Wall-climbing robot active anti-falling system and method based on state prediction and spatial trajectory planning

Abstract

The application discloses a wall-climbing robot active anti-falling system and method based on state prediction and space trajectory planning, and relates to the technical field of robot safety protection. The system comprises at least three active force control modules arranged in an asymmetric non-coplanar manner, a multi-source fusion perception terminal and an anti-falling central controller. The method is executed by the anti-falling central controller: obtaining the spatial coordinates of each force control module and generating a load capacity envelope, and importing a three-dimensional model of a working environment; constructing an extended Kalman filter by fusing UWB and IMU data and predicting the state forward; triggering pre-braking when the predicted state exceeds a safety threshold, planning a collision-free S-shaped buffer trajectory of the robot offsetting the center line of the pipeline in combination with the three-dimensional model; solving optimal tension distribution by using a constrained quadratic programming; and converting the target tension into target torque closed-loop control to execute variable acceleration braking until safe hovering. The application realizes pre-judgment zero-impact anti-falling, avoids secondary collision and improves the ultimate load capacity.

Description

Technical Field

[0001] This invention relates to the fields of robot safety protection, multi-flexible cable parallel robots, and intelligent control technology. Specifically, it relates to an active fall prevention system and method for wall-climbing robots based on state prediction and spatial trajectory planning. This system is applicable to intelligent active fall prevention for robots performing climbing tasks in complex three-dimensional confined spaces (such as curved tailrace pipes of pumped storage power stations, irregularly shaped pressure tanks, and the interior of wind turbine towers). Background Technology

[0002] When wall-climbing robots operate on vertical or irregularly curved surfaces, their adsorption systems (such as magnetic adsorption and negative pressure adsorption) often experience momentary failures due to changes in wall curvature, coating peeling, or oil contamination, leading to falls. Traditional passive speed-differential fall arrestors rely on fixed anchor points directly above the machine and require a certain fall speed (usually greater than 1.5 m / s) to mechanically lock via an internal centrifugal mechanism. These systems suffer from fatal flaws such as being unsuitable for bottom-up climbing scenarios, having severe response lag, and experiencing extremely high impact forces at the moment of locking (easily tearing the robot structure or damaging the wall).

[0003] In recent years, the industry has proposed active fall arrest schemes that utilize the coordinated control of multiple flexible ropes. However, in complex engineering applications, existing active fall arrest technologies still face the following insurmountable bottlenecks: (1) Non-standard layouts caused by environmental constraints are prone to control singularities: The actual work site (such as the tailrace pipe maintenance manhole) is often small and cannot arrange standard equilateral triangle anchor points. The anchor points are usually non-coplanar and asymmetric. Traditional simple tension distribution algorithms based on plane geometry are prone to Jacobian matrix singularities when faced with non-coplanar anchor points, resulting in no solution for tension distribution or instantaneous overload and burnout of a single motor; (2) Sensor physical delay leads to "missing the optimal braking window": The fall is a transient process on the order of milliseconds. Existing multi-source fusion methods mostly use simple threshold judgment, while UWB positioning has a communication delay of tens of milliseconds, and IMU (Inertial Measurement Unit) has integral drift error. Relying solely on instantaneous measurement values ​​to trigger braking often results in the robot having already acquired a large initial falling velocity when the system issues the braking command; (3) Linear braking is prone to secondary collisions in curved and confined spaces: Existing active braking only calculates the upward pulling force. In curved pipes (such as S-shaped wind tunnels and bends), if the robot is only subjected to the upward pulling force during the fall, it is very easy to directly hit the curved pipe wall due to the horizontal inertia, causing secondary damage to the equipment. (4) Constant acceleration braking leads to low end impact and load capacity: Existing flexible braking mostly adopts constant deceleration model, which lacks a smooth transition at the end of the fall, resulting in violent elastic oscillation of the rope at the moment of stopping; and does not make full use of the available fall space at the bottom, resulting in the high peak braking power required by the system.

