point. At this stage, precise attitude control is not pursued; the primary objective is to shorten the approach time and maintain a safe distance to avoid collisions.

[0087] Second-level fine-tuning approximation: After the device enters the pre-approximation spherical domain, the state is updated once per control cycle. The residual position vector between the current position and the target attachment point is calculated, and iterative correction is performed cycle by cycle to converge the device to an allowable error spherical domain with a radius of R2 (e.g., 3mm) centered on the target attachment point. Positioning accuracy is the primary objective in this stage.

[0088] (2) Residual calculation and approximation law:

[0089] Let P be the current relative position of the device in the m-th approximation control cycle. m The target point is P. target Define the residual position vector:

[0090]

[0091] in:

[0092] e m The approximation residual vector for the m-th period, in units of m;

[0093] P m : The real-time position vector of the device in the m-th cycle, in meters;

[0094] P target : Target position vector, in meters.

[0095] The approximation law in proportional-differential (PD) form generates the speed command:

[0096]

[0097] in:

[0098] v m : The desired speed command for the m-th control cycle, in m / s;

[0099] K p Proportional gain determines the degree to which the residual amplifies the velocity, measured in units of 1 / s;

[0100] K d Differential gain is used to dampen the rate of change of residual error and suppress overshoot. Its unit is a dimensionless coefficient with a time dimension.

[0101] : The change in residual between two adjacent periods, in meters;

[0102] dt: Control cycle duration, in seconds.

[0103] If and only if When the magnitude of the residual vector is less than or equal to the allowable error radius R2, the system determines that it has entered the allowable error sphere and allows it to enter the gripping contact stage.

[0104] (3) Compliant impedance attitude centering control:

[0105] Throughout the approach and contact process, it is essential to ensure that the device's gripping reference surface is correctly aligned with the conductor's axis to avoid uneven force distribution or damage to the conductor due to "oblique gripping." The system implements impedance control on the rotation angle (roll-off angle) θ around the conductor's trajectory axis, shaping it into a predetermined mass-damping-stiffness dynamic characteristic.

[0106] The target impedance relationship is as follows:

[0107]

[0108] in:

[0109] : Angle alignment error, i.e., the deviation between the current roll angle and the axial tilt angle of the guide wire, in rad;

[0110] : Rate of change of angular error, in rad / s;

[0111] : The second rate of change of the angle error (difference in angular acceleration), in rad / s²;

[0112] The set equivalent inertia ("mass") parameter determines the speed and compliance of the attitude response;

[0113] The set equivalent damping parameters determine the dissipation capacity of attitude motion, which is used to suppress jitter and impact.

[0114] The set equivalent stiffness parameters determine the strength of the attitude recovery centering tendency;

[0115] Equivalent external disturbance moment (equivalent quantity of wind load and UAV jitter coupling), unit N·m.

[0116] This impedance relationship dynamically "compliances" the device's attitude—allowing for appropriate deviations in attitude under external force without overshooting, and continuously converging towards the centering angle; by adjusting... , , It can simultaneously achieve the two contradictory goals of "following the vibration of the conductor" and "resisting impact", ensuring stable alignment and non-destructive contact.

[0117] Step S3: Adaptive Compliant Grip Control:

[0118] (1) Closed-loop adjustment of gripping force:

[0119] After the gripping mechanism contacts the wire, the contact normal force F is measured by the force sensing component. c The feedback quantity is F, and the target attachment pressure is F. ref The gripping mechanism output commands are generated using a combination of PID feedback and feedforward compensation.

[0120] Define pressure error:

[0121]

[0122] in:

[0123] e F Pressure error, in nanometers (N).

[0124] F ref : Set the target gripping pressure (determined based on the device's weight, wire diameter, and safety margin), in N;

[0125] F c Force sensor measured contact normal force, unit N.

[0126] PID control output:

[0127]

[0128] in:

[0129] The basic driving quantity of the PID control output;

[0130] Proportional gain;

[0131] Integral gain, used to eliminate steady-state error;

[0132] Differential gain is used to suppress pressure overshoot.

[0133] : Time integral of pressure error;

[0134] : Rate of change of pressure error over time.

[0135] The feedforward compensation term is used to offset the effects of gravity and wind load:

[0136]

[0137] in:

[0138] : Feedforward compensation driving quantity;

[0139] Mass of the testing device, in kg;

[0140] Gravitational acceleration, taken as 9.8 m / s²;

[0141] The device's current pitch angle, in rad, is used to accurately compensate for the projected component of gravity along the gripping direction.

[0142] : Map gravity onto the cosine projection of the gripping direction according to the attitude angle;

[0143] : Wind load feedforward coefficient, unit N·s / m;

[0144] Real-time wind speed, in m / s.

[0145] Final gripping mechanism output command:

[0146]

[0147] Among them, the feedback term ensures that the actual pressure accurately tracks the target, and the feedforward term eliminates the deterministic disturbances of gravity and wind load in advance. The sum of the two makes the fluctuation of the actual gripping force less than the set threshold.

[0148] (2) Compact and non-destructive pressing – staged pressurization:

[0149] To avoid impact damage caused by a single pressurization, a staged pressurization strategy is adopted: the target pressure is divided into several pressurization stages, and the current pressure is kept stable in each stage before entering the next stage.

[0150] Generation of the i-th level pressure step size:

[0151]

[0152] in:

[0153] : The increment of pressure during the i-th pressurization, in N;

[0154] : Maximum allowable pressurization increment in a single operation (set upper limit to prevent shock), unit: N;

[0155] : The error between the target and the current pressure at level i, in N;

[0156] The pressurization rate coefficient (0 < λ ≤ 1) determines the rate of pressurization.

[0157] Only when the contact force increment ΔF c (i.e., the change in actual pressure between two consecutive samplings) is lower than the set upper limit ΔFsafe Only when the contact force changes too much is the pressure allowed to proceed to the next stage; if the rate of change is too large, the pressure is paused and reversed, thus ensuring that the contact force on the conductor surface is always "compliantly adhered" rather than "rigidly impacted." This mechanism directly guarantees the non-damaging nature of the gripping process.

