Ship underwater intelligent cleaning control system and method based on multi-zone plume feedback

Abstract

The application discloses a ship underwater intelligent cleaning control system and method based on multi-region plume feedback, and the system comprises a multi-region plume sensing module, a data recording module, a control module and a high-pressure water jet supply module.

Description

Technical Field

[0001] This invention relates to the field of ship cleaning technology, and in particular to an intelligent underwater cleaning control system and method for ships based on multi-regional plume feedback. Background Technology

[0002] With the increasing prominence of marine organism attachment, fouling layer accumulation, and antifouling coating aging during long-term ship operation, underwater cleaning technology has gradually become an important research direction in the fields of ship operation and maintenance and marine equipment maintenance. Existing underwater hull cleaning methods mainly include mechanical brushing, high-pressure water jet cleaning, cavitation water jet cleaning, and automated underwater robotic cleaning. Among these, high-pressure water jet cleaning, due to its advantages of high cleaning efficiency, strong adaptability, and wide applicability to complex curved surfaces, has been widely used in underwater cleaning operations of ship hull plating, biofouling layers, and dock structures.

[0003] Currently, underwater cleaning systems for ships are gradually evolving from simple mechanical cleaning towards intelligent and automated processes. Some solutions already possess the capabilities of jet parameter adjustment, sensor detection, or cleaning path control, which has improved the efficiency and automation level of underwater ship cleaning to a certain extent. However, most existing technical solutions still rely on fixed-parameter jetting or local parameter adjustment, focusing primarily on the optimization of the cleaning device structure, the design of the jetting mechanism, or the motion control of the robot. There is a lack of systematic research on the complex plume diffusion behavior formed after high-pressure water jets act on the ship's hull surface, resulting in the following problems in actual underwater cleaning processes: First, existing technologies generally lack the ability to perceive the overall spatial distribution of plumes. After a high-pressure water jet impacts the hull wall, the attached material is stripped off at high speed, forming a large number of suspended particle plumes. These plumes exhibit complex diffusion behaviors due to local water flow, jet pressure, and the curved surface structure of the hull, including near-wall backflow, local retention, resuspension, and far-field diffusion. However, most existing solutions only use single-point turbidity detection or local image detection methods, which cannot distinguish whether the current turbidity state originates from the stripping of effective attached material, local backflow retention, or far-field diffusion, easily leading to misjudgment of the state.

[0004] Secondly, existing technologies lack the ability to continuously assess the dynamic changes in the plume. Due to factors such as water flow disturbances, bubble interference, particle obstruction, and changes in illumination in the underwater environment, underwater plumes exhibit significant time-varying and nonlinear characteristics. Most existing solutions adjust parameters based solely on instantaneous detection results, lacking the ability to dynamically identify plume trends and different cleaning stages. They cannot determine whether the current area is in an effective cleaning stage, a diffusion enhancement stage, or a cleaning completion stage, resulting in a lack of targeted adjustment of jet parameters.

[0005] Furthermore, existing jet control methods mostly employ fixed jet pressure or simple closed-loop regulation, lacking an adaptive control mechanism based on the plume state. In actual cleaning processes, while consistently high jet pressure can improve the removal of deposits, it also easily leads to a large number of particles diffusing into the far-field region, increasing the risk of secondary pollution in ports and docks. Conversely, when local backflow is enhanced, plume particles tend to remain persistently near the ship's hull, reducing image recognition effectiveness and potentially leading to repeated cleaning and energy waste. Therefore, it is difficult to achieve a balance between cleaning efficiency and diffusion suppression.

[0006] Therefore, how to establish an intelligent control system and method that can perceive the spatial distribution of plumes in multiple regions during underwater cleaning of ships and combine the dynamic changes of plumes to realize state estimation, state switching and adaptive closed-loop regulation has become a key technical problem that urgently needs to be solved in the field of underwater ship cleaning. Summary of the Invention

[0007] In view of the above-mentioned prior art, the present invention provides a ship underwater intelligent cleaning control system and method based on multi-region plume feedback, which mainly solves the technical problems existing in the above-mentioned background art.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a ship underwater intelligent cleaning control system based on multi-region plume feedback, comprising: A multi-region plume sensing module is used to establish multiple plume monitoring areas around the nozzle. Sensors are used in these multiple plume monitoring areas to perform spatial distribution sensing and obtain multi-region plume sensing data. The multiple plume monitoring areas include: the area in front of the nozzle, the area behind the nozzle, the near-wall backflow area, the far-field diffusion area, and the background reference area, which correspond to areas A, B, C, D, and E, respectively. The data recording module is used to acquire the injection parameter information data of the injection nozzle; The control module is used to analyze and process the multi-region plume sensing data and the jet parameter information data, generate a jet parameter adjustment vector, and convert the jet parameter adjustment vector into control commands for the jet parameters. A high-pressure water jet supply module is electrically connected to the control module and is used to receive control commands to adjust the injection pressure and flow rate of the high-pressure water jet; the pressure range of the high-pressure water jet supply module is 15~70MPa; The spray execution module is electrically connected to the control module and connected to the high-pressure water jet supply module through a water outlet pipe. It is used to receive the control commands and adjust the spray angle and spray duty cycle of the high-pressure water jet regulated by the high-pressure water jet supply module, as well as adjust the nozzle moving speed, to control the high-pressure water jet to act on the hull wall for cleaning the ship. The control module includes a data preprocessing unit, a plume state evaluation unit, a cleaning state discrimination unit, and an adaptive fuzzy PID control unit connected in sequence. The data preprocessing unit is used to preprocess the multi-region plume sensing data and construct plume state characteristic indicators based on the preprocessed multi-region plume sensing data. The plume state assessment unit is used to construct a plume state estimation model. The plume state characteristic index and the jetting parameter information data are input into the plume state estimation model, and the final plume state estimation result is output. The cleaning state discrimination unit is used to identify the current cleaning state based on the final plume state estimation result; The adaptive fuzzy PID control unit is used to generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and to convert the jet parameter adjustment vector into a control command for the jet parameters.

[0009] Optionally, the plume state characteristic indicators include: washing and shedding indicators, near-wall reflux indicators, far-field diffusion indicators, plume change rate indicators, and image clarity indicators; The expression for the washing and shedding index is:

[0010] in, To clean and remove indicators, This refers to the relative turbidity in region B, which is the area behind the nozzle. This refers to the relative turbidity of region A, i.e., the area in front of the nozzle. The expression for the near-wall reflux index is:

[0011] in, As a near-wall reflux indicator, The relative turbidity in region C, i.e., the near-wall reflux region, is... The relative turbidity in region D, i.e., the far-field diffusion region; The expression for the far-field diffusion index is:

[0012] in, As a far-field diffusion indicator; The expression for the plume change rate index is:

[0013] in, This is an indicator of the rate of change of the plume; The expression for the image sharpness index is:

[0014] in, As an indicator of image sharpness, The image gradient value. This represents the number of pixels in the image.

[0015] Optionally, the processing steps of the plume state assessment unit include: A plume state vector is constructed based on the plume state characteristic index and the jet parameter information data; A plume state estimation model is constructed, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on the radial basis function. The plume state vector is input into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state. The RBF neural network error compensation model performs error compensation on the extended Kalman filter corrected state to obtain the final plume state estimation result.

[0016] Optionally, the step of inputting the plume state vector into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state includes: Obtain the injection parameter adjustment vector of the previous cycle, and define the control input vector based on the injection parameter adjustment vector; The state vector of the plume at the previous moment and the control input vector of the previous cycle are input into the state transition function to predict the state and obtain the predicted value of the plume state vector at the current moment. Calculate the prediction error covariance matrix at the current time. Define an observation model, use the observation model to map the predicted value of the plume state vector to the predicted observation value, calculate the actual observation vector at the current time based on the predicted plume state vector at the current time, update the observation based on the predicted observation value and the actual observation vector, and obtain the innovation error at the current time. The Kalman gain is calculated based on the innovation error and the prediction error covariance matrix. The state is then corrected based on the Kalman gain and the innovation error to obtain the extended Kalman filter corrected state at the current time.

[0017] Optionally, the RBF neural network error compensation model performs error compensation on the extended Kalman filter corrected state to obtain the final plume state estimation result, including: The innovation error obtained during the extended Kalman filtering process is input into the RBF neural network, which calculates the output of the RBF neurons in the hidden layer through radial basis functions. The outputs of all RBF neurons are combined to form the hidden layer output vector, and the error compensation term is calculated based on the hidden layer output vector. The correction error is calculated using the error compensation term, and the extended Kalman filter correction state is compensated a second time based on the correction error to update the plume state and obtain the final plume state estimation result. The weights of the RBF neural network are updated based on the correction error using an online learning method.

