Grab bucket drainage device and method based on water pressure volume linkage and intelligent sensing control

Abstract

The present application belongs to the field of control system, and relates to a grab bucket drainage device and method based on water pressure volume linkage and intelligent sensing control. The device comprises four subsystems of physical drainage, intelligent sensing, edge computing control and cloud digital twin. Based on the mechanism that the water pressure decreases to drive the pre-inflatable air bag to expand and squeeze the water in the grab bucket during the lifting process of the grab bucket, the device supports zero energy consumption without air supplement and AI adaptive double mode operation with air supplement, and realizes working condition adaptive control by combining multi-source sensor fusion and deep reinforcement learning technology. The method relies on the natural pressure difference driving air bag drainage during the lifting process of the grab bucket, and cooperates with dynamic air supplement to optimize the efficiency. The present application upgrades the traditional grab bucket drainage scheme from multiple dimensions, can realize zero energy consumption drainage in shallow water and low density working conditions, greatly reduces the operating energy consumption, improves the solid-liquid separation efficiency through the adaptive extrusion of the air bag, reduces the invalid transportation, and cooperates with the edge AI to dynamically adapt to the working condition, so as to realize the cost reduction and efficiency increase and green upgrade of dredging operation.

Description

Technical Field

[0001] This invention belongs to the field of control system technology, specifically relating to a grab bucket drainage device and method based on water pressure volume linkage and intelligent sensing control. It is particularly suitable for grab bucket dredgers to achieve intelligent adaptive control of water discharge in the bucket during dredging operations through edge computing and deep reinforcement learning. Background Technology

[0002] Dredging is a core component of projects such as inland waterway maintenance, coastal port expansion, river and lake ecological dredging, and deep-sea mining. Grab dredgers are the most widely used type of equipment in dredging projects due to their strong adaptability to working conditions and controllable operating costs, accounting for more than 60% of the total stock of dredging equipment in China.

[0003] During the operation of grab dredgers, a large amount of water enters the bucket along with the sludge as the grab bucket opens and closes to excavate sludge. This water has no engineering value and must be discharged from the bucket before lifting and transporting the dredged material. Residual water in the bucket directly reduces the effective payload space and increases the weight of ineffective transport. Furthermore, the suspended sludge, heavy metals, and other pollutants carried by the water can easily cause secondary pollution if leaked during transport. Therefore, the efficiency, energy consumption, and environmental friendliness of the water discharge process have become one of the core bottlenecks restricting the improvement of dredging operation efficiency.

[0004] Currently, the vast majority of existing grab buckets use the most primitive gravity drainage system: a small slit is made at the bottom of the grab bucket, relying on gravity to allow water to seep out naturally. This system has inherent technical limitations: (1) Extremely low drainage efficiency: The drainage cycle generally takes 3 to 8 minutes, accounting for more than 30% of the total time of a single operation cycle, which seriously restricts construction efficiency; (2) Poor drainage effect: 30% to 50% of the water will remain in the bucket, which will increase the invalid load of a single transport by 15% to 40%, and the invalid transport cost will account for more than 20% of the total operating cost; (3) High environmental risks: The leakage process is not controlled and the water in the bucket is directly leaked into the working area, which is likely to cause the suspended solids in the water to exceed the standard, which does not meet the current water environment supervision requirements.

[0005] To address the shortcomings of gravity drainage, the industry has attempted to introduce improved drainage solutions such as pneumatic, hydraulic, and mechanical extrusion, but all of them have insurmountable drawbacks: Pneumatic drainage solution: Compressed air is introduced into the hopper to squeeze out the water. This requires an additional high-power air compressor and high-pressure air pipeline, which increases energy consumption by more than 100% compared to gravity drainage. In addition, the air pipeline is easily blocked by mud and sand, resulting in high maintenance costs. Hydraulic drainage solution: The water in the bucket is discharged by a hydraulic pusher driving the extrusion plate. The structure is complex and the failure rate of hydraulic components is high. Once the hydraulic oil leaks, it will directly cause oil pollution to the water. It is only suitable for scenarios with very low environmental protection requirements. Mechanical extrusion solution: It relies on the mechanical force of the grab opening and closing to link the extrusion structure for drainage. The wear rate of the extrusion components is more than 3 times that of ordinary grabs, the service life of the equipment is shortened by 40%, and it is only compatible with specific models of grabs and cannot be installed on existing equipment.

[0006] In addition, all of the above-mentioned improvement solutions have common intelligent defects: they lack real-time status perception, working condition self-adaptation and digital control capabilities, drainage operations rely entirely on the manual experience of crew members, drainage effect fluctuates greatly under different water depths and different silt viscosity conditions, and they cannot meet the current management requirements of smart dredging such as operation data retention, status visualization and compliance traceability.

[0007] Therefore, there is an urgent need to develop a grab bucket drainage device and method based on water pressure-volume linkage and intelligent sensing control to achieve the organic unity of efficient drainage, intelligent decision-making, digital management and control and green operation. Summary of the Invention

[0008] The purpose of this invention is to solve the above-mentioned problems in the prior art and to provide a grab bucket drainage device based on water pressure-volume linkage and intelligent sensing control.

[0009] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A grab bucket drainage device based on water pressure-volume linkage and intelligent sensing control includes a grab bucket body, a physical drainage subsystem, an intelligent sensing subsystem, an edge computing intelligent control subsystem, and a cloud-based digital twin subsystem. The device supports two operating modes: a no-air-supply mode and an air-supply AI adaptive mode. The physical drainage subsystem includes at least one deformable airbag, an air circuit assembly, and a drainage channel. The deformable airbag is fixed to the inner wall of the grab bucket, the air circuit assembly connects the airbag to the air replenishment control unit, and the drainage channel is opened at the bottom or side wall of the grab bucket. The intelligent sensing subsystem consists of multiple types of monitoring sensors, and all sensors are connected to the edge computing intelligent control subsystem via a waterproof industrial bus. The edge computing intelligent control subsystem includes an edge computing unit and a programmable logic controller. The edge computing unit has a built-in AI inference engine that can generate a gas replenishment control strategy based on real-time sensor data. The cloud-based digital twin subsystem communicates with the edge computing intelligent control subsystem via a wireless communication network for end-to-end operation management and model iteration.

[0010] Preferably, in the non-air replenishment mode, the pre-inflation pressure of the airbag is 105kPa~120kPa. After inflation, the air passage is sealed. The airbag can naturally expand as the external water pressure decreases during the lifting process of the grab bucket. The residual water in the squeeze bucket is discharged through the drainage channel without additional energy consumption. In the AI ​​adaptive mode of air replenishment, the AI ​​inference engine of the edge computing unit can adaptively output the optimal air replenishment pressure and air replenishment timing based on real-time water depth, mud-water mixing density, airbag internal pressure, and target drainage volume parameters using a deep reinforcement learning algorithm. The air replenishment pressure is 1.1 to 1.3 times the initial inflation pressure, and the air replenishment process is executed in 1 to 3 dynamic stages.

