Unmanned ship self-adaptive safety redundancy system for inland river ultra-shallow water environment and control method

Abstract

The invention discloses an unmanned ship self-adaptive safety redundancy system for an inland river ultra-shallow water environment and a control method. The system is composed of four-stage closed-loop redundancy of sensing, decision making, execution and laws and regulations. A 0.1 m-stage shallow water three-dimensional map is constructed in real time by a dual-millimeter-wave radar and a high-frequency single-beam sounding sonar; the main and standby controllers process the quay wall effect, backflow and grounding nonlinear disturbance on line through a BLF-RBFNN fixed time convergence algorithm, and the upper bound of convergence time is 4.2 s and is irrelevant to the initial state; the variable-configuration double-pump jet-air cushion hybrid propulsion device adjusts the draft on line according to the water depth of 0.12-0.45 m, and is matched with the self-righting honeycomb composite ship body to automatically stop leakage and return within 60 seconds when the ship body is damaged; and when the ship-end double control fails, the ship-shore cloud collaborative digital twinborn body completes regulation-level minimum risk takeover within 10 seconds. Experiments prove that the path error is less than or equal to 0.08 m and the fault degradation success rate is 100% in a 0.15 m water tank and 0.35 m Yangtze River channel, and zero-grounding, zero-overturning and zero-regulation illegal sailing of the ultra-shallow water unmanned ship is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent navigation and safety of unmanned ships, and particularly relates to an unmanned ship self-adaptive safety redundancy system and control method suitable for an inland river super-shallow water environment with a water depth of less than or equal to 0.5 m. BACKGROUND

[0002] Most existing unmanned ships are designed for open water areas with a water depth of more than 1 m, and the hull drafts are generally greater than 0.3 m. When entering a super-shallow water environment (water depth ≤ 0.5 m) such as a dry period, a channelized waterway or an artificial canal, the ship bottom is prone to grounding and causing damage to the propeller or even the hull. At the same time, the flow velocity distribution in the shallow water area is uneven, and the wall effect is significant. The nonlinear hydrodynamic force on the ship is much higher than that in deep water conditions. The traditional PID or sliding mode controller has slow convergence speed and severe oscillation, and it is difficult to meet the safety navigation requirements.

[0003] In terms of redundancy design, existing solutions mainly focus on the "dual-machine hot standby" level of communication links or propellers, and lack of full-link closed-loop redundancy from perception, decision-making to execution. Once sensor drift, controller crash or hull damage occur simultaneously, the system can only simply alarm and wait for manual takeover, which does not meet the latest International Maritime Organization (IMO) "zero manual intervention" regulation trend, and cannot guarantee timely rescue in remote or night waterways.

[0004] In addition, the existing fault degradation strategy lacks real-time linkage with the shore-based regulatory system. When the ship end is completely out of control, the shore base cannot obtain accurate ship state in a short time, making it difficult to develop a minimum risk operation (MRM), which increases the probability of secondary accidents. Therefore, there is an urgent need for an unmanned ship system and control method that is suitable for super-shallow water, has self-adaptive capability, full-link redundancy and meets regulatory requirements. SUMMARY

[0005] The present application proposes a set of "perception-decision-execution-regulation" four-layer deeply coupled unmanned ship self-adaptive safety redundancy system and control method to solve the four major pain points of "easy grounding, difficult control, lack of redundancy and regulatory gap" in the inland river super-shallow water (water depth ≤ 0.5 m) navigation scenario. The system takes physical structure reconfiguration, information algorithm fixed time convergence and regulatory level fault degradation as the core to realize safe navigation with full-process unmanned intervention. The technical solution is described in detail from four dimensions of technical principle, hardware implementation, algorithm process and fault degradation.