[0018] Optionally, the processing steps of the cleaning status determination unit include: The cleaning process is divided into four states: background stability, effective cleaning, large amount of shedding, local reflux, far-field diffusion, cleaning completion, and over-cleaning risk. A combination of primary and auxiliary criteria is used to continuously monitor the plume state characteristic indicators in the final plume state estimation results using a sliding time window. Using the washing and shedding index Near-wall reflux index and far-field diffusion index The absence of significant growth over multiple consecutive sampling periods is used as the primary criterion to determine whether the background is in a stable state. Using the washing and shedding index A sustained upward trend across multiple sampling periods was used as the primary criterion, combined with far-field diffusion indicators. The threshold value is used as an auxiliary criterion to determine whether the cleaning state is effective. Using the washing and shedding index The main criterion is to maintain a rapid upward trend in turbidity over multiple consecutive sampling cycles, and the auxiliary criterion is to combine the continuous increase in turbidity in the area behind the nozzle that is higher than the corresponding increase under effective cleaning conditions to determine whether it is in a state of large-scale shedding. Near-wall reflux index The primary criterion is to exceed a set threshold for multiple consecutive sampling cycles, and the secondary criterion is to combine the change in near-wall velocity as an auxiliary criterion to determine whether the flow is in a local backflow state. Far-field diffusion index The primary criterion is that multiple consecutive sampling periods are higher than the diffusion threshold, and the turbidity change rate in the far field region is used as an auxiliary criterion to determine whether the region is in a far field diffusion state. Using the plume change rate index Maintaining a value close to zero for multiple consecutive sampling periods is used as the primary criterion, combined with image sharpness indicators. Continuous recovery is used as an auxiliary criterion to determine whether the cleaning process is complete. With injection pressure, injection flow rate, and injection pulse duty cycle maintained at a consistently high level, and the cleaning and shedding indicators... The primary criterion for determining whether a state is at risk of over-cleaning is that there is no significant increase in multiple consecutive sampling cycles. If only one of the state criteria is satisfied, the state that satisfies the criterion is taken as the current cleaning state. When multiple state criteria are met simultaneously, the judgment is made in the order of priority: over-cleaning risk state, local backflow state, far-field diffusion state, effective cleaning state, and cleaning completed state, and the state with the highest priority is taken as the current cleaning state.

[0019] Optionally, the adaptive fuzzy PID control unit includes: a fuzzification submodule, a rule inference submodule, and a defuzzification output submodule; The fuzzification submodule is used to convert the plume state feature indicators in the final plume state estimation result into corresponding fuzzy linguistic variables using a preset membership function. The rule reasoning submodule is used to perform fuzzy rule reasoning based on the current cleaning state and the fuzzy linguistic variables corresponding to the characteristic indicators of each plume state, and generate fuzzy reasoning results. The defuzzy output submodule is used to defuzzify the fuzzy inference result using the centroid method to obtain the injection parameter adjustment vector, and to generate control commands for the injection parameters based on the injection parameter adjustment vector; the injection parameter adjustment vector includes: injection pressure adjustment amount, injection flow rate adjustment amount, injection angle adjustment amount, nozzle moving speed adjustment amount, and injection duty cycle adjustment amount.

[0020] Optionally, the high-pressure water jet supply module includes: an inlet pipe, a filter, a high-pressure pump, and an outlet pipe connected in sequence. A pressure sensor, a pressure regulating valve, and a flow valve are sequentially installed on the outlet pipe. The front end of the pressure regulating valve is connected to a bypass pressure relief pipeline, and a pressure relief valve is installed on the bypass pressure relief pipeline. The pressure regulating valve is used to adjust the injection pressure of the high-pressure water jet according to the control command; The flow valve is used to adjust the jet flow rate of the high-pressure water jet according to the control command.

[0021] Optionally, the injection execution module includes a pulse injection control unit, a nozzle attitude adjustment unit, a movement drive unit, and a nozzle. The pulse injection control unit, the nozzle attitude adjustment unit, and the movement drive unit are all connected to the control module, and the nozzle is connected to the nozzle attitude adjustment unit. The pulse jet control unit is used to adjust the high-pressure water jet opening duration, closing duration, and jet pulse duty cycle of the nozzle according to the control command of the control module. The nozzle attitude adjustment unit is used to adjust the spray angle of the nozzle according to the control command issued by the control module; The moving drive unit is used to adjust the moving speed of the nozzle along the surface to be cleaned according to the control commands issued by the control module.

[0022] Secondly, the present invention provides a ship underwater intelligent cleaning control method based on multi-region plume feedback, comprising the following steps: Step S1: Acquire multi-region plume sensing data and jet parameter information; Step S2: Construct a plume state estimation model, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on radial basis functions. Step S3: Process the multi-region plume sensing data and the jet parameter information data using the plume state estimation model to obtain the final plume state estimation result; identify the current cleaning state based on the final plume state estimation result; Step S4: Generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and generate a control command for the jet parameters based on the jet parameter adjustment vector; Step S5: Adjust the jet pressure, jet flow rate, jet angle and jet duty cycle of the high-pressure water jet according to the control command, and adjust the nozzle moving speed, so as to control the high-pressure water jet to act on the hull wall and clean the ship.

[0023] The beneficial effects of this invention are as follows: Compared with the prior art, the intelligent underwater cleaning control system for ships provided by this invention, based on multi-regional plume feedback, realizes dynamic perception, state identification, and active control of plume diffusion behavior during the underwater cleaning process of ships. By establishing multiple plume monitoring areas around the nozzle, such as the area in front of the nozzle, the area behind the nozzle, the near-wall backflow area, the far-field diffusion area, and the background reference area, the system can simultaneously monitor the plume state at different spatial locations. Compared with existing single-point turbidity detection or local image detection methods, it can effectively distinguish different states such as effective peeling of adhering materials, local backflow retention, and far-field diffusion, thereby improving the accuracy and stability of plume state identification.

[0024] This invention further establishes a plume state estimation model and a multi-state switching mechanism. By continuously analyzing the dynamic changes of the plume, it realizes the dynamic identification of effective cleaning state, local backflow state, far-field diffusion state, cleaning completion state, and over-cleaning risk state. This solves the problems of lack of continuous state judgment capability and difficulty in adapting to complex underwater working conditions in the prior art, and improves the system's adaptability in complex flow fields and high turbidity environments.

[0025] This invention introduces a dynamic threshold adjustment mechanism, enabling the system to automatically adjust the state criterion threshold based on operating parameters such as background turbidity, water flow disturbance intensity, jet pressure, and nozzle movement speed. This avoids the problem of misjudgment that can easily occur under different operating conditions in traditional fixed threshold control methods, thereby improving the robustness of system state recognition and control stability.

[0026] This invention also establishes an adaptive closed-loop control system based on plume feedback, enabling dynamic and coordinated adjustment of parameters such as jet pressure, jet flow rate, jet angle, jet pulse duty cycle, and nozzle movement speed. This allows the system to automatically optimize the jet control strategy based on real-time plume conditions, thereby reducing the risks of local backflow and far-field diffusion, and improving the stability and cleaning efficiency of the underwater cleaning process. Simultaneously, by employing a pulsed high-pressure water jet control method, it effectively reduces the large-scale turbidity plume and local particle accumulation phenomena formed under continuous high-pressure jetting conditions, lowering the risk of secondary pollution in port and dock areas.

[0027] This invention achieves proactive control of plume diffusion behavior during underwater ship cleaning by constructing a system-level intelligent control system that combines plume state perception, state estimation, state switching, dynamic threshold adjustment, and closed-loop control. This improves the intelligence and automation level of the underwater ship cleaning system and enhances its engineering application capabilities in complex underwater environments. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the overall structure of the intelligent underwater cleaning control system for ships provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the division of multiple plume monitoring areas provided in an embodiment of the present invention; Figure 3 This is a flowchart of the plume state estimation provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the adaptive fuzzy PID control unit provided in an embodiment of the present invention. Figure 5 This is a flowchart illustrating the intelligent underwater cleaning control method for ships provided in an embodiment of the present invention. Figure 6This is a schematic diagram illustrating the turbidity variation trend of a multi-region plume provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. In the following description, the expression "some embodiments" refers to a subset of all possible embodiments; however, it should be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with each other without conflict.

