A multi-sensor-based ROV constant-distance control method and system

Abstract

The application provides a kind of based on multi-sensor ROV constant distance control method and system adhering to, it is related to underwater robot control technical field, comprising the following steps: S1: multi-source heterogeneous sensor high frequency synchronous acquisition;S2: observation noise adaptive evaluation based on environmental disturbance;S3: multi-source data depth fusion and state estimation, with the inertial measurement unit data as time update leading input, the prior state estimation value of ROV is calculated by nonlinear kinematic state equation;S4: composite sliding mode control and thrust distribution, the difference between the posterior normal distance estimation value and the preset target target distance generates equidistance error signal;The equidistance error signal is input to the adaptive sliding mode controller containing jet reaction force feedforward compensation, and the six-degree-of-freedom thrust control command is solved output;The application dynamically allocates the trust weight of each sensor according to ultrasonic ranging variance and DVL beam correlation coefficient in real time.

Description

Technical Field

[0001] This invention relates to the field of underwater robot control technology, and in particular to a method and system for ROV wall-hugging constant distance control based on multiple sensors. Background Technology

[0002] With the rapid expansion of the global shipping industry and increasingly stringent environmental regulations such as the International Maritime Organization's guidelines on biofouling management, the treatment of biofouling on the underwater hull surfaces of ships has become a crucial step for the shipping industry to reduce operating costs. Barnacles, mussels, algae, and other organisms attached to ship hulls can increase navigation resistance by 15% to 40%, resulting in additional fuel consumption for the global shipping industry annually. Traditional manual underwater cleaning methods suffer from high safety risks, low efficiency, and significant limitations imposed by sea conditions, making them unsuitable for the industry's evolving needs.

[0003] Using a remotely operated underwater robot equipped with a cavitation jet cleaning device is currently recognized in the industry as an efficient and environmentally friendly underwater hull cleaning technology. Its cavitation jet achieves physical removal of bio-attachments through micro-jet and shock waves generated by bubble collapse, and has advantages such as high cleaning efficiency, minimal damage to coatings, and no chemical pollution.

[0004] However, in existing technologies, underwater robot constant-distance control methods often use a single ultrasonic rangefinder as the distance feedback source or a simple multi-sensor fusion scheme with fixed weights. These methods exhibit extremely poor robustness against bubble interference and cannot adapt to cavitation jet operation environments. Secondly, cavitation jets generate a normal thrust of 5–30 N in the direction of the ROV cleaning head. This thrust acts directly on the ROV body and is one of the main sources of disturbance causing it to deviate from the target distance. The disturbance caused by the jet's reaction force cannot be effectively suppressed. Under high-pressure operating conditions, the instantaneous changes in the jet's reaction force often exceed the controller's adjustment capabilities, causing the ROV to suddenly detach from the hull surface or violently impact the hull, resulting in costly damage to the antifouling coating and equipment failure. Thirdly, existing solutions typically use simplified state estimation models with insufficient state vector dimensions, low accuracy of integration methods, and limited measurement information. Furthermore, existing underwater robot control platforms often use general-purpose ARM architecture processors and non-real-time operating systems, resulting in limited computing power and an inability to prioritize critical control tasks, leading to insufficient real-time control and an inability to quickly respond to complex underwater disturbances. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-sensor-based ROV wall-hugging constant distance control method and system, which solves the technical problems existing in the prior art.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows:

[0007] A multi-sensor-based ROV wall-hugging constant distance control method and system includes the following steps: S1: High-frequency synchronous acquisition by multi-source heterogeneous sensors, acquiring ROV normal distance data between the ROV and the working surface output by an ultrasonic ranging array arranged on the ROV body, ROV triaxial translational velocity data relative to the working surface output by a Doppler log, ROV triaxial acceleration data and triaxial attitude angular rate data output by an inertial measurement unit, and working reaction force data output by a six-axis force / torque sensor; S2: Adaptive evaluation of observation noise based on environmental disturbance, calculating the bubble interference intensity of the current environment based on the statistical characteristics of the normal distance data within a preset sliding time window; calculating the bottom surface tracking quality based on the beam correlation coefficient of the Doppler log; dynamically adjusting the measurement noise covariance moment of the multi-source extended Kalman filter based on the bubble interference intensity and the bottom surface tracking quality. S3: Multi-source data deep fusion and state estimation. Using the inertial measurement unit data as the primary input for time update, the prior state estimate of the ROV is calculated through nonlinear kinematic state equations. The measurement noise covariance matrix adjusted in step S2 is introduced to update the normal distance data and three-axis translational velocity data, and output the posterior optimal normal distance estimate and three-dimensional attitude estimate of the ROV relative to the working surface. S4: Composite sliding mode control and thrust allocation. The posterior normal distance estimate is subtracted from the preset target distance to generate an equidistant error signal. The equidistant error signal is input to an adaptive sliding mode controller that includes jet reaction force feedforward compensation, and a six-degree-of-freedom thrust control command is calculated and output. The six-degree-of-freedom thrust control command is mapped to the ROV's vector thrusters through the thrust allocation matrix to drive the ROV to maintain a constant distance from the working surface.