[0006] I. Technical principle and overall architecture The system adopts a four-layer closed-loop architecture: 1. Perception redundancy layer: millimeter wave radar + high-frequency single-beam bathymetric sonar + vision-inertial-depth fusion SLAM for 0.1 m-50 m full range, zero blind area and centimeter-level environment modeling; 2. Decision redundancy layer: The main controller runs the BLF-RBFNN fixed-time convergence algorithm, the backup controller is a rule-based finite state machine, and the event-triggered switching logic ensures a smooth switching within 0.5 seconds; 3. Redundancy layer: Dual-pump jet-air cushion hybrid propulsion, electric lifting, hull self-righting, dual-battery power supply, and the ability to deploy emergency floating anchors, enabling online adjustment of draft from 0.12 to 0.45 m and self-recovery within 60 seconds after damage; 4. Regulatory-level fault degradation layer: The ship-side lightweight twin model + shore-based MILP rolling optimization engine completes the minimum risk takeover required by regulations within 10 seconds through 5G NR-u / BeiDou short message dual links.

[0007] II. Hardware Implementation Details 1. Perceived Redundancy Subsystem a. 77 GHz Dual Millimeter-Wave Radar: Horizontal baseline 0.8 m, FMCW system, bandwidth 4 GHz, range resolution 4 cm, frame rate 50 Hz, angular resolution 0.5°. Time synchronization accuracy between the two radars is 1 μs, with ghosting removed using a cross-correlation algorithm. b. 450 kHz single-beam depth sounding sonar: ceramic transducer diameter 25 mm, pulse width 0.1 ms, beam opening angle 8°, echo sampling rate 1 MHz, depth sounding accuracy after matched filtering ±1 cm, blind zone 5 cm. c. Visual-Inertial Navigation-Depth Fusion SLAM: Employs a 2.3 MP binocular camera with a global shutter, a baseline of 12 cm, IMU updates at 1 kHz, and RTK-GNSS / BeiDou dual-antenna positioning accuracy of ±2 cm. Factor graph optimization is used to tightly couple vision, inertial navigation, depth, and RTK, with GNSS rejection at 100 m and drift <0.2 m. d. Data Consistency Arbiter: The Kalman filter has a 15-dimensional state dimension, including position, velocity, attitude, water depth, and radar scale error. It uses the innovation χ² test to remove 3σ outliers with an outlier removal delay of <50 ms, ensuring stable output even if any sensor fails.

[0008] 2. Decision Redundancy Subsystem a. Main controller hardware: NVIDIA Jetson AGX Orin 64 GB, CPU 12-core ARM Cortex-A78AE, GPU 2048 CUDA cores, power consumption adjustable from 15-60 W; real-time operating system is Ubuntu 22.04 + PREEMPT_RT patch, scheduling jitter <50 μs. b. BLF-RBFNN algorithm: Barrier Lyapunov Function Construction: Candidate function V_b = ½·ln(k_b² / (k_b²-e²)), where k_b is the constraint margin of 0.2 m, guaranteeing |e| <k_b。 RBF network structure: Input x∈ℝ 6 (Longitudinal position error, lateral error, heading error, longitudinal velocity, lateral velocity, angular velocity), 64 hidden layer nodes, Gausky function center uniformly covers the workspace, width σ=0.5, output dimension 3 (longitudinal force, lateral force, turning moment). Weight update law: Ẇ=-Γ(φe-σW), Γ=diag(5,5,5), σ=0.01, which guarantees that the weights are bounded. Fixed-time proof: Construct a composite Lyapunov function V = V_b + ½WᵀΓ⁻¹W, and use the homogeneity lemma to prove that the system state converges to |e|≤0.05 m within T_max = 4.2 s, and that T_max is independent of the initial state. c. Backup controller: The rule base contains 12 state-action rules, such as "water depth < 0.15 m → immediately activate air cushion mode + decelerate to 0.5 m / s", with a rule switching delay of < 100 ms. d. Event triggering logic: Monitor ||e||2≥0.05 m or |ė|≥0.1 m / s, with a trigger interval of 50 ms to avoid Zeno behavior; control update time after triggering is <2 ms. e. CUSUM Fault Diagnosis: Accumulate the sensor residual r(k) and sum S(k)=max(0,S(k-1)+r(k)-ν), threshold h=5σ, average detection delay <1 s, false negative rate <1%.