[0158] Step S4: Multi-source fusion attachment status determination:

[0159] (1) Acquisition and normalization of three types of information:

[0160] After attachment and compression are completed, the system needs to answer "has it now stably gripped the conductor?". Relying on a single piece of information (such as pressure alone) is unreliable under strong electromagnetic interference and wind disturbances; therefore, a fusion of three types of heterogeneous information is used for determination:

[0161] Grafting force information: Current contact normal force F c This measures whether the stress level has reached a safe threshold. Stress satisfaction is defined as follows:

[0162]

[0163] Where: P F Normalized stress satisfaction (values ​​around 0 to 1); F c Current positive pressure; F ref For target pressure. P F The closer to or greater than 1, the more sufficient the pressure.

[0164] Attitude stability information: After attachment, the attitude of the device should tend to stabilize. Attitude stability is defined as follows:

[0165]

[0166] in: The magnitude of the deviation vector between the device's current attitude angle and the initial attitude angle of the gripping device; The maximum allowable deviation for the set attitude; The value is between 0 and 1, with the value closer to 1 indicating a more stable posture and better centering.

[0167] Visual fit information: Define visual fit:

[0168]

[0169] Where: δ v For visual calculation, the attachment deviation (centering angle deviation or lateral deviation) of the reference plane relative to the conductor axis is measured in mm or rad; δ v，max The maximum allowable deviation for the set visual fit. V The closer it is to 1, the more precise the fit.

[0170] (2) Weighted composition and stability criterion:

[0171] The three normalized values ​​are weighted and combined to form a weighted evaluation metric for attachment fit:

[0172]

[0173] in:

[0174] Q: Graft fit evaluation metric (dimensionless, value approximately 0 to 1);

[0175] w1, w2, w3: These are the weight coefficients for the pressure, attitude, and vision channels, respectively, and they satisfy w1+w2+w3=1. The weights are preset based on the reliability of the sensor under various operating conditions (e.g., the weight of the pressure channel is appropriately increased during strong winds).

[0176] Stable attachment criterion: When Q continuously exceeds the set stability threshold Q thr And maintain the set duration t hold If the capture is successful (e.g., after 2 seconds), the capture is considered successful.

[0177]

[0178] That is: starting from t0 and lasting for t hold Within the time window, the fit evaluation score remains no lower than the threshold. If this is not met, return to step S3 to continue applying pressure and adjustment until the criteria are met or the abandonment logic is triggered.

[0179] (3) Intelligent feasibility assessment:

[0180] Based on full weighting and fusion, if the system fails to bring Q to the threshold within the set number of retries, or if the initial prediction indicates that the current operating conditions (such as strong winds, continuous vibration, or visual degradation caused by electromagnetic interference) are unfavorable, the edge computing unit integrates multi-source data streams, prediction results, and historical failure samples to assess the feasibility of the attachment. When the assessed feasibility is lower than the set threshold, an abandonment command is generated, the control device exits the attachment process, and safely returns to standby, avoiding risky attachment under adverse operating conditions that could damage equipment or cables.

[0181] Step S5: Actively maintain control after attachment is complete, specifically including:

[0182] (1) Maintain dynamic adjustment of pressure according to wind load:

[0183] After successful attachment, the system enters the active holding phase, the core of which is: instead of relying on static friction for "passive" fixation, it "actively" applies adjustable pressure according to environmental changes to resist slippage and detachment.

[0184] The pressure to maintain the target is composed of two parts: static and dynamic.

[0185]

[0186] in:

[0187] F hold The current target remains under pressure, in units of N;

[0188] F hold，0 Static foundation holding pressure (considering the device's own weight and minimum safe grip redundancy), unit N;

[0189] K w，h : Wind speed regulation coefficient for maintaining pressure as a function of wind speed, in N·s / m;

[0190] v wind Real-time wind speed, in m / s;

[0191] v hold，0 : Wind speed threshold. Only when the wind speed exceeds this value will the pressure be significantly increased; otherwise, the basic pressure will be maintained. Unit: m / s.

[0192] According to the above formula: as wind speed increases, pressure increases synchronously, thus offsetting the slippage trend; as wind speed decreases, pressure is adjusted accordingly to avoid long-term excessive pressure affecting the conductor, balancing stability and line safety.

[0193] (2) Anti-slip and anti-detachment deflection compensation:

[0194] Under wind load, the device may slip along the conductor or deflect (roll) around the conductor. The system continuously monitors the device's deflection angle using attitude sensors, and when a deflection trend is detected, incremental compensation is applied to the holding force in that direction.

[0195]

[0196] in:

[0197] : The incremental compensating pressure applied to counteract deflection, in N;

[0198] Attitude-pressure compensation coefficient, in N·s / rad;

[0199] : Measured attitude deflection rate (angular velocity), unit rad / s;

[0200] || Take the absolute value, which means only the order of magnitude of the deflection rate is taken;

[0201] : Duration of the continuous observation period of deflection, in seconds, used to accumulate the deflection rate into an estimate of the compensable displacement.

[0202] The superposition of compensation pressure and foundation holding pressure constitutes the final gripping and security holding force, thereby effectively suppressing the slippage, deflection and detachment of the device on the conductor, and significantly improving the holding stability under a level 12 wind load (wind speed 32.7m / s).

[0203] (3) Parameter deviation monitoring and automatic correction:

[0204] Throughout the holding phase, the system compares each holding parameter (actual holding pressure, attitude angle, wind speed) with the set target value in real time. When any parameter deviates beyond the allowable range, it automatically performs closed-loop correction to keep the system at the set working point until the detection operation is completed. Then, it executes the safety release procedure (unloading pressure smoothly in reverse order of gripping and attaching, releasing the centering constraint, and disconnecting from the guide wire).

[0205] Additionally, the system deploys an edge computing unit locally on the detection device to complete the feasibility assessment of attachment, temporary storage of multi-source data, and matching of optimal attachment parameters. Through 4G / 5G or satellite communication, status data and operation records are uploaded to the cloud to achieve experience sharing among multiple devices, historical statistical analysis, and continuous iteration of strategy models. This enables the system to have adaptive capabilities across voltage levels, clamp models, and climate conditions, forming a positive feedback loop of "one-time attachment, experience accumulation, and continuous optimization".

[0206] Additionally, the present invention also provides a dynamic gripping control system for strong electromagnetic environments, comprising:

[0207] The environmental perception unit is used to acquire the relative pose information between the detection device and the target conductor, the motion trend information of the target conductor under environmental load, and the suspension attitude information of the detection device, and outputs the fused relative pose estimate and the target attachment point.

[0208] The docking control unit is used to drive the detection device to perform graded and progressive approximation based on the target attachment point, and to perform compliant centering of the device attitude through impedance adjustment.