[0030] Example 1 Please refer to the attached document. Figure 1 This invention provides a ship underwater intelligent cleaning control system based on multi-region plume feedback, comprising: A multi-region plume sensing module is used to establish multiple plume monitoring areas around the nozzle. Sensors are used in these multiple plume monitoring areas to perform spatial distribution sensing and obtain multi-region plume sensing data. The multiple plume monitoring areas include: the area in front of the nozzle, the area behind the nozzle, the near-wall backflow area, the far-field diffusion area, and the background reference area, which correspond to areas A, B, C, D, and E, respectively. The data recording module is used to acquire the injection parameter information data of the injection nozzle; The control module is used to analyze and process the multi-region plume sensing data and the jet parameter information data, generate a jet parameter adjustment vector, and convert the jet parameter adjustment vector into control commands for the jet parameters. A high-pressure water jet supply module is electrically connected to the control module and is used to receive control commands to adjust the injection pressure and flow rate of the high-pressure water jet; the pressure range of the high-pressure water jet supply module is 15~70MPa; The spray execution module is electrically connected to the control module and connected to the high-pressure water jet supply module through a water outlet pipe. It is used to receive the control commands and adjust the spray angle and spray duty cycle of the high-pressure water jet regulated by the high-pressure water jet supply module, as well as adjust the nozzle moving speed, to control the high-pressure water jet to act on the hull wall for cleaning the ship. The control module includes a data preprocessing unit, a plume state evaluation unit, a cleaning state discrimination unit, and an adaptive fuzzy PID control unit connected in sequence. The data preprocessing unit is used to preprocess the multi-region plume sensing data and construct plume state characteristic indicators based on the preprocessed multi-region plume sensing data. The plume state assessment unit is used to construct a plume state estimation model. The plume state characteristic index and the jetting parameter information data are input into the plume state estimation model, and the final plume state estimation result is output. The cleaning state discrimination unit is used to identify the current cleaning state based on the final plume state estimation result; The adaptive fuzzy PID control unit is used to generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and to convert the jet parameter adjustment vector into a control command for the jet parameters.

[0031] It should be noted that this technical solution uses high-pressure water jet cleaning to clean the ship's hull underwater. By using a high-pressure water pump to pressurize the water to extremely high pressure, and then forming a high-speed, high-energy-density dense water jet through a special nozzle, it directly impacts the marine organisms, dirt, rust layers and other pollutants attached to the underwater surface of the ship's hull, causing them to break down, peel off and be carried away by the water flow, thus achieving the effect of effectively cleaning the ship's hull. In this cleaning process, a plume will inevitably be generated. The plume is a density-driven or momentum-driven turbulent diffuser formed by the thorough mixing of the peeled-off dirt particles (biological debris, rust particles, residual paint layer) with the surrounding water. Specifically, this invention uses the turbid plume formed during the underwater cleaning process of ships as the feedback object. By setting up multiple functional monitoring areas around the nozzle, various data information from different areas are collected in real time. The various data information are processed and analyzed to determine the current plume state and identify the current cleaning state. Subsequently, the parameters are dynamically adjusted based on the final plume state estimation result and the cleaning state discrimination result, so that the cleaning system can achieve adaptive cleaning and dynamic control under different attachment states and different underwater disturbance conditions. Specifically: Please refer to the attached document. Figure 2 The multi-region plume sensing module is used to acquire multi-region plume sensing data and divide the space around the nozzle into multiple regions, including: the area in front of the nozzle (A region), the area behind the nozzle (B region), the near-wall backflow area (C region), the far-field diffusion area (D region), and the background reference area (E region). Sensors were deployed in multiple plume monitoring areas to collect data and obtain multi-region plume sensing data. The sensors included turbidity sensors, image acquisition units, local flow velocity sensors, and distance sensors. Specifically: a forward turbidity sensor was installed in the area in front of the nozzle (Area A), facing the area to be cleaned, to detect the background water turbidity before the spray, providing information on the local water state of the area before the current spray; a backward turbidity sensor was installed in the area behind the nozzle (Area B), facing the main diffusion direction of the plume, to detect the high-concentration plume turbidity formed by the shedding of deposits, characterizing the intensity of deposit shedding; a lateral turbidity sensor was installed in the near-wall backflow area (Area C). Near-wall backflow plume concentration, velocity, and direction changes are detected by sensors and local velocity sensors mounted on the side of the nozzle and facing the hull wall. This characterizes near-wall backflow and local stagnation. In the far-field diffusion zone (Zone D), far-field turbidity and local velocity sensors are installed on the outside of the nozzle and facing away from the hull wall. This detects the far-field diffusion degree and diffusion flow field state, characterizing far-field diffusion intensity. In the background reference zone (Zone E), a background turbidity sensor is installed on the side of the moving drive unit away from the spray direction. This acquires the ambient background turbidity unaffected by the current spray, helping to determine whether the current water turbidity originates from effective cleaning, local backflow, or far-field diffusion. An image acquisition unit is installed near the nozzle and facing the hull wall to acquire images of the hull surface. A distance sensor is installed at the nozzle tip along the spray direction to obtain the distance between the nozzle and the hull wall. The above data together constitute multi-regional plume sensing data, namely: turbidity data, local velocity and direction data, image clarity data, and distance data between the nozzle and the ship's wall in each plume monitoring area, which provide input information for subsequent plume state estimation and cleaning state determination.

[0032] It should be noted that zones A through E are all plume monitoring areas established around the same nozzle, used to characterize the plume state at different spatial locations around the nozzle, and do not indicate the installation of multiple nozzles; the image acquisition unit and the distance sensor are both installed with the same nozzle as the installation reference and do not belong to any of the zones A through E.

[0033] After the high-pressure water jet is sprayed onto the hull, the multi-area plume sensing module acquires multi-area plume sensing data reflecting changes in the water and pollutants. Then, the data recording module acquires the spray parameter information of the nozzle. This spray parameter information reflects the parameter status information set by the nozzle, including: spray pressure data, spray flow rate data, spray angle data, spray pulse duty cycle, and nozzle movement speed. Pressure and flow sensors in the high-pressure water jet supply module are used to collect the high-pressure water jet pressure and flow rate, respectively; an angle sensor in the nozzle attitude adjustment unit is used to collect the current spray angle; the pulse spray control unit records the high-pressure water jet's on and off durations and calculates the current spray pulse duty cycle; and a speed sensor in the movement drive unit collects the nozzle's movement speed along the surface to be cleaned. Next, the control module analyzes the multi-region plume sensing data and jet parameter information data, and provides the adjustment amount of the jet parameter information data for the next moment to generate a jet parameter adjustment vector. Based on the jet parameter adjustment vector, the control command for the jet parameters is obtained and sent to the high-pressure water jet supply module and the jet execution module. The high-pressure water jet supply module can adjust the jet pressure and jet flow rate of the high-pressure water jet according to the control command. The jet execution module can adjust the jet angle, jet duty cycle, and nozzle movement speed of the high-pressure water jet. Since the jet execution module and the high-pressure water jet supply module are connected through the water outlet pipe, after the high-pressure water jet supply module adjusts the jet pressure and jet flow rate of the high-pressure water jet according to the control command, the high-pressure water jet will enter the jet execution module through the water outlet pipe. In the jet execution module, the jet angle, jet duty cycle, and nozzle movement speed of the high-pressure water jet are adjusted. Finally, the adjusted high-pressure water jet will exit through the nozzle and act on the hull wall to clean the hull wall of the ship. In other words, the control module analyzes and processes the data in real time, and continuously issues control commands for the next injection parameters. The high-pressure water jet supply module and the injection execution module adjust according to the control commands, including the injection parameter adjustment vector, thereby controlling the output of the high-pressure water jet from the nozzle.

[0034] It should be noted that the intelligent underwater cleaning control system for ships proposed in this invention can be deployed in underwater robots, wall-hull cleaning robots, and remotely operated underwater vehicles, as well as in manually assisted underwater cleaning platforms or semi-automated hull cleaning devices; the modules in the system can be implemented using either a centralized control structure or a distributed control structure.

[0035] In another embodiment, in addition to using a multi-region monitoring method consisting of a turbidity sensor, an image acquisition unit, and a local velocity sensor, the multi-region plume sensing module can also employ underwater laser scattering detection, acoustic turbidity detection, underwater sonar imaging, or underwater multispectral detection to sense the spatial distribution of the plume. The type, number, and arrangement of sensors in different regions can also be adjusted according to the actual ship structure, water environment, and cleaning conditions. As long as the identification of the spatial distribution, local backflow, and far-field diffusion of the plume can be achieved, it falls within the scope of protection of this invention.

[0036] As an optional implementation method, please refer to the appendix. Figure 3 The plume state characteristic indicators include: washing and shedding indicators, near-wall reflux indicators, far-field diffusion indicators, plume change rate indicators, and image clarity indicators. The expression for the washing and shedding index is:

[0037] in, To clean and remove indicators, This refers to the relative turbidity in region B, which is the area behind the nozzle. This refers to the relative turbidity of region A, i.e., the area in front of the nozzle. The expression for the near-wall reflux index is:

[0038] in, As a near-wall reflux indicator, The relative turbidity in region C, i.e., the near-wall reflux region, is... The relative turbidity in region D, i.e., the far-field diffusion region; The expression for the far-field diffusion index is:

[0039] in, As a far-field diffusion indicator; The expression for the plume change rate index is:

[0040] in, This is an indicator of the rate of change of the plume; The expression for the image sharpness index is:

[0041] in, As an indicator of image sharpness, The image gradient value. This represents the number of pixels in the image.