[0008] Furthermore, the dynamic adjustment of the measurement noise covariance matrix in step S2 includes the following steps: T1: When the bubble interference intensity exceeds a preset bubble density threshold, increase the elements of the measurement noise covariance matrix corresponding to the ultrasonic ranging array according to a first monotonically increasing function; T2: When the bottom surface tracking quality is lower than a preset tracking quality threshold, increase the elements of the measurement noise covariance matrix corresponding to the Doppler speedometer according to a second monotonically increasing function; T3: Combine the adjusted ultrasonic measurement noise covariance matrix and the Doppler measurement noise covariance matrix in a block diagonal form to obtain the final measurement noise covariance matrix. .

[0009] Furthermore, the bubble interference intensity in step S2 is determined by the variance of the normal distance data within the sliding time window. The bottom surface tracking quality is characterized by the cross-correlation coefficients of the four acoustic beams of a Doppler log. Characterization; the measurement noise covariance matrix Satisfying the function mapping relationship: ,in Follow Monotonic and non-decreasing, following Monotonic and non-increasing.

[0010] Furthermore, the control law of the adaptive sliding mode controller including jet reaction force feedforward compensation in step S4 is as follows: ,in, This is the equivalent control term based on the ROV hydrodynamic model; This refers to a switching control term employing a boundary layer saturation function; the switching control term use Form, among which, To switch gain, This is the boundary layer thickness, used to suppress and control chattering; This is the compensation term for ocean current and swell disturbances output by the nonlinear disturbance observer; The normal reverse thrust of the cleaning jet is collected based on a six-axis force / torque sensor. The generated feedforward compensation term.

[0011] Furthermore, in step S4, the sliding surface of the adaptive sliding mode controller is designed as follows: ,in, This is the difference between the posterior optimal normal distance estimate and the preset target distance. The convergence rate parameter is greater than 0.

[0012] Furthermore, the multi-source extended Kalman filtering in step S3 includes the following steps: Q1: Define the system state vector It includes 10 state variables, including attitude quaternions, three-axis velocity, and three-dimensional position; Q2: In the time update stage, the fourth-order Runge-Kutta method is used to integrate the nonlinear kinematic state equations to obtain the prior state estimates. and prior covariance matrix Q3: During the measurement update phase, a 7-dimensional measurement vector is constructed, containing 4 ultrasonic distance signals and 3 Doppler velocities. Calculate the Kalman gain matrix Q4: Correct the prior state estimate based on the Kalman gain matrix to obtain the posterior state estimate. and post-validation contrast matrix .

[0013] Furthermore, the ultrasonic ranging array in step S1 includes at least four ultrasonic rangefinders arranged symmetrically in a rectangle at the four corners of the ROV body, with a sampling frequency of not less than 100Hz; the inertial measurement unit is arranged near the center of mass of the ROV, with a sampling frequency of not less than 500Hz; the Doppler odometer is arranged directly below the bottom of the ROV, with a sampling frequency of not less than 10Hz; and the six-axis force / torque sensor is installed on the mounting surface of the ROV cleaning head, with a sampling frequency of not less than 500Hz.

[0014] Furthermore, steps S3 and S4 are both executed in real time on the edge computing master node based on the RISC-V architecture, with a total computation latency of no more than 2ms per cycle.

[0015] Furthermore, it also includes a permanent magnet adsorption module, which is integrated into the four drive wheels of the ROV chassis, with a single wheel permanent magnet adsorption force of not less than 100N; the permanent magnet adsorption module works in cooperation with the vector propulsion execution module, with the permanent magnet adsorption module providing passive and stable adsorption force and the vector propulsion execution module providing dynamic disturbance compensation force.