[0209] The compliant gripping unit is used to perform closed-loop regulation and feedforward compensation of the contact pressure with real-time gripping force as feedback, and to constrain the single pressure increment and pressure change conditions during the staged pressing process.

[0210] The state determination unit is used to normalize and weight the gripping force information, device posture information and visual fit information to synthesize the gripping fit evaluation quantity, and determine whether the device has reached a stable gripping state based on the time window criterion.

[0211] The active holding unit is used to apply an active and adjustable holding pressure and deflection increment compensation to the attached detection device based on environmental load and device attitude information, to maintain the attachment until the detection operation is completed and the device is safely released.

[0212] The edge computing unit, located locally on the detection device, is used to perform feasibility assessment of attachment and local processing of multi-source data, and to collaborate with the cloud to update the attachment strategy.

[0213] In summary, the above technical solution, in practical application, specifically includes the following five operational logic steps:

[0214] Step 1: Perception and Prediction

[0215] Real-world scenario: A drone equipped with an X-ray inspection device is slowly approaching the crimped conductor of a high-voltage power transmission line. At this time, a strong electromagnetic field exists near the power line (the electric field strength of 66kV-220kV lines can reach over 10kV / m), causing the drone to exhibit visible shaking while hovering, and the conductor itself is also oscillating at a low frequency due to wind. The core challenge facing the system is that the target itself is moving, while each status update inevitably involves a delay.

[0216] If the system approximates the position solely based on the "currently measured conductor position," then by the time it reaches the calculated position, the conductor has already shifted, inevitably causing a docking deviation. Therefore, this invention establishes a dual-channel collaborative logic of "current state estimation + future trend prediction":

[0217] Current state estimation channel (visual-inertial tight coupling): The traverse image provided by the multi-view camera and the device's own inertial data provided by the IMU are mutually corrected within a tight coupling framework. Even if strong electromagnetic interference degrades visual positioning, the inertial navigation system can still maintain short-term state stability due to its high-frequency output above 100Hz; conversely, the inertial navigation system experiences continuous drift, while the vision system periodically "calibrates and corrects" it, ensuring that the relative pose estimation of the device remains reliable even during dynamic jitter.

[0218] Future trend prediction channel (recursive prediction + wind load correction): Considering that the conductor swing has short-term quasi-periodicity and the historical trajectory shape contains sufficient extrapolation information, it is possible to obtain a sufficiently accurate forward estimate without introducing a highly complex model. Therefore, the system extrapolates the conductor displacement sequence of several past sampling periods in an attenuated weighting method of "the closer to the present, the greater the weight", to obtain the predicted position of the conductor in the next sampling period. At the same time, the measured wind speed is used as an external correction input - the higher the wind speed, the greater the correction amplitude of the predicted displacement.

[0219] When the predicted position of the conductor is unreliable (such as when a sudden change in wind speed causes an increase in the prediction error), the system automatically reduces the prediction component and increases the weight of the "current actual position" by using the confidence weight coefficient β, thereby adaptively balancing "following the trend" and "maintaining stability".

[0220] Step Two: Two-Level Progression

[0221] Real-world scenario: The hovering jitter of a drone is approximately ±5cm, while X-ray imaging requires the device to achieve millimeter-level alignment accuracy. If the system is required to track a jittering target with millimeter precision from the outset, the controller will either overshoot and oscillate or exert excessive force, causing a dangerous collision.

[0222] The system decomposes the approximation process into two two-stage phases with clearly defined target bits and priorities:

[0223] Level 1 (Pre-approach): The control logic prioritizes "entering the safe zone as quickly as possible and avoiding collisions," rapidly delivering the device to a spherical area with a radius of R1 (centimeter-level) centered on the target's gripping point. At this stage, meticulous adjustments are deliberately avoided, as accuracy cannot be guaranteed at this point, and excessive adjustments could introduce oscillations.

[0224] Level 2 (Fine Adjustment): After entering the pre-approximation sphere, the control logic switches to "aiming for millimeter-level convergence". In each control cycle, the system calculates the residual vector between the current position of the device and the target attachment point, generates a speed command based on the proportional-derivative law of PD approximation, and compresses the residual cycle by cycle until the residual magnitude converges to within the allowable error sphere radius R2 (millimeter level).

[0225] The "sphere radius" is used as a clear switching criterion between the two stages to ensure that the switching timing is unique and unambiguous; the differential term in the PD approximation law provides a damping effect, suppressing overshoot during the convergence process and preventing the device from oscillating back and forth near the target point.

[0226] Meanwhile, throughout the approach and contact process, the system implements "compliant alignment" of the device's rolling attitude through compliant impedance control—shaping the attitude dynamics to the set mass-damping-stiffness characteristics. When external wind loads or drone vibrations attempt to "bump" the device off course, the attitude system does not resist rigidly but allows for moderate compliance and smooth return to center, thus avoiding the impact of hard collisions on the structure and ensuring continuous alignment between the gripping reference plane and the conductor axis.

[0227] Step 3: Smoothing and Pressing

[0228] Actual operating scenario: The device's gripping mechanism is about to contact the conductor. If full pressure is applied directly at this moment, the polyurethane gripper head will cause a momentary impact on the conductor surface, posing a risk of scratches, indentations, or even damage to the conductor's insulation layer. This system designs the gripping and pressure application process as a "graded, progressive, force feedback closed-loop":

[0229] Target pressure closed-loop: The positive pressure measured by the force sensor is used as feedback, and compared with the set target gripping pressure to form an error. This error is then used to generate the basic drive through a PID controller. Simultaneously, feedforward compensation is superimposed—the projection of gravity along the gripping direction is accurately calculated based on the current pitch angle of the device and then canceled out. Wind load is then compensated based on real-time wind speed. The feedback term ensures accurate and stable final pressure; the feedforward term pre-calculates and eliminates deterministic disturbances such as "gravity + wind load," allowing the feedback to only address minute residuals, thus enabling fast and accurate pressure control.

[0230] Engineering safeguards for non-destructive pressing: Each pressurization increment is limited to a set upper limit, and the next pressurization stage is only permitted if the actual contact force increment between two adjacent samples does not exceed the safety limit. If an excessively rapid change in contact force is detected, the system immediately pauses pressurization and initiates a callback.