[0004] Therefore, there is an urgent need for an active fall prevention system and method for wall-climbing robots that can overcome the limitations of coplanar anchor point arrangement, achieve state advance prediction, and plan collision-free flexible buffer trajectories in confined bending spaces, so as to achieve intelligent safety fall prevention with high load and zero impact. Summary of the Invention

[0005] To address the shortcomings of existing technologies, such as non-standard anchor point layouts leading to control singularities, sensor delays causing missed braking windows, linear braking causing secondary collisions, and large impacts at the end of constant acceleration braking, this invention provides an active fall prevention system and method for wall-climbing robots based on state prediction and spatial trajectory planning. It achieves state advance prediction through extended Kalman filtering, solves the singularity problem of tension distribution at non-coplanar anchor points through constrained quadratic programming, and plans a collision-free S-shaped flexible buffer trajectory within a confined bending space, thus achieving intelligent safety fall prevention with high load and zero impact.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an active fall prevention method for a wall-climbing robot based on state prediction and spatial trajectory planning. The method is executed by a fall prevention central controller and includes the following steps: S10. System initialization self-calibration: Obtain the spatial three-dimensional coordinates of at least three active force control modules. The at least three active force control modules are allowed to be arranged asymmetrically and non-coplanarly in three-dimensional space. Each active force control module is connected to the wall-climbing robot via a flexible rope. Calculate the system load capacity limit envelope under the current non-coplanar configuration based on the spatial three-dimensional coordinates and import it into the three-dimensional digital model of the working environment. S20, State Advance Prediction: By fusing low-frequency UWB positioning data with high-frequency IMU data to construct an extended Kalman filter, the current pose of the wall-climbing robot is calculated, and the predicted state for future periods is obtained by extrapolating forward through the state transition matrix. S30, Dynamic safety trajectory planning: When the predicted state exceeds the safety threshold, pre-braking is triggered. The maximum available safe fall space is calculated in combination with the three-dimensional digital model. If a curved pipe wall is detected below, the radial offset is calculated to generate a collision-free three-dimensional S-shaped buffer trajectory that smoothly guides the wall-climbing robot to offset towards the center line of the pipe. S40. Optimization of tension distribution with constraints: Calculate the instantaneous three-dimensional braking acceleration vector required by the wall-climbing robot based on the target buffer trajectory, construct a mechanical equilibrium equation including the direction vectors of non-coplanar anchor points, and use a quadratic programming algorithm to solve for the optimal tension distribution under the premise of satisfying the maximum torque constraints of each motor and the constraint that the rope cannot be compressed. S50, Closed-loop Flexible Braking: The target tension obtained by solving is combined with the real-time rope drum radius to convert it into target torque. Combined with the feedback from the dynamic tension sensor, PID torque closed-loop control is performed to drive the servo motor to perform variable acceleration braking until the wall-climbing robot comes to a safe stop.

[0007] Secondly, the present invention provides an active fall prevention system for a wall-climbing robot based on state prediction and spatial trajectory planning, including a non-standard multi-point anchoring base station, a multi-source fusion sensing terminal, and a fall prevention central controller. The fall prevention central controller includes a state prediction unit, a spatial trajectory planning unit, and a tension distribution optimization unit, which work together to execute the above-mentioned active fall prevention method for the wall-climbing robot.

[0008] Compared with the prior art, the present invention has the following significant advantages: (1) Breaking geometric limitations and highly adaptable to engineering: The introduction of a quadratic programming tension distribution algorithm completely solves the singularity problem caused by non-coplanar and asymmetric anchor points, allowing the anti-fall module to be fixed in place at any level of the maintenance port edge on site. (2) "Predictive" zero-delay braking: Based on the extended Kalman filter, the advanced prediction mechanism intervenes in advance when the robot experiences a slight slip and has not yet fully formed the "precursor of weightlessness", smoothing out the physical response delay of the traditional system (about 100ms) and greatly shortening the total fall distance. (3) Eliminate secondary collisions in confined spaces: Creatively introduce spatial envelope control, dynamically superimpose radial compensation force in vertical braking force, forcibly pull the robot back to the safe center of the pipeline, and avoid collision damage inside complex irregular structures; (4) Improve the system's ultimate load and equipment lifespan: Based on the S-shaped variable acceleration control with remaining space, peak shaving and valley filling significantly reduce the peak impact on the motor and reducer during braking (the peak torque is reduced by about 40%), which greatly improves the system's maximum anti-fall load capacity under the same hardware conditions. (5) Multi-level safety redundancy: The dynamic tension sensor provides real tension feedback at the kilohertz level to form a torque closed loop. The centrifugal mechanical locker forcibly intervenes and locks when the control system is completely powered off and the rope pull-out speed exceeds 2m / s, forming a double bottom line protection of the control layer and the mechanical layer. Attached Figure Description

