A ship bulwark drainage intelligent management system combined with multi-sensor data monitoring

Abstract

The application discloses a kind of ship bulwark drainage intelligent management systems combined with multi-sensor data monitoring, and the application relates to ship intelligent control and automation technical field, comprising: sensor group, dynamic trust chain fusion decision module, multi-modal cross-validation module, fault-tolerant execution and reverse verification module and central controller.The ship bulwark drainage intelligent management systems combined with multi-sensor data monitoring, by constructing dynamic trust chain fusion decision model and multi-modal cross-validation mechanism, the accuracy and timeliness of multi-source sensor data fusion under complex sea conditions are improved;Sensor misjudgment caused by mechanical vibration, bubble interference and the like is effectively inhibited, combined with digital twin preplay and reverse verification closed loop, reduces the mis-triggering rate of drainage instruction, shortens response delay, simultaneously by two-stage screening and majority voting mechanism, the cooperative fault tolerance of sensor cluster is realized, to ensure the continuous and stable operation of system under single-point fault or abnormal interference.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent control and automation of ships, in particular to a ship bulwark drainage intelligent management system combining multi-sensor data monitoring. BACKGROUND

[0002] In the development process of ship bulwark drainage systems, traditional management schemes generally rely on fixed thresholds or single sensor signals to trigger drainage actions. Due to the complexity of the ship's sailing environment, especially when encountering waves, sharp turns, or sudden load changes, the dramatic changes in the ship's attitude will cause the water accumulation state in the bulwark area to exhibit highly dynamic characteristics. For example, the liquid level sensor may produce a false high water level signal due to the inclination of the ship body, and the flow rate sensor may have data jumps due to the vortex interference of the drainage pipe.

[0003] Although the prior art attempts to improve monitoring comprehensiveness by increasing the number of sensors, the processing of multi-source heterogeneous data still uses simple weighted averaging or static logic judgment, lacking a dynamic evaluation mechanism for the credibility of sensor data. When multiple sensors produce contradictory data due to environmental interference, the system cannot adjust the data fusion strategy according to real-time working conditions. For example, in the ship body roll state, the weight of the side wall liquid level sensor susceptible to mechanical vibration is not reduced, or the instantaneous abnormal values of the drainage pipe pressure sensor due to air bubbles are ignored, ultimately leading to delayed, false triggering, or redundant execution of drainage commands. The existing improvement direction is mostly focused on optimizing hardware deployment or execution mechanism response speed, but it does not solve the problem of adaptive multi-sensor collaborative decision-making, making it difficult for the system to meet the management needs of modern intelligent ships in terms of robustness and real-time performance in complex sea conditions. SUMMARY

[0004] (I) Technical problems solved

[0005] To address the shortcomings of the prior art, the present application provides a ship bulwark drainage intelligent management system combining multi-sensor data monitoring, which solves the problem of how to realize dynamic fusion and priority decision of multi-source sensor data to improve the accuracy and timeliness of drainage strategies in complex sea conditions.

[0006] (II) Technical solutions

[0007] To achieve the above purpose, the present application realizes the following technical solutions: a ship bulwark drainage intelligent management system combining multi-sensor data monitoring, comprising:

[0008] a sensor group, a dynamic trust chain fusion decision module, a multi-modal cross-validation module, a fault-tolerant execution and reverse verification module, and a central controller;

[0009] The sensor group includes a liquid level sensor installed on the inner side of the ship's bulwark, a flow rate sensor and a pressure sensor in the drainage pipeline, a ship deck attitude sensor, and an external environment monitoring sensor, each sensor is connected with a dynamic trust chain fusion decision module through a data bus; in the implementation of the sensor group, the liquid level sensor is vertically installed on the inner side of the bulwark near the drainage port of the bulkhead surface, the probe thereof is packaged with corrosion-resistant material, real-time collection of water depth data in the bulwark area is realized, and the data is connected to the data bus through a waterproof cable; the flow rate sensor and the pressure sensor in the drainage pipeline are respectively embedded in the upstream and the bending part of the inner wall of the pipeline, the flow rate sensor measures the water flow rate based on the ultrasonic Doppler effect, the pressure sensor detects the fluid pressure fluctuation in the pipeline through a pressure-sensitive element, and the signals of the two are synchronously transmitted to the dynamic trust chain fusion decision module after being processed by an electromagnetic interference shielding layer; the ship deck attitude sensor is fixed on the deck structure near the center of gravity of the ship, integrates a three-axis accelerometer and a gyroscope, and outputs roll angle, pitch angle and angular acceleration data at a sampling frequency of 100Hz; the external environment monitoring sensor is arranged at the top of the ship mast, including a wind speed and direction instrument and a wave height radar, and continuously collects external wind and wave parameters.

