A raw water pipeline hydraulic dredging robot and an adaptive control method thereof

Abstract

The present application relates to the technical field of pipeline dredging, and particularly relates to a raw water pipeline hydraulic dredging robot and a self-adaptive control method thereof, which comprises a mobile platform, a control system and a multi-joint mechanical arm carried on the mobile platform, and the end of the multi-joint mechanical arm is provided with a dredging head comprising a nozzle and a touch sensor. Real-time tactile information when the dredging head interacts with silt is collected by the touch sensor, and the internal parameters of a preset tactile digital twin model are updated in the control system based on the real-time tactile information and a current control instruction. The predicted control results of a plurality of candidate control instructions in future time periods are obtained through the updated preset tactile digital twin model. A target control instruction is selected from the plurality of predicted control results according to the plurality of predicted control results. The dredging action of the multi-joint mechanical arm and the nozzle is controlled according to the target control instruction, effectively solving the problems of low dredging operation efficiency and low adaptability in the control of the hydraulic dredging robot.

Description

Technical Field

[0001] This invention relates to the field of pipeline dredging technology, specifically to a hydraulic dredging robot for raw water pipelines and its adaptive control method. Background Technology

[0002] Raw water pipelines, as a crucial infrastructure of urban water supply systems, suffer from severe water conveyance capacity and water quality due to silt buildup, necessitating regular cleaning and maintenance. Currently, hydraulic dredging robots are commonly used to replace manual labor, thereby improving dredging efficiency and operational safety. The mainstream control methods for these robots often employ preset programs or simple feedback control strategies. This involves pre-planning the robotic arm's movement trajectory and nozzle parameters based on a pipeline map, or adjusting fixed rules during operation by detecting single pieces of information such as silt thickness using sensors.

[0003] However, due to the complex internal environment of raw water pipelines, silt often exhibits non-uniform and time-varying distribution at different locations in the pipeline. Existing control methods lack the ability to perceive and model the interaction between the sludge removal head and the silt. Their fixed control logic or parameters are difficult to adapt to this complex and ever-changing work object in real time. As a result, a single control command cannot maintain optimal sludge removal efficiency in a changing environment, often leading to insufficient sludge removal in certain areas or energy waste, which restricts the improvement of the overall efficiency and adaptability of sludge removal operations. Summary of the Invention

[0004] To address the technical problems of low dredging efficiency and low adaptability in existing hydraulic dredging robots, this application provides a hydraulic dredging robot for raw water pipelines and its adaptive control method.

[0005] The technical solution provided in this application for a hydraulic dredging robot for raw water pipelines and its adaptive control method is as follows: A hydraulic sludge removal robot for raw water pipelines includes a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is equipped with a sludge removal head, which includes a nozzle and a touch sensor. Among them, the tactile sensor is used to sense the interaction with the silt.

[0006] This application also provides an adaptive control method for a hydraulic dredging robot for raw water pipelines. This adaptive control method is applied to the aforementioned hydraulic dredging robot for raw water pipelines, which includes a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is equipped with a dredging head, which includes a nozzle and a touch sensor. The adaptive control method includes: After collecting real-time tactile information when the dredging head interacts with the sludge using a tactile sensor, the internal parameters of the preset tactile digital twin model are updated in the control system based on the real-time tactile information and the current control commands. By using the updated preset tactile digital twin model, predictive control results for various candidate control commands in future time periods are obtained; From a variety of predictive control results, the candidate control command corresponding to the predictive control result that conforms to the preset control result is selected as the target control command; The sludge removal actions of the multi-joint robotic arm and nozzle are controlled according to the target control instructions.

[0007] Furthermore, the control system stores the dredging water jet control signal, the real-time motion status information of the multi-joint robotic arm, the pipe structure of the raw water pipeline, and the real-time pose of the hydraulic dredging robot. The steps for collecting real-time tactile information when the dredging head interacts with the sludge via tactile sensors include: The initial tactile signal is collected by a tactile sensor when the dredging head interacts with the sludge, and the initial tactile signal is compensated and filtered based on the dredging water jet control signal to obtain the first tactile information. Based on real-time motion state information, the first tactile information is decoupled from motion to obtain the second tactile information; Based on the pipe structure and real-time pose, the second tactile information is mapped onto the pipe spatial coordinate system to obtain real-time tactile information.

[0008] Furthermore, based on the pipe structure and real-time pose, the steps to map the second tactile information onto the pipe spatial coordinate system to obtain real-time tactile information include: The real-time pose of the hydraulic dredging robot is obtained through its spatial position and attitude measurement unit. Based on real-time pose, the nominal position and normal vector of the tactile sensor under the pipe structure are calculated. Based on the nominal position and normal vector, establish a rigid transformation relationship between the sensor coordinate system and the pipeline space coordinate system; Based on the rigid transformation relationship, coordinate transformation and vector decomposition are performed on the second tactile information, and the second tactile information is mapped to the pipeline space coordinate system to obtain the interaction forces that the sludge removal head experiences in the axial, radial and circumferential directions along the pipeline structure. Among them, the real-time tactile information is the interactive force that the sludge removal head experiences in the axial, radial and circumferential directions along the pipeline structure.

[0009] Furthermore, the steps for updating the internal parameters of the preset tactile digital twin model in the control system based on real-time tactile information and current control commands include: Input the current control command into the preset tactile digital twin model to obtain the model's predicted tactile information; By comparing the tactile information predicted by the model with the real-time tactile information, the tactile residual vector is obtained; Based on the tactile residual vector, sensitivity analysis is performed on the parameters in the preset tactile digital twin model to screen out the internal parameters to be updated. Based on the internal parameters to be updated and the sensitivity corresponding to each internal parameter, and combined with preset adjustment compensation and preset forgetting factor, the internal parameters to be updated are updated to obtain the updated preset tactile digital twin model.

[0010] Furthermore, the steps for obtaining the predictive control results of various candidate control commands in future time periods through the updated preset tactile digital twin model include: Based on the current control command and historical control commands, generate multiple candidate control commands; Each candidate control command is sequentially input into the updated preset tactile digital twin model, and forward rolling simulation is performed to obtain the tactile response prediction sequence corresponding to each candidate control command in the future time period. Calculate the uncertainty measure corresponding to each tactile response prediction sequence at each prediction step; The tactile response prediction sequence and uncertainty metric corresponding to each candidate control command are integrated to form the predictive control result corresponding to each candidate control command.

[0011] Furthermore, the preset control results include preset pipeline safety constraints, preset motion state constraints, and preset comprehensive evaluation values. The step of selecting the candidate control command corresponding to the predicted control result that conforms to the preset control results from multiple predictive control results as the target control command includes: From the tactile response prediction sequences included in each predictive control result, the tactile feature vector and the predicted dredging efficiency are obtained by analysis; The tactile feature vectors are compared with the preset pipeline safety constraints and preset motion state constraints respectively, and the first preset control result corresponding to the tactile feature vectors that meet the preset pipeline safety constraints and preset motion state constraints is selected. Based on the predicted dredging efficiency and uncertainty measure corresponding to the first preset control results, the comprehensive evaluation value is calculated. The candidate control command corresponding to the first preset control result whose comprehensive evaluation value is greater than the preset comprehensive evaluation value is determined as the target control command.

[0012] Furthermore, according to the target control instructions, the steps for controlling the sludge removal actions of the multi-joint robotic arm and nozzle include: The target control command is analyzed to obtain the desired end effector trajectory of the multi-joint robotic arm and the desired injection parameters of the nozzle. After converting the desired end-effector trajectory into a sequence of desired angles for each joint in the multi-joint robotic arm, a coordinated control signal sequence is generated by combining it with the desired injection parameters. Based on the coordinated control signal sequence, the hydraulic actuators of each joint and the adjustment mechanism of the nozzle are driven to perform the dredging action.

[0013] Furthermore, the steps following the dredging operation include: The execution status of the dredging action is monitored by the joint encoder of the multi-joint robotic arm and the pressure sensor of the nozzle, and status feedback information is obtained. Based on the state feedback information and the cooperative control signal sequence, the execution deviation of the current control cycle is calculated, and the execution deviation is used to compensate and correct the cooperative control signal sequence of the next control cycle.