[0042] Specifically, due to various factors in the underwater environment, such as bubble disturbance, water flow fluctuations, local eddies, and particle obstruction, turbidity data in different regions can fluctuate significantly. Therefore, it is necessary to perform time synchronization, data filtering, outlier removal, and background subtraction on the turbidity data of each region to reduce the impact of environmental disturbances and background turbidity fluctuations on the calculation results of state characteristic indicators. Specifically: First, the turbidity data of each plume monitoring area in the multi-region plume sensing data are synchronized in time to ensure temporal alignment; then, a combined sliding window filtering and median filtering method is used for data smoothing, expressed as:

[0043] in, Indicates the first The turbidity value after passing through the sliding window average filter at any given time. Indicates the first The original turbidity measurement values ​​at each sampling time. This represents the length of the sliding window, which is the number of sampling points used in the averaging calculation. Indicates the current sampling time; Next, outlier removal is performed, with the expression as follows:

[0044] in, Indicates the current number Turbidity sample values ​​at each sampling time. This represents the average value of the sampled values ​​within the current sliding window. This represents the standard deviation of the sampled values ​​within the current sliding window. Represents the outlier discrimination coefficient. This represents the absolute value of the deviation between the current sampled value and the window mean. When the deviation between the current sampled value and the window mean exceeds... When the sample value exceeds one standard deviation, it is considered an outlier and is removed. The real-time turbidity of each region is defined as follows:

[0045] in, Turbidity in the area in front of the nozzle. Turbidity in the area behind the nozzle. Turbidity in the near-wall reflux region. Turbidity in the far-field diffusion region Turbidity of the background reference area; Next, background subtraction is performed using the following expression:

[0046] in, For the first Regional relative turbidity, For the real-time turbidity of the corresponding area, Turbidity of the background reference area; The preprocessed turbidity data, along with local velocity and direction data, image clarity data, and distance data between the nozzle and the ship's wall from the multi-region plume sensing data, are used to construct plume state characteristic indicators. These indicators include: washing and shedding indicators, near-wall backflow indicators, far-field diffusion indicators, plume change rate indicators, and image clarity indicators. Image gradient values The gradient value is calculated from the ship's surface image acquired by the image acquisition unit. Specifically, the ship's surface image is first converted to grayscale. Then, the image gradient operator is used to calculate the grayscale change rate of each pixel in the grayscale-converted ship's surface image, thus obtaining the image gradient value of the corresponding pixel. Finally, the image sharpness index is calculated based on the gradient values ​​of all pixels. ; Image sharpness index A higher value indicates clearer image edges and texture features, corresponding to a lower degree of occlusion from ship surface attachments; image sharpness index The smaller the value, the higher the degree of image blurring, corresponding to a higher degree of plume occlusion or attachment coverage; that is, the image gradient value reflects the richness of image edge and texture information. When the attachments on the hull surface are removed, the image edge features become more obvious, the image gradient value increases, and the corresponding image clarity index improves. Therefore, it can be used to characterize the cleaning effect of the current area.

[0047] The expression for the plume state vector is:

[0048] in, For the first Time-limited plume state vector This is a washing and shedding index used to characterize the degree of peeling off of attached substances. The near-wall recirculation index is used to characterize the degree of plume recirculation and local stagnation near the hull. It is a far-field diffusion index used to characterize the extent to which the plume diffuses away from the ship's hull. The plume change rate index is used to characterize the trend of plume state changes. Image sharpness is an index used to characterize the visual observability of a current area. For the first The injection pressure at any moment, For the first Jet flow rate at any time For the first The nozzle movement speed at any given moment For the first The duty cycle of the jet pulse at any given moment.

[0049] The plume state characteristic index is used to determine the current cleaning stage and control objective, while the injection parameter information data is used to characterize the current injection execution state and adjustment range. The plume state characteristic index and injection parameter information data are input into the plume state estimation model for processing to obtain the final plume state estimation result and identify the current cleaning state. Then, the adaptive fuzzy PID control unit is used to generate control commands for the injection parameters, and the injection parameters of the injection nozzle are adjusted according to the control commands.

[0050] As an optional implementation, the processing steps of the plume state assessment unit include: A plume state vector is constructed based on the plume state characteristic index and the jet parameter information data; A plume state estimation model is constructed, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on the radial basis function. The plume state vector is input into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state. The RBF neural network error compensation model performs error compensation on the extended Kalman filter corrected state to obtain the final plume state estimation result.

[0051] Specifically, the plume state estimation model includes two stages for state prediction: Extended Kalman Filter (EKF) state prediction and RBF neural network residual compensation. The EKF state prediction model first predicts the state based on the previous plume state and control input. Then, it compares the real-time observation data with the predicted observations, calculates the innovation error, and corrects the predicted state using Kalman gain, resulting in the EKF-corrected state. The RBF neural network step follows the EKF-corrected state, further utilizing the prediction residuals for online learning to compensate for any nonlinear errors not yet compensated by the EKF-corrected state, thereby further improving the state estimation accuracy and yielding the final plume state estimation result. By using the EKF state prediction model for plume state prediction and the RBF neural network for error correction, the dynamic recognition capability of the plume state can be improved, thus enhancing the accuracy of plume state prediction.

[0052] It should be noted that the extended Kalman filter is suitable for predicting the continuous evolution of plume state over time, while the RBF neural network can compensate for prediction errors caused by nonlinear diffusion, local backflow, and particle occlusion in complex underwater environments. Compared with simple threshold judgment or single filtering methods, the plume state estimation model constructed in this invention, which includes an extended Kalman filter state prediction model and an RBF neural network error compensation model, can simultaneously take into account real-time performance and nonlinear compensation capabilities.

[0053] As an optional implementation, the step of inputting the plume state vector into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state includes: Obtain the injection parameter adjustment vector from the previous cycle, and define a control input vector based on the injection parameter adjustment vector, with the following expression:

[0054] in, For the first The control input vector at time t, For the first The amount of injection pressure adjustment at any given time. For the first The amount of jet flow rate adjustment at any given time. For the first The amount of adjustment for the nozzle movement speed at any given time. For the first The amount of injection angle adjustment at any given time. For the first The duty cycle adjustment of the injection pulse at any given moment; The plume state vector from the previous moment and the control input vector from the previous cycle are input into the state transition function for state prediction. The predicted value of the plume state vector at the current moment is obtained, expressed as:

[0055] in, For the first Predicted value of the plume state vector at time t. For the first The feather state vector at time 1. For the first Time to the The control input vectors that have been executed between time points, This is the state transition function; The prediction error covariance matrix at the current time is calculated using the following expression:

[0056] in, For the first The covariance matrix of the prediction error at any given time. For the first The time-state estimation error covariance matrix, Let be the state Jacobian matrix. for The transpose of the matrix, This is the process noise covariance matrix; this formula is used to estimate the uncertainty of the state prediction results and provides a basis for subsequent Kalman gain calculation. Define an observation model, and use the observation model to map the predicted value of the plume state vector to the predicted observed value. Calculate the actual observed vector at the current time based on the predicted plume state vector at the current time. Update the observation based on the predicted observed value and the actual observed vector to obtain the innovation error at the current time, expressed as:

[0057] in, For the first The actual observed vector at time t, The observation function is used to establish the mapping relationship between state variables and sensor observations. For the first The feather state vector at time 1. For the first The observation noise vector at any given time is used to characterize sensor measurement errors and random disturbances; the observation vector includes turbidity data, flow velocity data, flow direction data, image clarity data, and distance data between the nozzle and the hull wall; this observation model is used to establish the correspondence between theoretically predicted states and actual detection data, providing observational basis for subsequent state correction;

[0058] in, For the first Constantly innovate and make mistakes. For the first The actual observed vector at any given time. These are the predicted observations obtained based on the predicted state; The Kalman gain is calculated based on the covariance matrix of the innovation error and the prediction error. State correction is then performed based on the Kalman gain and the innovation error to obtain the extended Kalman filter corrected state at the current time, expressed as:

[0059] in, For the first Moment-time Kalman gain, To observe the Jacobian matrix, for The transpose of the matrix, To observe the noise covariance matrix;

[0060] in, For the first The state is adjusted by the extended Kalman filter at any given time.

[0061] It should be noted that "the plume state vector of the previous moment" refers to the k-th moment. The plume state vector at time 1; "the control input vector of the previous time step" refers to the k-th time step. The injection parameter adjustment vector executed between time 1 and time k includes the injection pressure adjustment, injection flow rate adjustment, nozzle movement speed adjustment, injection angle adjustment, and injection pulse duty cycle adjustment; "previous time" corresponds to k. Time 1; "Current Time" corresponds to Time K.