[0011] Preferably, the AI ​​inference engine built into the edge computing unit integrates three types of functional modules: (1) The working condition adaptive classification module adopts a CNN and GRU, temporal convolutional network architecture to classify the current working condition, and the working condition classification accuracy is ≥95%; (2) The time series prediction module adopts the LSTM, GRU and Transformer time series architecture to predict the drainage trend of the next 10 seconds based on the drainage data of the previous 30 seconds, with a prediction accuracy of ≥92%; (3) Anomaly detection module: using isolated forest, single-class support vector machine, local anomaly factor and its variants equipment health predictive maintenance model, autoencoder architecture to analyze operating data to achieve early warning of faults.

[0012] Preferably, the cloud-based digital twin subsystem has the following functions: (1) Real-time three-dimensional digital twin mapping based on the actual geometric parameters of the grab bucket is used to realize the visual monitoring of the operation; (2) The AI ​​model OTA remote update function can continuously iterate and optimize the control strategy based on the full amount of operation data; (3) Predictive maintenance reminder function; (4) Automatic generation function of compliance report for water discharge in the bucket.

[0013] Preferably, the monitoring sensors of the intelligent sensing subsystem include an airbag internal pressure sensor, a water depth / external pressure sensor, a mud-water density sensor, a drainage flow sensor, a six-axis attitude IMU sensor, and a temperature sensor, which can realize full-dimensional perception of the grab bucket's operating status, drainage status, and equipment health status.

[0014] Preferably, the deformable airbag is made of highly elastic, wear-resistant, and corrosion-resistant special rubber or composite material, and has a U-shaped, annular, or multi-chamber structure, with its outer surface coated with polytetrafluoroethylene or silicone anti-fouling coating.

[0015] Preferably, the drainage channel is an inclined or conical drainage hole with a diameter ranging from 80mm to 150mm. The drainage hole has a replaceable filter screen with a mesh size of 1mm to 10mm, which can be adjusted according to the particle size of the mud and sand being processed.

[0016] Another objective of this invention is to provide a grab bucket drainage method based on water pressure-volume linkage and intelligent sensing control.

[0017] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A grab bucket drainage method based on water pressure-volume linkage and intelligent sensing control includes the following steps: S1 system initialization: After the sensor self-test is completed, the airbag is filled with gas at a preset initial pressure, and the non-gas replenishment mode or the gas replenishment AI adaptive mode is selected according to the working conditions. S2 Excavation Operation: The grab bucket is launched into the water to excavate mud. External water pressure compresses the airbag to its minimum volume, and the intelligent sensing subsystem continuously collects all-dimensional operation data. S3 Drainage Trigger: After the six-axis attitude IMU sensor detects the lifting action of the grab bucket, the drainage control process is automatically triggered; S4 Drainage Execution: Drainage is completed according to the selected operating mode. After drainage is completed, the system automatically records the operation data and uploads it to the cloud digital twin subsystem. S5 Reset Cycle: After the grab bucket unloads mud, the airbag automatically deflates and resets, ready to enter the next working cycle.

[0018] Preferably, the drainage execution logic in step S4 is set to correspond to two operating modes respectively: (1) When the non-air replenishment mode is selected, the airbag expands naturally by squeezing the water out during the lifting process due to the pressure difference between the inside and outside. The edge computing unit continuously collects sensor data and performs multi-source data fusion to verify the drainage status through extended Kalman filter algorithm. (2) When the AI ​​adaptive mode for air replenishment is selected, the edge computing unit calls the deep reinforcement learning strategy network to output the air replenishment strategy. The programmable logic controller controls the air pump to replenish air in stages according to the preset time sequence. After each stage of air replenishment is completed, the pressure sensor provides feedback verification. The edge computing unit corrects the subsequent air replenishment parameters in real time.

[0019] Preferably, the cloud-based digital twin subsystem periodically optimizes and trains the AI ​​inference model at the edge based on full historical operation data. After training, updates are sent to the edge computing unit via OTA to achieve continuous iteration of the control strategy.

[0020] Due to the adoption of the above technical solution, the beneficial effects obtained by the present invention include: 1. This invention enables a non-air-supply mode in shallow water or low water content conditions, utilizing water pressure difference to drive the airbag to expand naturally for drainage. The entire process consumes no additional energy, reducing operating costs by more than 30% compared to traditional continuous air supply solutions and significantly improving energy efficiency. Simultaneously, in deep water or high-viscosity material conditions, it switches to an air-supply AI adaptive mode. The edge computing unit dynamically outputs the optimal air-supply pressure (1.1 to 1.3 times the initial pressure) and the air-supply sequence of 1 to 3 stages based on real-time water depth, mud-water density, airbag internal pressure, and other parameters through deep reinforcement learning. This improves drainage efficiency by more than 40% compared to traditional methods and effectively avoids the risk of overpressure. 2. This invention integrates an adaptive classification module for working conditions, which automatically identifies the working environment based on sensor data and enables seamless switching between two modes; at the same time, it combines cloud-based digital twins and OTA updates to support continuous iterative optimization of control strategies, forming a closed-loop intelligent system of "perception-decision-execution-evolution", which greatly improves the autonomous operation capability and reliability of equipment throughout its entire life cycle. 3. This invention uses composite rubber with a built-in mesh reinforcement structure to enhance the airbag's pressure and tear resistance, adapting to deep-water high pressure and frequent mechanical stress; its surface is coated with an antifouling coating, which can effectively inhibit the attachment of marine organisms and ensure the airbag's flexibility and response speed during long-term operation. Attached Figure Description

[0021] Figure 1 This is an overall architecture diagram of an embodiment of the grab bucket drainage device based on water pressure-volume linkage and intelligent sensing control of the present invention.

[0022] Figure 2 This is a schematic diagram of an embodiment of the physical drainage mechanism of the present invention.

[0023] Figure 3 This is a flowchart of one embodiment of two working modes of the present invention.

[0024] The attached figures are labeled as follows: 1. Grab body; 2. Airbag; 3. Air circuit assembly; 4. Drainage channel; 5. Air replenishment control unit. Detailed Implementation

[0025] Please see Figure 1-3As shown, this invention mainly provides a grab bucket drainage device based on water pressure-volume linkage and intelligent sensing control. It includes a grab bucket body 1, a physical drainage subsystem, an intelligent sensing subsystem, an edge computing intelligent control subsystem, and a cloud-based digital twin subsystem. The device supports two operating modes: a non-air-supply mode and an air-supply AI adaptive mode. In the non-air-supply mode, the airbags naturally expand and squeeze out water based on water pressure changes during grab bucket lifting, achieving zero-energy operation. In the air-supply AI adaptive mode, the edge computing system combines real-time sensor data and uses AI algorithms to dynamically adjust the air-supply strategy, adapting to the efficient drainage needs under complex working conditions, providing a cost-effective and intelligent tailwater discharge solution for grab bucket operations.

[0026] like Figure 1 As shown, the physical drainage subsystem in this embodiment includes at least one deformable airbag 2, an air circuit assembly 3, and a drainage channel 4. The deformable airbag 2 is fixed to the inner wall of the grab body 1. The air circuit assembly 3 connects the airbag 2 to the air replenishment control unit 5. The drainage channel 4 is opened at the bottom or side wall of the grab body 1. All components work together to achieve efficient separation of ore or mud and water in the grab.