[0014] Beneficial effects achieved: This application provides an adaptive control method for a hydraulic dredging robot for raw water pipelines. The hydraulic dredging robot includes a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is equipped with a dredging head, which includes a nozzle and a tactile sensor. The adaptive control method includes: collecting real-time tactile information when the dredging head interacts with sludge through the tactile sensor; updating the internal parameters of a preset tactile digital twin model in the control system based on the real-time tactile information and the current control command; obtaining predictive control results for multiple candidate control commands in future time periods through the updated preset tactile digital twin model; selecting the candidate control command corresponding to the predictive control result that conforms to the preset control result from the multiple predictive control results as the target control command; and controlling the dredging actions of the multi-joint robotic arm and the nozzle according to the target control command.

[0015] In this application, real-time tactile information acquired through a tactile sensor is used to establish a perception of the sludge state, and a closed-loop adaptive control loop of "perception-prediction-optimization-execution" is constructed based on this. Specifically, firstly, the internal parameters of a preset tactile digital twin model are updated online based on real-time tactile information and current control commands, enabling the preset tactile digital twin model to reflect the current interaction state between the dredging head and the sludge. This solves the problem of mismatch between fixed models or parameters caused by the time-varying and non-uniform characteristics of sludge, providing a realistic and reliable interactive foundation for adaptive control. Secondly, using the updated preset tactile digital twin model, predictive simulations are performed on the predictive control results of various candidate control commands in future time periods, thereby identifying potentially more efficient or safer dredging action schemes in the upcoming time period, changing the existing approach that relies solely on... The system overcomes the limitations of reacting instantly to current or past information. It then selects the candidate control command that best matches the preset control result from multiple predictive control results as the final target control command to be executed. This ensures that each selected control action is the optimal or near-optimal solution for the target performance index under the current cognitive environment, thereby improving the dredging efficiency of a single dredging operation. Finally, the system drives the multi-joint robotic arm and nozzle to perform dredging actions according to the target control command, forming a complete closed loop. Since the control command for the next cycle will be re-optimized and generated based on new tactile information and an updated model, the entire system can continuously perceive environmental changes, update its internal cognitive model, and adjust its control strategy in real time. This achieves efficient, autonomous, and continuous adaptation to complex and variable sludge environments, fundamentally improving the overall efficiency of dredging operations and the adaptability of the control system. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the modules of the hydraulic dredging robot for raw water pipelines in this application; Figure 2 This is a flowchart illustrating the steps of an adaptive control method for a hydraulic dredging robot for raw water pipelines, as described in this application. 10. Mobile platform; 20. Control system; 30. Multi-joint robotic arm; 40. Dredging head; 50. Nozzle; 60. Touch sensor. Detailed Implementation

[0017] The following combination Figure 1 and Figure 2 This application will be described in further detail.

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described 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 the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back, etc.), the directional indications are only used to explain the relative positional relationship and movement of the components in a specific posture. If the specific posture changes, the directional indications will also change accordingly.

[0020] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the use of "and" or "and / or" throughout the text includes three parallel solutions. For example, "A and B" includes solution A, solution B, or a solution that simultaneously satisfies A and B. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

[0021] This application discloses a hydraulic dredging robot for raw water pipelines.

[0022] Please refer to Figure 1 The hydraulic dredging robot for raw water pipelines proposed in this embodiment includes a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is equipped with a dredging head, which includes a nozzle and a touch sensor. The touch sensor is used to sense interaction with the sludge.

[0023] In the hydraulic dredging robot for raw water pipelines proposed in this embodiment, the included mobile platform serves as the mobile load-bearing foundation for the hydraulic dredging robot, responsible for its movement and positioning within the pipeline, providing spatial accessibility for the dredging action; the included control system serves as the core processing and decision-making unit, responsible for receiving sensor signals, executing control algorithms, and generating drive commands to coordinate the overall operation of the hydraulic dredging robot; the multi-joint robotic arm mounted on the mobile platform serves as the core work execution mechanism, and its joint degrees of freedom allow the dredging head at the end to reach and conform to different positions on the inner wall of the raw water pipeline, achieving multi-angle coverage of complex siltation areas; the dredging head at the end of the robotic arm is the terminal that directly interacts with the silt, and its included nozzles generate high-pressure water jets to impact, cut, and transport the silt, serving as the direct action component for completing the dredging function; the integrated tactile sensor is used to perceive in real time the mechanical interaction information generated between the dredging head and the silt under the impact of the water jet, such as reaction force and pressure distribution, thereby providing the control system with feedback signals describing the physical state of the work object and its contact with the environment.

[0024] Based on the collaborative work between the components, the hydraulic dredging robot can dynamically adjust the posture of the multi-joint robotic arm and the jet parameters of the nozzle through the control system based on the interactive information provided by the touch sensor. This enables precise, efficient and adaptive dredging operations on unstructured pipeline sludge with varying physical properties, which can actively adapt to environmental changes.

[0025] The tactile sensor can be a six-dimensional force / torque sensor from the ATI Industrial Automation - Delta series.

[0026] This application also discloses an adaptive control method for a hydraulic dredging robot for raw water pipelines.

[0027] Please refer to Figure 2 The adaptive control method for a hydraulic dredging robot for raw water pipelines proposed in this embodiment includes steps S10 to S40: Step S10: After collecting real-time tactile information when the sludge cleaning head interacts with the sludge through the tactile sensor, the internal parameters of the preset tactile digital twin model are updated in the control system based on the real-time tactile information and the current control command.

[0028] The purpose of this step is to build a cognitive model for the hydraulic dredging robot that can reflect the current dredging interaction process, so as to solve the fundamental problem of the static model being out of touch with the control reality due to the non-uniform and time-varying physical properties of silt.

[0029] Among them, the real-time tactile information collected by the sludge dredging head when interacting with the sludge is transformed into real-time tactile information that the control system can process and analyze in real time by converting the physical signals such as reaction force and pressure distribution generated when the high-pressure water jet impacts the sludge, which are directly related to the mechanical state of the sludge, such as hardness, viscosity and density, into the most direct sensory input of the current working environment and sludge for the entire control system.

[0030] In the control system, based on real-time tactile information and current control commands, the internal parameters of the preset tactile digital twin model are updated. That is, by utilizing the actual correspondence between the perceived interaction information (i.e., real-time tactile information) and the control cause that triggered the interaction (i.e., the current control command), the preset tactile digital twin model describing the interaction law of "water jet-sludge" is corrected and optimized. This allows the predicted output of the preset tactile digital twin model to continuously and closely approximate the mechanical feedback measured by the sensor. By transforming a static model that was originally built based on prior knowledge or calibration data and may gradually become mismatched with changes in time and environment into a dynamic model that can learn online and evolve adaptively, the preset tactile digital twin model can be ensured to most realistically represent the current specific dredging interaction state at any time. This provides a real-time and accurate virtual experimental basis for subsequent prediction and control decision optimization using the preset tactile digital twin model.

[0031] Among them, the preset tactile digital twin model is a dynamic mathematical model that is pre-established within the control system to simulate and predict the mechanical interaction between the sludge and the dredging head under the impact of the water jet. Its internal parameters are updated online according to real-time tactile information to continuously approximate the real physical interaction process.

[0032] Step S20: Using the updated preset tactile digital twin model, predictive control results for various candidate control commands in future time periods are obtained.

[0033] Based on the updated preset tactile digital twin model, control decisions will be transformed from passive execution based on historical and current information to proactive planning and rehearsals for multiple future possibilities.

[0034] Specifically, the updated preset tactile digital twin model yields predictive control results for various candidate control commands in future time periods. Utilizing a real-time calibrated preset tactile digital twin model that can confidently characterize the causal relationship between "control command and tactile response," it serves as a virtual prediction engine. Before the control system actually executes the dredging action, it performs multi-step forward simulations of various potential candidate control strategies. This allows for the pre-calculation of the dredging effect, mechanical load, and other predictive control results that each control strategy may cause in the future. By constructing a predictive window for the future state of the physical world in a digital virtual space, the control system can pre-evaluate the potential effects and risks of different control commands without costly trial and error in the actual physical system. This provides a comprehensive and reliable comparative basis for intelligent decision optimization based on quantitative prediction results in subsequent steps.