[0231] The goal of "non-destructive" is implemented through three determinable control constraints: "gradual pressure increase + bounded contact force increment + real-time force feedback and callback". It does not simply make the claw material soft (soft does not mean it is not harmful; soft claws can still scratch if applied too much pressure), but rather ensures through the control mechanism that the force on the wire is within a safe range at all times.

[0232] Step 4: Multi-source determination:

[0233] Real-world scenario: The gripping mechanism has been pressurized. However, relying solely on pressure for judgment may lead to a misjudgment of "secure grip" under strong electromagnetic interference or sudden gusts of wind, causing subsequent vibrations, blurred images, or even the device falling. This system combines three pieces of evidence into a single judgment:

[0234] Pressure evidence (tactile channel): Whether the actual positive pressure reaches the target pressure;

[0235] Attitude evidence (inertial channel): Whether the attitude of the device has become stable and converged after the device has gripped it;

[0236] Visual evidence (visual channel): Check whether the reference plane and the conductor axis are precisely aligned.

[0237] The three pieces of evidence are each normalized (mapped to a dimensionless range of 0 and 1), and then weighted and synthesized to obtain a fixation fit evaluation quantity Q, which is approximately between 0 and 1. The system only determines "fixation successful" when Q is continuously higher than the threshold within a set time.

[0238] The vague concept of "successful capture" is transformed into a quantifiable and definable precise event: "three types of heterogeneous evidence continuously satisfy the weighted joint criterion within a time window." Simultaneously, the weights w1, w2, and w3 are adaptively adjusted according to operating conditions. Even if one sensor is temporarily blinded by strong electromagnetic interference, as long as the other two sensors remain reliable, the system can still make a correct judgment, fundamentally solving the vulnerability of existing single-sensor solutions where "all are blind if one is blind."

[0239] Furthermore, when repeated adjustments fail to bring Q to the target level, or when initial predictions indicate severe operating conditions, the system is willing to "abandon unnecessary attachments," proactively exiting and returning to standby mode. This logic enables the device to proactively choose safety over risk in dangerous conditions such as strong winds and thunderstorms, avoiding the serious consequences of equipment falling and wire damage.

[0240] Step 5: Dynamic Maintenance

[0241] Actual operating scenario: The device has stably attached to the guide wire, but X-ray imaging often takes tens of seconds to several minutes. During this time, the wind force may increase or the wind direction may change abruptly. If the device is merely "passively holding on by friction," it will slip, deflect, or even detach once the wind load exceeds the friction limit. This system immediately switches to active holding mode after successful attachment.

[0242] Pressure fluctuates with the wind: Real-time wind speed is compared with threshold wind speed. If the wind speed increases, the pressure is increased synchronously according to the wind force adjustment coefficient to maintain the pressure and counteract the slippage trend caused by wind load. If the wind speed decreases, the pressure decreases accordingly to avoid long-term overpressure affecting line safety.

[0243] Instant deflection compensation: The device's deflection angular velocity is continuously monitored by attitude sensors. Once a deflection trend is detected (an early sign of impending slippage or rolling), incremental compensation is immediately applied to the holding force in that direction.

[0244] The system has been upgraded from "static friction passive holding" to "closed-loop holding with dynamic active compensation of pressure based on wind speed and attitude." It internalizes "maintaining stability" into a continuously operating closed-loop control process (measuring wind speed and attitude → calculating compensation pressure → applying force → re-measuring), rather than a one-time action. This allows the device to "actively grip" without detaching in gale-force winds of up to level 12, while also "appropriately loosening" after the gale to protect the conductor.

[0245] To better understand the technical solution of this invention, the following specific embodiments are provided: Dynamic capture of the entire process of live X-ray detection of 66kV transmission line crimping clamps:

[0246] I. The basic parameters for the task are shown in Table 2 below:

[0247] Table 2: Basic Parameters of the Operation

[0248]

[0249] II. Step S1: Pre-attachment environmental perception and state prediction:

[0250] (a) SA1 Visual-Inertial Tightly Coupled Positioning:

[0251] The drone hovered 1.5m beside the power line. The binocular camera captured images of the power line at 30fps, detecting the center coordinates of the power line in each frame; the IMU output the three-axis angular velocity and linear acceleration of the device at 100Hz.

[0252] The state vector of the device relative to the point where it grips the conductor:

[0253]

[0254] Substituting the current actual measurement (assuming the device is currently stationary and suspended 10cm directly above the conductor, with no attitude deflection):

[0255]

[0256] x=0.002m=2mm: Longitudinal deviation of the device relative to the attachment point along the conductor;

[0257] y=0.015m=15mm: Lateral (direction of conductor swing) deviation of the device relative to the gripping point;

[0258] z=0.100m=100mm: The vertical approximation distance of the device relative to the conductor;

[0259] θ = 0.5°: Roll angle of the device (around the axis of the conductor's direction);

[0260] =1.0°: Device pitch angle;

[0261] ψ=0.3°: Yaw angle of the device.

[0262] Time update (inertial navigation prediction):

[0263] The angular velocity measured by the IMU is ω = [-0.01, 0.02, 0.01]. T rad / s and acceleration a = [0.05, 0.03, 9.79] T m / s 2 The calculation is recursively performed within a control period dt = 0.02 s. Position is obtained by the second integral of acceleration, and attitude is obtained by the first integral of angular velocity.

[0264] Taking the lateral displacement y as an example, substitute:

[0265]

[0266] =0.015m: lateral position at the previous moment;

[0267] =0m / s: Current lateral velocity (drone hovering);

[0268] =0.03m / s 2 IMU measured lateral acceleration;

[0269] =0.02s: Control cycle.

[0270] Substitute: ;

[0271] The inertial navigation system predicted a lateral displacement of 15.006 mm, which is only 0.006 mm forward, demonstrating the IMU's high-frequency accurate recursion within a short period.

[0272] Visual measurement correction:

[0273] The visual camera has calculated the lateral position of the device relative to the wire to be 15.2 mm, i.e., z. vis =0.0152m;

[0274] There is a 0.2mm discrepancy between visual measurements and inertial navigation predictions. Due to the slightly larger variance of visual noise under strong electromagnetic interference, the system assigns weights based on the noise covariance: Let the visual weight be w. vis =0.4, inertial navigation weight w imu =0.6, then the lateral estimation after fusion: ;

[0275] Substitute into the calculation: ;

[0276] Results: The fusion concept—combining visual and inertial navigation constraints—means that a disturbance in one component does not lead to overall inaccuracy. At this point, the measured lateral deviation of the device is approximately 15.08 mm. Approximate state estimation vector: ;

[0277] (ii) The SA2 recursive prediction model provides advanced prediction of conductor motion trends:

[0278] Operating condition: Under wind load, the conductor is oscillating laterally at approximately 0.3 Hz with an amplitude of ±3 cm. The system needs to predict in advance where the conductor will move next to avoid the situation where "the target has already moved away when its position is seen".