[0027] Specifically: The deformable airbag 2, as the core actuator of the physical drainage subsystem, can be made of high-strength flexible rubber composite material. It is custom-cut according to the curvature of the inner wall of the grab bucket body 1, possessing excellent compressive strength, wear resistance, and corrosion resistance, and can adapt to the complex mud and water environment and pressure changes during grab bucket operations. The airbag is tightly attached to the inner wall of the grab bucket using special fasteners (such as a high-temperature resistant and corrosion-resistant adhesive), typically arranged symmetrically on the sides or top inner wall of the grab bucket. This arrangement does not affect the normal loading space of the grab bucket, while maximizing the compression of residual water inside the bucket during expansion. The design of the airbag 2 fully integrates the grab bucket's volume with the operating conditions. Initially, it is flat, occupying minimal space within the grab bucket. When the grab bucket is raised and the external water pressure decreases, the airbag can rapidly expand, reaching a maximum volume of 30%-40% of the grab bucket's internal space. This effectively compresses the ore or slurry on the grab bucket's inner wall, ensuring efficient removal of residual water. Simultaneously, the airbag surface undergoes a special anti-slip treatment to prevent relative sliding with the grab bucket's inner wall or materials, ensuring long-term stable operation.

[0028] The air circuit assembly 3 is the hub connecting the deformable airbag and the 2 air replenishment control unit 5. It consists of components such as a high-pressure air hose, a solenoid valve, a pressure sensor, and quick connectors (not explicitly shown in the diagram for simplicity). The high-pressure air hose is a polyurethane flexible tube with embedded steel wire, which can be seamlessly connected to the airbag inlet and the air replenishment control unit via quick connectors. The solenoid valve, as the control switch of the air circuit, adopts a high-response pilot structure, which can complete the switching of the air circuit on and off in milliseconds, and precisely control the inflation and deflation process of the airbag. The pressure sensor monitors the air pressure inside the airbag in real time. When the air pressure reaches the preset threshold, it automatically triggers the air replenishment or deflation command and transmits the data synchronously to the edge computing intelligent control subsystem to provide accurate pressure parameters for the AI ​​algorithm and realize the dynamic adjustment of the airbag pressure. The quick connectors ensure convenient installation and maintenance of the air circuit assembly, and the disassembly and replacement of the tubing can be completed in a short time.

[0029] Drainage channels 4 can be located at the bottom or side wall of the grab bucket body, depending on the type of grab bucket and the operating scenario. They typically employ an inclined or conical drainage hole design, forming an angle of 15°-30° with the inner wall of the grab bucket, facilitating rapid water discharge under the combined effects of gravity and airbag compression. The diameter of the drainage holes within the drainage channels ranges from 80mm to 150mm, and each hole contains a replaceable filter screen with a mesh size of 1mm to 10mm. The filter screen aperture is customized according to the particle size of the material being handled by the grab bucket, effectively intercepting mud, sand, gravel, and other debris to prevent clogging of the drainage channels and avoid material loss with the water.

[0030] The number and size of drainage channels 4 are designed reasonably according to the grab bucket volume and drainage requirements. Small and medium-sized grab buckets are usually equipped with 2-4 drainage channels, while large grab buckets are equipped with 4-6 channels. The cross-sectional area of ​​a single channel ensures that it will not generate excessive resistance when draining at high flow rates. The inner wall of the channel is smoothed to reduce water flow resistance and further improve drainage efficiency.

[0031] The working principle of the physical drainage subsystem is based on the combination of air pressure drive and gravity drainage. By expanding and contracting deformable airbags, the internal spatial structure of the grab bucket is changed, which enhances the squeezing and flow of water between the ore or mud, and finally achieves solid-liquid separation through the drainage channel.

[0032] After the grab bucket completes the grabbing operation of ore or mud, the air replenishment control unit receives the grab bucket closing signal and immediately activates the air circuit assembly to inject high-pressure gas into the deformable airbag. Under the action of air pressure, the airbag expands rapidly, forming a uniform extrusion force on the mud on the inner wall of the grab bucket. Under the action of extrusion force, the free water between the ore or mud particles is separated from the ore or mud particles and converges into water flowing to the drainage channel at the bottom or side wall of the grab bucket. At this time, the pressure sensor provides real-time feedback on the air pressure inside the airbag. When the air pressure reaches the preset 2MPa, the flow regulating valve automatically adjusts the air intake speed to maintain the airbag pressure stable and avoid excessive compaction of ore or mud or damage to the airbag due to excessive pressure.

[0033] During the drainage process, the air supply control unit monitors the moisture changes of the ore or slurry in real time through the humidity sensor built into the grab bucket. When the moisture of the ore or slurry drops below a preset threshold, the solenoid valve switches to the venting state, and the deformable airbag quickly contracts, returning to its initial flat state to avoid affecting subsequent unloading operations. The entire drainage process takes only 30-60 seconds, improving drainage efficiency by more than 60% compared to traditional natural drainage methods.

[0034] While the grab bucket is unloading, the drainage channels continue to function. As the grab bucket tilts, moisture remaining on the surface of the ore or slurry is discharged along the drainage channels under gravity. At the same time, the deformable airbags can inflate and contract intermittently, causing slight vibrations to the inner wall of the grab bucket, reducing ore or slurry residue and further improving the cleanliness of the unloading.

[0035] In this embodiment, the intelligent sensing subsystem serves as the core front-end for data acquisition and consists of multiple types of monitoring sensors. All sensors are connected to the edge computing intelligent control subsystem via a waterproof industrial bus. Specifically: The monitoring sensors of the intelligent sensing subsystem include an airbag pressure sensor, a water depth / external pressure sensor, a mud-water density sensor, a drainage flow sensor, a six-axis attitude IMU sensor, and a temperature sensor, which can realize full-dimensional perception of the grab bucket's operating status, drainage status, and equipment health status. in; Airbag internal pressure sensor: A high-precision diffused silicon pressure sensor is installed inside the deformable airbag to monitor changes in gas pressure inside the airbag in real time, providing core pressure data for AI control strategies; the sensor has IP68 waterproof performance and can work stably in the humid and dusty environment inside the grab bucket for a long time.

[0036] External water pressure sensor: Installed on the outer wall of the grab bucket, it calculates real-time water depth data by measuring the static pressure of the water body. The measurement range is 0-100m, and the response time is ≤0.5s, accurately capturing the dynamic changes of external water pressure during the lifting process of the grab bucket.

[0037] Drainage flow sensor: It adopts a non-contact ultrasonic flow meter, which is installed outside the drainage channel. It can accurately measure the drainage flow without entering the water flow and can provide real-time feedback on drainage efficiency, providing a basis for AI algorithm to adjust the air replenishment strategy.

[0038] Slurry density sensor: It adopts the vibration density measurement principle and is installed on the inner wall of the grab bucket to monitor the density change of the slurry mixture in the bucket in real time. The measurement range is 1.0-2.0 g / cm³, which helps the system to judge the material loading status and the proportion of water residue.

[0039] Six-axis attitude sensor (IMU): Integrating a three-axis accelerometer and a three-axis gyroscope, it is installed at the center of the grab body to monitor the grab's tilt angle, motion acceleration and rotational angular velocity in real time, ensuring that the system accurately grasps the grab's operating attitude.

[0040] Ambient temperature and humidity sensor: Installed on the outside of the grab bucket, it monitors changes in temperature and humidity of the working environment, providing environmental parameter support for sensor data compensation and equipment status assessment.