[0035] Step S30: Select the candidate control command corresponding to the predictive control result that conforms to the preset control result from a variety of predictive control results as the target control command.

[0036] Based on the various predictive control results provided in step S20, the forward-looking situation assessment is transformed into a clear target control decision. Specifically, it systematically compares and evaluates the predictive control result corresponding to each candidate control command with the pre-set preset control target. Through this comparison and evaluation, it identifies and selects the predictive control result that best achieves the preset dredging efficiency index while fully satisfying all safety constraints from all candidate control commands. Finally, it selects the candidate control command corresponding to the optimal predictive result as the target control command to be actually executed. This ensures that every control action of the hydraulic dredging robot is not random or empirical, but rather the optimal or satisfactory solution obtained after virtual simulation verification and multi-objective optimization. Thus, the information predicted by the preset tactile digital twin model is directly transformed into decision commands driving the actual action, enabling the entire dredging action to continuously evolve autonomously towards the preset optimal direction of high efficiency and low energy consumption while meeting hard safety constraints, achieving a leap from "perception-prediction" to "decision-optimization".

[0037] Step S40: Control the sludge removal actions of the multi-joint robotic arm and nozzle according to the target control command.

[0038] As the execution terminal of the entire adaptive control loop, the target control command representing the optimal dredging strategy is parsed into the specific motion parameters of each joint of the multi-joint robotic arm and the adjustment amount of the nozzle. It then drives the corresponding hydraulic and mechanical execution units to coordinate their actions, so that the high-pressure water jet can act on the specific siltation area on the inner wall of the raw water pipeline with the optimal impact angle, position and pressure predicted in the forecast. This ensures that the decisions made in the early stage based on real-time perception and model prediction are accurately reflected in the physical world. The dredging action is no longer a product of a fixed program or simple feedback, but an optimal solution for the current complex silt environment, thereby directly and efficiently maximizing dredging efficiency and optimizing constraints such as energy consumption.

[0039] At the same time, the execution result of this step will immediately generate new interaction with the raw water pipeline and be captured by the touch sensor again, forming new real-time tactile information. This provides new input for the "perception-update-prediction-optimization" closed loop of the next control cycle, enabling the entire control system to continuously adjust itself according to the action effect, and ultimately achieve continuous, autonomous, and high-performance operation of dredging in a dynamically changing environment.

[0040] In one feasible implementation, step S10 may specifically include steps S11 to S17, wherein the control system stores the dredging water jet control signal, the real-time motion status information of the multi-joint robotic arm, the pipe structure of the raw water pipeline, and the real-time pose of the hydraulic dredging robot. Step S11: The initial tactile signal when the sludge removal head interacts with the sludge is collected by the tactile sensor, and the initial tactile signal is compensated and filtered based on the sludge removal water jet control signal to obtain the first tactile information.

[0041] When the nozzle of the sludge removal head sprays high-pressure water jets to impact and remove sludge from the inner wall of the raw water pipe, the multi-axis force / torque sensor (i.e., tactile sensor) integrated on the mechanical structure of the sludge removal head directly measures the forces and torques it experiences from various directions. These forces and torques are the initial tactile signals. Essentially, it is a data stream of multi-dimensional forces and torques in the sensor's coordinate system, which includes the coupling of multiple sources such as water jet impact, robotic arm movement, structural vibration, and the reaction force of the target sludge.

[0042] Next, the initial tactile signal is compensated and filtered based on the dredging water jet control signal. The process involves the control system first acquiring the real-time control signal driving the nozzle. Based on this real-time control signal and the pre-calibrated flow-pressure characteristic curve obtained experimentally, the jet pressure and flow parameters at the nozzle outlet are directly determined. Then, based on the determined jet pressure and flow parameters, combined with the fixed geometric dimensions of the nozzle, the average velocity and impact dynamic pressure of the water jet at the outlet are calculated using Bernoulli's equation and the continuity equation of fluid mechanics. Based on the calculated average velocity, impact dynamic pressure, cross-sectional area of ​​the nozzle outlet, and the momentum theorem for the jet impacting a fixed smooth surface, the total impact force generated by the jet on the theoretical impact surface is calculated. Then, based on the known mechanical structure design parameters of the sludge removal head, namely the fixed three-dimensional position and orientation relationship of the nozzle outlet relative to the tactile sensor mounting base, the calculated total impact force is decomposed to the force center of the sensor mounting base according to the principle of statics. Finally, through a transformation matrix (i.e., force-torque transfer function) obtained by pre-calibration offline after assembly, which describes the fixed relationship between the force on the mounting base and the output of each measurement channel of the sensor, the force and torque decomposed to the base are converted into theoretical readings in the measurement coordinate system of the tactile sensor. This theoretical reading is the predicted theoretical interference force component that should be generated on the tactile sensor when a pure water jet impacts a fixed surface under clear water or no sludge load conditions.

[0043] The predicted theoretical interference force component is subtracted in real time from the actual acquired initial tactile signal, thereby initially separating the force signal mainly generated by the reaction of external sludge. Then, the separated signal is further filtered (such as by using an adaptive filter or notch filter) to suppress the remaining system resonant frequency noise and random interference. Finally, the first tactile information that has been purified and can directly reflect the real physical interaction between the sludge head and the sludge is obtained. This effectively removes the strong interference inherent in the high-pressure water jet itself and that is unrelated to the sludge being dredged. This allows the tactile information on which subsequent processing and analysis depend to more purely and accurately characterize the physical properties of the sludge (such as hardness and viscosity). This provides a high-fidelity and low-noise input basis for the accurate updating of the preset tactile digital twin model and subsequent adaptive control decisions, significantly improving the accuracy and reliability of the entire control system's environmental perception.

[0044] Step S12: Perform motion decoupling processing on the first tactile information based on the real-time motion state information to obtain the second tactile information.

[0045] Sensors, such as encoders and inertial measurement units, installed at each joint of the multi-joint robotic arm acquire its motion state information in real time, including the angles, angular velocities, and angular accelerations of each joint, thus obtaining the real-time motion state. Then, based on the known dynamic model of the robotic arm (including parameters such as the mass, center of mass position, and inertia tensor of each link) and the currently acquired real-time motion state, the control system uses dynamic calculation methods such as the Newton-Euler equations or the Lagrange equations to calculate in real time the theoretical inertial force and torque generated by the mass and inertia of the sludge removal head under its own acceleration, deceleration, and the influence of gravity, acting on the touch sensors under pure motion conditions without external sludge contact. This part of the force is collectively referred to as the motion force. Pseudo-force; then, in this step, the first tactile information typically contains force components in three orthogonal directions and torque components in three-dimensional space. The motion pseudo-force is also a vector with the same dimension and coordinate orientation. In each control cycle, the first tactile information and the calculated motion pseudo-force are directly subtracted element-by-element in the sensor coordinate system. The result of this subtraction operation is the second tactile information after mathematically removing the influence of the robotic arm's own motion dynamics. This provides high-fidelity tactile feedback that is only related to environmental interaction for the subsequent update of the preset tactile digital twin model and decision control, greatly improving the accuracy of tactile perception and the reliability of judging the state of silt.

[0046] Step S13: Based on the pipe structure and real-time pose, the second tactile information is mapped to the pipe spatial coordinate system to obtain real-time tactile information.

[0047] First, the positioning and navigation system of the hydraulic dredging robot acquires the position and attitude of the robot body in the global reference system in real time to determine the real-time pose. At the same time, the control system has a pre-stored digital structural model of the target raw water pipeline. This model defines a pipeline spatial coordinate system with the pipeline axis as the reference. It usually takes a certain starting point of the pipeline as the origin, the pipeline axis as the Z axis, and the radial and circumferential axes as the X and Y axes, respectively.