[0279] Data sequence values: Take the lateral displacement sequence (mm) of the conductor from the most recent N=10 sampling periods (20ms per period, 0.2s in total): ;

[0280] That is, the conductor is swinging to the right (lateral), and the displacement increases from 5.2mm to 17.8mm.

[0281] Recursive prediction formula:

[0282]

[0283] Stepwise substitution (assigning weights from nearest to farthest):

[0284] Recent data d k =17.8, weight α 0 =1, contribution 17.8×1=17.8;

[0285] d k-1 =16.2, weight α 1 =0.9, contribution 16.2×0.9=14.58;

[0286] d k-2 =15.0, weight α 2 =0.81, contribution 15.0 × 0.81 = 12.15;

[0287] d k-3 =13.4, weight α 3 =0.729, contribution 13.4×0.729=9.77;

[0288] d k-4 =12.0, weight α 4 =0.656, contribution 12.0 × 0.656 = 7.87;

[0289] d k-5 =10.9, weight α 5 =0.590, contribution 10.9×0.590=6.43;

[0290] d k-6 =9.5, weight α 6 =0.531, contribution 9.5 × 0.531 = 5.05;

[0291] d k-7 =8.1, weight α 7 =0.478, contribution 8.1×0.478=3.87;

[0292] d k-8 =6.8, weight α 8 =0.430, contribution 6.8×0.430=2.93;

[0293] d k-9 =5.2, weight α 9 =0.387, contribution 5.2×0.387=2.01;

[0294] Numerator (weighted sum):

[0295]

[0296] ;

[0297] Substituting into the formula (note that the denominator of the formula is N, i.e., 10 points, and the weighted sum is divided by N): d k+1 =82.46 / 10=8.246mm;

[0298] Result: The point value output by the system is a weighted "trend representative value". Since the latest displacement value is still increasing during the oscillation, the weighted average will fall at a position slightly lower than the latest point, but representing the overall trend. That is, the system predicts that the lateral displacement of the conductor at the next moment will be about 8.25mm (a slight decrease compared to the latest value, reflecting that the oscillation is not a unidirectional sudden increase). In fact, if the conductor is periodically falling from a high level, this prediction will guide the device to converge towards the middle of the oscillation, avoiding chasing the peak of the oscillation.

[0299] Wind load correction:

[0300] Substitute: ;

[0301] d k+1 =8.25mm;

[0302] K w =0.5: Wind force correction factor. For every 1 m / s increase in wind speed, the conductor displacement increases by approximately 0.5 mm.

[0303] v wind =6m / s: Measured wind speed;

[0304] v0=3m / s: The threshold wind speed at which the conductor swings significantly.

[0305] calculate: ;

[0306] After wind load correction, the predicted lateral displacement of the conductor at the next moment is approximately 9.75 mm.

[0307] (III) SA3 state fusion generates target attachment points:

[0308] The current approximation state estimate is weighted and fused with the predicted displacement: ;

[0309] Substitute:

[0310] (The current position of the device relative to the conductor);

[0311] =9.75mm: Corrected predicted displacement;

[0312] n: Unit vector in the lateral direction of the conductor's swing, taken as n=[0,1,0] T (Along the y-axis);

[0313] β=0.7: Prediction confidence weight (higher value when wind conditions are stable).

[0314] Only the y-component (horizontal) is affected. Calculate the y-component: ;

[0315] The x and z components remain unchanged (the prediction mainly affects the lateral oscillation): ;

[0316] Target capture point: ;

[0317] Analysis: The system did not set the target at "the current measured position of the conductor at 15.08mm", but actively moved it forward to 21.9mm. This was because the prediction showed that the conductor was swinging to the right. When the device actually arrived, the conductor would move to a position of about 21.9mm, which was also the key to converging from ±5cm drone jitter to millimeter-level alignment.

[0318] III. Step S2: Dynamic docking approximation control:

[0319] (a) First-order pre-approximation:

[0320] The system uses R1=5cm as the pre-approach radius of the sphere, and the control command is "enter the vicinity of the target as soon as possible and avoid collision", without pursuing millimeter-level correction.

[0321] The current device position follows Pm=[2,15.08,100]mm, and the target P target =[2,21.9,100]mm.

[0322] Residual vector: ;

[0323] Residual modulus: ;

[0324] Comparing R1=5cm=50mm: 6.82mm≤50mm;

[0325] The criterion is met: the device is now within the pre-approach sphere, and the system has switched from "rapid approximation" to "millimeter-level convergence" mode.

[0326] (ii) Second-order fine-tuning approximation (PD approximation law):

[0327] After entering the fine-tuning stage, the device needs to converge from a residual of 6.82mm to within R2=3mm.

[0328] First fine-tuning cycle (m=1):

[0329] Position P1 = [2, 15.08, 100] mm, residual along the y-direction: ;

[0330] Speed ​​command (proportional-derivative):

[0331]

[0332] K p =8;K d =3; dt=0.02s;

[0333] First cycle e y,prev =0 (considered as 0 for historical residuals).

[0334] Substitute:

[0335]

[0336]

[0337] The speed command is approximately -1.08 m / s, and the direction is negative y (moving towards the target point on the traverse).

[0338] Analysis: The differential term (second term) in PD is dominant (-1.023 vs -0.0546). Because the residual changes drastically, strong damping is needed to suppress overshoot and prevent the device from overshooting the target point.

[0339] After executing multiple cycles (assuming it's the 10th cycle, m=10):

[0340] Assume the residual has been compressed to e at this point. y,10 =0.0025m=2.5mm, residual change rate (Positive, the risk of overshoot is reduced). .

[0341] The velocity turning into a small positive quantity indicates that it has approached the target and is slowly converging back to positive.

[0342] Convergence criterion:

[0343] Residual modulus: ;

[0344] Comparing R2=3mm: 2.5mm≤3mm;

[0345] The criterion is met: the device has entered the permissible error range and is allowed to enter the contact and attachment stage.