[0062] Specifically, the extended Kalman filter state prediction model is used for extended Kalman filter state prediction and observation correction, with the first... The feather state vector at time t and the first Time to the Using the control input vectors already executed between time points as input, for the th time... The state of the plume at any given time is predicted and corrected; this includes the state prediction stage, the prediction error covariance prediction stage, the observation update stage, and the state correction stage. During the state prediction phase, the k-th... The plume state vector at time 1 and the k-th time The control input vectors executed between time 1 and time k are input into the state transition function to predict the plume state vector at time k. The control input vectors include injection pressure adjustment, injection flow rate adjustment, nozzle movement speed adjustment, injection angle adjustment, and injection pulse duty cycle adjustment, which characterize the influence of the adjustment actions of the injection actuator in the previous control cycle on the current plume state evolution process.

[0063] In the prediction error covariance prediction stage, the state Jacobian matrix is ​​constructed based on the first-order partial derivative of the state transition function with respect to the state variables. Combined with the state estimation error covariance matrix and the process noise covariance matrix of the previous time step, the prediction error covariance matrix at time k is calculated to characterize the uncertainty of the predicted state.

[0064] During the observation update phase, the predicted value of the plume state vector is mapped to the predicted observation value according to the observation model. The observation data collected in real time at time k (the plume state vector at time k) is mapped to the actual observation vector according to the observation model. Then, the predicted observation value is compared with the actual observation vector to obtain the innovation error. The Kalman gain is calculated based on the covariance matrix of the innovation error and the prediction error.

[0065] In the state correction stage, the predicted plume state vector at time k is corrected using Kalman gain to obtain an estimated plume state vector at time k. This estimated plume state vector at time k is used as the corrected state of the extended Kalman filter (EPF), which is the output of the EPF state prediction model. This is further input into the RBF neural network for online learning of the prediction residual and error compensation. Through state prediction, prediction error covariance prediction, observation update, and state correction processes, the impact of measurement noise and environmental disturbances on the state estimation results can be reduced, improving the accuracy of plume state estimation.

[0066] As an optional implementation, the RBF neural network error compensation model compensates for the error in the extended Kalman filter corrected state to obtain the final plume state estimation result, including: The innovation error obtained during the extended Kalman filtering process is input into the RBF neural network. The RBF neural network calculates the output of the RBF neurons in the hidden layer using radial basis functions, as expressed by:

[0067] in, For the first The output of each RBF neuron To innovate the error vector, For the first Each neuron center, For the first The width of each radial basis function, Represents the Euclidean norm; The outputs of all RBF neurons are combined to form the hidden layer output vector, and the expression for the hidden layer output vector is:

[0068] in: Let be the hidden layer output vector of the RBF neural network at time k. These represent the outputs of the 1st to mth RBF neurons in the RBF hidden layer, respectively. This represents the number of RBF neurons; The error compensation term is calculated based on the hidden layer output vector, and the expression is as follows:

[0069] in, For error compensation term, The weight matrix of the output layer of the RBF neural network. This is the output vector of the hidden layer; The correction error is calculated using the aforementioned error compensation term. Based on this correction error, the extended Kalman filter correction state is then subjected to secondary compensation to update the plume state, yielding the final plume state estimation result, expressed as:

[0070] in, To correct the error, To innovate errors, This is the error compensation term;

[0071] in, For the final state estimation results, For predicting the state vector, The Kalman gain matrix is... To correct for errors; The RBF neural network weights are updated based on the correction error using an online learning method, expressed as:

[0072] in, The updated network weight matrix, This is the current network weight matrix. For learning rate, To correct for errors.

[0073] Specifically, since the actual underwater plume diffusion process is affected by local backflow, particle obstruction, water flow disturbance, and nonlinear diffusion behavior, relying solely on the Extended Kalman Filter (EBF) is insufficient to completely eliminate model errors. Therefore, an RBF neural network is further utilized to learn and compensate for the innovative errors generated by the EBF state prediction model online, thereby improving the accuracy of the plume state estimation results. Specifically, the innovative errors obtained during the EBF process are used as input to the RBF neural network. The hidden layer output is calculated using radial basis functions, and an error compensation term is calculated based on the hidden layer output and network weights. Subsequently, this error compensation term is used to perform secondary compensation on the EBF corrected state, yielding the final plume state estimation result. Simultaneously, an online learning method is employed to update the RBF neural network weights based on the corrected error, enabling it to adapt to nonlinear disturbances and model errors during the underwater plume diffusion process.

[0074] As an optional implementation, the processing steps of the cleaning state determination unit include: The cleaning process is divided into four states: background stability, effective cleaning, large amount of shedding, local reflux, far-field diffusion, cleaning completion, and over-cleaning risk. A combination of primary and auxiliary criteria is used to continuously monitor the plume state characteristic indicators in the final plume state estimation results using a sliding time window. Using the washing and shedding index Near-wall reflux index and far-field diffusion index The absence of significant growth over multiple consecutive sampling periods is used as the primary criterion to determine whether the background is in a stable state. Using the washing and shedding index A sustained upward trend across multiple sampling periods was used as the primary criterion, combined with far-field diffusion indicators. The threshold value is used as an auxiliary criterion to determine whether the cleaning state is effective. Using the washing and shedding index The main criterion is to maintain a rapid upward trend in turbidity over multiple consecutive sampling cycles, and the auxiliary criterion is to combine the continuous increase in turbidity in the area behind the nozzle that is higher than the corresponding increase under effective cleaning conditions to determine whether it is in a state of large-scale shedding. Near-wall reflux index The primary criterion is to exceed a set threshold for multiple consecutive sampling cycles, and the secondary criterion is to combine the change in near-wall velocity as an auxiliary criterion to determine whether the flow is in a local backflow state. Far-field diffusion index The primary criterion is that multiple consecutive sampling periods are higher than the diffusion threshold, and the turbidity change rate in the far field region is used as an auxiliary criterion to determine whether the region is in a far field diffusion state. Using the plume change rate index Maintaining a value close to zero for multiple consecutive sampling periods is used as the primary criterion, combined with image sharpness indicators. Continuous recovery is used as an auxiliary criterion to determine whether the cleaning process is complete. With injection pressure, injection flow rate, and injection pulse duty cycle maintained at a consistently high level, and the cleaning and shedding indicators... The primary criterion for determining whether a state is at risk of over-cleaning is that there is no significant increase in multiple consecutive sampling cycles. If only one of the state criteria is satisfied, the state that satisfies the criterion is taken as the current cleaning state. When multiple state criteria are met simultaneously, the judgment is made in the order of priority: over-cleaning risk state, local backflow state, far-field diffusion state, effective cleaning state, and cleaning completed state, and the state with the highest priority is taken as the current cleaning state.

[0075] Specifically, the sampling period is the time interval between the sensor completing one data acquisition and state update. In this embodiment, the sampling period is 0.5 to 2 seconds, preferably 1 second. Multiple consecutive sampling periods mean that N consecutive sampling periods satisfy the corresponding criterion conditions, where N is 3 to 10, preferably 5. The cleaning state discrimination unit uses a combination of primary and auxiliary criteria to determine the current cleaning stage. It continuously monitors the plume state characteristic indicators in the final plume state estimation results using a sliding time window. When the corresponding state indicator meets the preset conditions within multiple consecutive sampling periods, the corresponding state is determined to be valid. The primary criterion is used to identify the main characteristics of the current plume state, while the auxiliary criterion is used to verify the rationality of the state discrimination results and suppress misjudgments caused by environmental disturbances. Specifically, at the beginning of the hull cleaning, the spray nozzles move along the hull wall and continuously spray high-pressure water jets onto the hull wall. At this time, the attached material has not yet been significantly peeled off, the turbidity in each area remains at a low level, and the cleaning and detachment index, near-wall backflow index and far-field diffusion index are all at a low level, indicating that the current area is in a stable background state. As the spraying continues, the deposits begin to fall off gradually, and the turbidity in the area behind the nozzle gradually increases. When the cleaning and detachment index continues to rise for several consecutive sampling cycles, and the far-field diffusion index does not exceed the set threshold, it indicates that a large number of deposits have fallen off in the high-pressure water jet impact area, and it is determined that the effective cleaning state has been entered. Subsequently, a large number of detached particles fell off the surface of the ship. At that time, the cleaning and detachment index rose rapidly in several consecutive sampling cycles, and the turbidity in the area behind the nozzle continued to increase, indicating that it had entered a state of large-scale detachment. Some detached particles and plumes accumulate near the hull surface and form local backflow. When the near-wall backflow index exceeds the set threshold for multiple consecutive sampling cycles, and the flow velocity in the near-wall area fluctuates significantly or the flow direction changes, it indicates that the plume is stagnating or backflowing near the hull wall, and it is determined to have entered a local backflow state. If the plume spreads further away from the ship and causes the far-field diffusion index to exceed the diffusion threshold for multiple consecutive sampling cycles, it indicates that the plume has begun to spread significantly into the far-field region. Combined with the turbidity change rate in the far-field region as an auxiliary criterion, when the turbidity in the far-field diffusion region continues to increase, it is determined that the plume has entered the far-field diffusion state.