[0009] Figure 1This is a flowchart illustrating the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning in this application; Figure 2 This is a diagram showing the overall structure and topology of the active fall prevention system in a non-standard curved space, based on the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning in this application. Figure 3 This is a schematic diagram of the S-shaped target trajectory planning principle considering wall collision constraints in a confined space, based on the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning in this application. Figure 4 This is a flowchart of the quadratic programming (QP) asymmetric tension allocation algorithm of the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning in this application; Figure 5 This is a structural block diagram of the first embodiment of the active fall prevention system for wall-climbing robots based on state prediction and spatial trajectory planning in this application; Figure 6 This is a physical diagram of the fall protection system of the first embodiment of the active fall protection system for wall-climbing robots based on state prediction and spatial trajectory planning in this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0011] This application provides an active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning in this application.

[0012] Reference Figure 2 , Figure 2This diagram illustrates the overall structure and topology of the first embodiment of the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning, within a non-standard curved space. The embodiment uses an S-shaped curved tailrace pipe of a pumped-storage power station as the operational scenario. The non-standard multi-point anchoring base station of the fall prevention system is deployed at the maintenance port at the bottom of the tailrace pipe. Due to obstacles such as valves and pipelines around the maintenance port, the three active force control modules A, B, and C cannot be arranged on the same horizontal plane, forming a non-standard spatial configuration with coordinates P1(x1,y1,z1), P2(x2,y2,z2), and P3(x3,y3,z3). Each active force control module comprises three hardware layers: the drive layer consists of a high-power AC servo motor (with 3 times overload capacity) and a low-backlash planetary reducer; the sensing layer consists of an absolute encoder mounted at the motor's tail (for calculating the released rope length and the real-time drum radius) and a dynamic tension sensor mounted at the rope guide wheel (providing kilohertz-level real tension feedback); the safety layer is a purely mechanical centrifugal ratchet and pawl mechanism. When the control system is completely powered off and the rope extraction speed exceeds 2 m / s, the mechanical locking device forcibly intervenes and locks the bottom line for protection. A multi-source fusion sensing terminal is mounted on the wall-climbing robot body, including a UWB tag module, an IMU module, and a laser or ultrasonic array. The characteristics of the multi-source sensors are shown in Table 1.

[0013] Table 1. Sensor characteristics of multi-source fusion sensing terminal In this embodiment, the active fall prevention method for wall-climbing robots based on state prediction and spatial trajectory planning includes the following steps: Step S10: System initialization, self-calibration, and spatial envelope generation.

[0014] During system initialization, the non-standard three-dimensional spatial coordinates P of each active force control module are acquired. i (x_i, y_i, z_i) (i = 1, 2, 3), the coordinates can be obtained through laser ranging calibration or manual measurement and input; based on the spatial three-dimensional coordinates of each active force control module, the rope direction and the rated maximum tension of each motor, calculate the system load capacity limit envelope diagram under the current non-coplanar configuration, and calibrate the maximum load range that can be protected under the current anchor point layout; at the same time, import the three-dimensional digital model (CAD model) of the working environment, the three-dimensional digital model is used to describe the bending direction of the tailrace pipe, the coordinates of the pipe wall centerline and the clearance size, and provide environmental constraints for subsequent trajectory planning.

[0015] Step S20: Multi-source information fusion and state advance prediction based on extended Kalman filter.

[0016] To eliminate the shortcomings of both UWB positioning (which has a delay of approximately 30ms and an update rate of 20Hz) and IMU (high frequency 400Hz but suffers from integral drift), the fall arrestor central controller incorporates an extended Kalman filter (EKF) prediction model. The algorithm logic architecture is based on... Figure 2 As shown.

[0017] (1) Definition of state variables: The system state vector is defined as follows: (1) in, These represent the robot's three-dimensional positions, They represent three-dimensional velocities, These represent three-dimensional accelerations.