[0346] (III) Compliant Impedance Attitude Centering Control

[0347] Approximately the entire range, the system implements impedance control on the roll angle θ (around the conductor's trajectory axis), shaping it into a mass-damping-stiffness characteristic:

[0348]

[0349] Substitute parameters:

[0350] M d =0.5kg⋅m 2 (Equivalent inertia, unit: kg·m²);

[0351] B d =8N⋅m⋅s / rad (equivalent damping, unit: Newton-meter-second per radian);

[0352] K d =20N⋅m / rad (equivalent stiffness, unit: Newton-meter-radian);

[0353] =0 rad / s 2 (Angular acceleration deviation, which is 0 under steady-state assumptions);

[0354] =0.1 rad / s (angular velocity deviation);

[0355] =0.02rad (angular deviation, approximately 1.15°);

[0356] Suppose a gust of wind causes the device's roll angle to deviate by e. θ =0.02rad (approximately 1.15°), angular velocity deviation =0.1rad / s, and the angular acceleration is approximately 0.

[0357] The equivalent disturbance can be obtained from the torque balance: F ext =0.5×0+8×0.1+20×0.02=0+0.8+0.4=1.2N⋅m.

[0358] The system considers the dominant disturbance at this moment to be a wind load equivalent torque of 1.2 N·m. Due to the use of compliant impedance, the device will not rigidly resist this 1.2 N·m transient disturbance, but will allow for moderate compliance (due to damping B). d (Absorbs vibration energy), after the wind passes, it is due to stiffness K d =20 drives a smooth return to the center position. This strategy avoids the impact of hard collisions on the structure and damage to the conductors.

[0359] IV. Step S3: Adaptive Compliant Grab Control

[0360] Operating condition: The device has entered the 3mm tolerance range, and the mechanical claw is about 3mm away from the surface of the wire, ready to make contact.

[0361] (a) Setting target pressure:

[0362] Based on the device's own weight and safety redundancy, the target attachment pressure is: F ref =220N;

[0363] (ii) Pressure Feedback Closed Loop (PID):

[0364] At the initial contact moment, the force sensor measured the normal force F. c =0N (Not yet pressed).

[0365] Pressure error: ;

[0366] PID control output (take K) P =2,K I =0.5,K D =0.01, initial integral term is 0, differential term is 0): u PID =2×220+0.5×0+0.01×0=440;

[0367] Basic driving quantity u PID =440 (used to quickly build up pressure).

[0368] (iii) Feedforward compensation:

[0369] Calculate the component of gravity along the gripping direction and wind load compensation:

[0370]

[0371] Substitute:

[0372] m=30kg, g=9.8m / s 2 ;

[0373] Current pitch angle β of the device a =1.0°, cos1.0°=0.99985;

[0374] K w =2N⋅s / m、v wind =6m / s.

[0375] Gravitational projection: mgcosβ a =30×9.8×0.99985=294×0.99985=293.956N;

[0376] Wind load compensation: K w ′v wind =2×6=12N;

[0377] Total feedforward: u FF =293.956+12=305.956N≈306N;

[0378] The feedforward term indicates that gravity alone has a component of approximately 294N along the gripping direction under the current attitude of the device, plus approximately 12N of wind load at 6m / s, totaling approximately 306N. Without feedforward, relying solely on the PID controller to catch up would result in continuous large outputs and oscillations; with feedforward, this 306N is "calculated and eliminated in advance," and the PID controller only needs to handle the small residuals, resulting in fast and stable pressure control.

[0379] (iv) Total output and staged pressurization:

[0380] Total output command: u=u PID +u FF =440+306=746;

[0381] Driven by the overall command, the mechanical gripper of the device begins to press, and the pressure gradually increases from 0.

[0382] Staged pressurization:

[0383] The target pressure of 220N is divided into 5 levels, each with a target pressure of approximately 44N. The pressure increment for each level is constrained by the following formula:

[0384]

[0385] ΔFmax=20N: Maximum single pressurization increment;

[0386] λ=0.5: Pressure rate coefficient.

[0387] Calculation for Level 1 pressurization:

[0388] Current pressure F c =0, error : ;

[0389] Take the smaller value: ΔF1 = min(20, 110) = 20N;

[0390] That is, the maximum pressure applied in the first stage is 20N, with the pressure increasing from 0 to approximately 20N.

[0391] Level 2: Assuming the pressure has reached 18N (due to a slight slew in the system output and feedback), the error... : ΔF2=min(20,0.5×202)=min(20,101)=20N;

[0392] The pressure rose to approximately 38 N.

[0393] Level 3 (Approaching the Target): Assuming a cumulative error of 160N, the error... : ΔF3=min(20,0.5×60)=min(20,30)=20N;

[0394] The pressure rose to approximately 180 N.

[0395] Level 4: Pressure 180N, Error 40N: ΔF4=min(20,0.5×40)=min(20,20)=20N;

[0396] The pressure rises to approximately 200 N.

[0397] Finally, it converges to 220N: pressure 200N, error 20N: ΔF5=min(20,0.5×20)=min(20,10)=10N;

[0398] At this point, the step size automatically decreases to 10N, the pressure rises to 210N, and then the PID is finely adjusted to 220N.

[0399] Contact force increment criterion: Before each pressurization stage, the system checks the actual pressure change ΔF between two adjacent samplings. c Is it below the safety limit ΔF? safe (Let ΔF) safe =25N / s). Since the maximum increment per stage is limited to 20N and the pressurization process is controlled by the sampling period (20ms), the pressure rise rate is approximately 20N / 0.02s = 1000N / s.

[0400] The system does not apply 20N of pressure instantaneously, but rather smoothly over a cycle through force tracking. Assuming the execution layer smoothly completes each 20N step within 50 control cycles (1 second), the actual pressure rise rate = 20N / 1s = 20 N / s < 25N / s, satisfying the upper limit. Therefore, the staged pressurization mechanism, through "limiting single-stage increments + smooth execution," ensures that the rate of change of contact pressure on the conductor surface remains below the safety threshold, achieving non-destructive pressing.

[0401] Results: The claw tip gradually presses the wire to 220N in millimeter-level and second-level increments, forming a soft contact and slow compression with the polyurethane claw tip on the wire surface. Actual testing on the NY-120 wire clamp showed that the scratch depth on the wire surface was <0.01mm, meeting the requirements for non-destructive testing.

[0402] V. Step S4: Multi-source fusion attachment status determination:

[0403] Operating condition: The robotic gripper has applied 220N of pressure, and the system needs to determine whether it has "truly gripped firmly".