[0016] Furthermore, it includes the following modules: a multi-source sensor module, including an ultrasonic ranging array, an inertial measurement unit, a Doppler log, and a six-axis force / torque sensor, used to collect multi-source heterogeneous operational data of the ROV; an edge computing main control module, based on a RISC-V architecture with a computing power of at least 40 TOPS, used to run multi-source extended Kalman filter fusion algorithm and adaptive sliding mode control algorithm; a real-time operating system module, mounted on the edge computing main control module, including a PREEMPT-RT real-time scheduling kernel patch, used to allocate independent high-priority scheduling strategies for the filtering fusion thread and the control calculation thread, and to achieve high-precision clock synchronization of multi-source sensor data; a coprocessor microcontroller module, which communicates at high speed with the edge computing main control module, used for the acquisition of raw sensor signals at the underlying level and the output of thruster PWM control signals; and a vector propulsion execution module, including at least four vector thrusters and their bidirectional electronic speed controllers, used to receive and execute the six-degree-of-freedom thrust control commands output by the edge computing main control module.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] (i) The present invention dynamically allocates the trust weights of each sensor in real time based on the ultrasonic ranging variance and the DVL beam correlation coefficient. When the high-density bubbles generated by the cavitation jet cause severe jumps in the ultrasonic signal, the system automatically reduces its dependence on ultrasonic data and instead uses IMU inertial integral and DVL velocity constraints as the main factors for state estimation. When the DVL bottom surface tracking quality deteriorates, the DVL weight is automatically reduced and the fusion weight of ultrasonic and IMU is enhanced.

[0019] (ii) This invention utilizes a six-axis force / torque sensor installed on the cleaning head mounting surface to collect the normal reverse thrust of the cleaning jet in real time, and injects it directly into the control law as a feedforward compensation quantity, thereby eliminating the disturbance of the jet reaction force on the ROV attitude from the root and replacing the conservative approach of passively increasing the control gain in the traditional solution.

[0020] (III) This invention constructs a 10-dimensional system state vector containing attitude quaternions, three-axis velocity and three-dimensional position. It uses the fourth-order Runge-Kutta method to perform high-precision integration of the nonlinear kinematic equations and constructs a 7-dimensional measurement vector containing four ultrasonic distances and three Doppler velocities for optimal fusion. This achieves deep complementarity of multi-source heterogeneous data. Even when a single sensor fails or enters the near-wall blind zone, the system can still maintain continuous and stable pose estimation through the redundant information of other sensors, avoiding the control open-loop and attitude instability problems caused by the failure of a single sensor in the traditional scheme.

[0021] (iv) This invention assigns an independent real-time high-priority scheduling strategy to the extended Kalman filter fusion thread and the sliding mode controller solution thread, and controls the synchronization error of multi-source sensor data to within 1ms through a high-precision clock synchronization mechanism, which can quickly respond to complex and ever-changing underwater disturbances and ensure the dynamic performance of constant distance control.

[0022] (v) This invention combines non-contact multi-sensor distance measurement technology and is applicable to hulls made of various materials such as steel, aluminum alloy, and fiberglass. Its versatility is significantly better than the pure magnetic track solution that is only applicable to steel hulls. Attached Figure Description

[0023] Figure 1 The flowchart below shows a multi-sensor-based ROV wall-hugging constant distance control method and system disclosed in this invention.

[0024] Figure 2 This is a framework diagram of a multi-sensor-based ROV wall-hugging constant distance control method and system disclosed in this invention. Detailed Implementation

[0025] To make the content of this invention easier to understand, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Identical components are represented by the same reference numerals. It should be noted that the terms "front," "rear," "left," "right," "up," and "down" used in the following description refer to directions in the accompanying drawings, while the terms "inner" and "outer" refer to directions toward or away from the geometric center of a specific component, respectively.

[0026] Example 1:

[0027] like Figure 1-2As shown, this embodiment provides a multi-sensor-based ROV wall-hugging constant distance control method and system, including the following steps:

[0028] S1: High-frequency synchronous acquisition by multi-source heterogeneous sensors, acquiring data including the normal distance between the ROV and the working surface output by the ultrasonic ranging array arranged on the ROV body, the three-axis translational velocity data of the ROV relative to the working surface output by the Doppler log, the three-axis acceleration data and three-axis attitude angular rate data of the ROV output by the inertial measurement unit, and the working reaction force data output by the six-axis force / torque sensor; the ultrasonic ranging array in step S1 includes at least 4 ultrasonic rangefinders, arranged symmetrically in a rectangle at the four corners of the ROV body, with a sampling frequency of not less than 100Hz; the inertial measurement unit is arranged near the center of mass of the ROV, with a sampling frequency of not less than 500Hz; the Doppler log is arranged directly below the bottom of the ROV, with a sampling frequency of not less than 10Hz; the six-axis force / torque sensor is installed on the mounting surface of the ROV cleaning head, with a sampling frequency of not less than 500Hz; specifically, in this embodiment, 4 MaxBotix MaxSonar WR sensors are used. An MB7360 underwater ultrasonic rangefinder, with a range of 20–700 mm, an accuracy of ±1 mm, and a sampling frequency of 125 Hz, is positioned at the four corners of the ROV chassis, forming a rectangular symmetrical array. A WaterLinked DVL-A50 Doppler speedometer, with a range of 0.05–10 m and a sampling frequency of 10 Hz, is positioned directly below the bottom of the ROV. A VectorNav VN-100 9-axis IMU, with a sampling frequency of 800 Hz and an attitude accuracy of 0.3° RMS, is positioned near the ROV's center of mass. An ATI sensor is also used. The Mini45 waterproof modified version features a six-axis force / torque sensor with a range of 100N and a sampling frequency of 500Hz, mounted on the cleaning head's mounting surface. After precise mechanical alignment and external parameter calibration, the multi-source sensors aggregate their collected data via SPI, CAN bus, or Gigabit Ethernet to the co-processing MCU, which then forwards it to the edge computing master node. It should be noted that all sensor data is aligned using a high-precision clock synchronization mechanism within the real-time operating system, with clock synchronization errors controlled within 1ms.