[0041] All of the above sensors are connected to the edge computing intelligent control subsystem via a waterproof industrial bus to build a stable and reliable data transmission network.

[0042] In this embodiment, the intelligent sensing subsystem adopts a "distributed acquisition, centralized processing" working mode. Each sensor independently collects data and transmits it to the edge computing unit in real time through an industrial bus. The edge computing intelligent control subsystem performs fusion analysis on the collected multi-source data. It can remove noise interference and extract effective feature parameters through algorithms such as Kalman filtering and wavelet analysis to provide high-quality input data for the AI ​​inference engine.

[0043] In non-air replenishment mode, sensor data is mainly used to monitor the inflation status of the airbag and the drainage progress to ensure that the drainage process is stable and controllable. In air replenishment AI adaptive mode, sensor data is input into the AI ​​inference engine in real time. The deep reinforcement learning algorithm dynamically adjusts the air replenishment pressure and timing based on real-time water pressure, flow rate, density and other data to achieve optimal control of drainage efficiency.

[0044] In this embodiment, the edge computing intelligent control subsystem includes an edge computing unit and a programmable logic controller. The edge computing unit has a built-in AI inference engine that can generate a gas replenishment control strategy based on real-time sensor data. Specifically: The edge computing unit adopts an industrial-grade embedded hardware platform, equipped with a high-performance processor and a dedicated AI acceleration chip, possessing powerful local data processing and AI inference capabilities. The unit has a built-in lightweight AI inference engine that can efficiently run deep reinforcement learning algorithm models. Taking water depth, mud-water density, airbag internal pressure, and target displacement as input features, it employs deep reinforcement learning algorithms, including but not limited to Proximal Policy Optimization (PPO), Advantageous Actor-Critic (A3C), Flexible Actor-Critic (SAC), Deep Q-Network (DQN), and their variants, to adaptively calculate the optimal air injection pressure and timing. Based on real-time sensor data collected by the intelligent sensing subsystem, it quickly generates the optimal air injection control strategy.

[0045] As the core of the system's execution control, the programmable logic controller (PLC) adopts a modular design and can be flexibly configured with input / output modules, communication modules, etc., according to actual needs. The PLC is responsible for receiving the air replenishment control strategy generated by the edge computing unit and converting it into specific control instructions, which are then used to precisely control the air replenishment process of the airbag through the solenoid valves of the air circuit components.

[0046] Furthermore, the AI ​​inference engine integrates three types of functional modules: 1. The adaptive classification module for operating conditions includes, but is not limited to, deep learning models such as a hybrid architecture of convolutional neural networks (CNN) and gated recurrent units (GRU) (CNN+GRU), CNN+LSTM, ResNet+BiLSTM, and temporal convolutional networks (TCN). For complex and ever-changing operating scenarios, it can automatically capture key features and dynamic change patterns in operating condition data, achieving accurate classification of equipment operating conditions (shallow water / deep water, low viscosity / high viscosity), with a stable classification accuracy rate of over 95%. This high-accuracy classification capability provides a solid foundation for subsequent intelligent control and decision-making, ensuring that the equipment always operates within the optimal operating condition range. 2. The time series prediction module, including but not limited to architectures such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer time series network, and Informer, possesses powerful time series data processing and trend analysis capabilities. Based on real-time drainage data from the past 30 seconds, it can deeply mine the potential correlations and evolutionary trends behind the data and accurately predict the drainage trend for the next 10 seconds, with a prediction accuracy of no less than 92%. By understanding the operational status of the drainage system in advance, maintenance personnel can take timely and targeted control measures to effectively avoid equipment failures and safety hazards caused by drainage anomalies, thereby improving the operational stability and reliability of the drainage system. 3. The anomaly detection module integrates Isolation Forest, One-Class SVM, Autoencoder, Local Outlier Factor (LOF), and their variants into a predictive maintenance model for equipment health, constructing a multi-dimensional, multi-layered anomaly detection system. This system can detect anomalies in multi-dimensional time-series data such as airbag internal pressure, air pump current, and temperature in real time, issuing accurate early warnings 7-14 days before equipment failure occurs. Through early warnings, the maintenance team has ample time to conduct equipment inspection and maintenance, eliminating potential faults in their infancy, significantly reducing equipment downtime risks and maintenance costs, and significantly improving equipment lifespan and overall efficiency.

[0047] The working mechanism of the AI ​​inference engine is as follows: The AI ​​inference engine can receive multi-source sensor data transmitted by the intelligent sensing subsystem in real time, including airbag internal pressure, external water pressure, drainage flow rate, mud and water density, and grab bucket posture. Before performing inference calculations, the engine preprocesses the raw data, including data cleaning, normalization, and feature extraction, to remove noise interference and extract effective feature parameters, providing high-quality input data for subsequent strategy generation. Meanwhile, the engine's built-in deep reinforcement learning algorithm models [including but not limited to Proximal Policy Optimization (PPO), Advantageous Actor-Critic (A3C), Flexible Actor-Critic (SAC), Deep Q-Network (DQN) and other algorithms, adaptively generate air replenishment strategies (air replenishment pressure P_add, number of segments N_segments, duration of each segment T_segment)], continuously learning and optimizing through continuous interaction with the environment; during the grab bucket drainage operation, the algorithm takes the current sensor data as the state input, the air replenishment pressure, air replenishment timing and other parameters as the action output, and the drainage efficiency, energy consumption and other parameters as the reward signal, and gradually generates the optimal air replenishment control strategy through continuous trial and error and learning; The algorithm model has adaptive adjustment capabilities for different operating conditions. For example, in deep-water operations, the algorithm will appropriately increase the air supply pressure based on real-time water pressure data to enhance the airbag expansion capacity. In high-viscosity mud operations, the algorithm will optimize the air supply sequence and extend the air supply time to ensure that water is fully discharged.

[0048] In addition, the air replenishment control strategy generated by the AI ​​inference engine is output to the PLC in the form of digital instructions. The PLC controls the solenoid valve of the air circuit component according to the instructions to achieve precise air replenishment of the airbag. At the same time, the engine continuously receives feedback data during the drainage process, evaluates the effect of strategy execution, and continuously optimizes the algorithm model parameters through an online learning mechanism to improve the adaptability and effectiveness of the strategy.

[0049] In non-inflated mode, the edge computing unit is mainly responsible for real-time monitoring and analysis of the data collected by the intelligent sensing subsystem, and uses the AI ​​inference engine to determine the inflation status and drainage progress of the airbag. When abnormal pressure inside the airbag or low drainage efficiency is detected, the system will issue an early warning signal in a timely manner to remind the operator to check and handle the situation. The PLC is in standby mode and will only perform safety protection actions when the system encounters an emergency.

[0050] In the AI ​​adaptive mode for air replenishment, the edge computing unit and PLC work together to achieve closed-loop control. The intelligent sensing subsystem collects operation data in real time and transmits it to the edge computing unit. The AI ​​inference engine generates an air replenishment control strategy based on the data and sends it to the PLC. The PLC controls the air circuit components to perform air replenishment operations according to the strategy. At the same time, the intelligent sensing subsystem transmits feedback data during the air replenishment process to the edge computing unit in real time. The AI ​​inference engine dynamically adjusts the strategy based on the feedback data to ensure that the drainage efficiency is always at its optimal state.