[0076] The plume change rate index can reflect the dynamic changes in the process of removing attached materials. As the attached materials in the current area are gradually removed, when the plume change rate index approaches zero for several consecutive sampling cycles, it indicates that the process of removing attached materials is gradually ending. At the same time, when the image clarity index continues to recover and remains stable, it is determined that the cleaning is completed. Since the image clarity index can reflect the image clarity of the hull wall, when the hull is in an effective cleaning state, a local backflow state, or a far-field diffusion state, the plume is constantly changing when the high-pressure water jet is sprayed onto the hull. At this time, the image of the hull wall is not clear. Therefore, using the image clarity index as an auxiliary criterion can further help to judge the current cleaning status and improve the accuracy of the judgment results. Furthermore, under any of the above conditions, if the injection pressure, injection flow rate, and injection pulse duty cycle are detected to remain at a consistently high level, while the cleaning and detachment indicators do not show a significant increase for several consecutive sampling cycles, then the system is deemed to be in an over-cleaning risk state. The over-cleaning risk state is an abnormal state throughout the entire process and can occur at any stage. When the injection energy is detected to remain consistently high while the cleaning and detachment effect does not significantly improve, the system is deemed to be in an over-cleaning risk state. Since this state has the highest priority, the controller prioritizes reducing the injection pressure adjustment, injection flow rate adjustment, and injection duty cycle adjustment to avoid energy waste and excessive scouring of the hull surface.

[0077] It should be noted that the threshold values ​​for each indicator are not fixed and can be dynamically adjusted based on background water turbidity, local water flow disturbance intensity, jet pressure, jet flow rate, and nozzle movement speed.

[0078] In terms of cleaning state discrimination, in addition to using a state machine approach based on thresholds and state switching mechanisms, another embodiment can also use a probability model, fuzzy rule base, neural network classifier, or reinforcement learning decision model to achieve dynamic discrimination and state switching for different cleaning stages; the state criterion threshold can also be adjusted using a fixed threshold, adaptive threshold, phased threshold, or online learning update method.

[0079] Please refer to the attached document. Figure 6 The turbidity change trend of the multi-region plume can be divided into three stages: Stage I, Stage II, and Stage III. Stage I is the initial stage of cleaning, which mainly corresponds to the stable background state, the effective cleaning state, and the state of large-scale detachment. In this stage, the jet nozzles begin to clean the hull wall with high-pressure water jets, the attached materials are gradually peeled off, and the turbidity in the area behind the nozzles increases rapidly. The main control objective is to ensure the effective removal of the attached materials. Phase II is the plume development and control phase, mainly corresponding to the local backflow state and the far-field diffusion state. In this phase, some plumes form local stagnation and backflow near the hull wall, and the turbidity in the near-wall backflow zone remains high. At the same time, the turbidity in the far-field diffusion zone begins to gradually increase, and the main control objective shifts from improving cleaning efficiency to suppressing local backflow and far-field diffusion. Phase III is the later stage of cleaning, mainly corresponding to the cleaning completion state. In this stage, as the deposits are gradually removed, the turbidity in the area behind the nozzle decreases, image clarity is restored, and the spray energy is further reduced while the nozzle movement speed is increased, thus entering the next cleaning area. The over-cleaning risk state is an abnormal state throughout the entire process and can occur in any stage. It is triggered when the system detects that the spray energy remains high while the cleaning and removal effect has not significantly improved, and it has the highest control priority. Based on the above stage division, Figure 6 This shows the turbidity variation trend in different areas and also illustrates the correspondence between the cleaning state discrimination and control target switching of the present invention.

[0080] As an optional implementation method, please refer to the appendix. Figure 4 The adaptive fuzzy PID control unit includes: a fuzzification submodule, a rule reasoning submodule, and a defuzzification output submodule; The fuzzification submodule is used to convert the plume state feature indicators in the final plume state estimation result into corresponding fuzzy linguistic variables using a preset membership function. The rule reasoning submodule is used to perform fuzzy rule reasoning based on the current cleaning state and the fuzzy linguistic variables corresponding to the characteristic indicators of each plume state, and generate fuzzy reasoning results. The defuzzy output submodule is used to defuzzify the fuzzy inference result using the centroid method to obtain the injection parameter adjustment vector, and to generate control commands for the injection parameters based on the injection parameter adjustment vector; the injection parameter adjustment vector includes: injection pressure adjustment amount, injection flow rate adjustment amount, injection angle adjustment amount, nozzle moving speed adjustment amount, and injection duty cycle adjustment amount.

[0081] Specifically, the fuzzification submodule presets fuzzy linguistic variables corresponding to plume state characteristic indicators, including: the fuzzy linguistic variables corresponding to the cleaning and shedding indicator include "decrease", "stable", and "increase"; the fuzzy linguistic variables corresponding to the near-wall backflow indicator include "decrease", "stable", and "increase"; the fuzzy linguistic variables corresponding to the far-field diffusion indicator include "low", "medium", and "high"; the fuzzy linguistic variables corresponding to the plume change rate indicator include "decline", "stable", and "increase"; and the fuzzy linguistic variables corresponding to the image clarity indicator include "fuzzy", "normal", and "clear". Next, a preset membership function is used, which can be a triangular membership function or a trapezoidal membership function, to calculate the membership degree of each indicator and obtain the real-time indicator value of each indicator. The real-time indicator value is then converted into the corresponding fuzzy linguistic variable.

[0082] Next, the rule reasoning submodule performs fuzzy rule reasoning based on the current cleaning status and the fuzzy linguistic variables corresponding to each indicator, and obtains the fuzzy reasoning result. For example, when the cleaning shedding indicator is "increased" and the far-field diffusion indicator is "low", the fuzzy reasoning result is to increase the injection pressure adjustment amount ΔP and the injection flow rate adjustment amount ΔQ; when the near-wall backflow indicator is "increased", the fuzzy reasoning result is to increase the injection angle adjustment amount Δθ and decrease the injection pulse duty cycle adjustment amount ΔD; when the far-field diffusion indicator is "high", the fuzzy reasoning result is to decrease the injection pressure adjustment amount ΔP, the injection flow rate adjustment amount ΔQ, and the injection pulse duty cycle adjustment amount ΔD; when the image clarity indicator is "clear" and the plume change rate indicator is "stable", the fuzzy reasoning result is to decrease the injection pressure adjustment amount ΔP and increase the nozzle movement speed adjustment amount ΔV.

[0083] The defuzzy output submodule uses the centroid method to defuzzify the fuzzy inference results. The centroid method calculates the position of the centroid corresponding to the area of ​​the fuzzy output membership function, and converts the inference results corresponding to the fuzzy linguistic variables into specific numerical outputs, thereby obtaining the injection pressure adjustment amount ΔP, injection flow adjustment amount ΔQ, injection angle adjustment amount Δθ, nozzle moving speed adjustment amount ΔV, and injection pulse duty cycle adjustment amount ΔD.

[0084] After defuzzing, the injection parameter adjustment vector is obtained:

[0085] Where ΔU is the injection parameter adjustment vector. This is the injection pressure adjustment amount. For injection flow rate adjustment, This is the amount of spray angle adjustment. This refers to the adjustment amount of the nozzle movement speed. This refers to the injection duty cycle adjustment amount.

[0086] Then, the target injection parameter value is calculated based on the current injection parameter value and the corresponding adjustment amount. The value range of each adjustment amount is preset according to the equipment parameters. For example, the injection pressure adjustment amount corresponds to the pressure change range, the injection flow rate adjustment amount corresponds to the flow rate change range, the injection angle adjustment amount corresponds to the nozzle attitude adjustment range, the nozzle moving speed adjustment amount corresponds to the moving speed change range of the moving drive unit, and the injection pulse duty cycle adjustment amount corresponds to the pulse injection duty cycle change range.

[0087] The output value obtained after defuzzification using the center-of-gravity method directly corresponds to the specific numerical value of each adjustment variable and serves as the basis for subsequent calculation of target injection parameters. The formula for calculating the target injection parameter value is as follows: P target =P current +ΔP Q target =Q current +ΔQ θ target =θ current +Δθ V target =V current +ΔV D target =D current +ΔD Among them, P target For the target injection pressure, Q target For the target jet flow rate, θ target V is the target spray angle. target D represents the target movement speed of the nozzle. target The target jet pulse duty cycle; P current Q is the current injection pressure. current θ is the current jet flow rate. current V is the current injection angle. current D represents the current moving speed of the nozzle. current This represents the current duty cycle of the jet pulse.