[0018] (2) State transition equation (prediction step): The system state is updated using high-frequency IMU data. Let the sampling time interval be... : (2) in, Let k be the prior state estimate at time k. Let F be the posterior state estimate at time K1, and F be the Newtonian kinematic state transition matrix. This represents process noise. The system simultaneously updates the covariance matrix: (3) in, Let Q be the prior estimate of the covariance matrix, and let Q be the process noise covariance matrix.

[0019] (3) Measurement Update Equation (Correction Step): When low-frequency UWB coordinate data is received, correction is performed. The UWB observation vector is: (4) Calculate the Kalman gain: (5) Update the state vector: (6) in, For Kalman gain, Let H be the observation vector at time k, H be the observation matrix (only location information is extracted), and R be the measurement noise covariance matrix.

[0020] (4) State advance prediction and triggering: At the current time k, the controller uses the state transition matrix F to extrapolate N cycles forward (e.g., extrapolate the next 50ms): (7) in, This is the predicted state vector obtained by extrapolating N periods backward from time k. If the vertical acceleration component in the predicted state vector... When the free fall threshold approaches (e.g., below -8.0 m / s²), the system determines that the predicted state has exceeded the safety threshold and will issue pre-tensioning and braking commands in advance before the "actual fall" occurs. This allows the system to intervene in advance during the "precursor of weightlessness" stage when the robot experiences slight slippage and has not yet fully formed a free fall, thus eliminating the physical response delay of about 100 ms in the sensing and communication links.

[0021] Step S30: Dynamic safety trajectory planning and S-shaped speed control in confined space.

[0022] Once braking is triggered, the system extracts the robot's initial state (initial position). Initial velocity Combining the imported 3D digital model of the tailrace pipe, envelope trajectory planning is performed. The trajectory planning principle refers to... Figure 3 As shown.

[0023] (1) Calculation of buffer space and radial offset: Scan the three-dimensional digital model downwards to calculate the maximum available clearance. And set a safe buffer distance: (8) That is, the maximum available safe fall space is calculated based on a preset safety factor of 0.8 and the maximum available clearance. If the pipe below bends (i.e., the pipe wall centerline coordinates ( (Due to a deviation), the system calculates a normal guiding force, and the target hovering point... The geometric center of the pipe is set at the height of the final fall. This requires that the braking acceleration is no longer just vertical, but a full three-dimensional vector: (9) in, Let be the instantaneous three-dimensional braking acceleration vector at time t. The horizontal radial component is used to guide the robot to deflect towards the centerline of the pipe. The vertical component is used to provide upward braking force.

[0024] (2) S-shaped variable acceleration curve: To eliminate the oscillation at the end of braking, trajectory planning with limited jerk (Jerk, denoted as j) is adopted. Taking vertical deceleration along the Z-axis as an example, the speed... and acceleration Satisfying piecewise polynomial smooth transition: In the initial stage (setup period), the acceleration rapidly increases from -g to the maximum positive braking acceleration. Mid-term (constant period), maintained Stable deceleration; in the final stage (smooth end), as the speed approaches 0, the acceleration also smoothly decays to 0 according to an S-shaped curve, achieving a true zero-impact "soft landing".

[0025] Step S40: Asymmetric tension distribution and constrained quadratic programming optimization.

[0026] This step addresses the core pain point caused by "non-standard anchor points." The algorithm flow is as follows: Figure 4 As shown.

[0027] (1) Establish a three-dimensional spatial mechanical model: Let the unit direction vector of the i-th rope be: (10) in, , , These are the direction angles of the i-th rope relative to the three coordinate axes, obtained from the robot's real-time position and the corresponding anchor point coordinates. This is to achieve the planned instantaneous acceleration. The system's force equilibrium equations are: (11) Where U is the direction matrix composed of the unit direction vectors of each rope, and T is the tension vector composed of the tension of each rope. To determine the desired resultant force, we expand the above equation along three coordinate axes: (12) in, , , These are the three components of the unit direction vector of the first rope, and so on for the rest; , , The values ​​represent the tension of the three ropes; m is the robot's mass; and the vertical component includes the gravitational acceleration of 9.8 m / s².