[0051] In this embodiment, the cloud-based digital twin subsystem communicates with the edge computing intelligent control subsystem via a wireless communication network, constructing an "edge-cloud" collaborative closed loop for end-to-end operation management and model iteration. In this architecture, the edge side is responsible for real-time data acquisition, AI inference, and local control decisions, ensuring low-latency response in grab bucket drainage operations; while the cloud-based digital twin subsystem constructs a high-fidelity virtual mapping model based on sensor data, control logs, and operation results uploaded from the edge, enabling full lifecycle monitoring and in-depth analysis of the physical system.

[0052] Specifically, it includes: 1. Data Interoperability Mechanism and Functional Division of Labor Uplink data stream (edge ​​→ cloud): Real-time upload of key process data, including raw sensor data such as water depth h, mud-water mixture density ρmix, airbag volume Vcur, pressure Pcur, and discharge Q; Upload AI decision logs, such as operating condition classification results, predicted trends, generated gas replenishment strategies (gas replenishment pressure Padd, gas replenishment timing Tseq), and actual execution deviations. Fault warning information: Equipment health status assessment, abnormal detection trigger records and maintenance suggestions.

[0053] After being compressed and encrypted, the data is transmitted to the cloud via highly reliable wireless communication networks such as 5G or industrial Wi-Fi 6, ensuring a stable connection in complex port environments.

[0054] Downlink data stream (cloud → edge): Deploys optimized AI model parameters: an updated inference model trained on global job data, improving the accuracy of edge condition identification and prediction; Synchronous control strategy templates: a library of best practice strategies refined for typical working conditions (such as high-viscosity sludge and gravel mixtures); Remote configuration commands: system parameter adjustment, security threshold update, firmware upgrade package, etc.

[0055] 2. End-to-end operation control capabilities Real-time 3D digital twin mapping and visualization monitoring: Based on the actual geometric parameters of the grab (such as dimensions, airbag distribution, sensor layout, etc.), a high-fidelity dynamic 3D digital model is constructed in the cloud. This model is driven by data uploaded in real time from the edge, realizing the synchronous evolution of the physical grab and the virtual model, supporting remote multi-view visualization monitoring of the operation status, and facilitating operators to promptly detect abnormal working conditions; Operation traceability and auditing: Completely record the input characteristics, control process and output effect of each round of operation to meet the requirements of quality traceability and environmental compliance; Multi-machine collaborative optimization: When multiple grabs are operating in parallel, the cloud can coordinate resource allocation and operation rhythm to avoid mutual interference and improve overall efficiency.

[0056] 3. Model Iteration and Continuous Evolution Data-driven learning: New data generated by each assignment is automatically injected into the cloud training pool for retraining and validating AI models, forming a "deployment-feedback-optimization" closed loop; Federated learning support: While protecting data privacy, it supports joint training of models across sites, enabling equipment in different ports to share intelligent experiences; Version management and canary release: After the new model is verified in the cloud, it is pushed to edge devices in batches to ensure that system upgrades are safe and controllable.

[0057] 4. Predictive maintenance reminder function Based on historical data and machine learning algorithms (such as isolated forest and autoencoder), the cloud performs trend analysis on potential faults such as airbag aging, seal leakage, and sensor drift, and issues maintenance warnings 7 to 14 days in advance, reducing the risk of sudden downtime and improving equipment availability.

[0058] Automatic generation of compliance reports for water discharge into the container The system automatically records key parameters such as drainage volume, drainage time, mud-water mixing density, and ambient water depth for each operation, and generates standardized emission compliance reports based on environmental regulatory requirements. It supports one-click export or remote upload to the regulatory authority's platform, helping enterprises meet ecological compliance needs.

[0059] This system transforms physical operations into measurable, traceable, and optimizable digital processes through a closed loop of "perception-decision-execution-optimization," significantly improving operational efficiency and management transparency. Furthermore, the "cloud-edge" collaboration mechanism not only enables precise control over individual operations but also drives the entire operational system towards self-adaptation and self-evolution through continuous learning and global optimization, significantly enhancing its intelligence level and operational efficiency.

[0060] In this embodiment, the dual-mode intelligent drainage control strategy is as follows: In non-air replenishment mode, for conventional working conditions such as shallow water operations and materials with low moisture content, the system adopts a zero-energy passive drainage solution. Before operation, the deformable airbags on the inner wall of the grab bucket are pre-inflated, with the pressure precisely controlled within the range of 105kPa to 120kPa. After inflation, the air passage is immediately sealed. When the grab bucket grabs and lifts material, the external water pressure decreases synchronously with the decrease in water depth. The airbags naturally expand due to the pressure difference between the inside and outside, continuously squeezing the space inside the bucket and forcing the residual water to be discharged through the drainage channels at the bottom or side wall. No additional energy input is required throughout the process, which can reduce equipment operating costs by more than 30%.

[0061] For example, in the "no gas replenishment mode", a physical drainage mechanism is adopted. Based on the ideal gas law, under isothermal conditions, P1V1= P2V2 is satisfied, that is, the pressure of the gas in the airbag is inversely proportional to the volume.

[0062] As the grab bucket is lifted from a depth h to the surface, the external water pressure P2 = Patm + ρgh gradually decreases. Meanwhile, the gas inside the airbag, being sealed and at a constant temperature, expands according to Boyle's Law. Initially (underwater), the volume is P1V1. As P2 decreases, the airbag volume V2 naturally increases, compressing the remaining water inside the bucket and pushing it out through the drainage channel.

[0063] Furthermore, this process requires no additional energy consumption, relying entirely on the gas expansion effect driven by water pressure changes to achieve efficient and energy-saving passive drainage. It is suitable for working conditions with less silt and low drainage resistance, and is the core working principle of the system in "non-air replenishment mode".

[0064] In AI adaptive air replenishment mode: For complex working conditions such as deep-water operations and materials with high moisture content, the system activates an active and precise drainage mechanism. The AI ​​inference engine built into the edge computing unit can adaptively output the optimal air replenishment pressure and timing using deep reinforcement learning algorithms, based on real-time water depth, mud-water mixing density, current airbag volume, target drainage volume, and current internal pressure parameters of the airbag. The optimal air replenishment pressure output by the algorithm is strictly limited to 1.1 to 1.3 times the initial inflation pressure, ensuring sufficient expansion power while avoiding over-inflation damage to the airbag. The air replenishment process is adaptively divided into 1 to 3 dynamic stages according to the working conditions. For example, a three-stage strategy of "rapid pressure increase - pressure stabilization and maintenance - pressure replenishment and termination" is adopted for highly viscous materials, while a two-stage strategy of "initial air replenishment - secondary fine-tuning" is adopted for conventional complex working conditions. The drainage efficiency is improved by more than 40% compared with the traditional mode.

[0065] Among them, the air replenishment AI adaptive mode realizes intelligent drainage control through edge computing and deep reinforcement learning (DRL). Its working principle can be summarized as a closed-loop dynamic control process of "perception-decision-execution".

[0066] Specifically: The mechanism consists of three parts: the perception layer, the decision-making layer, and the execution layer, forming a real-time feedback closed loop.