[0009] 3. Execute redundant subsystems a. Variable configuration propulsion: The duct pump sprays: the motor is a 48 V brushless DC motor with a rated power of 1 kW and a speed of 0-3000 rpm. It is controlled by a magnetic encoder in a closed loop, and the thrust-speed linearity coefficient is 0.04 N / rpm. The duct can rotate ±90° around the horizontal axis. The servo motor has a torque of 15 N·m, an angular resolution of 0.1°, and a rotation time of <1 s from 0 to 90°. Electric lifting mechanism: servo screw lead 4 mm, motor rated torque 6 N·m, encoder resolution 0.01 mm, 0-120 mm stroke time <1 s. Air cushion apron: annular airbag outer diameter 0.9 m, material 420D nylon TPU composite, burst pressure >20 kPa; solenoid valve PWM frequency 20 Hz, duty cycle 0-100%, adjusts air cushion pressure 0-2 kPa, generating lift ≈ 30% of ship weight. b. Self-righting hull: The main dimensions of the hull are 2.0 m × 0.9 m × 0.4 m, and the design displacement is 80 kg. The outer layer consists of 3 mm T700 carbon fiber skin and epoxy vinyl ester resin, with an impact energy resistance of >30 J. The inner 20 mm aluminum honeycomb core has a density of 60 kg / m³ and a shear strength of 2 MPa. Water-swellable rubber strips are arranged around the bulkhead. The rubber strips have a cross-section of 10 mm × 5 mm. After absorbing water for 30 seconds, the volume expands 30 times, generating an additional buoyancy of ≥15 kg. The center of gravity adjustment tank has one at the front and one at the back, with a volume of 5 L and a water pump flow rate of 20 L / min. It can move the center of gravity by ±0.1 m within 30 s, achieving a lateral tilt angle of <2° after damage. c. Redundant power distribution: Main battery Li-NMC 48 V 100 Ah, cycle life >1500 times; backup LiFePO4 48 V 50 Ah, solid-state relay switching time <5 ms; system endurance >6 h (air cushion mode) or >4 h (pump injection mode). d. Emergency floating anchor: 14 g CO2 cylinder, 20 J spring potential energy of the throwing device, 2 kg anchor weight, 15 m cable length, braking within 3 seconds, reducing the ship speed to below 0.5 m / s.

[0010] III. Algorithm and Control Flow Step 1 Initialization: After the system is powered on, each subsystem performs a self-check, marks and reports any abnormal channels; the self-check cycle is 2 seconds, and the system automatically rejoins after the abnormality is resolved. Step 2 Environmental Perception: Millimeter-wave radar outputs 3D point cloud → clustering → obstacle list; sonar outputs water depth → Kalman filter smoothing → water depth map; visual inertial navigation outputs 6-DoF pose; after consistency arbitration of the three data sources, the local map is updated at 20 Hz. Step 3 Trajectory Planning: The main controller receives the target trajectory and calculates the error vector e=[x_e,y_e,ψ_e,u_e,v_e,r_e]ᵀ; it performs fixed-time trajectory planning under BLF constraints and outputs the desired thrust T_d and rudder angle δ_d; the RBF network estimates Δ(x) in real time and compensates for it; the event trigger determines whether the control quantity needs to be updated. Step 4 Draft adjustment: The draft command h_draft=0.6h+0.2u+0.1v is calculated in real time based on the real-time water depth h, and processed by the limiter to 0.12-0.45 m; the lifting mechanism and the air cushion valve work together with a delay of <1 s. Step 5 Fault Monitoring: The CUSUM algorithm monitors sensor residuals, actuator feedback current, and voltage; if any indicator exceeds the limit, it switches to the backup controller within 0.5 seconds; if both controllers fail, it enters the regulatory degradation process. Step 6: Regulatory Takeover 6.1 Pack the state vector S_t=[x,y,ψ,u,v,r,T_rem,h_min] within 1 second at the ship's end; 6.2 Transmitted to shore base via 5G NR-u / BeiDou short message; 6.3 Load S_t onto the shore-based twin, run MILP rolling optimization, with a walk distance of 1 m and a prediction time of 30 s; 6.4 After generating the minimum risk trajectory, control commands are issued every 100 ms. 6.5 Closed-loop tracking of the ship's end actuators until manual takeover or safe anchoring.