[0048] Next, based on the real-time pose, combined with the fixed dimensional parameters of the robot's mechanical structure and the installation position of the tactile sensors at the end of the robotic arm, the control system generates a homogeneous transformation matrix containing translation vectors and rotation matrices, based on the real-time pose of the robot body in the global reference frame. And the transformation matrix from the robot base coordinate system to the end flange coordinate system of the robotic arm, obtained by forward kinematics calculation. Then, combined with the pre-measured and calibrated fixed installation transformation matrix from the end flange coordinate system to the touch sensor coordinate system, By performing matrix multiplication operations sequentially, that is Thus, the three-dimensional coordinates of the origin of the sensor coordinate system in the global world coordinate system are obtained. and the directions of its coordinate axes.

[0049] Because the pre-stored digital structure model of the pipeline not only defines the pipeline spatial coordinate system, but also provides a pose description of the origin of the pipeline spatial coordinate system in the same global world coordinate system. By obtaining the inverse matrix describing the pose This allows us to obtain the transformation relationship from the world coordinate system to the pipeline space coordinate system.

[0050] Representing the origin of the sensor coordinate system in the homogeneous coordinates of the world coordinate system Multiply by the inverse transformation matrix, i.e. The resulting vector The first three components are the precise three-dimensional coordinates of the sensor coordinate system origin in the pipe space coordinate system; simultaneously, the rotation matrix of the sensor relative to the world coordinate system... take Rotation part ,get ,in It is the direction cosine matrix of each coordinate axis of the sensor coordinate system relative to each coordinate axis of the pipeline space coordinate system. Each column represents the unit direction vector of the X, Y, and Z axes of the sensor coordinate system in the pipeline space coordinate system.

[0051] Then, using the homogeneous transformation matrix formed by this precise three-dimensional coordinates and direction cosine matrix, the second tactile information (i.e., a six-dimensional vector containing three force components and three torque components) represented in the sensor coordinate system is transformed. This transformation process involves multiplying the force and torque vectors by the corresponding rotation matrix, thereby transforming the expression reference of force and torque from the sensor local coordinate system to the global pipe space coordinate system aligned with the pipe's geometric features. After the transformation, real-time tactile information expressed in the pipe space coordinate system is obtained. This provides a standardized, interpretable, and positionally defined mechanical state input for subsequent determination of the force direction (e.g., whether the impact force is axial propulsion or radial compression), association with specific pipe segment structures, and the realization of pre-set tactile digital twin model updates and adaptive control decisions based on spatial location.

[0052] Step S14: Input the current control command into the preset tactile digital twin model to obtain the model's predicted tactile information.

[0053] It should be noted that the preset tactile digital twin model is a virtual dynamic mathematical model pre-built in the control system. It is based on physics and data and can accurately simulate the mechanical relationship between the sludge head and the sludge through online dynamic parameter updates, thereby predicting tactile response and guiding control decisions.

[0054] The current control command is input into a preset tactile digital twin model to obtain the model's predicted tactile information. This generates a theoretical prediction value of the expected interaction result based on the current cognitive state of the preset tactile digital twin model, thus providing a benchmark for comparison with the actual measurement results in subsequent steps. Using the preset tactile digital twin model with an embedded "control command-tactile response" mapping relationship, a forward calculation or simulation is performed. That is, based on the input current control command, the preset tactile digital twin model deduces and outputs the model's predicted tactile information that it believes should be generated when the sludge head interacts with the sludge under the current control command.

[0055] Furthermore, step S14 may also include steps S141 to S144: Step S141: Obtain the real-time pose through the spatial position and attitude measurement unit of the hydraulic dredging robot.

[0056] The spatial position and attitude measurement unit typically consists of an odometer, an inertial measurement unit, and a lidar or pipeline feature matching vision module. The odometer provides relative displacement increment observations, the inertial measurement unit provides angular velocity and acceleration observations to calculate attitude changes, and the lidar or pipeline feature matching vision module provides absolute pose constraint observations by scanning the features of the inner wall of the raw water pipeline and matching them with pre-stored or online constructed pipeline maps.

[0057] The state of the robot's mobile platform coordinate system relative to the global reference coordinate system is defined as a state vector. This state vector typically includes three-dimensional position, three-dimensional attitude angles (roll, pitch, and yaw), and their corresponding velocities and angular velocities. In each filtering cycle, angular velocity and acceleration observations measured by the inertial measurement unit are used as inputs, combined with the robot's kinematic model, to update the state vector and its covariance from the previous moment, obtaining the current state prediction. Subsequently, relative displacement increment observations from the odometry and absolute pose constraint observations from LiDAR or vision are simultaneously introduced. By establishing a nonlinear observation model between the observations from each sensor and the state prediction, the residual between the actual observation and the expected observation based on the state prediction is calculated. The state prediction and its covariance are then weighted and corrected using Kalman gain, which is dynamically calculated based on the state prediction covariance and observation noise to ensure short-term relative accuracy of the odometry. High accuracy and reliable absolute precision of the feature matching sensor are achieved, but an optimal balance must be struck between these two aspects. Through this iterative prediction and update, the extended Kalman filter can effectively fuse high-frequency but potentially drifting inertial data, relatively reliable mileage increment data, and low-frequency but absolutely accurate pipeline feature matching data. This results in a stable and continuous output of an optimally estimated three-dimensional position coordinate and three-dimensional attitude angle of the robot's mobile platform coordinate system relative to the global reference coordinate system, i.e., the required real-time pose. This provides the entire control system with reliable, continuous, and real-time state information about the absolute spatial position of the hydraulic dredging robot inside the raw water pipeline. This real-time pose is the geometric basis for all subsequent perception mapping, motion planning, and decision control based on the pipeline spatial coordinate system. It ensures that the interaction mechanical information between the dredging head and the sludge can be accurately correlated to the specific physical position and structural direction of the pipeline, thus enabling adaptive control decisions to have clear spatial correlation and environmental consistency.

[0058] Step S142: Based on the real-time pose, calculate the nominal position and normal vector of the tactile sensor under the pipe structure.

[0059] The control system first calls the pre-stored digital structure model of the pipeline. This digital structure model not only defines the spatial coordinate system of the pipeline, but also includes the parameterized equation of the pipeline centerline and the pipeline structure on the inner wall surface of the pipeline.

[0060] Using real-time pose, the actual three-dimensional position of the touch sensor in the global coordinate system is calculated through the robot's kinematic model. The actual three-dimensional position of the sensor is then vertically projected onto the center line of the pipe defined by the pipe structure, thereby finding the projection point on the pipe center line that is closest to the sensor. The coordinates of this projection point in the pipe space coordinate system are defined as the nominal position of the sensor in the pipe structure.

[0061] While determining the nominal position, based on the geometric properties of the pipeline's digital structural model at the nominal position, the unit direction vector perpendicular to the inner wall surface of the pipeline and pointing towards the pipeline axis is calculated. This vector is the normal vector of the sensor at the nominal position, providing a stable geometric reference for establishing a precise transformation relationship between the sensor coordinate system and the pipeline spatial coordinate system in subsequent steps. This ensures that regardless of the robot's pose, the spatial mapping of tactile information can be based on a repeatable reference frame consistent with the theoretical geometry of the pipeline, greatly improving the spatial consistency and physical interpretability of the force perception results.

[0062] Step S143: Based on the nominal position and normal vector, establish a rigid transformation relationship between the sensor coordinate system and the pipeline space coordinate system.

[0063] The nominal position is the known three-dimensional coordinate of the sensor in the pipeline spatial coordinate system, which is directly used as the translation vector in the transformation relationship from the sensor coordinate system to the pipeline spatial coordinate system.

[0064] Based on the calculated normal vector and the inherent definition of the pipeline space coordinate system, the pipeline axis is usually taken as the Z-axis. A complete rotation matrix is ​​constructed through vector operations. Specifically, the normal vector is normalized and used as a basis axis of the pipeline space coordinate system, such as the Z-axis. Then, the cross product of the normal vector with the pipeline axis unit vector is used to obtain the second orthogonal basis axis, such as the X-axis. Finally, the cross product of the first two vectors is used to obtain the third orthogonal basis axis, such as the Y-axis. The 3x3 matrix formed by these three orthogonal unit column vectors known in the pipeline space coordinate system is the rotation matrix from the sensor coordinate system to the pipeline space coordinate system.