[0029] S2: Adaptive assessment of observation noise based on environmental disturbances, calculating the bubble interference intensity of the current environment based on the statistical characteristics of the normal distance data within a preset sliding time window; calculating the bottom tracking quality based on the beam correlation coefficient of the Doppler log; dynamically adjusting the measurement noise covariance matrix of the multi-source extended Kalman filter based on the bubble interference intensity and the bottom tracking quality;

[0030] The dynamic adjustment of the measurement noise covariance matrix includes the following steps:

[0031] T1: When the bubble interference intensity exceeds the preset bubble density threshold, the elements of the measurement noise covariance matrix corresponding to the ultrasonic ranging array are increased according to the first monotonically increasing function.

[0032] This invention uses the variance of ultrasonic ranging data within a sliding time window. As a quantitative indicator of bubble interference intensity, its physical essence lies in the fact that microbubbles generated by cavitation jets scatter and absorb ultrasonic echo energy, causing random jumps and fluctuations in ranging values. The higher the bubble density, the greater the dispersion and variance of the ranging data. The larger the value, the better; the preset sliding time window length is set to N=50, which was determined through extensive experimental optimization. This length can reflect the bubble interference level at the current moment in real time, and effectively suppress the instantaneous influence of random noise, avoiding frequent and drastic fluctuations in the measurement noise covariance; preset bubble density threshold. The typical value is ,when > When the condition is determined to be a severe bubble interference condition, the reliability of the ultrasonic ranging data decreases significantly.

[0033] The first monotonically increasing function takes the form of a linear amplification, and its specific expression is as follows:

[0034] ,

[0035] in, The basic measurement noise covariance for ultrasonic ranging corresponds to the inherent measurement error in a bubble-free environment. The noise amplification factor is used to control the rate at which the covariance increases with the interference intensity. The function ensures that the measurement noise covariance is monotonically non-decreasing with the bubble interference intensity, that is, the stronger the interference, the lower the system's confidence in the ultrasonic data.

[0036] T2: When the bottom surface tracking quality is lower than the preset tracking quality threshold, the elements of the measurement noise covariance matrix corresponding to the Doppler speed meter are increased according to the second monotonically increasing function.

[0037] In this embodiment, a sliding time window of length N=50 is defined; the Doppler log measures the triaxial translational velocity of the ROV relative to the hull surface by emitting four acoustic beams underwater and receiving the reflected echoes, and the cross-correlation coefficients of the four beams are also measured. This directly reflects the quality of the bottom echo signal and the accuracy of the velocity measurement; The closer it is to 1, the better the consistency of the echo signals of the four beams and the higher the bottom surface tracking quality. The lower the value, the worse the bottom surface reflection conditions. This is especially true if the water is turbid, has a rough biofilm layer, or is blocked by air bubbles. The larger the speed measurement error, the greater the preset tracking quality threshold. The typical value is 0.7, when < At that time, it was determined to be Track operating conditions that cause quality degradation;

[0038] The second monotonically increasing function also adopts a linear amplification form, and its specific expression is as follows:

[0039] ,in, for The basic measurement noise covariance of speed measurement corresponds to the inherent measurement error under ideal bottom surface tracking conditions; This is the noise amplification factor; this function ensures that the measurement noise covariance monotonically does not increase with the bottom surface tracking quality, meaning that the worse the tracking quality, the weaker the system's response to noise. The lower the trust level of the data;

[0040] T3: Combine the adjusted ultrasonic measurement noise covariance matrix and the Doppler measurement noise covariance matrix in a block diagonal form to obtain the final measurement noise covariance matrix. ;

[0041] Due to ultrasonic ranging and The measurement noise from velocity measurements is physically independent and uncoupled; therefore, a block diagonal matrix structure can accurately reflect the statistical characteristics of multi-source measurement noise. The final synthesized noise... It is a 7×7 block diagonal matrix, where the first 4 diagonal elements correspond to the measurement noise of the 4 ultrasonic rangefinders, and the last 3 diagonal elements correspond to... The measurement noise of the three-axis velocity is 0 for all off-diagonal elements.