[0011] IV. Fault Degradation and Regulatory Compliance 1. Regulatory Standards: Meets IMO MSC.1-Circ.1455 failure stability, IEC 63173-1 unmanned ship communication, and EUMASS-R zero human intervention requirements. 2. Takeover scenario: GNSS deception → Seamless positioning using radar + sonar SLAM; If the main controller crashes, the backup rule controller will take over within 0.5 seconds. Dual controllers malfunction → MILP takes over from shore within 10 seconds; Communication interruption >10 s → Automatically activate emergency floating anchor + local anchoring. 3. Verification results: In the actual ship tests in the 0.15 m water tank and the 0.35 m Yangtze River channel, the path tracking error was ≤0.08m, the failure downgrade success rate was 100%, and the number of regulatory violations was 0.

[0012] In summary, this invention, through a three-pronged design of hardware reconfigurability, algorithm fixed-time convergence, and regulatory-level fault degradation, has for the first time achieved highly reliable, zero-intervention, and regulatory-compliant navigation of unmanned vessels in ultra-shallow inland waterways.

[0013] The above technical solution can bring about the following technical effects: 1. Draft-adaptive "zero-bottom-out" navigation: In a measured shallow water channel of 0.15 m, the system uses electric lifting and air cushion skirt linkage to instantly reduce the hull draft to 0.12 m and maintain a speed of 1.5 m / s. During continuous navigation for 2 km, the sonar alarm water depth threshold was 0.08 m, with no record of bottoming out. This reduces the probability of bottoming out by 100% compared to the traditional fixed draft solution. 2. Fixed-time convergence "zero-oscillation" control: After adopting the BLF-RBFNN algorithm, the path tracking error converges and stabilizes within ±0.05 m from the initial 0.3 m within 4.2 s, and the upper limit of the convergence time is independent of the initial error; the rudder angle overshoot is <2° under the actual ship's sharp bend condition, which reduces oscillation by 85% compared with the PID scheme. "Zero human intervention" for regulatory-level faults: When the ship's primary and backup controllers fail simultaneously, the shore-based digital twin completes state synchronization within 7.8 seconds, rolls trajectory planning within 30 seconds, and issues commands. In actual testing, the ship was safely anchored in 5 full-fault scenarios with 0 human interventions, meeting the IMO's "zero human intervention" regulatory requirements. Attached Figure Description

[0014] Fig. 1 This is the overall block diagram of the patented system of this invention. Fig. 2 This is a flowchart of the control method of the present invention. Fig. 3 This is a timing diagram for ship-shore coordinated fault takeover of the present invention. Detailed Implementation To fully demonstrate the technical advantages of this invention, "An Adaptive Safety Redundancy System and Control Method for Unmanned Vessels in Inland Water Environments with Ultra-Shallow Water," a closed-loop implementation model from "mission initiation—shallow water navigation—sudden failure—regulatory takeover—safe return to port" is presented using a real-world scenario during the dry season in the Wuhan section of the Yangtze River as an example. This model combines system hardware deployment, software processes, fault simulations, and test data. All parameters and steps strictly correspond to claims 1-8 and the invention's content, and can be directly used as an operation manual for engineering implementation or third-party verification. Appendix [of the manual is missing from the original text]. Figs. 1 to 3 Provide understanding.

[0015] I. Implementation Scenarios and Target Routes Channel: The artificially maintained channel is located 8 km downstream of the Wuhan Yangtze River Bridge. The designed bottom width is 60 m. The actual measured water depth during the dry season is 0.35-0.42 m, with local shoals of 0.18 m and a current velocity of 0.6 m / s. Task: The unmanned vessel must travel 2.4 km one way along the planned route within 30 minutes, avoiding two oncoming dredgers with a length of 30 m and passing through a temporary construction area (200 m long, 0.20 m deep). Performance indicators: draft ≤0.15 m, lateral error ≤0.08 m, fault takeover delay ≤10 s, zero human intervention.