[0404] (a) Normalization of three sources of evidence:

[0405] 1. Pressure Channel (Tactile): Measured positive pressure Fc =220N, target F ref =220N:P F =F c / F ref =220 / 220=1.0;

[0406] 2. Attitude Channel (Inertia):

[0407] After the device grips, its attitude deflects approximately relative to its initial state. =0.5°, maximum allowable =5°:P A =1 - 0.5 / 5 = 1 - 0.1 = 0.9;

[0408] 3. Visual channel:

[0409] The contact deviation δ between the reference surface and the conductor axis v =0.8mm, maximum allowable δ v,max =3mm:P V =1 - 0.8 / 3 = 1 - 0.267 = 0.733;

[0410] (ii) Weighted composition:

[0411] We assign weights w1=0.5 (pressure), w2=0.3 (posture), and w3=0.2 (visual), and w1+w2+w3=1: Q=0.5×1.0+0.3×0.9+0.2×0.733;

[0412] Substituting the values ​​into the equation: Q = 0.5 + 0.27 + 0.1466 = 0.9166;

[0413] The adhesion fit evaluation metric Q≈0.917.

[0414] (III) Stability Criterion: Set a stability threshold Q thr =0.85, duration t hold =2s:Q=0.917≥Q thr =0.85;

[0415] The system continuously samples, maintaining Q≥0.85 for 2 seconds.

[0416] Result: Capture successful.

[0417] (iv) Robustness analysis (gust scenario):

[0418] Suppose that 2.5 seconds after the judgment, a sudden gust of wind of 9 m / s occurs, causing a slight sway in the device's attitude. The attitude stability decreases to 0.5, the fit deviation slightly increases to 1.2 mm, and the visual channel value drops to 0.6. At this time: Q new=0.5×1.0+0.3×0.5+0.2×0.6=0.5+0.15+0.12=0.77;

[0419] Q new =0.77<0.85, falling below the threshold at a single moment.

[0420] However, since the criterion requires "maintaining for 2 seconds," the system will not immediately fail due to a single instantaneous disturbance. The system actively maintains the state (step S5) to rapidly increase the maintenance pressure, allowing the attitude to recover and Q to rise above 0.9, thus re-satisfying the continuity criterion. The combination of multi-source fusion and the continuity duration criterion ensures that short-term disturbances will not lead to misjudgment, demonstrating robustness in judgment under strong electromagnetic and strong wind environments.

[0421] (v) Abandon the logic of holding on (stop loss):

[0422] As a control, if a signal in the environment continuously degrades (e.g., vision is blocked by corona halos for a long time, or posture is constantly jittering), causing Q to fail to reach 0.85 for an extended period, and Q remains between 0.6 and 0.7 after repeated adjustments, the edge computing unit, by integrating multi-source data streams, prediction results, and historical samples, determines that the feasibility of the capture is below the 60% threshold. In this case, the system actively abandons the capture attempt, and the control device exits and returns to standby.

[0423] This "loss prevention logic" enables the device to proactively avoid risks under adverse operating conditions, preventing blind attachment that could lead to falls or damage to the wires, thus demonstrating the engineering rationality of control.

[0424] VI. Step S5: Actively maintain control after attachment is complete:

[0425] Operating conditions: Attachment successful, X-ray imaging begins, requiring 30-60 seconds, during which wind conditions may change.

[0426] (i) Maintain pressure dynamically adjusted according to wind load

[0427] Real-time wind speed v wind Gradually increasing from 6 m / s to 10 m / s:

[0428]

[0429] F hold,0 =150N: Static foundation holding pressure (covering self-weight + safety redundancy);

[0430] K w,h =15N⋅s / m: Wind force adjustment coefficient;

[0431] v hold,0 =5m / s: Wind speed threshold.

[0432] When the wind speed is 6 m / s: F hold=150+15×(6-5)=150+15=165N;

[0433] When the wind speed is 10 m / s: F hold =150+15×(10-5)=150+75=225N;

[0434] As the wind speed increases from 6 to 10 m / s, the pressure simultaneously increases from 165 N to 225 N—actively "clamping" the system to counteract the slippage trend as the wind load increases. When the wind speed drops below 5 m / s (e.g., 3 m / s), then v wind <v hold,0 F hold =150N, returning to the baseline value to avoid prolonged overvoltage affecting the conductor. Upgraded from "static friction passive holding" to "pressure dynamic active compensation with wind speed".

[0435] (ii) Deflection compensation (preventing instability in its early stages):

[0436] The attitude sensor continuously monitors the deflection angular velocity of the device. When a deflection trend is detected:

[0437]

[0438] K θ =30N⋅s / rad: Attitude-pressure compensation coefficient;

[0439] Assuming deflection angular velocity =0.005 rad / s;

[0440] Observation period t obs =2s.

[0441] Calculate: F comp =30×0.005×2=0.3N;

[0442] The compensation during the first observation period was only 0.3N, a small amplitude; however, the system continuously monitored and compensated. If the deflection trend widens (e.g....), Increased to 0.05 rad / s (i.e., deflection speeds up), compensation amount increases: F comp =30×0.05×2=3N;

[0443] The amount of compensation is strongly correlated with the deflection trend—the more dramatic the deflection, the stronger the compensation, thus nipping any emerging instability trend in the bud. Ultimately, this maintains pressure: F final =F hold +F comp ;

[0444] If there is a deflection trend in a wind of 10 m / s: F final =225+3=228N;

[0445] Under steady wind conditions of 6 m / s: F final =165 + 0.3 ≈ 165.3 N;

[0446] The entire logic remains the same: Under a level 12 wind load (32.7 m / s), theoretically, when the wind speed is much greater than the threshold: F hold =150+15×(32.7-5)=150+15×27.7=150+415.5=565.5N;

[0447] Maintaining the pressure at approximately 566 N, combined with deflection compensation, is sufficient to keep the device stationary even in extreme winds, and the measured device failure rate can be controlled to within 3%.

[0448] VII. In this embodiment, using a 66kV line, wind speeds of 6-10m / s, and a 30kg detection device as the actual operating conditions, the technical solution of the present invention is achieved as follows:

[0449] From ±5cm drone hovering jitter to ±3mm millimeter-level alignment, relying on visual-inertial tight coupling + advanced prediction + dual-ring approximation;

[0450] The depth of scratches on the conductor surface is less than 0.01mm, achieved through graded pressure application, bounded contact force increments, and feedforward compensation.