[0042] The bubble interference intensity is expressed as the variance of the normal distance data within a sliding time window. The bottom surface tracking quality is characterized by the cross-correlation coefficients of the four acoustic beams of a Doppler log. Characterization;

[0043] The measurement noise covariance matrix Satisfying the function mapping relationship: ,in Follow Monotonic and non-decreasing, following Monotonically non-increasing;

[0044] Specifically, the measurement noise covariance matrix Satisfying a unified function mapping relationship:

[0045] ,in, Follow Monotonic and non-decreasing, following Monotonically non-increasing; this mapping relationship enables dynamic intelligent allocation of sensor confidence: when bubble interference increases, the system automatically reduces its reliance on ultrasonic data and instead relies more on IMU inertial integrals and... Speed ​​constraint; when When tracking quality deteriorates, the system automatically reduces the focus. The reliance on data is transformed into an enhanced fusion weight of ultrasound and IMU; the above mechanism fundamentally solves the problem of accuracy degradation of traditional fixed-weight fusion schemes under dynamic interference environment, and greatly improves the robustness of the system to strong bubble interference.

[0046] S3: Deep fusion of multi-source data and state estimation. Using the inertial measurement unit data as the primary input for time update, the prior state estimate of the ROV is calculated through nonlinear kinematic state equations. The measurement noise covariance matrix adjusted in step S2 is introduced to update the normal distance data and three-axis translational velocity data, and the posterior optimal normal distance estimate and three-dimensional attitude estimate of the ROV relative to the working surface are output.

[0047] The multi-source extended Kalman filter includes the following steps:

[0048] Q1: Define the system state vector It has 10 dimensions, including 10 state variables: attitude quaternion, three-axis velocity, and three-dimensional position;

[0049] Q2: During the time update phase, the triaxial acceleration and angular rate output from the IMU are used as the primary inputs. The fourth-order Runge-Kutta method is used to integrate the nonlinear kinematic state equation at 500Hz to obtain the prior state estimate. and prior covariance matrix ;

[0050] Q3: During the measurement update phase, a 7-dimensional measurement vector is constructed, containing 4 ultrasonic distance signals and 3 Doppler velocities. Calculate the Kalman gain matrix ; Among them, the first 4 dimensions For 4-channel ultrasonic ranging, unit: mm; unit: mm; then 3D. This is the DVL three-axis velocity, in m / s, directly outputting the measured velocity value;

[0051] Q4: Correct the prior state estimate based on the Kalman gain matrix to obtain the posterior state estimate. and post-validation contrast matrix Kalman gain calculation: Post-hoc state update: Synchronously update the covariance matrix: From the posterior state The paper proposes: Normal distance estimate That is, the 10th dimension of the state vector; the three-dimensional attitude estimate. The output is normalized and then sent to the next stage for ROV navigation and control.

[0052] Both steps S3 and S4 are executed in real time on the edge computing master node based on the RISC-V architecture, with a total computation latency of no more than 2ms per cycle.

[0053] S4: Composite sliding mode control and thrust distribution: The posterior normal distance estimate is subtracted from the preset target distance to generate an equidistant error signal; the equidistant error signal is input to an adaptive sliding mode controller that includes jet reaction force feedforward compensation, and a six-degree-of-freedom thrust control command is calculated and output; the six-degree-of-freedom thrust control command is mapped to the ROV's vector thruster through the thrust distribution matrix, driving the ROV to maintain a constant distance from the working surface;

[0054] In this embodiment, the target distance is set. (Optimal operating distance for cavitation jet); the sliding surface design of the adaptive sliding mode controller is as follows:

[0055] ,in, This is the difference between the posterior optimal normal distance estimate and the preset target distance. is a convergence rate parameter greater than 0; where, This is the convergence rate parameter;

[0056] The control law of the adaptive sliding mode controller for jet reaction force feedforward compensation is as follows:

[0057] ,in, This is the equivalent control term based on the ROV hydrodynamic model; This refers to a switching control term employing a boundary layer saturation function; the switching control term use Form, among which, To switch gain, This is the boundary layer thickness, used to suppress and control chattering; This is the compensation term for ocean current and swell disturbances output by the nonlinear disturbance observer; The normal reverse thrust of the cleaning jet is collected based on a six-axis force / torque sensor. The generated feedforward compensation term; ultimately, the total control torque. The thrust commands for the four vector thrusters are obtained by inverse calculation of the 4×6 thrust distribution matrix, and then executed by the BlueRobotics T200 vector thrusters driven by the bidirectional electronic speed controller.

[0058] The permanent magnet adsorption module is integrated into the four drive wheels of the ROV chassis, with a single wheel permanent magnet adsorption force of no less than 100N. The permanent magnet adsorption module works in conjunction with the vector propulsion execution module, with the permanent magnet adsorption module providing passive and stable adsorption force and the vector propulsion execution module providing dynamic disturbance compensation force. When facing the steel hull surface, the permanent magnet adsorption module provides passive and stable adsorption force, and the vector propulsion execution module provides dynamic disturbance compensation force. The two work together to achieve stable wall-hull operation.