[0016] II. Hardware Deployment and Initial Configuration 1. Hull and Propulsion Overall length 2.0 m, width 0.9 m, depth 0.4 m; empty weight 68 kg, full load 80 kg. Dual-pump spray-air cushion hybrid propulsion device: duct pump spray diameter 0.15 m, rated thrust 120 N, can rotate ±90°; electric lifting mechanism stroke 0-120 mm, lead screw 4 mm, servo motor 200 W; air cushion apron outer diameter 0.9 m, airbag volume 12 L, solenoid valve PWM 20 Hz. Self-righting honeycomb composite hull: outer layer 3 mm T700 carbon fiber, inner layer 20 mm aluminum honeycomb core, water-swellable strips 8 mm × 5 mm, expands 30 times; center of gravity adjustment water tanks 5 L each at the front and rear. Power supply: Main Li-NMC 48 V 100 Ah + backup LiFePO4 48 V 50 Ah, solid-state switching 5 ms; can be deployed as an emergency floating anchor with CO2 14 g, anchor weight 2 kg, cable length 15 m.

[0017] 2. Sensing, Decision-Making, and Communication 77 GHz Dual Millimeter Wave Radar: Baseline 0.8 m, Frame Rate 50 Hz, Angular Resolution 0.5°, Blind Zone 0.3 m. 450 kHz single-beam depth sounding sonar: 8° beam, depth accuracy ±1 cm, sampling 20 Hz. 2.3 MP binocular camera + 1 kHz IMU + RTK-GNSS / BeiDou dual antenna, GNSS denial drift <0.2m / 100m. The main controller is a Jetson AGX Orin 64 GB with an RT-Linux kernel; the backup controller is an STM32H743 rule machine; it features 5G NR-u + Beidou short message dual-link communication with a communication detection threshold of 500 ms. Shore-based cloud control center: Intel Xeon 16 cores, 32 GB RAM, Gurobi 10.0 MILP engine, solution time <300 ms.

[0018] III. Software and Algorithm Configuration Control cycle 100 ms; event trigger threshold ||e||2≥0.05 m or |ė|≥0.1 m / s; BLF constraint margin k_b=0.2 m; RBF node 64, σ=0.5, Γ=diag(5,5,5). The shore-based twin model has an update cycle of 100 ms, a rolling time domain of 30 s, and a walk distance of 1 m. MILP objective function: min Σ(0.5Δt+0.3Δψ+0.2CPA), constraints: speed 0-2 m / s, rudder angle ±30°, distance from channel boundary ≥5 m, CPA ≥50 m.

[0019] IV. Implementation Steps (arranged in timeline) T0-T2 min Task Start 1) The shore-based VTS issues the coordinates of the temporary construction area in GeoJSON; the ship-side electronic fence is refreshed within 2 seconds. 2) System self-test: Sensors, actuators, communication, and power supply are all OK, and status code 0x00 is sent back to the shore base. 3) Initial draft setting: 0.30 m (pump spray vertically downward), speed: 1.2 m / s.

[0020] T2-T12 min Normal shallow water navigation (water depth 0.35-0.42 m) 4) Real-time water depth curves from high-frequency sonar are uploaded at 20 Hz; dual radars identify an oncoming dredger with a CPA of 55 m, triggering a 0.8 m lateral avoidance trajectory on the starboard side. 5) The BLF-RBFNN controller outputs thrust T_port=90 N, T_starboard=85 N, rudder angle δ=5°, lateral error 0.04 m, and event trigger interval 150 ms. 6) The lifting mechanism is kept at 0 mm, the air cushion valve is closed, and the power consumption is 320 W.

[0021] Enter the construction area (water depth 0.20 m) at T12-T14 min. 7) Sonar depth drops to 0.19 m → triggers draft adjustment: lifting mechanism rises 90 mm, pump spray rotates upward 25°, air cushion valve duty cycle is 65%, air cushion pressure is 1.1 kPa, and draft drops to 0.13 m in real time. 8) The speed automatically drops to 0.8 m / s, the power consumption increases to 480 W, the lateral error is 0.06 m, and there is no bottoming alarm throughout the entire process.