[0065] Finally, the rotation matrix and translation vector are combined to form a 4x4 homogeneous transformation matrix. This homogeneous transformation matrix defines a rigid transformation relationship between the sensor coordinate system and the pipeline spatial coordinate system. This allows the real-time tactile information originally expressed in the sensor's local coordinate system to be transformed into an absolute reference system that is perfectly aligned with the pipeline geometry. This ensures that all subsequent control decisions and analyses based on spatial position and orientation are built on a unified and stable geometric reference, greatly enhancing the spatial consistency of the entire control system's environmental perception and the physical interpretability of control decisions.

[0066] Step S144: Based on the rigid transformation relationship, the second tactile information is subjected to coordinate transformation and vector decomposition, and the second tactile information is mapped to the pipeline space coordinate system to obtain the interaction forces of the sludge removal head along the axial, radial and circumferential directions of the pipeline structure; wherein, the real-time tactile information is the interaction forces of the sludge removal head along the axial, radial and circumferential directions of the pipeline structure.

[0067] First, the 3x3 rotation matrix is ​​extracted from the rigid transformation relationship. This rotation matrix defines the rotational transformation from the sensor coordinate system to the pipe space coordinate system. Next, the second tactile information is multiplied by the rotation matrix to complete the pure rotational coordinate transformation, thus obtaining the general expression of the force vector in the pipe space coordinate system.

[0068] Then, vector decomposition is performed. Since the pipeline space coordinate system is usually predefined with its axial (e.g., Z-axis), radial (e.g., X-axis), and circumferential (e.g., Y-axis) orthogonal bases, the generalized expression obtained after transformation is projected onto these three coordinate axes respectively. The projection length (i.e., the dot product of the vector with each base axis) is the scalar magnitude of the interaction forces experienced by the dredging head in the axial, radial, and circumferential directions of the pipeline structure. This maps and decomposes the second tactile information from its original local coordinate system, which is fixed to the sensor, to the pipeline space coordinate system, which is completely consistent with the geometric characteristics of the working environment. This transforms the tactile perception from an abstract force vector coupled with the robot's posture into mechanical components (i.e., axial force, radial force, and circumferential force) that are directly bound to the pipeline's physical structure and have clear engineering semantics. This allows the control system to intuitively judge the mechanical state of the dredging operation (e.g., whether the impact force is mainly used for axial propulsion or may cause radial compression). This provides clear, accurate, and directly usable input information for subsequent adaptive control decisions based on pipeline structural characteristics (e.g., adjusting the posture to avoid excessive radial force damaging the pipe wall).

[0069] Step S15: Compare the tactile information predicted by the model with the real-time tactile information to obtain the tactile residual vector.

[0070] First, it is ensured that the model-predicted tactile information and the real-time tactile information are within the same mathematical framework, meaning both are force vectors defined in the pipe's spatial coordinate system and possessing the same dimensions and physical meaning (e.g., both decomposed into axial, radial, and circumferential components). Next, in each control cycle, the control system performs element-wise algebraic subtraction between the real-time tactile information and the model-predicted tactile information. Each element of the resulting operation represents the numerical difference between the actual measured force and the model-predicted force in a specific direction of the pipe structure—the tactile residual vector. This vector reveals in real-time the deviation between the predicted output of the current preset tactile digital twin model and the actual feedback from the physical world. As a key indicator of model prediction accuracy, this tactile residual vector provides a precise and reliable input signal for subsequent steps, including analyzing the sources of model errors and driving selective online updates of the model's internal parameters.

[0071] Step S16: Based on the tactile residual vector, perform sensitivity analysis on the parameters in the preset tactile digital twin model and select the internal parameters to be updated.

[0072] For numerous adjustable parameters in a preset tactile digital twin model (such as physical parameters characterizing sludge viscosity, cohesion, or certain fitting coefficients), the local influence of each parameter on the model's predicted tactile information at the current model operating point (defined by the current control command and model state) is evaluated. This is usually achieved by calculating the partial derivative (gradient) of the preset tactile digital twin model output with respect to each parameter or by applying a small parameter perturbation and observing the magnitude of the change in the model's prediction. The result is a list of sensitivity coefficients, which quantifies the amount of change in the model's predicted tactile information caused by a unit change in each parameter.

[0073] Next, this list of sensitivity coefficients is combined with the current tactile residual vector for analysis. By comparison, such as calculating the dot product or projection of the sensitivity coefficient vector and the residual vector, the parameters whose direction of change can most effectively explain or compensate for the current residual are identified. Finally, based on a preset sensitivity threshold, parameters with high sensitivity coefficients and whose adjustment direction is consistent with the residual compensation direction are selected from all parameters and marked as internal parameters to be updated. This locks in the main contributing parameters that cause the current prediction error, rather than blindly averaging all parameters, avoiding the instability risks caused by redundant or reverse adjustments. This ensures that the preset tactile digital twin model can adapt to the changing sludge interaction dynamics in the fastest and most robust way.

[0074] Step S17: Based on the internal parameters to be updated and the sensitivity corresponding to each internal parameter to be updated, and combined with the preset adjustment compensation and preset forgetting factor, update the internal parameters to be updated to obtain the updated preset tactile digital twin model.

[0075] A parameter update equation is constructed, the core of which is the iterative form of the recursive least squares method. Its input is the tactile residual vector, and the key design lies in the improvement of the standard algorithm.

[0076] Specifically, the parameter update algorithm will independently calculate an adaptive step size for each internal parameter to be updated. This adaptive step size is positively correlated with the sensitivity of the corresponding internal parameter to be updated. Parameters with high sensitivity will get a larger step size for fast and large correction, while parameters with low sensitivity will have a smaller step size, thereby achieving a match between the update intensity and the parameter's influence.

[0077] Preset adjustment compensation is introduced as an additional stabilization term, which dynamically fine-tunes the step gain based on the historical statistical characteristics (such as the rate of change) of the tactile residual vector. When the tactile parameter vector fluctuates drastically, the step size is appropriately reduced to prevent overshoot, and the step size is increased to accelerate convergence when the residual is consistently large.

[0078] Meanwhile, a preset forgetting factor is applied to the covariance matrix update of the parameter update algorithm. Its role is to reduce the weight of historical statistical characteristics in the current parameter estimation in an exponentially decaying manner, so that the preset tactile digital twin model can gradually forget old interaction information that may no longer be applicable, thereby enhancing the ability to track recent new interaction characteristics.

[0079] Finally, by solving the recursive least squares equations that incorporate sensitivity, preset adjustment compensation, and preset forgetting factors, an optimal set of parameter corrections was calculated and applied to the preset tactile digital twin model, thereby obtaining the updated preset tactile digital twin model. This achieves a model online self-correction mechanism that is both fast-responding and robustly convergent, ensuring that the updated preset tactile digital twin model can closely and reliably adapt to the current real "water jet-sludge" interactive dynamic state, providing a high-fidelity virtual benchmark for prediction and control decisions in subsequent steps.

[0080] In one feasible implementation, step S20 may specifically include steps S21 to S24: Step S21: Generate multiple candidate control commands based on the current control commands and historical control commands.

[0081] First, the sequence of control commands that have been executed in recent times and are arranged in chronological order is recorded and stored to form historical control commands. These historical control commands reflect the recent dredging action trends and dredging operation context of the hydraulic dredging robot.

[0082] Using the current control command as a reference point, diverse candidate control commands are generated by purposefully applying various pre-defined bounded disturbances in the control variable space (including robotic arm joint motion parameters, nozzle adjustment parameters, etc.). These disturbances typically include slightly increasing or decreasing the amplitude of the current control command, changing the coordination relationship between different control variables in the command, or making short-term extrapolations based on the changing trends of historical control commands. This generates a series of candidate control commands that are both consistent with the current and historical control commands in terms of action and have controllable differences. This provides a set of strategy samples for subsequent prediction and simulation steps that are both centered around the current operating state and cover a variety of possibilities within a reasonable range of actions. This ensures that the search process for the optimal control command is not a blind random attempt, but a targeted and efficient exploration based on recent operational experience. This fundamentally enhances the continuity of control decisions and the intelligence of exploration, which is a key prerequisite for achieving forward-looking rolling optimization in complex environments.