[0088] The control module generates control commands based on the target injection parameter values ​​and sends them to the high-pressure water jet supply module and the injection execution module. The pressure regulating valve is used to regulate the injection pressure, the flow valve is used to regulate the injection flow, the nozzle attitude adjustment unit is used to regulate the injection angle, the movement drive unit is used to regulate the nozzle movement speed, and the pulse injection control unit is used to regulate the injection pulse duty cycle, thereby completing the conversion of the injection parameter adjustment vector into control commands.

[0089] After completing one parameter adjustment, the system continues to sense the status of the plume in each region in real time, and re-enters the data acquisition, status estimation, status discrimination and control adjustment stages, thus forming a continuous closed-loop control process of "sensing-estimation-discrimination-control-feedback".

[0090] It should be noted that the adaptive fuzzy PID control unit does not directly use all parameters in the plume state vector as fuzzy control inputs simultaneously. Instead, it identifies the current cleaning state based on the cleaning and shedding indices, near-wall backflow indices, far-field diffusion indices, plume change rate indices, and image clarity indices from the plume state estimation results. It then combines these with the current jet pressure, jet flow rate, nozzle movement speed, jet angle, and jet pulse duty cycle to generate the corresponding jet parameter adjustment vector. For velocity data, flow direction data, image clarity data, and nozzle-to-hull wall distance data, these are primarily observed and directly participate in the state correction process of the plume state estimation model.

[0091] As an optional implementation, the high-pressure water jet supply module includes: an inlet pipe, a filter, a high-pressure pump, and an outlet pipe connected in sequence. A pressure sensor, a pressure regulating valve, and a flow valve are sequentially installed on the outlet pipe. The front end of the pressure regulating valve is connected to a bypass pressure relief pipeline, and a pressure relief valve is installed on the bypass pressure relief pipeline. The pressure regulating valve is used to adjust the injection pressure of the high-pressure water jet according to the control command; The flow valve is used to adjust the jet flow rate of the high-pressure water jet according to the control command.

[0092] Specifically, during operation, water enters the filter through the inlet pipe, where impurities are filtered out, and then enters the high-pressure pump for pressurization. The pressurized high-pressure water is then transported through the outlet pipe. A pressure sensor monitors the pipeline pressure in real time, a pressure regulating valve adjusts the high-pressure water jet pressure, and a flow valve adjusts the high-pressure water jet flow rate. When the system pressure exceeds the set value, some of the high-pressure water enters the pressure relief valve through the bypass pressure relief pipeline to release pressure, ensuring the stable operation of the high-pressure water jet supply system. The outlet end of the outlet pipe is connected to the jet actuator, and the high-pressure water, after being regulated by the pressure regulating valve and the flow valve, is finally transported to the jet actuator through the outlet end of the outlet pipe.

[0093] As an optional implementation, the injection execution module includes a pulse injection control unit, a nozzle attitude adjustment unit, a movement drive unit, and a nozzle. The pulse injection control unit, the nozzle attitude adjustment unit, and the movement drive unit are all connected to the control module, and the nozzle is connected to the nozzle attitude adjustment unit. The pulse jet control unit is used to adjust the high-pressure water jet opening duration, closing duration, and jet pulse duty cycle of the nozzle according to the control command of the control module. The nozzle attitude adjustment unit is used to adjust the spray angle of the nozzle according to the control command issued by the control module; The moving drive unit is used to adjust the moving speed of the nozzle along the surface to be cleaned according to the control commands issued by the control module.

[0094] Specifically, the control module generates corresponding control commands based on the injection parameter adjustment vector and sends them to the pressure regulating valve actuator, flow valve actuator, pulse injection control unit, nozzle attitude adjustment unit, and movement drive unit, respectively, to achieve dynamic adjustment of injection pressure, injection flow rate, injection angle, injection duty cycle, and nozzle movement speed.

[0095] Example 2 Please refer to the attached document. Figure 5 This invention provides a ship underwater intelligent cleaning control method based on multi-region plume feedback, comprising the following steps: Step S1: Acquire multi-region plume sensing data and jet parameter information; Step S2: Construct a plume state estimation model, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on radial basis functions. Step S3: Process the multi-region plume sensing data and the jet parameter information data using the plume state estimation model to obtain the final plume state estimation result; identify the current cleaning state based on the final plume state estimation result; Step S4: Generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and generate a control command for the jet parameters based on the jet parameter adjustment vector; Step S5: Adjust the jet pressure, jet flow rate, jet angle and jet duty cycle of the high-pressure water jet according to the control command, and adjust the nozzle moving speed, so as to control the high-pressure water jet to act on the hull wall and clean the ship.

[0096] Specifically, before starting work, the operator first initializes the spray parameters based on the ship type, type of deposits, condition of the antifouling coating, turbidity of the current working water area, intensity of water flow disturbance, and allowable cleaning intensity. The initialization parameters mainly include spray pressure, spray flow rate, nozzle moving speed, spray angle, and spray pulse duty cycle. The above initial parameters can be set by the operator based on experience, or recommended initial values ​​can be provided by the historical cleaning condition database. Among them, the nozzle moving speed refers to the linear velocity of the nozzle moving along the preset cleaning path direction of the hull wall, which is used to characterize the distance the nozzle sweeps across the hull wall per unit time, and is controlled by the movement drive unit. Next, a background turbidity benchmark is established, obtained based on the current water background reference turbidity. Then, jet cleaning is initiated, acquiring multi-regional plume sensing data and jet parameter information. Based on the multi-regional plume sensing data, data preprocessing and background subtraction calculations are performed to obtain plume state characteristic indicators, and a plume state estimation model is constructed. This model processes the plume state characteristic indicators and jet parameter information to obtain the final plume state estimation result, and the current cleaning state is determined based on this result. Finally, a jet parameter adjustment vector is generated based on the final plume state estimation result and the current cleaning state, and the jet parameters are adjusted accordingly. The parameter adjustment vector conversion yields the control command for the injection parameters. Finally, based on the control command, the injection pressure, injection flow rate, injection angle, injection duty cycle, and nozzle movement speed are adjusted to complete the adjustment of the injection parameters at the current moment. After adjustment, it is determined whether the cleaning completion conditions are met. If not, the next moment's online closed-loop adjustment begins. That is, the control command for the next moment is obtained, the injection parameters are adjusted according to the control command, and it is determined whether the cleaning completion conditions are met. This process is repeated continuously until the cleaning completion conditions for that area are met, then the process continues in the next cleaning area until the ship's cleaning is finally completed.

[0097] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A ship underwater intelligent cleaning control system based on multi-regional plume feedback, characterized in that, include: The multi-region plume sensing module is used to establish multiple plume monitoring areas around the nozzle, and to use sensors in the multiple plume monitoring areas to perform spatial distribution sensing to obtain multi-region plume sensing data. The multiple plume monitoring areas include: the area in front of the nozzle, the area behind the nozzle, the near-wall backflow area, the far-field diffusion area, and the background reference area; A rearward turbidity sensor is installed in the rear area of ​​the nozzle, which is mounted on the rear side of the nozzle and faces the main diffusion direction of the plume. It is used to detect the high concentration of plume turbidity formed by the shedding of the attached material and can characterize the peeling strength of the attached material. The background reference area is equipped with a background turbidity sensor, which is installed on the side of the moving drive unit away from the spray direction. It is used to obtain the environmental background turbidity that is not affected by the current spray, which can help determine whether the current water turbidity is caused by effective cleaning, local backflow or far-field diffusion. The data recording module is used to acquire the injection parameter information data of the injection nozzle; The control module is used to analyze and process the multi-region plume sensing data and the jet parameter information data, generate a jet parameter adjustment vector, and convert the jet parameter adjustment vector into control commands for the jet parameters. A high-pressure water jet supply module is electrically connected to the control module and is used to receive control commands to adjust the injection pressure and flow rate of the high-pressure water jet; the pressure range of the high-pressure water jet supply module is 15~70MPa; The spray execution module is electrically connected to the control module and connected to the high-pressure water jet supply module through a water outlet pipe. It is used to receive the control commands and adjust the spray angle and spray duty cycle of the high-pressure water jet regulated by the high-pressure water jet supply module, as well as adjust the nozzle moving speed, to control the high-pressure water jet to act on the hull wall for cleaning the ship. The control module includes a data preprocessing unit, a plume state assessment unit, a cleaning state discrimination unit, and a fuzzy control unit connected in sequence. The data preprocessing unit is used to preprocess the multi-region plume sensing data and construct plume state characteristic indicators based on the preprocessed multi-region plume sensing data. The plume state assessment unit is used to construct a plume state estimation model. The plume state characteristic index and the jetting parameter information data are input into the plume state estimation model, and the final plume state estimation result is output. The cleaning state discrimination unit is used to identify the current cleaning state based on the final plume state estimation result; The fuzzy control unit is used to generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and to convert the jet parameter adjustment vector into a control command for the jet parameters.

2. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 1, characterized in that, The plume state characteristic indicators include: washing and shedding indicators, near-wall reflux indicators, far-field diffusion indicators, plume change rate indicators, and image clarity indicators. The expression for the washing and shedding index is: in, To clean and remove indicators, This refers to the relative turbidity in region B, which is the area behind the nozzle. This refers to the relative turbidity of region A, i.e., the area in front of the nozzle. The expression for the near-wall reflux index is: in, As a near-wall reflux indicator, The relative turbidity in region C, i.e., the near-wall reflux region, is... The relative turbidity in region D, i.e., the far-field diffusion region; The expression for the far-field diffusion index is: in, As a far-field diffusion indicator; The expression for the plume change rate index is: in, This is an indicator of the rate of change of the plume; The expression for the image sharpness index is: in, As an indicator of image sharpness, The image gradient value. This represents the number of pixels in the image.

3. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 2, characterized in that, The processing steps of the plume state assessment unit include: A plume state vector is constructed based on the plume state characteristic index and the jet parameter information data; A plume state estimation model is constructed, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on the radial basis function. The plume state vector is input into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state. The RBF neural network error compensation model performs error compensation on the extended Kalman filter corrected state to obtain the final plume state estimation result.

4. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 3, characterized in that, The step of inputting the plume state vector into the extended Kalman filter state prediction model for plume state prediction and correction to obtain the extended Kalman filter corrected state includes: Obtain the injection parameter adjustment vector of the previous cycle, and define the control input vector based on the injection parameter adjustment vector; The state vector of the plume at the previous moment and the control input vector of the previous cycle are input into the state transition function to predict the state and obtain the predicted value of the plume state vector at the current moment. Calculate the prediction error covariance matrix at the current time. Define an observation model, use the observation model to map the predicted value of the plume state vector to the predicted observation value, calculate the actual observation vector at the current time based on the predicted plume state vector at the current time, update the observation based on the predicted observation value and the actual observation vector, and obtain the innovation error at the current time. The Kalman gain is calculated based on the innovation error and the prediction error covariance matrix. The state is then corrected based on the Kalman gain and the innovation error to obtain the extended Kalman filter corrected state at the current time.

5. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 4, characterized in that, The RBF neural network error compensation model compensates for the errors in the extended Kalman filter corrected state to obtain the final plume state estimation result, including: The innovation error obtained during the extended Kalman filtering process is input into the RBF neural network, which calculates the output of the RBF neurons in the hidden layer through radial basis functions. The outputs of all RBF neurons are combined to form the hidden layer output vector, and the error compensation term is calculated based on the hidden layer output vector. The correction error is calculated using the error compensation term, and the extended Kalman filter correction state is compensated a second time based on the correction error to update the plume state and obtain the final plume state estimation result. The weights of the RBF neural network are updated based on the correction error using an online learning method.

6. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 2, characterized in that, The processing steps of the cleaning status determination unit include: The cleaning process is divided into four states: background stability, effective cleaning, large amount of shedding, local reflux, far-field diffusion, cleaning completion, and over-cleaning risk. A combination of primary and auxiliary criteria is used to continuously monitor the plume state characteristic indicators in the final plume state estimation results using a sliding time window. Using the washing and shedding index Near-wall reflux index and far-field diffusion index The absence of significant growth over multiple consecutive sampling periods is used as the primary criterion to determine whether the background is in a stable state. Using the washing and shedding index A sustained upward trend across multiple sampling periods was used as the primary criterion, combined with far-field diffusion indicators. The threshold value is used as an auxiliary criterion to determine whether the cleaning state is effective. Using the washing and shedding index The main criterion is to maintain a rapid upward trend in turbidity over multiple consecutive sampling cycles, and the auxiliary criterion is to combine the continuous increase in turbidity in the area behind the nozzle that is higher than the corresponding increase under effective cleaning conditions to determine whether it is in a state of large-scale shedding. Near-wall reflux index The primary criterion is to exceed a set threshold for multiple consecutive sampling cycles, and the secondary criterion is to combine the change in flow velocity in the near-wall recirculation zone to determine whether it is in a local recirculation state. Far-field diffusion index The primary criterion is that multiple consecutive sampling periods are higher than the diffusion threshold, and the turbidity change rate in the far-field diffusion region is used as an auxiliary criterion to determine whether the region is in a far-field diffusion state. Using the plume change rate index Maintaining a value close to zero for multiple consecutive sampling periods is used as the primary criterion, combined with image sharpness indicators. Continuous recovery is used as an auxiliary criterion to determine whether the cleaning process is complete. With injection pressure, injection flow rate, and injection pulse duty cycle maintained at a consistently high level, and the cleaning and shedding indicators... The primary criterion for determining whether a state is at risk of over-cleaning is that there is no significant increase in multiple consecutive sampling cycles. If only one of the state criteria is satisfied, the state that satisfies the criterion is taken as the current cleaning state. When multiple state criteria are met simultaneously, the judgment is made in the order of priority: over-cleaning risk state, local backflow state, far-field diffusion state, effective cleaning state, and cleaning completed state, and the state with the highest priority is taken as the current cleaning state.

7. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 2, characterized in that, The fuzzy control unit includes: a fuzzification submodule, a rule reasoning submodule, and a defuzzification output submodule; The fuzzification submodule is used to convert the plume state feature indicators in the final plume state estimation result into corresponding fuzzy linguistic variables using a preset membership function. The rule reasoning submodule is used to perform fuzzy rule reasoning based on the current cleaning state and the fuzzy linguistic variables corresponding to the characteristic indicators of each plume state, and generate fuzzy reasoning results. The defuzzy output submodule is used to defuzzify the fuzzy inference result using the centroid method to obtain the injection parameter adjustment vector, and to generate control commands for the injection parameters based on the injection parameter adjustment vector; the injection parameter adjustment vector includes: injection pressure adjustment amount, injection flow rate adjustment amount, injection angle adjustment amount, nozzle moving speed adjustment amount, and injection duty cycle adjustment amount.

8. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to any one of claims 1-7, characterized in that, The high-pressure water jet supply module includes: an inlet pipe, a filter, a high-pressure pump, and an outlet pipe connected in sequence. A pressure sensor, a pressure regulating valve, and a flow valve are installed in sequence on the outlet pipe. The front end of the pressure regulating valve is connected to a bypass pressure relief pipeline, and a pressure relief valve is installed on the bypass pressure relief pipeline. The pressure regulating valve is used to adjust the injection pressure of the high-pressure water jet according to the control command; The flow valve is used to adjust the jet flow rate of the high-pressure water jet according to the control command.

9. The intelligent underwater cleaning control system for ships based on multi-regional plume feedback according to claim 8, characterized in that, The injection execution module includes a pulse injection control unit, a nozzle attitude adjustment unit, a movement drive unit, and a nozzle. The pulse injection control unit, the nozzle attitude adjustment unit, and the movement drive unit are all connected to the control module, and the nozzle is connected to the nozzle attitude adjustment unit. The pulse jet control unit is used to adjust the high-pressure water jet opening duration, closing duration, and jet pulse duty cycle of the nozzle according to the control command of the control module. The nozzle attitude adjustment unit is used to adjust the spray angle of the nozzle according to the control command issued by the control module; The moving drive unit is used to adjust the moving speed of the nozzle along the surface to be cleaned according to the control commands issued by the control module.

10. A method for intelligent underwater cleaning control of ships based on multi-regional plume feedback, the method being used to implement the intelligent underwater cleaning control system for ships based on multi-regional plume feedback as described in claim 1, characterized in that... Includes the following steps: Step S1: Acquire multi-region plume sensing data and jet parameter information; Step S2: Construct a plume state estimation model, which includes an extended Kalman filter state prediction model based on the extended Kalman filter algorithm and an RBF neural network error compensation model based on radial basis functions. Step S3: Process the multi-region plume sensing data and the jet parameter information data using the plume state estimation model to obtain the final plume state estimation result; The current cleaning state is identified based on the final plume state estimation result; Step S4: Generate a jet parameter adjustment vector based on the final plume state estimation result and the current cleaning state, and generate a control command for the jet parameters based on the jet parameter adjustment vector; Step S5: Adjust the jet pressure, jet flow rate, jet angle and jet duty cycle of the high-pressure water jet according to the control command, and adjust the nozzle moving speed, so as to control the high-pressure water jet to act on the hull wall and clean the ship.

Images