[0067] 1. Perception Layer: Construction of Multidimensional State Space The AI ​​inference engine collects the following key feature vectors as input in real time: Environmental parameters: real-time water depth h (determines external hydrostatic pressure Pext=Patm+ρgh), mud-water mixture density ρmix (determines drainage resistance).

[0068] Equipment status: Current airbag volume Vcu, current internal pressure Pcur, airflow rate.

[0069] Task objectives: target water displacement Qtarget, remaining operation time window.

[0070] 2. Decision Layer: Deep Reinforcement Learning (DRL) Policy Generation The AI ​​models built into the edge computing unit (such as PPO or DQN algorithms) map the above states into an action space: Action output: Optimal air replenishment pressure setting Pset and air replenishment timing Tseq.

[0071] Pressure range constraint: The replenishment pressure is strictly controlled between 1.1 and 1.3 times the initial pressure to ensure a balance between safety and effectiveness.

[0072] Phase division logic: Phase 1 (Rapid Response Period): When the grab bucket has just started lifting or is in deep water, the external water pressure is high. The AI ​​determines that a basic pressure difference needs to be established quickly, outputs a higher pressure increment (such as 1.3 times the initial pressure), rapidly expands the airbag to overcome the hydrostatic pressure, and starts drainage.

[0073] Phase Two (Fine-tuning Period): As water depth decreases and external pressure drops, the natural expansion tendency of the airbags intensifies. The AI ​​detects that the drainage rate is stabilizing or approaching the target and switches to medium-pressure replenishment (e.g., 1.2 times) to maintain drainage power while preventing over-inflation.

[0074] Phase 3 (Pressure Holding / Fine-Tuning Period): Approaching the water surface or nearing the end of drainage, the resistance mainly comes from the viscosity of the mud. The AI ​​outputs low-pressure fine-tuning (e.g., 1.1 times or stops air replenishment), utilizing the residual elasticity of the airbags to complete the final venting and avoid gas escape and waste.

[0075] 3. Execution layer: Segmented closed-loop control The gas path components execute according to the timing instructions issued by the AI: Dynamic valve control: The proportional control valve precisely adjusts the intake flow rate according to Pset, rather than a simple on / off control.

[0076] Real-time feedback correction: After each stage, the sensors reassess Pcur and Vcur. If the actual drainage effect is lower than predicted (e.g., due to a sudden increase in resistance caused by mud blockage).

[0077] This mode is suitable for complex working conditions such as high-density mud or varying water depths, and has stronger adaptability and operational efficiency compared to the passive mode.

[0078] It should be noted that this invention employs a non-air-supply mode in shallow water or low-moisture conditions, utilizing water pressure difference to drive the airbag to naturally expand and drain water. The entire process consumes no additional energy, reducing operating costs by over 30% compared to traditional continuous air supply solutions and significantly improving energy efficiency. Simultaneously, in deep water or high-viscosity material conditions, it switches to an AI-adaptive air-supply mode. The edge computing unit dynamically outputs the optimal air-supply pressure (1.1-1.3 times the initial pressure) and a 1-3 stage air-supply sequence based on real-time water depth, mud-water density, and airbag internal pressure through deep reinforcement learning. This improves drainage efficiency by over 40% compared to traditional methods and effectively avoids overpressure risks. Furthermore, the system integrates an adaptive operating condition classification module that automatically identifies the operating environment based on sensor data, enabling seamless switching between the two modes. Combined with cloud-based digital twins and OTA updates, it supports continuous iterative optimization of control strategies, forming a closed-loop intelligent system of "perception-decision-execution-evolution," significantly enhancing the autonomous operation capability and reliability of the equipment throughout its entire lifecycle.

[0079] The operation method of this invention is as follows: S1 system initialization: After the sensor self-test is completed, the airbag is filled with gas at a preset initial pressure, and the non-gas replenishment mode or the gas replenishment AI adaptive mode is selected according to the working conditions. S2 Excavation Operation: The grab bucket is launched into the water to excavate mud. External water pressure compresses the airbag to its minimum volume, and the intelligent sensing subsystem continuously collects all-dimensional operation data. S3 Drainage Trigger: After the six-axis attitude IMU sensor detects the lifting action of the grab bucket, the drainage control process is automatically triggered; S4 Drainage Execution: Drainage is completed according to the selected operating mode. After drainage is completed, the system automatically records the operation data and uploads it to the cloud digital twin subsystem. S5 Reset Cycle: After the grab bucket unloads mud, the airbag automatically deflates and resets, ready to enter the next working cycle.

[0080] In this embodiment, the drainage execution logic in step S4 is set to correspond to two operating modes respectively: (1) When the non-air replenishment mode is selected, the airbag expands naturally by squeezing the water out during the lifting process due to the pressure difference between the inside and outside. The edge computing unit continuously collects sensor data and performs multi-source data fusion to verify the drainage status through extended Kalman filter algorithm. Specifically: In this mode, the system relies entirely on physical mechanisms for drainage, with a focus on the accurate monitoring and verification of the drainage status; Natural expansion drainage: The air passage remains sealed. As the grab bucket is lifted, the external water pressure gradually decreases. The airbag expands naturally using the pressure difference between the inside and outside, squeezing out the residual water in the bucket and discharging it through the drainage channel. The entire process consumes no additional energy. Multi-source data fusion verification: The edge computing unit continuously collects multi-dimensional sensor data such as water depth, airbag volume, and pressure, and uses the Extended Kalman Filter (EKF) algorithm to perform multi-source data fusion. This algorithm can effectively filter out sensor noise, estimate the true volume change rate of the airbag and the water discharge rate in real time, thereby verifying whether the drainage status meets expectations and ensuring the reliability of the drainage effect in zero-energy mode.

[0081] (2) When the AI ​​adaptive mode for air replenishment is selected, the edge computing unit calls the deep reinforcement learning strategy network to output the air replenishment strategy, the programmable logic controller controls the air pump to replenish air in stages according to the preset time sequence, and the pressure sensor provides feedback verification after each stage of air replenishment is completed. The edge computing unit corrects the subsequent air replenishment parameters in real time. Specifically: In this mode, the system achieves efficient drainage through closed-loop control of "perception-decision-execution-feedback"; AI strategy generation: The edge computing unit calls the built-in deep reinforcement learning strategy network to output the optimal air replenishment pressure and the staged air replenishment sequence based on real-time operating conditions (water depth, mud density, etc.); PLC execution control: The programmable logic controller (PLC) receives strategy instructions and precisely controls the air pump to replenish air in stages (stages 1 to 3) according to a preset sequence.

[0082] Real-time feedback correction: After each stage of air replenishment is completed, the pressure sensor immediately feeds back the actual internal pressure value; the edge computing unit compares the feedback value with the target value, and if there is a deviation, it corrects the air replenishment parameters of the subsequent stages in real time (such as adjusting the inflation rate or duration), forming a dynamic closed-loop control to ensure that the final drainage effect meets the target requirements.

[0083] In this embodiment, the cloud-based digital twin subsystem periodically optimizes and trains the AI ​​inference model at the edge based on full historical operation data. After training, the update is sent to the edge computing unit via OTA to achieve continuous iteration of the control strategy.