[0059] A multi-sensor-based ROV wall-hugging constant-distance control system includes the following modules:

[0060] The multi-source sensor module, including an ultrasonic ranging array, an inertial measurement unit, a Doppler log, and a six-axis force / torque sensor, is used to collect multi-source heterogeneous operational data of the ROV.

[0061] The edge computing main control module, based on the RISC-V architecture, has a computing power of at least 40 TOPS and is used to run multi-source extended Kalman filter fusion algorithm and adaptive sliding mode control algorithm; it has at least 8GB of onboard RAM and supports a wide-temperature industrial-grade operating environment of -40℃ to 85℃; this module adopts the design concept of heavy edge and light cloud, and executes all core control and fusion computing tasks completely locally, thus completely avoiding the impact of underwater communication delay on the control closed loop.

[0062] The real-time operating system module, mounted on the edge computing main control module, includes a PREEMPT-RT real-time scheduling kernel patch, which is used to allocate independent high-priority scheduling strategies for the filtering and fusion thread and the control calculation thread, and to achieve high-precision clock synchronization of multi-source sensor data; it also prioritizes sensor data receiving interrupts and control command output interrupts to ensure real-time transmission and reception of critical data.

[0063] The coprocessor microcontroller module communicates at high speed with the edge computing master control module for the acquisition of raw signals from the underlying sensors and the output of PWM control signals for the actuators. This module uses a high-performance STM32H7 series microcontroller with a main frequency of no less than 400MHz and rich built-in peripheral interfaces. Specifically, it drives the acquisition of raw data from each sensor through hard real-time interrupts, preprocesses and converts analog signals to digital, and sends the data to the edge computing master control module after packaging it with a unified timestamp.

[0064] The vector propulsion execution module includes at least four vector thrusters and their bidirectional electronic speed controllers. It receives and executes six-degree-of-freedom thrust control commands output from the edge computing main control module. The vector thrusters are waterproof brushless DC vector thrusters with a maximum thrust of at least 5.1 kgf per unit and a response time of less than 10 ms. The four thrusters are arranged in an X-shape on the ROV chassis, achieving complete decoupling control of the six degrees of freedom through thrust vector synthesis. They can independently generate thrust in any direction and torque around any axis. The bidirectional electronic speed controller is a waterproof electronic speed controller supporting bidirectional speed control with a response time of less than 5 ms and overcurrent, overvoltage, and overheat protection functions. This speed controller receives PWM signals output from the coprocessor microcontroller unit and precisely adjusts the thruster's speed and direction to achieve continuous and smooth thrust adjustment. The thrust distribution matrix system has a built-in 4×6 order thrust distribution matrix. It uses a pseudo-inverse method to inversely decompose the six-degree-of-freedom thrust control commands into independent thrust commands for each thruster, ensuring optimal thrust output distribution and minimum energy consumption.

[0065] Example 2:

[0066] Based on Example 1, this example provides a complete hardware architecture for a multi-sensor-based ROV wall-hugging constant distance control system.

[0067] Edge computing main control module

[0068] It adopts an industrial-grade edge computing core board based on the Pingtouge XuanTie C906 or equivalent RISC-V architecture, with at least 8GB of RAM, a computing power of at least 40 TOPS, and support for an industrial-grade temperature range of -40℃ to 85℃. This node runs an open-source operating system based on a heavily customized Debian architecture with real-time PREEMPT-RT patches, and the scheduling latency of critical threads is controlled within 100%. Within this range, the system assigns independent real-time high-priority scheduling strategies to the extended Kalman filter fusion thread and the sliding mode controller solution thread to ensure real-time computation.

[0069] Coprocessor microcontroller module

[0070] The system employs an STM32H743VIT6 microcontroller with a main frequency of 480MHz. It drives the sampling of raw signals from each sensor through hard real-time interrupts and sends the data to the edge computing master node at a 500Hz cycle via a gigabit Ethernet bus. The coprocessor MCU is also responsible for generating high-precision PWM control signals for the vector thruster and magnetic drive wheel motor. The PWM resolution is 16 bits and the frequency is 20kHz.

[0071] Multi-source sensor module

[0072] include:

[0073] Ultrasonic ranging array: 4 MaxBotix MaxSonar WR MB7360 underwater ultrasonic rangefinders, arranged symmetrically in a rectangle at the four corners of the ROV chassis;

[0074] Inertial Measurement Unit: VectorNav VN-100 9-axis IMU, located at the geometric center of the ROV;

[0075] Doppler log: WaterLinked DVL-A50 model, installed at the front bottom of the ROV;

[0076] Six-axis force / torque sensor: ATI Mini45 waterproof modified version, installed at the connection flange between the cleaning head and the ROV body.