[0451] It does not misjudge or fall under gust disturbances, relying on multi-source fusion (Q=0.917) + duration criterion + active maintenance (pressure rises to a maximum of 566N).

[0452] The device failure rate is ≤3% and the attachment success rate is ≥95%, which meets the project's target indicators.

[0453] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic gripping control method under strong electromagnetic environment, characterized in that, Includes the following steps: Before attachment, acquire the relative pose information between the detection device and the target conductor, the motion trend information of the target conductor under environmental load, and the suspension attitude information of the detection device; The relative pose information is processed by multi-source perception fusion, and the motion trend information is processed by trend prediction and environmental load correction. The current relative pose estimation and the corrected trend prediction results are weighted and fused to generate the target attachment point and approximation trajectory. Based on the target gripping point, the driving detection device performs a graded progressive approximation: first, it converges to the first region around the target gripping point; then, it iteratively corrects the relative pose based on real-time updates in each control cycle until it converges to the second region around the target gripping point with higher accuracy than the first region. During the approach process, the attitude of the detection device is smoothly aligned by impedance adjustment, so that the gripping reference surface of the detection device and the axis of the target wire are kept in a set alignment relationship. When the detection device enters the second area and the gripping mechanism contacts the target wire, the contact pressure applied by the gripping mechanism is adjusted in a closed loop based on the real-time gripping force collected as feedback. The environmental load and gravity components are compensated for in advance. At the same time, an upper limit constraint is set on the single pressure increment, and the pressure is only increased when the actual contact pressure change at adjacent times meets the set safety conditions, so that the gripping mechanism can perform graded, non-destructive, and compliant pressing of the target wire. Once the contact pressure reaches the set safety condition, the gripping force information, device posture information, and visual fit information are normalized and weighted to obtain the gripping fit evaluation quantity. When the evaluation quantity continuously meets the stable attachment criterion within the set time, the attachment is considered successful; otherwise, the attachment parameters are adjusted until the criterion is met or an abandonment decision is triggered. After successful attachment, based on environmental load information and device attitude information, an active adjustable holding pressure related to external disturbances is applied to the detection device, and incremental compensation is made for the deflection trend of the detection device caused by external disturbances, so that the detection device maintains attachment under dynamic environmental loads until the detection operation is completed and a safe release is performed.

2. The dynamic gripping control method under strong electromagnetic environment according to claim 1, characterized in that, Multi-source sensing fusion involves tightly coupling and filtering the high-frequency output of inertial sensing with the pose observation of visual sensing: when any sensing channel experiences measurement degradation in a strong electromagnetic environment, the availability of relative pose information is maintained through another sensing channel.

3. The dynamic gripping control method under strong electromagnetic environment according to claim 2, characterized in that: Trend prediction involves weighted extrapolation of the relative pose sequence of the target traverse over a historical period, and then correcting the weighted extrapolation results based on the deviation between the environmental load and the set reference.

4. The dynamic gripping control method under strong electromagnetic environment according to claim 3, characterized in that: Impedance adjustment dynamically shapes the attitude of the detection device around the target conductor's traverse axis into a set inertia-damping-stiffness characteristic, enabling the detection device to gradually adapt and smoothly return to acclimatization under the action of external disturbance torque.

5. The dynamic gripping control method under strong electromagnetic environment according to claim 4, characterized in that: The closed-loop regulation of the contact pressure is a proportional-integral-derivative feedback regulation; the feedforward compensation includes the projection compensation of gravity along the gripping direction under the current attitude of the detection device, as well as the compensation for real-time environmental load.

6. The dynamic gripping control method under strong electromagnetic environment according to claim 5, characterized in that: The normalized weighted summation is achieved by normalizing the gripping force information, device attitude information, and visual fit information to a unified dimension, and then summing them according to a set weight coefficient. The weight coefficient is adjusted based on the reliability of each piece of information under the current working conditions.

7. The dynamic gripping control method under strong electromagnetic environment according to claim 6, characterized in that: The active adjustable holding pressure is dynamically adjusted based on the deviation between the real-time environmental load and the set threshold, so that the holding pressure increases with the increase of external disturbance and decreases with the decrease of external disturbance, so as to suppress slippage and avoid long-term overpressure. The incremental compensation is generated based on the attitude deflection accumulated by the detection device during the set observation period.

8. The dynamic gripping control method under strong electromagnetic environment according to claim 7, characterized in that, It also includes feasibility assessment of the grabbing process: comprehensively analyze the multi-source sensing data, trend prediction results, approximation trajectory and control parameters during the grabbing process. When the assessment shows that the feasibility of grabbing under the current working conditions is lower than the set threshold, a decision to abandon the grabbing is generated and the detection device is controlled to exit the grabbing process and return to the standby state.

9. The dynamic gripping control method under strong electromagnetic environment according to claim 8, characterized in that: During the gripping and holding process, the deviation between the actual value and the set value of the holding parameters of the detection device is monitored in real time. When the deviation exceeds the allowable range, correction is made. After the detection operation is completed, the contact pressure is gradually unloaded, the attitude centering constraint is released, and the detection device is released from the target guide in the reverse order of gripping and pressing.

10. A dynamic gripping control system for strong electromagnetic environments, characterized in that, include: The environmental perception unit is used to acquire the relative pose information between the detection device and the target conductor, the motion trend information of the target conductor under environmental load, and the suspension attitude information of the detection device, and outputs the fused relative pose estimate and the target attachment point. The docking control unit is used to drive the detection device to perform graded and progressive approximation based on the target attachment point, and to perform compliant centering of the device attitude through impedance adjustment. The compliant gripping unit is used to perform closed-loop regulation and feedforward compensation of the contact pressure with real-time gripping force as feedback, and to constrain the single pressure increment and pressure change conditions during the staged pressing process. The state determination unit is used to normalize and weight the gripping force information, device posture information and visual fit information to synthesize the gripping fit evaluation quantity, and determine whether the device has reached a stable gripping state based on the time window criterion. The active holding unit is used to apply an active and adjustable holding pressure and deflection increment compensation to the attached detection device based on environmental load and device attitude information, to maintain the attachment until the detection operation is completed and the device is safely released. The edge computing unit, located locally on the detection device, is used to perform feasibility assessment of attachment and local processing of multi-source data, and to collaborate with the cloud to update the attachment strategy.

Images