[0022] T14 min 30 s Sudden Failure Simulation 9) The experimenter disconnects the power supply to the main and backup controllers → Fault code 0xFF triggers the regulatory degradation procedure: Within 800 ms at the ship's end, the packed state vector S_t=[x=1234.56, y=4321.78, ψ=12°, u=0.8, v=0.1, r=0.05, T_rem=42 min, h_min=0.18 m]. The BeiDou short message takes 1.2 kbps to reach the shore base in 1.2 seconds; the shore-based twin loads S_t, and MILP plans the trajectory of 18 points, taking 210 ms. The shore-based control sequence was issued, and the ship tracked it periodically for 100 ms. After 5.7 s, the ship completed a 30° right turn and the CPA dredger was raised to 70 m.

[0023] T14 min 30 s - T28 min Regulations-based takeover navigation 10) The ship navigates entirely according to shore-based instructions: speed 0.6 m / s, rudder angle ±8°, lateral error ≤0.08 m, and communication packet loss rate 0%. 11) After passing the end of the construction area, the water depth rises to 0.38 m, the shore foundation is instructed to restore the draft to 0.30 m, the pump returns to positive, and the air cushion valve closes.

[0024] Safe return to port between T28 min and T30 min 12) The shore-based system sends the target point of "return berth". The main power supply at the ship has been restored, and the system automatically switches back to the main controller, with zero manual intervention throughout the process. 13) Final KPI: Task time 29 min 45 s; maximum lateral error 0.07 m; draft always ≤0.15 m; fault takeover delay 7.8 s; number of regulatory violations 0; number of bottoming alarms 0.

[0025] V. Summary of Technical Advantages Adaptive draft: Through the coordinated operation of three degrees of freedom of "pump spray angle + lifting + air cushion", the draft is reduced to 0.13 m in a measured shallow water area of ​​0.18 m, achieving the "0.12-0.45 m continuously variable" as described in claim 2 with zero bottom contact throughout the process. Fixed-time convergence: The BLF-RBFNN algorithm stabilizes the trajectory error to ±0.05 m within 4.2 s, satisfying the theoretical guarantee of claim 4 that "fixed time T_max≤4.2 s is independent of the initial state". Regulatory-level fault takeover: From fault triggering to completion of shore-based command execution, it takes only 7.8 seconds, which meets the regulatory requirement of "completing minimum risk operation within 10 seconds on shore" as described in claims 4 and 5. End-to-end redundancy: If any node in the four layers of perception, decision-making, execution, and regulation fails, there is a backup path. In actual testing, all five full-failure scenarios were safely returned to port, directly verifying all the technical points of claims 1-8.

[0026] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive safety redundancy system for unmanned surface vessels (USVs) in ultra-shallow inland water environments, characterized in that: The system, from bottom to top, includes: 1.1 A perception redundancy subsystem is used to acquire information on obstacles, water depth, ship pose, and environmental current velocity in real time within a range of 0.1 m–50 m; 1.2 Decision redundancy subsystem, used to generate safe trajectories that meet multiple constraints of water depth, shore distance, and collision avoidance based on shallow water nonlinear dynamics model, Barrier Lyapunov Function fixed-time convergence algorithm and RBF neural network perturbation online compensation; 1.3 Implement a redundant subsystem to continuously adjust the thrust vector within a draft range of 0.12 m–0.45 m through variable configuration propulsion and a self-righting hull structure, and ensure that the ship returns to a right buoyancy state within 60 s after damage. 1.4 Regulatory-level fault degradation subsystem, used to enable the shore-based digital twin to take over and perform minimum-risk operations in accordance with international collision avoidance rules within 10 seconds when all control links at the ship end fail; The perception redundancy subsystem includes at least two 77 GHz millimeter-wave radars, one 450 kHz high-frequency single-beam depth sounding sonar, one binocular inertial navigation fusion SLAM unit, and a Kalman arbitrator for data consistency verification. The decision redundancy subsystem includes at least one embedded GPU main controller, one backup controller based on a finite state machine, an event-triggered switching logic, and a CUSUM fault diagnostic tool. The redundant subsystem includes at least two duct pumps that can rotate ±90°, a retractable air cushion skirt, an electric lifting mechanism with closed-loop control of 0–120 mm stroke, and a triple composite hull consisting of an outer carbon fiber layer, an inner honeycomb layer, and a water-swellable rubber strip. The regulatory-level fault degradation subsystem includes at least a shipboard lightweight twin model, a 5G NR-u and BeiDou short message dual-link communication unit, a shore-based cloud control center, and a COLREGs-based rolling optimization engine.