[0028] (2) Quadratic Programming Optimization Solution: In the case of non-coplanarity, the aforementioned direction matrix U is prone to ill-conditioned behavior (maximum condition number). Direct inversion can lead to negative tension in a rope (the rope cannot be compressed) or unrealizable tension of tens of thousands of Newtons. This invention transforms it into a quadratic programming (QP) problem with inequality constraints, with the objective function being to minimize energy consumption and tension extrema: (13) Where W is a diagonal weight matrix, the weights can be dynamically adjusted according to the current temperature and margin of each motor. The equality constraint is: (14) That is, to make the greatest effort to satisfy the desired trajectory; the inequality constraints are: (15) in, Set to 50N (to keep the rope slightly tensioned and prevent it from slipping off the trough). The rated maximum tension of the motor is used. The controller calls the embedded QP solver (such as the effective set method) to calculate the optimal tension distribution that satisfies all the above physical constraints within a 1ms control cycle. .

[0029] Step S50: Torque closed loop and underlying execution.

[0030] Calculate the target tension Then, the system needs to convert it into execution commands for the motor. Considering that the rope is laid out in multiple layers on the drum, the effective radius of the drum... It is the length of the rope released. dynamic function The formula for calculating the target motor torque is: (16) in, Let the target torque of the i-th motor be the latter term. To compensate for the rotational inertia of the motor rotor and the reducer, a dynamic feedforward torque is provided. For the equivalent moment of inertia, This represents the angular velocity of the motor.

[0031] Subsequently, the target torque command By combining the real feedback values ​​from the hardware tension sensor, a closed loop is formed through the PID control algorithm, and the data is sent to each servo driver via the EtherCAT industrial Ethernet bus to achieve millisecond-level flexible and robust takeover of the falling robot until the robot safely hovers at the target hovering point.

[0032] This invention also provides an active fall prevention system for wall-climbing robots based on state prediction and spatial trajectory planning, referring to... Figure 5 The system includes a non-standard multi-point anchored base station 10, a multi-source fusion sensing terminal 20, and a fall protection central controller 30.

[0033] The non-standard multi-point anchoring base station includes at least three independently arranged main power control modules. The main power control modules are allowed to be arranged asymmetrically and non-coplanarly in three-dimensional space. Each main power control module includes a servo motor, an absolute encoder, a dynamic tension sensor, and a redundant centrifugal mechanical lock. Each main power control module is connected to the wall-climbing robot via a flexible rope. The multi-source fusion sensing terminal is mounted on the wall-climbing robot body and includes a UWB tag module for acquiring absolute spatial coordinates, an IMU module for acquiring high-frequency three-dimensional acceleration and angular velocity, and a laser or ultrasonic array for detecting near-field wall distance. The fall protection central controller is communicatively connected to the non-standard multi-point anchoring base station and the multi-source fusion sensing terminal. It includes a state prediction unit, a spatial trajectory planning unit, and a tension distribution optimization unit, which respectively execute the relevant functions of steps S20, S30, and S40 to S50. A physical diagram of the fall protection system is shown below. Figure 6 As shown.

[0034] By acquiring the spatial coordinates of each force control module and generating a load capacity envelope, the system imports the data into a 3D model of the operating environment. An extended Kalman filter is constructed by fusing UWB and IMU data, and the predicted state is extrapolated forward. When the predicted state exceeds a safety threshold, pre-braking is triggered. A collision-free S-shaped buffer trajectory is planned and guided towards the centerline of the pipeline using the 3D model. Constrained quadratic programming is employed to solve for the optimal tension distribution. The target tension is converted into a target torque, and a closed-loop control system is used to control the servo motor to perform variable acceleration braking until a safe stop is achieved. This results in predictive zero-impact fall prevention, avoiding secondary collisions and improving ultimate load capacity.

[0035] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this application. In practical applications, those skilled in the art can select some or all of it to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.

[0036] In addition, for technical details not described in detail in this embodiment, please refer to the method for active fall prevention of wall-climbing robots based on state prediction and spatial trajectory planning provided in any embodiment of this application, which will not be repeated here.