[0083] Step S22: Input each candidate control command into the updated preset tactile digital twin model in sequence, perform forward rolling simulation, and obtain the tactile response prediction sequence corresponding to each candidate control command in the future time period.

[0084] Using the robot's current real-time state (including the pose of the multi-joint robotic arm and the spatial position of the sludge removal head) as the initial conditions for simulation, the control variables of each candidate control command within the first prediction time step, such as the target joint angle of the multi-joint robotic arm at the next moment and the target opening of the nozzle, are input into the updated preset tactile digital twin model. Based on its current parameters, the updated preset tactile digital twin model calculates the predicted tactile response value that the sludge removal head and the multi-joint robotic arm will produce at the end of the first prediction time step after executing the control variable. This predicted tactile response value is then used as the initial condition for the next prediction time step, and the above process is repeated, inputting the control variable corresponding to the second prediction time step from the same candidate control command sequence again. This process is iterated until all predicted time steps within the entire preset future time period are calculated. Finally, a set of tactile response prediction values ​​containing multiple time steps is obtained, arranged in chronological order. This set is the tactile response prediction sequence corresponding to the candidate control command. This complete forward rolling simulation process is executed independently for each candidate control command, thereby obtaining a set of non-tactile response prediction sequences that correspond one-to-one with the candidate control commands. This allows for a virtual trial run under every possible control command throughout the entire time period without the hydraulic dredging robot actually performing any physical actions. This enables precise quantitative comparison of the long-term execution effects of different control commands, providing forward-looking data for subsequent optimization decisions based on the prediction results.

[0085] Step S23: Calculate the uncertainty measure corresponding to each tactile response prediction sequence at each prediction step length.

[0086] Among them, the uncertainty measure is a scalar or vector metric used to quantify the reliability of tactile response prediction sequences.

[0087] First, the prediction error characteristics of the preset tactile digital twin model itself are evaluated and tracked. For example, by recording the residuals between the historical tactile response prediction sequence and the corresponding historical real-time tactile information online, and statistically analyzing their mean and covariance, the corresponding prediction error is formed. When performing the forward rolling simulation in step S22, the control system not only transmits the predicted tactile response value, but also transmits a covariance matrix that characterizes the current prediction compensation prediction uncertainty.

[0088] At each prediction step, two calculations are performed: First, at each prediction step, the preset tactile digital twin model is linearized at the current prediction state (including the dredging head pose, tactile state, sludge parameter estimation, etc.) and control input (the current step component of the candidate control command). After calculating the Jacobian matrix of the model state transition function with respect to the state, the state uncertainty output from the previous prediction step (represented by a covariance matrix) is left-multiplied by this Jacobian matrix and then right-multiplied by its transpose. This linearly propagates the uncertainty of the previous prediction step to the current prediction step according to the first-order approximation of the model dynamics, obtaining the prior uncertainty estimate after dynamic propagation. Second, a preset... The inherent inaccuracy of the tactile digital twin model is modeled as a zero-mean Gaussian process noise. Its covariance matrix is ​​obtained by online statistical analysis of the historical tactile residuals of the preset tactile digital twin model and is regarded as the model's inherent prediction error at the current operating point. Finally, this covariance matrix representing the model's own error is added to the aforementioned prior uncertainty covariance matrix after dynamic propagation to obtain the overall uncertainty measure that integrates state transfer uncertainty and model's own error at the current prediction step (i.e., the updated posterior covariance moment). The trace or the largest eigenvalue of this posterior covariance matrix, or other scalar derived indices, can be used as the specific uncertainty measure of the prediction value at this step. By calculating an evolving and quantified uncertainty measure for each prediction step in the future time period during the rolling simulation, a confidence label is attached to each tactile response prediction sequence. This allows subsequent decision-making steps to not only know "what the possible results of executing a certain control command" but also "how likely this control command is to be inaccurate." This greatly enhances the ability to perceive the cognitive limitations of the preset tactile digital twin model itself and the time-varying nature of the environment.

[0089] Step S24: Integrate the tactile response prediction sequence and uncertainty measure corresponding to each candidate control command to form the prediction control result corresponding to each candidate control command.

[0090] For each candidate control command, a structured data object is created. This data object contains two core fields: one is the tactile response prediction sequence corresponding to the candidate control command, and the other is an uncertainty measure that is strictly aligned with the tactile response prediction value at each step in the tactile response prediction sequence. These two data are strictly synchronized and aligned in the time domain, so that in the subsequent decision-making process, for any specific prediction time step in the future, the tactile response prediction value and the uncertainty measure of the tactile response prediction value can be obtained simultaneously and with the same index, thereby forming the corresponding predictive control result.

[0091] In one feasible implementation, step S30 may specifically include steps S31 to S34, wherein the preset control results include preset pipeline safety constraints, preset motion state constraints, and preset comprehensive evaluation values: Step S31: From the tactile response prediction sequences included in each prediction control result, the tactile feature vector and the predicted dredging efficiency are obtained by parsing.

[0092] For each candidate control command corresponding to the tactile response prediction sequence, two parallel parsing processes are performed. On the one hand, the parsing of the tactile feature vector is accomplished by extracting features from the tactile response prediction sequence in the time and frequency domains. For example, statistical and morphological indicators such as peak force, average force, force change gradient, and distribution of the main force direction of the tactile response prediction sequence during key periods are calculated, and these indicators are arranged in a fixed order to form a tactile feature vector. This tactile feature vector concisely describes the overall force, impact, and pattern of the expected interaction between the dredging head and the sludge when the candidate control command is executed from a mechanical perspective. On the other hand, the analysis of predicted dredging efficiency is based on a preset mapping model that can map the predicted sequence of tactile response to the dredging effect. This preset mapping model combines the predicted sequence of tactile response over the entire prediction period with the predicted movement speed of the hydraulic dredging robot. By integration or weighted accumulation, it estimates the total volume of sludge expected to be removed or the average dredging rate per unit time under the preset control command, thereby obtaining a quantitative expected dredging efficiency. This provides a refined, comparable, and physically meaningful input for subsequent steps of safety constraint comparison and multi-objective comprehensive optimization, greatly improving the processing efficiency of the decision-making system and the pertinence of the evaluation.

[0093] Step S32: Compare the tactile feature vectors with the preset pipeline safety constraints and preset motion state constraints respectively, and select the first preset control result corresponding to the tactile feature vectors that meet the preset pipeline safety constraints and preset motion state constraints.

[0094] It should be noted that the preset pipeline safety constraints are pre-set based on the material, design standards, and historical health status of the raw water pipeline, and are the mechanical limits that must not be exceeded during dredging operations. Specifically, they are quantified as the absolute maximum value of radial impact force, the threshold of circumferential shear force, and the potential continuous pressure threshold on the pipe wall in the pipeline spatial coordinate system. The preset motion state constraints are pre-set based on the physical limits of the robot arm and hydraulic system, and are the kinematic and dynamic boundaries that ensure stable operation. Specifically, they are quantified as the maximum allowable torque of each joint, the maximum angular velocity, and the stability margin of the overall posture of the dredging head.

[0095] In this step, the comparison and screening process involves first extracting feature components directly corresponding to the preset pipeline safety constraints and preset motion state constraints from the tactile feature vector. These components include, for example, the peak radial force, the average circumferential force, and the estimated equivalent load torque of each joint. Then, these extracted feature components are compared numerically with the preset thresholds of the corresponding constraints. This comparison is achieved through a series of logical judgments. Only when all the tested components in the tactile feature vector corresponding to a candidate control command simultaneously satisfy the corresponding preset pipeline safety constraints and preset motion state constraints will the predictive control result corresponding to that candidate control command be marked as feasible and classified as the first preset control result. This forms a subset of feasible solutions that have passed the safety screening, thereby ensuring that subsequent optimization decisions are made only within an absolutely safe range. This eliminates potential equipment damage and operational accidents caused by pursuing dredging efficiency from the root of the decision-making process, fundamentally guaranteeing the physical safety and operational reliability of the high-risk dredging operation.