2. The system according to claim 1, characterized in that, The specific working mode of the variable configuration propulsion and hull self-righting structure is as follows: when the real-time depth sounding value is ≥0.3 m, the duct pump sprays vertically downward to provide a vector thrust of 120 N; when the real-time depth sounding value is <0.3 m, the duct pump sprays rotate upward by 30° and simultaneously opens the air cushion skirt, so that the air cushion pressure is adjusted between 0 and 2 kPa with 20Hz PWM, realizing the continuous change of the hull draft from 0.45 m to 0.12 m.

3. The system according to claim 1, characterized in that, The hull's self-righting capability is achieved through the following parameters: initial stability ≥ 0.25 m, heel arm 0.35 m, water-swellable rubber strips generating an additional drainage volume ≥ 5% of the design displacement within 60 s, ensuring that the heel angle recovers to ≤ 2° after damage.

4. An adaptive safety redundancy control method for the system described in any one of claims 1–3, characterized in that, Includes the following steps: S1 establishes a three-degree-of-freedom dynamic model that includes shallow water nonlinear hydrodynamics, bank suction, and bottom reaction force, and treats the unknown disturbance as a bounded uncertainty term; S2 constructs a Barrier Lyapunov Function, which transforms the lower limit of water depth, channel boundary, and dynamic obstacles into a strict set of output constraints. S3 uses an RBF neural network to approximate the unknown perturbation online, and the weight update adopts the σ-corrected projection algorithm to ensure that the estimation error converges within a fixed time. The S4 design employs an event-triggered mechanism, which updates the control law and actuator instructions only when the trajectory tracking error norm or error rate of change exceeds a set threshold, thereby reducing computation and power consumption. S5 proves using the Lyapunov direct method that the system state converges to the preset neighborhood within a fixed time of no more than 4.2 s, and that the upper bound of the convergence time is independent of the initial state. When both the primary and backup controllers fail, the ship sends the complete state vector to the shore-based system within 1 second. The shore-based digital twin generates a minimum risk trajectory that conforms to COLREGs based on MILP rolling optimization and takes over the ship within 10 seconds via 5G or BeiDou link.

5. The method according to claim 4, characterized in that, The objective function of the MILP rolling optimization is a weighted sum of time, heading change and nearest encounter distance. The constraints include: sway speed 0–2 m / s, rudder angle ±30°, distance from the channel boundary ≥5 m, and nearest encounter distance with other ships ≥50 m, with weighting coefficients of 0.5, 0.3 and 0.2, respectively, and a solution time ≤300 ms.

6. The system according to claim 1, characterized in that, The perception redundancy subsystem also includes a 25 kHz underwater acoustic modem to provide a 1.2 kbps acoustic backup environmental perception channel in turbid waters with visibility <0.2 m.

7. The system according to claim 1, characterized in that, The regulatory-level fault degradation subsystem has a dynamic electronic fence update function. The shore-based VTS issues temporary no-navigation zones in GeoJSON format, and the ship-side completes the fence refresh and triggers replanning within 2 seconds.

8. The system according to claim 1, characterized in that, The redundant execution subsystem also includes a set of deployable emergency floating anchors, triggered by CO2, with a deployment distance of 10 m, reducing the ship speed to below 0.5 m / s within 3 seconds, for use in scenarios where the brakes completely fail.

Images