[0084] The specific operational logic is as follows: 1. Cloud-based Global Optimization and Model Training: The cloud acts as the "central brain," aggregating massive amounts of historical operational data from all edge nodes. Leveraging its powerful computing resources, the cloud performs deep analysis and global strategy optimization. For example, by constructing high-precision Physical Information Neural Networks (PINNs) or reinforcement learning models, large-scale simulation training can be conducted in a virtual digital twin. This training can identify hidden energy-saving opportunities, predict equipment failures, or optimize production scheduling, generating new AI models or control strategies that are more accurate and have a more global perspective than local models at the edge.

[0085] 2. Model Lightweighting and Adaptation: Due to the limited computing power of edge computing units (e.g., limited CPU / GPU resources), highly complex models trained in the cloud cannot be directly deployed. Therefore, before OTA deployment, model compression is usually required, including knowledge distillation, pruning, and quantization (e.g., converting FP32 to INT8), to transform large models into lightweight versions suitable for edge devices while ensuring that accuracy loss is within an acceptable range.

[0086] 3. OTA Delivery and Edge Updates: Through OTA (Over-The-Air) technology, the cloud securely pushes optimized lightweight models or strategy parameters to the edge computing unit. Upon receiving the update, the edge device replaces the old model. This process achieves a shift from "static preset" to "dynamic evolution," ensuring that the edge controller can adapt to changes in equipment wear and tear, environmental disturbances, or production demands.

[0087] 4. Edge Real-Time Inference and Feedback: The updated edge AI model is immediately put into operation, responsible for millisecond-level real-time data acquisition, local processing, and control command execution. The edge device uploads new operational data and control effects back to the cloud in real time, forming a closed-loop data flow of "perception-decision-optimization-update". This continuous iteration mechanism enables the system to continuously self-correct and improve, achieving an upgrade from passive response to proactive predictive maintenance and adaptive control.

[0088] This architecture effectively balances the powerful computing capabilities of the cloud with the low latency requirements of the edge, solving the problem that traditional control systems are inadequate when facing the complex and ever-changing physical world. It is a key technological path for achieving intelligent control in Industry 4.0 and smart cities.

[0089] Based on the above analysis, the implementation will be carried out below, combining theoretical and specific experimental data, as follows: Example 1: Small and medium-sized grab buckets Application scenarios: Inland waterway maintenance, dredging of small ports, with a working water depth of ≤10m.

[0090] (1) Grab bucket volume: 4m³ (2) Airbag assembly: 2 U-shaped airbags (0.4m in diameter and 2m in length), reinforced silicone rubber (tensile strength ≥48MPa, elongation at break ≥480%, thickness 5mm), initial air pressure 112kPa, and surface coated with 0.1mm polytetrafluoroethylene coating.

[0091] (3) Drainage channel: 6 inclined drainage holes (diameter 110mm, inclination angle 15°), with built-in filter screen (mesh size 3mm).

[0092] (4) Target water depth: 20m (5) Test results:

[0093] Test results show that, compared to traditional gravity drainage or unassisted drainage methods, the introduction of an airbag-assisted drainage mechanism significantly improves operational efficiency. No-air-replenishment mode: Relying solely on natural expansion due to water pressure changes, it achieves a 68% reduction in residual water and a 78% reduction in drainage time. This demonstrates that even with zero additional energy consumption, physical structure optimization can bring significant efficiency gains.

[0094] AI Adaptive Air Supplement Mode: By actively supplementing air (increasing pressure from 112kPa to 130kPa), the residual water reduction rate is further increased to 80%, and the drainage time is shortened by 90%. This means that within the same work cycle, the effective load ratio is higher, the effective earthwork volume per excavation is increased, and the unit volume operation cost and time cost are significantly reduced.

[0095] Meanwhile, the system's dual-mode operation strategy allows the equipment to freely switch between "ultimate energy saving" and "ultimate efficiency" based on actual operating conditions: In routine maintenance where the water depth is shallow (≤10m) and the drainage speed requirements are not extremely stringent, the non-air replenishment mode is adopted. By utilizing the initial air pressure of 112kPa and the natural physical properties of the U-shaped airbag, efficient drainage without additional energy consumption is achieved, which greatly extends the equipment's endurance and reduces operating costs.

[0096] When dealing with high-viscosity mud or high-intensity operations requiring rapid turnover, the air replenishment mode can be activated to maximize drainage performance with minimal energy input (increasing the air pressure to 1.1 to 1.3 times the normal level), demonstrating extremely high energy efficiency.

[0097] In addition, in Example 1, the replenishment pressure was set to 130 kPa (approximately 1.16 times the initial pressure of 112 kPa), which is within the preset optimal range of 1.1 to 1.3 times. This result verifies that the AI ​​adaptive algorithm can accurately identify the operating conditions and output a reasonable replenishment strategy, avoiding the safety risks and energy waste caused by overpressure, and overcoming the problem of incomplete drainage caused by underpressure, thus achieving precise control.

[0098] Example 2: Large Grab Bucket Application scenarios: coastal port expansion and waterway maintenance, with operating water depth ≤30m.

[0099] (1) Grab bucket volume: 8m³ (2) Airbag assembly: 3 multi-chamber airbags (2 sidewalls, 1 bottom), composite rubber (tensile strength ≥55MPa, elongation at break ≥500%, thickness 8mm), initial air pressure 120kPa, built-in mesh reinforcement structure, and surface coated with anti-fouling coating.

[0100] (3) Drainage channel: 10 inclined drainage holes (150mm in diameter, 20° inclination angle), with built-in filter screen (5mm mesh).

[0101] (4) Target water depth: 30m (5) Test results:

[0102] Test results show that the system maintains high drainage capacity even under high external pressure conditions with a water depth of up to 30m. Non-air replenishment mode: Relying on an initial air pressure of 120 kPa and a multi-chamber airbag design, the residual water volume is reduced by 75% and the drainage time is shortened by 85%, far exceeding the traditional gravity drainage method, which verifies the effectiveness of the passive drainage mechanism in deep water conditions.

[0103] AI Adaptive Air Supplement Mode: The air supplement pressure is increased to 140 kPa (1.17 times the initial pressure), which is in the optimal range of 1.1 to 1.3 times. The residual water reduction rate is further increased to 85%, and the drainage time is shortened by 95%, significantly improving the net load and circulation efficiency of a single operation.

[0104] Meanwhile, the three independent chambers (two sidewalls + one bottom) enable multi-point coordinated expansion, uniformly squeezing the mud-water mixture within the bucket, avoiding uneven drainage in certain areas, and improving drainage integrity. The use of composite rubber (tensile strength ≥55MPa, elongation at break ≥500%) with a built-in mesh reinforcement structure enhances the airbag's pressure and tear resistance, adapting to deep-water high pressure (≤30m) and frequent mechanical stress. The surface is coated with an antifouling coating, effectively inhibiting marine organism attachment and ensuring the airbag's flexibility and response speed during long-term operation. It is equipped with 10 drainage holes with a diameter of 150mm and an inclination angle of 20°, further expanding the flow area compared to Example 1, adapting to the high drainage requirements of an 8m³ large-capacity grab bucket. The built-in 5mm mesh filter prevents large particle clogging while balancing flow efficiency and filtration accuracy, making it suitable for marine sediment environments containing impurities such as gravel and shells.