[0037] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0038] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0039] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of this application. The above are only preferred embodiments of this application and do not limit the patent scope of this application. All equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A method for active fall prevention of a wall-climbing robot based on state prediction and spatial trajectory planning, characterized in that, The method is executed by the fall protection central controller and includes: S10. System initialization self-calibration: Obtain the spatial three-dimensional coordinates of at least three active force control modules. The at least three active force control modules are allowed to be arranged asymmetrically and non-coplanarly in three-dimensional space. Each active force control module is connected to the wall-climbing robot via a flexible rope. Calculate the system load capacity limit envelope under the current non-coplanar configuration based on the spatial three-dimensional coordinates and import it into the three-dimensional digital model of the working environment. S20, State Advance Prediction: By fusing low-frequency UWB positioning data with high-frequency IMU data to construct an extended Kalman filter, the current pose of the wall-climbing robot is calculated, and the predicted state for future periods is obtained by extrapolating forward through the state transition matrix. S30, Dynamic safety trajectory planning: When the predicted state exceeds the safety threshold, pre-braking is triggered. The maximum available safe fall space is calculated in combination with the three-dimensional digital model. If a curved pipe wall is detected below, the radial offset is calculated to generate a collision-free three-dimensional S-shaped buffer trajectory that smoothly guides the wall-climbing robot to offset towards the center line of the pipe. S40. Optimization of tension distribution with constraints: Calculate the instantaneous three-dimensional braking acceleration vector required by the wall-climbing robot based on the target buffer trajectory, construct a mechanical equilibrium equation including the direction vectors of non-coplanar anchor points, and use a quadratic programming algorithm to solve for the optimal tension distribution under the premise of satisfying the maximum torque constraints of each motor and the constraint that the rope cannot be compressed. S50, Closed-loop Flexible Braking: The target tension obtained by solving is combined with the real-time rope drum radius to convert it into target torque. Combined with the feedback from the dynamic tension sensor, PID torque closed-loop control is performed to drive the servo motor to perform variable acceleration braking until the wall-climbing robot comes to a safe stop.

2. The active fall prevention method for wall-climbing robots according to claim 1, characterized in that: In step S10, each of the active force control modules includes a servo motor, an absolute encoder for calculating the released rope length and the real-time radius of the drum, a dynamic tension sensor for providing real tension feedback, and a redundant centrifugal mechanical lock. The system load capacity limit envelope is calculated based on the spatial three-dimensional coordinates of each active force control module, the rope direction, and the rated maximum tension of each motor, and is used to calibrate the maximum load range that can be protected under the current non-coplanar configuration.

3. The active fall prevention method for wall-climbing robots according to claim 1, characterized in that: In step S20, the system state vector of the extended Kalman filter consists of the three-dimensional position, three-dimensional velocity, and three-dimensional acceleration of the wall-climbing robot; in the prediction step, the system state vector is updated using high-frequency IMU data at sampling time intervals, and the covariance matrix is ​​updated synchronously; in the correction step, when low-frequency UWB positioning data is received, the Kalman gain is calculated, and the system state vector is updated according to the deviation between the UWB observed coordinates and the predicted pose; the update rate of the UWB positioning data is lower than the update rate of the IMU data.

4. The active fall prevention method for wall-climbing robots according to claim 3, characterized in that: In step S20, the forward extrapolation is to use the state transition matrix to extrapolate N control cycles from the current moment, and the future time period is 50 milliseconds of forward extrapolation; when the vertical acceleration component in the predicted state vector is lower than -8.0 meters per second squared and approaches the free fall threshold, it is determined that the predicted state has exceeded the safety threshold, and pre-tensioning and braking commands are issued in advance before the actual fall occurs.

5. The active fall prevention method for a wall-climbing robot according to claim 1, characterized in that: In step S30, the maximum available safe fall space is determined by multiplying the maximum available clearance obtained by scanning the three-dimensional digital model downward from the current position of the self-climbing robot by a preset safety factor. The recommended value of the preset safety factor is 0.

8. The target hovering point is set at the geometric center of the pipe at the height of the fall endpoint. The instantaneous three-dimensional braking acceleration vector is a full three-dimensional vector containing horizontal radial components and vertical components. The horizontal radial component is used to guide the wall-climbing robot to deviate towards the centerline of the pipe.

6. The active fall prevention method for a wall-climbing robot according to claim 5, characterized in that: In step S30, the collision-free three-dimensional S-shaped buffer trajectory adopts a variable acceleration curve with limited jerk, including an initial segment, a middle segment, and a final segment: the initial segment is the setup period, where the braking acceleration climbs from the free fall acceleration to the maximum braking acceleration; the middle segment is the constant period, where the maximum braking acceleration is maintained for stable deceleration; the final segment is the smooth termination period, where the braking acceleration smoothly decays to zero according to the S-shaped curve as the velocity approaches zero, achieving a zero-impact soft landing.