[0077] Vector propulsion execution module

[0078] It employs four BlueRobotics T200 vector thrusters, each with a maximum thrust of 5.1 kgf, and four BlueRobotics Basic ESC R3 bidirectional electronic speed controllers, with a response time of <5 ms. The four thrusters are distributed in an X-shape on the ROV chassis, forming a 6-DOF decoupled vector thrust field, capable of independently generating thrust along the X, Y, and Z axes and torque around the X, Y, and Z axes.

[0079] Example 3:

[0080] To verify the effectiveness of the invention, a control experiment was conducted in a wave pool measuring 10m × 4m × 2m; the experimental target surface was a 3m × 2m steel plate coated with antifouling paint, and the target distance was... The standard deviation of the equidistant error of the method of this invention was compared with that of the existing technology of single ultrasonic wave + fixed weight EKF under three operating conditions: still water, low-frequency swell of Sea State 1 wave generator, and medium-frequency swell of Sea State 2 wave generator. The experimental results are shown in the table below.

[0081]

[0082] In addition, the cavitation jet cleaning system was activated to test the jet reaction force feedforward term described in this invention. The effect; the equidistant error of ROV when feedforward is not enabled. Rise to 15-20mm; after enabling feedforward, The relative improvement rate is about 60% when the jet reaction force is controlled within the range of 6 to 8 mm; this data proves the significant effectiveness of the active feedforward compensation mechanism of the present invention.

[0083] Meanwhile, in the computing power test, the edge computing master node based on the RISC-V architecture described in this invention was used to perform complete multi-source EKF fusion and ASMC control law calculation. The single cycle time was stable between 1.2 and 1.8 ms, with a peak value of no more than 2 ms, which meets the hard real-time requirements of the 500Hz control closed loop.

[0084] The method and system described in this invention are not only applicable to underwater hull cleaning scenarios, but can also be extended to the following underwater operation scenarios requiring high-precision constant distance control:

[0085] (1) Underwater non-destructive testing of ship hulls: ROVs need to maintain a constant distance from the ship hull surface to obtain uniform ultrasonic or eddy current testing signals;

[0086] (2) Underwater weld inspection and maintenance: The ROV travels along the weld and maintains a constant distance to perform video inspection or local repair;

[0087] (3) Maintenance of offshore wind turbine foundations: ROV performs wall cleaning, coating inspection or cathodic protection maintenance on the underwater part of the offshore wind turbine facility;

[0088] (4) Underwater archaeology: ROV performs high-precision photogrammetric modeling on the surface of underwater archaeological sites;

[0089] (5) Cleaning of aquaculture cages: ROVs are used to remove biofouling from the surface of large cages.

[0090] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-sensor-based ROV wall-hugging constant distance control method, characterized in that: Includes the following steps: S1: High-frequency synchronous acquisition by multi-source heterogeneous sensors, acquiring the normal distance data between the ROV and the working surface output by the ultrasonic ranging array arranged on the ROV body, the three-axis translational velocity data of the ROV relative to the working surface output by the Doppler log, the three-axis acceleration data and three-axis attitude angular rate data of the ROV output by the inertial measurement unit, and the working reaction force data output by the six-axis force / torque sensor; S2: Adaptive assessment of observation noise based on environmental disturbances, calculating the bubble interference intensity of the current environment based on the statistical characteristics of the normal distance data within a preset sliding time window; calculating the bottom tracking quality based on the beam correlation coefficient of the Doppler log; dynamically adjusting the measurement noise covariance matrix of the multi-source extended Kalman filter based on the bubble interference intensity and the bottom tracking quality; S3: Deep fusion of multi-source data and state estimation. Using the inertial measurement unit data as the primary input for time update, the prior state estimate of the ROV is calculated through nonlinear kinematic state equations. The measurement noise covariance matrix adjusted in step S2 is introduced to update the normal distance data and three-axis translational velocity data, and the posterior optimal normal distance estimate and three-dimensional attitude estimate of the ROV relative to the working surface are output. S4: Composite sliding mode control and thrust distribution: The posterior normal distance estimate is subtracted from the preset target distance to generate an equidistant error signal; the equidistant error signal is input to an adaptive sliding mode controller that includes jet reaction force feedforward compensation, and a six-degree-of-freedom thrust control command is calculated and output; the six-degree-of-freedom thrust control command is mapped to the ROV's vector thruster through the thrust distribution matrix, driving the ROV to maintain a constant distance from the working surface.

2. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1, characterized in that: The dynamic adjustment of the measurement noise covariance matrix in step S2 includes the following steps: T1: When the bubble interference intensity exceeds the preset bubble density threshold, the elements of the measurement noise covariance matrix corresponding to the ultrasonic ranging array are increased according to the first monotonically increasing function. T2: When the bottom surface tracking quality is lower than the preset tracking quality threshold, the elements of the measurement noise covariance matrix corresponding to the Doppler speed meter are increased according to the second monotonically increasing function. T3: Combine the adjusted ultrasonic measurement noise covariance matrix and the Doppler measurement noise covariance matrix in a block diagonal form to obtain the final measurement noise covariance matrix. .

3. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1 or 2, characterized in that: The bubble interference intensity in step S2 is measured by the variance of the normal distance data within the sliding time window. The bottom surface tracking quality is characterized by the cross-correlation coefficients of the four acoustic beams of a Doppler log. Characterization; The measurement noise covariance matrix Satisfying the function mapping relationship: ,in Follow Monotonic and non-decreasing, following Monotonic and non-increasing.

4. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 3, characterized in that: The control law of the adaptive sliding mode controller that includes jet reaction force feedforward compensation in step S4 is as follows: ,in, This is the equivalent control term based on the ROV hydrodynamic model; This refers to a switching control term employing a boundary layer saturation function; the switching control term use Form, among which, To switch gain, This is the boundary layer thickness, used to suppress and control chattering; This is the compensation term for ocean current and swell disturbances output by the nonlinear disturbance observer; The normal reverse thrust of the cleaning jet is collected based on a six-axis force / torque sensor. The generated feedforward compensation term.

5. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 4, characterized in that: The sliding surface design of the adaptive sliding mode controller in step S4 is as follows: ,in, This is the difference between the posterior optimal normal distance estimate and the preset target distance. The convergence rate parameter is greater than 0.

6. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1, characterized in that: Multi-source extended Kalman filtering in step S3 Includes the following steps: Q1: Define the system state vector It includes 10 state variables, including attitude quaternions, three-axis velocity, and three-dimensional position; Q2: During the time update phase, the fourth-order Runge-Kutta method is used to integrate the nonlinear kinematic state equations to obtain the prior state estimates. and prior covariance matrix ; Q3: During the measurement update phase, a 7-dimensional measurement vector is constructed, containing 4 ultrasonic distance signals and 3 Doppler velocities. Calculate the Kalman gain matrix ; Q4: Correct the prior state estimate based on the Kalman gain matrix to obtain the posterior state estimate. and post-validation contrast matrix .

7. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1, characterized in that: The ultrasonic ranging array in step S1 includes at least four ultrasonic rangefinders arranged symmetrically in a rectangle at the four corners of the ROV body, with a sampling frequency of not less than 100Hz; the inertial measurement unit is arranged near the center of mass of the ROV, with a sampling frequency of not less than 500Hz; the Doppler odometer is arranged directly below the bottom of the ROV, with a sampling frequency of not less than 10Hz; and the six-axis force / torque sensor is installed on the mounting surface of the ROV cleaning head, with a sampling frequency of not less than 500Hz.

8. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1, characterized in that: Both steps S3 and S4 are executed in real time on the edge computing master node based on the RISC-V architecture, with a total computation latency of no more than 2ms per cycle.

9. The ROV wall-hugging constant distance control method based on multiple sensors according to claim 1, characterized in that: It also includes a permanent magnet adsorption module, which is integrated into the four drive wheels of the ROV chassis, with a single wheel permanent magnet adsorption force of not less than 100N; the permanent magnet adsorption module works in coordination with the vector propulsion execution module, with the permanent magnet adsorption module providing passive and stable adsorption force and the vector propulsion execution module providing dynamic disturbance compensation force.

10. A multi-sensor-based ROV wall-hugging constant-distance control system, used to execute the method as described in any one of claims 1-9, characterized in that, Includes the following modules: The multi-source sensor module, including an ultrasonic ranging array, an inertial measurement unit, a Doppler log, and a six-axis force / torque sensor, is used to collect multi-source heterogeneous operational data of the ROV. The edge computing main control module, based on the RISC-V architecture, has a computing power of at least 40 TOPS and is used to run multi-source extended Kalman filter fusion algorithm and adaptive sliding mode control algorithm. The real-time operating system module, mounted on the edge computing main control module, includes a PREEMPT-RT real-time scheduling kernel patch, which is used to allocate independent high-priority scheduling strategies for the filtering and fusion thread and the control and calculation thread, and to achieve high-precision clock synchronization of multi-source sensor data. The coprocessor microcontroller module communicates at high speed with the edge computing main control module and is used for the acquisition of raw signals from the underlying sensors and the output of PWM control signals for the thrusters. The vector propulsion execution module includes at least four vector thrusters and their bidirectional electronic speed controllers, which are used to receive and execute the six-degree-of-freedom thrust control commands output by the edge computing master control module.

Images