[0096] Step S33: Calculate the comprehensive evaluation value based on the predicted dredging efficiency and uncertainty measure corresponding to the first preset control result.

[0097] First, for each preset control result, its two core inputs are extracted: predicted dredging efficiency and uncertainty measure. Then, a preset comprehensive evaluation function is used to calculate the comprehensive evaluation value. This comprehensive evaluation function is usually designed as: Comprehensive Evaluation Value = Efficiency Weight × Predicted Dredging Efficiency - Risk Weight × Uncertainty Measure. Here, the efficiency weight and risk weight are positive coefficients preset according to the operational task preferences, used to adjust the trade-off between pursuing dredging efficiency and avoiding predicted risks. This allows the control system to intelligently select the control command that achieves the best balance between "maximum efficiency" and "maximum stability" from all safe feasible solutions, realizing the scientific trade-off and optimization decision-making between dredging efficiency and operational risks in complex and uncertain environments.

[0098] Before calculation, the predicted dredging efficiency and uncertainty measure of all candidate control commands are usually normalized. Specifically, firstly, all predicted dredging efficiencies are extracted from the data corresponding to all first preset control commands in the current control cycle to form an efficiency set, and all uncertainty measure values ​​are extracted to form an uncertainty set. Then, for the efficiency set, the system calculates the maximum and minimum values ​​in the set, and uses the formula "(predicted dredging efficiency of the current command - minimum predicted dredging efficiency value in the set) / (maximum efficiency value in the set - minimum predicted dredging efficiency value in the set)" to calculate each efficiency value, thereby linearly normalizing all predicted dredging efficiencies to the interval [0, 1], where 1 represents the highest expected dredging efficiency in the current feasible command set, and 0 represents the lowest efficiency. Similarly, for the uncertainty set, the system also calculates its maximum and minimum values, and uses the formula "(uncertainty measure of the current command - minimum uncertainty value in the set) / (maximum uncertainty value in the set - minimum uncertainty value in the set)" to calculate each efficiency value. The "minimum uncertainty within the set" is used to calculate each uncertainty measure, thereby linearly normalizing all uncertainty measures to the interval [0, 1]. Here, 1 represents the highest level of uncertainty in the current feasible instruction set, and 0 represents the lowest level of risk. After this processing, the predicted dredging efficiency value and uncertainty value measure, which may have had large differences in dimensions and numerical range, are uniformly converted into comparable values ​​with dimensionless and consistent scale. This provides a standardized input basis for subsequent comprehensive weighted scoring, eliminating the influence of dimensions and ensuring fair comparison.

[0099] Step S34: The candidate control command corresponding to the first preset control result whose comprehensive evaluation value is greater than the preset comprehensive evaluation value is determined as the target control command.

[0100] It should be noted that the preset comprehensive evaluation value in this step refers to the second highest comprehensive evaluation value among all the comprehensive evaluation values ​​corresponding to the first preset control results.

[0101] All the comprehensive evaluation values ​​corresponding to the first preset control results are sorted in descending order. After selecting the candidate control command corresponding to the first preset control result with the highest comprehensive evaluation value, the candidate control command is determined as the target control command. In this way, the optimal pursuit of the "efficiency-risk" balance point under the hard constraint of safety is achieved in each decision, and the dredging operation can be continuously and stably operated at the highest comprehensive efficiency level.

[0102] In one feasible implementation, step S40 may specifically include steps S41 to S45: Step S41: Analyze the target control command to obtain the desired end-effector trajectory of the multi-joint robotic arm and the desired injection parameters of the nozzle.

[0103] Read and unpack the data packet corresponding to the target control command, and decompose it into two core data streams with timestamp alignment. One part is the expected position and attitude sequence of the robotic arm end effector (dredging head) in the pipeline spatial coordinate system at each time step in the future period. This expected position and attitude sequence constitutes the expected end trajectory of the multi-joint robotic arm.

[0104] The other part consists of the expected hydraulic valve control signals (such as valve core opening) and working mode (such as pulse frequency) parameters corresponding to each time step on the same time axis that drive the adaptive sludge removal head nozzle. These are the expected injection parameters of the nozzle, which provide a unique and definite execution basis for subsequent steps to transform intelligent decisions into coordinated and precise physical movements and operational actions, ensuring the fidelity and consistency of information transmission between decision-making and execution.

[0105] Step S42: After converting the desired end trajectory into a sequence of desired angles for each joint of the multi-joint robotic arm, a coordinated control signal sequence is generated by combining the desired injection parameters.

[0106] Based on the geometric parameters of the multi-joint robotic arm, such as link length and joint offset, the inverse kinematic equations are solved in real time for each pose point on the desired end effector trajectory, i.e., the position and orientation of the sludge removal head in the pipeline spatial coordinate system at each moment. This usually involves a numerical iterative algorithm (such as using the pseudo-inverse of the Jacobian matrix for gradient descent search). This algorithm takes the current joint angle of the robotic arm as the initial value and quickly calculates a set of joint angles required for the end effector to reach the desired pose by minimizing the error between the current pose and the desired pose. This calculation process is executed on all pose points on the trajectory in time sequence, thus obtaining the sequence of desired angles of each joint that strictly corresponds to the desired end effector trajectory in time.

[0107] Next, a unified timeline is created, and the desired joint angles and desired injection parameters corresponding to each same timestamp are packaged into a structured data frame to form a cooperative control signal sequence. Each unit of this cooperative control signal sequence explicitly specifies the target state that each joint and nozzle should reach at a precise future moment, completing the key mapping from the end trajectory to the joint angle. The continuous movement of the robotic arm and the jet control of the nozzle are synchronized under a unified time reference, thereby generating a complete motion script that can drive all actuators of the hydraulic dredging robot to cooperate precisely in time and space. This provides the most direct instruction basis for finally realizing the high dynamic and high-precision trajectory tracking and precise control of the jet parameters of the dredging head.

[0108] Step S43: Based on the coordinated control signal sequence, drive the hydraulic actuators of each joint and the adjustment mechanism of the nozzle to perform the dredging action.

[0109] Each data frame in the collaborative control signal sequence is sent to the corresponding underlying servo controller in real time. Specifically, for the hydraulic actuator of a multi-joint robotic arm, each joint's independent hydraulic servo controller receives the current target angle value in its desired angle sequence and adjusts the flow and pressure into the joint hydraulic cylinder or hydraulic motor through a high-response electro-hydraulic proportional valve or servo valve to drive the piston rod or motor shaft to move precisely, so that the actual angle of the joint tracks the desired angle sequence and the dredging head moves along the desired end trajectory.

[0110] For the nozzle adjustment mechanism, its hydraulic servo controller controls the opening of the hydraulic valve in front of the nozzle according to the desired injection parameters in the coordinated control signal sequence, thereby precisely modulating the flow rate, pressure and impact mode of the high-pressure water jet.

[0111] All of the aforementioned underlying driving processes are performed under a unified clock synchronization, ensuring that the spatial movement of the robotic arm and the spraying action of the nozzle are strictly coordinated and executed according to the spatiotemporal relationship planned by the target control command at every moment. This ensures that the actual effect of the dredging operation can accurately match or even approach the optimal performance predicted by the model, and is the key physical execution link to achieve efficient, accurate and reliable operation of the entire adaptive control system.

[0112] Step S44: The execution status of the dredging action is monitored by the joint encoder of the multi-joint robotic arm and the pressure sensor of the nozzle, and status feedback information is obtained.

[0113] The actual rotation angle of each joint shaft is measured in real time by a joint encoder installed at each joint, and this actual rotation angle, together with the angular velocity signal obtained by its derivative, is fed back to the control system; at the same time, the pressure sensor integrated into the nozzle hydraulic circuit measures the actual working pressure of the water jet in real time.