7. The active fall prevention method for a wall-climbing robot according to claim 1, characterized in that: In step S40, the unit direction vector of each flexible rope is obtained by the real-time position of the wall-climbing robot and the coordinates of the anchor point of the corresponding active force control module; the mechanical equilibrium equation is that the resultant force of the tension of each rope along the corresponding unit direction vector is equal to the product of the mass of the wall-climbing robot and the instantaneous three-dimensional braking acceleration vector, wherein the vertical component is superimposed with gravitational acceleration.

8. The active fall prevention method for a wall-climbing robot according to claim 7, characterized in that: In step S40, the objective function of the quadratic programming algorithm is to minimize the weighted sum of squares of the tension of each rope, so as to simultaneously constrain energy consumption and tension extrema. The weight matrix is ​​a diagonal matrix and can be dynamically adjusted according to the current temperature and margin of each motor. The equality constraint of the quadratic programming algorithm is the mechanical equilibrium equation, and the inequality constraint is that the target tension of each rope is not less than the preset minimum tension and not greater than the rated maximum tension of the motor. The recommended value of the preset minimum tension is 50 N, which is used to keep the rope slightly tensioned to prevent derailment. The quadratic programming algorithm is solved using the effective set method, and the solution control period is 1 millisecond.

9. The active fall prevention method for a wall-climbing robot according to claim 1, characterized in that: In step S50, the effective radius of the rope drum is a dynamic function of the rope release length; the target torque is composed of the product of the target tension and the effective radius, plus a dynamic feedforward torque that compensates for the rotational inertia of the motor rotor and the reducer; the target torque command, combined with the real feedback value of the dynamic tension sensor, forms a closed loop through a PID control algorithm and is sent to each servo driver via an industrial Ethernet bus; the method also includes bottom-line protection: real-time monitoring of the rope pull-out speed, and when the control system is powered off and the rope pull-out speed exceeds 2 meters per second, triggering the centrifugal mechanical lock to forcibly lock.

10. An active fall prevention system for a wall-climbing robot based on state prediction and spatial trajectory planning, characterized in that, include: The non-standard multi-point anchoring base station includes at least three independently arranged main force control modules. The main force control modules are allowed to be arranged asymmetrically and non-coplanarly in three-dimensional space. Each main force control module includes a servo motor, an absolute encoder, a dynamic tension sensor, and a redundant centrifugal mechanical lock. Each main force control module is connected to the wall-climbing robot via a flexible rope. The multi-source fusion sensing terminal, mounted on the wall-climbing robot, includes a UWB tag module for acquiring absolute spatial coordinates, an IMU module for acquiring high-frequency three-dimensional acceleration and angular velocity, and a laser or ultrasonic array for detecting near-field wall distance. The fall prevention central controller is communicatively connected to the non-standard multi-point anchoring base station and the multi-source fusion sensing terminal, and includes a state prediction unit, a spatial trajectory planning unit and a tension distribution optimization unit; the state prediction unit is used to fuse low-frequency UWB positioning data and high-frequency IMU data to construct an extended Kalman filter, calculate the current pose of the wall-climbing robot, and obtain the predicted state for future time periods through forward extrapolation of the state transition matrix. The spatial trajectory planning unit is used to trigger pre-braking when the predicted state exceeds the safety threshold. It calculates the maximum available safe fall space by combining the three-dimensional digital model of the working environment and generates a collision-free three-dimensional S-shaped buffer trajectory that smoothly guides the wall-climbing robot to deviate towards the center line of the pipe. The tension distribution optimization unit is used to calculate the instantaneous three-dimensional braking acceleration vector based on the target buffer trajectory, construct a mechanical equilibrium equation that includes the direction vectors of non-coplanar anchor points, and use a quadratic programming algorithm to solve for the optimal tension distribution under the premise of satisfying the maximum torque constraints of each motor and the non-compression constraint of the rope. The target tension is combined with the real-time rope drum radius to convert it into target torque. Combined with the feedback of the dynamic tension sensor, PID torque closed-loop control is performed to drive the servo motor to perform variable acceleration braking until the wall-climbing robot is safely suspended.

Images