[0105] In addition, the initial air pressure is set at 120 kPa, which is at the upper limit of the system's recommended range (105~120 kPa) and matches the high-pressure background of deep-water operations. The replenishment air pressure of 140 kPa is dynamically generated by an AI adaptive algorithm, which reflects the system's ability to compensate for the characteristics of large grab buckets, such as large inertia and slow response, to ensure that the replenishment air process is smooth and shock-free and avoid structural damage.

[0106] In summary, this invention, through the deep integration of structural reinforcement, channel optimization, and intelligent control in large grab applications, successfully overcomes challenges such as deep water pressure, large volume, and complex media, demonstrating excellent engineering applicability and technological scalability.

[0107] The foregoing descriptions and embodiments are provided to enable those skilled in the art to understand and apply the present invention. It will be apparent to those skilled in the art that various modifications can be easily made to these contents, and the general principles described herein can be applied to other embodiments without creative effort. Therefore, the present invention is not limited to the foregoing descriptions and embodiments. Improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from its scope should be within the protection scope of the present invention.

Claims

1. A grab bucket drainage device based on water pressure-volume linkage and intelligent sensing control, comprising a grab bucket body, characterized in that, It also includes a physical drainage subsystem, an intelligent sensing subsystem, an edge computing intelligent control subsystem, and a cloud-based digital twin subsystem. The device supports two operating modes: a no-air-replenishment mode and an AI-adaptive air-replenishment mode. The physical drainage subsystem includes at least one deformable airbag, an air circuit assembly, and a drainage channel. The deformable airbag is fixed to the inner wall of the grab bucket, the air circuit assembly connects the airbag to the air replenishment control unit, and the drainage channel is opened at the bottom or side wall of the grab bucket. The intelligent sensing subsystem consists of multiple types of monitoring sensors, and all sensors are connected to the edge computing intelligent control subsystem via a waterproof industrial bus. The edge computing intelligent control subsystem includes an edge computing unit and a programmable logic controller. The edge computing unit has a built-in AI inference engine that can generate a gas replenishment control strategy based on real-time sensor data. The cloud-based digital twin subsystem communicates with the edge computing intelligent control subsystem via a wireless communication network for end-to-end operation management and model iteration.

2. The grab bucket drainage device according to claim 1, characterized in that, In the non-air replenishment mode, the airbag pre-inflation pressure is 105kPa~120kPa. After inflation, the air passage is sealed. The airbag can naturally expand as the external water pressure decreases during the lifting process of the grab bucket. The residual water in the squeeze bucket is discharged through the drainage channel without additional energy consumption. In the AI ​​adaptive mode of air replenishment, the AI ​​inference engine of the edge computing unit can adaptively output the optimal air replenishment pressure and air replenishment timing based on real-time water depth, mud-water mixing density, airbag internal pressure, and target drainage volume parameters using a deep reinforcement learning algorithm. The air replenishment pressure is 1.1 to 1.3 times the initial inflation pressure, and the air replenishment process is executed in 1 to 3 dynamic stages.

3. The grab bucket drainage device according to claim 2, characterized in that, The edge computing unit's built-in AI inference engine integrates three types of functional modules: (1) The working condition adaptive classification module adopts a CNN and GRU, temporal convolutional network architecture to classify the current working condition, and the working condition classification accuracy is ≥95%; (2) The time series prediction module adopts the LSTM, GRU and Transformer time series architecture to predict the drainage trend of the next 10 seconds based on the drainage data of the previous 30 seconds, with a prediction accuracy of ≥92%; (3) Anomaly detection module: using isolated forest, single-class support vector machine, local anomaly factor and its variants equipment health predictive maintenance model, autoencoder architecture to analyze operating data to achieve early warning of faults.

4. The grab bucket drainage device according to claim 1, characterized in that, The cloud-based digital twin subsystem has the following functions: (1) Real-time three-dimensional digital twin mapping based on the actual geometric parameters of the grab bucket is used to realize the visual monitoring of the operation; (2) The AI ​​model OTA remote update function can continuously iterate and optimize the control strategy based on the full amount of operation data; (3) Predictive maintenance reminder function; (4) Automatic generation function of compliance report for water discharge in the bucket.

5. The grab bucket drainage device according to claim 1, characterized in that, The monitoring sensors of the intelligent sensing subsystem include an airbag internal pressure sensor, a water depth / external pressure sensor, a mud-water density sensor, a drainage flow sensor, a six-axis attitude IMU sensor, and a temperature sensor, which can realize full-dimensional perception of the grab bucket's operating status, drainage status, and equipment health status.

6. The grab bucket drainage device according to claim 1, characterized in that, The deformable airbag is made of highly elastic, wear-resistant, and corrosion-resistant special rubber or composite material, and has a U-shaped, ring-shaped, or multi-chamber structure. The outer surface is coated with polytetrafluoroethylene or silicone anti-fouling coating.

7. The grab bucket drainage device according to claim 1, characterized in that, The drainage channel is an inclined or conical drainage hole with a diameter ranging from 80mm to 150mm. The drainage hole has a replaceable filter screen with a mesh size of 1mm to 10mm, which can be adjusted according to the particle size of the mud and sand being processed.

8. A grab bucket drainage method based on water pressure-volume linkage and intelligent sensing control, employing the device described in any one of claims 1 to 7, characterized in that, Includes the following steps: S1 system initialization: After the sensor self-test is completed, the airbag is filled with gas at a preset initial pressure, and the non-gas replenishment mode or the gas replenishment AI adaptive mode is selected according to the working conditions. S2 Excavation Operation: The grab bucket is launched into the water to excavate mud. External water pressure compresses the airbag to its minimum volume, and the intelligent sensing subsystem continuously collects all-dimensional operation data. S3 Drainage Trigger: After the six-axis attitude IMU sensor detects the lifting action of the grab bucket, the drainage control process is automatically triggered; S4 Drainage Execution: Drainage is completed according to the selected operating mode. After drainage is completed, the system automatically records the operation data and uploads it to the cloud digital twin subsystem. S5 Reset Cycle: After the grab bucket unloads mud, the airbag automatically deflates and resets, ready to enter the next working cycle.

9. The grab bucket drainage method according to claim 8, characterized in that, The drainage execution logic in step S4 is set up for two different operating modes: (1) When the non-air replenishment mode is selected, the airbag expands naturally by squeezing the water out during the lifting process due to the pressure difference between the inside and outside. The edge computing unit continuously collects sensor data and performs multi-source data fusion to verify the drainage status through extended Kalman filter algorithm. (2) When the AI ​​adaptive mode for air replenishment is selected, the edge computing unit calls the deep reinforcement learning strategy network to output the air replenishment strategy. The programmable logic controller controls the air pump to replenish air in stages according to the preset time sequence. After each stage of air replenishment is completed, the pressure sensor provides feedback verification. The edge computing unit corrects the subsequent air replenishment parameters in real time.

10. The grab bucket drainage method according to claim 8, characterized in that, The cloud-based digital twin subsystem regularly optimizes and trains the AI ​​inference model at the edge based on full historical operation data. After training, updates are distributed to the edge computing unit via OTA to achieve continuous iteration of control strategies.

Images