[0114] The control system synchronously collects measurement data from all joint encoders and pressure sensors at a fixed sampling period, associates them with the desired angle and desired injection parameters at corresponding moments in the coordinated control signal sequence, and packages them into structured state feedback information. This state feedback information includes key execution state variables such as the actual position and speed of each joint and the actual pressure of the nozzle, thus constructing a closed-loop information flow from "instruction issuance" to "effect perception". This enables the control system to know whether there is a deviation between the actual action and the planned intention and the magnitude of the deviation, providing an indispensable data foundation for the accurate calculation and dynamic compensation of the execution deviation in the next step.

[0115] Step S45: Based on the state feedback information and the cooperative control signal sequence, calculate the execution deviation of the current control cycle, wherein the execution deviation is used to compensate and correct the cooperative control signal sequence of the next control cycle.

[0116] First, the expected values ​​for the corresponding moment within the current control cycle are extracted from the coordinated control signal sequence, namely the expected angle and expected injection parameters. Simultaneously, the corresponding actual state values ​​are obtained from the synchronously acquired state feedback information, namely the actual angles of each joint measured by the joint encoder and the actual pressure measured by the nozzle pressure sensor. Then, the system compares the expected value of each control variable with the actual state value one by one, calculating the difference through simple algebraic subtraction. For example, the angle deviation of each joint is the expected angle minus the actual angle, and the pressure deviation of the nozzle is the expected pressure minus the actual pressure. The calculated difference constitutes the execution deviation, which quantifies the performance of the control cycle in the current control period. During the previous control cycle, the errors between the actual execution results and the planned instructions in key states such as position and pressure provide the control system with real-time, quantitative execution error feedback. This enables the control system to accurately perceive the action deviations caused by the dynamic lag of the actuator, model mismatch, or external unknown disturbances. This provides precise error input for the forward-looking compensation and correction of the coordinated control signal sequence in the next control cycle (such as feedforward compensation or feedback adjustment). Ultimately, a closed-loop error correction mechanism is formed, which significantly improves the actual trajectory tracking accuracy of the dredging action and the stability of the jet parameter control, ensuring the physical execution effect of the optimized decision.

[0117] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A hydraulic dredging robot for raw water pipelines, characterized in that, The system includes a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is equipped with a sludge removal head, which includes a nozzle and a touch sensor. The tactile sensor is used to sense interaction with the silt.

2. An adaptive control method for a hydraulic dredging robot for raw water pipelines, characterized in that, The adaptive control method for the hydraulic dredging robot for raw water pipelines is applied to the hydraulic dredging robot for raw water pipelines as described in claim 1, comprising a mobile platform, a control system, and a multi-joint robotic arm mounted on the mobile platform. The end of the multi-joint robotic arm is provided with a dredging head, which includes a nozzle and a touch sensor. The adaptive control method includes: After collecting real-time tactile information from the interaction between the dredging head and the sludge using the tactile sensor, the internal parameters of the preset tactile digital twin model are updated in the control system based on the real-time tactile information and the current control command. By using the updated preset tactile digital twin model, predictive control results for various candidate control commands in future time periods are obtained; From the various predictive control results, the candidate control command corresponding to the predictive control result that conforms to the preset control result is selected as the target control command; The sludge removal actions of the multi-joint robotic arm and the nozzle are controlled according to the target control command.

3. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 2, characterized in that, The control system stores the dredging water jet control signal, the real-time motion status information of the multi-joint robotic arm, the pipe structure of the raw water pipeline, and the real-time pose of the hydraulic dredging robot. The step of collecting real-time tactile information when the dredging head interacts with the sludge through the tactile sensor includes: The initial tactile signal when the dredging head interacts with the sludge is collected by the tactile sensor, and the initial tactile signal is compensated and filtered based on the dredging water jet control signal to obtain the first tactile information; Based on the real-time motion state information, the first tactile information is subjected to motion decoupling processing to obtain the second tactile information; Based on the pipe structure and the real-time pose, the second tactile information is mapped onto the pipe spatial coordinate system to obtain the real-time tactile information.

4. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 3, characterized in that, The step of mapping the second tactile information to the pipe space coordinate system based on the pipe structure and the real-time pose to obtain the real-time tactile information includes: The real-time pose is obtained through the spatial position and attitude measurement unit of the hydraulic dredging robot; Based on the real-time pose, the nominal position and normal vector of the tactile sensor under the pipe structure are calculated; Based on the nominal position and the normal vector, a rigid transformation relationship is established between the sensor coordinate system and the pipeline space coordinate system; Based on the rigid transformation relationship, coordinate transformation and vector decomposition are performed on the second tactile information, and the second tactile information is mapped to the pipeline spatial coordinate system to obtain the interaction forces that the sludge removal head experiences along the axial, radial and circumferential directions of the pipeline structure. The real-time tactile information refers to the interactive forces experienced by the sludge removal head along the axial, radial, and circumferential directions of the pipeline structure.

5. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 2, characterized in that, The step of updating the internal parameters of the preset tactile digital twin model in the control system based on the real-time tactile information and the current control command includes: The current control command is input into the preset tactile digital twin model to obtain the model's predicted tactile information; By comparing the tactile information predicted by the model with the real-time tactile information, a tactile residual vector is obtained; Based on the tactile residual vector, sensitivity analysis is performed on the parameters in the preset tactile digital twin model to filter out the internal parameters to be updated. Based on the internal parameters to be updated and the sensitivity corresponding to each of the internal parameters to be updated, and in combination with preset adjustment compensation and preset forgetting factor, the internal parameters to be updated are updated to obtain the updated preset tactile digital twin model.

6. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 2, characterized in that, The step of obtaining the predictive control results of multiple candidate control commands in future time periods through the updated preset tactile digital twin model includes: Based on the current control command and the historical control commands, the various candidate control commands are generated; Each candidate control command is sequentially input into the updated preset tactile digital twin model, and forward rolling simulation is performed to obtain the tactile response prediction sequence corresponding to each candidate control command in the future time period. Calculate the uncertainty measure corresponding to each of the tactile response prediction sequences at each prediction step size; The tactile response prediction sequence and uncertainty metric corresponding to each candidate control command are integrated to form the predictive control result corresponding to each candidate control command.

7. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 6, characterized in that, The preset control results include preset pipeline safety constraints, preset motion state constraints, and preset comprehensive evaluation values. The step of selecting the candidate control command corresponding to the predicted control result that conforms to the preset control results from the multiple predicted control results as the target control command includes: From the tactile response prediction sequences included in each of the aforementioned predictive control results, the tactile feature vector and the predicted dredging efficiency are obtained by parsing. The tactile feature vectors are compared with the preset pipeline safety constraints and the preset motion state constraints respectively, and the first preset control result corresponding to the tactile feature vectors that meet the preset pipeline safety constraints and the preset motion state constraints is selected. Based on the predicted dredging efficiency and uncertainty measure corresponding to the first preset control result, a comprehensive evaluation value is calculated. The candidate control command corresponding to the first preset control result whose comprehensive evaluation value is greater than the preset comprehensive evaluation value is determined as the target control command.

8. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 2, characterized in that, The step of controlling the dredging actions of the multi-joint robotic arm and the nozzle according to the target control command includes: The target control command is analyzed to obtain the desired end effector trajectory of the multi-joint robotic arm and the desired injection parameters of the nozzle; After converting the desired end trajectory into a sequence of desired angles for each joint of the multi-joint robotic arm, a coordinated control signal sequence is generated by combining the desired injection parameters. Based on the coordinated control signal sequence, the hydraulic actuators of each joint and the adjustment mechanism of the nozzle are driven to perform the dredging action.

9. The adaptive control method for the hydraulic dredging robot for raw water pipelines according to claim 8, characterized in that, The step of performing the dredging action includes: The execution status of the dredging action is monitored by the joint encoder of the multi-joint robotic arm and the pressure sensor of the nozzle, and status feedback information is obtained. Based on the state feedback information and the cooperative control signal sequence, the execution deviation of the current control cycle is calculated, wherein the execution deviation is used to compensate and correct the cooperative control signal sequence of the next control cycle.

Images