[0010] The dynamic trust chain fusion decision module receives real-time data of the sensor group, dynamically adjusts the confidence weight of each sensor based on the ship motion attitude parameters and environmental parameters; the dynamic trust chain fusion decision module obtains roll angular velocity and pitch angle data in real time through the ship deck attitude sensor, simultaneously receives surge period and wind speed information collected by the external environment monitoring sensor, and establishes a correlation model of ship motion attitude and environmental disturbance.

[0011] The multi-modal cross-validation module performs two-level screening on the data output by the dynamic trust chain fusion decision module through a physical constraint model and a historical behavior database; in the two-level screening process, the physical constraint model verifies the rationality of the data first, and the historical behavior analysis performs secondary filtering on the data that passes the physical verification but has potential abnormalities, finally generates a fusion data package with a verification mark and transmits it to the central controller, and records the conflict data characteristics in the screening process for subsequent sensor health evaluation.

[0012] The fault-tolerant execution and reverse verification module preforms the execution effect of the drainage instruction through a digital twin model, and performs reverse verification with the actual drainage result; in the reverse verification process, the module synchronously updates the parameter calibration coefficient of the digital twin model, dynamically adjusts the fluid resistance coefficient and pump efficiency curve according to the actual drainage effect, and ensures that the model prediction accuracy is continuously optimized with the system running. All verification results and operation records are transmitted to the decision log of the central controller through an encrypted link, providing data support for subsequent strategy optimization.

[0013] The central controller generates a drainage instruction according to the output signals of the dynamic trust chain fusion decision module, the multi-modal cross-validation module and the fault-tolerant execution and reverse verification module, and controls the action of the execution mechanism. The finally generated drainage instruction is transmitted to the execution mechanism through the CAN bus, and operation logs are sent to the ship central monitoring system synchronously, recording instruction time stamp, weight allocation parameter and reverse verification error value, forming a closed-loop control evidence chain.

[0014] Preferably, the dynamic trust chain fusion decision module comprises:

[0015] The environment perception unit is connected with the ship attitude sensor and the environment monitoring sensor, and is used for acquiring the ship roll angular velocity, the trim angle and the surge period in real time.

[0016] The trust chain reconstruction unit is connected with the liquid level sensor, the flow rate sensor and the pressure sensor, and is used for time domain decomposition and frequency domain filtering processing of the sensor signals.

[0017] The weight dynamic allocation engine receives the output signals of the environment perception unit and the filtered data of the trust chain reconstruction unit, and generates dynamic confidence weights of each sensor.

[0018] Preferably, the environment perception unit generates a dynamic attenuation coefficient of the liquid level sensor according to the correlation between the ship roll angular velocity and the surge period, and specifically:

[0019] When the ship roll angular velocity exceeds a preset threshold value, the dynamic attenuation coefficient of the liquid level sensor is reduced according to a linear function.

[0020] When the surge period coincides with the inherent rolling frequency of the ship, the credibility threshold of the flow rate sensor is increased to a preset safety value.

[0021] Preferably, the trust chain reconstruction unit comprises:

[0022] The liquid level signal processing subunit separates the mechanical vibration interference component in the liquid level sensor signal by using a wavelet packet decomposition algorithm.

[0023] The pressure signal processing subunit identifies the transient abnormal pulse in the pressure sensor data by using a sliding window entropy value calculation method, and the transient abnormal pulse is caused by bubble breakage in the drainage pipeline.

[0024] In the implementation of the trust chain reconstruction unit, the liquid level signal processing subunit carries out multi-scale wavelet packet decomposition on the original signal of the liquid level sensor installed on the inboard side of the bulwark. First, the collected time domain signal is divided into 8 sub-bands. Through energy spectrum analysis, the interference components overlapping with the mechanical vibration frequency of the ship are identified. The signals of the third to fifth sub-bands are extracted as the vibration noise base. The corresponding frequency band components are removed from the original data, and the low-frequency trend signal reflecting the true liquid level is retained. For the pressure sensor data processing subunit, a sliding time window is used to intercept a pressure data sequence of 50 consecutive sampling points. The Shannon entropy value of the data in each window is calculated. When the window entropy value suddenly increases by more than twice the standard deviation of the historical mean value, it is determined that a transient abnormal pulse caused by bubble rupture has occurred. The window data is automatically removed and replaced with linear interpolation of the adjacent windows.

[0025] Preferably, the weight dynamic allocation engine performs the following operations under the condition of high-frequency ship body shaking:

[0026] Adjust the dynamic confidence weight of the side wall liquid level sensor to the interval of 0.2-0.3;

[0027] Increase the dynamic confidence weight of the drain pipe pressure sensor to the interval of 0.7-0.8;

[0028] Based on the majority voting result of the majority voting mechanism of at least three sensors in the same monitoring area, the final output value of the dynamic confidence weight is corrected.

[0029] When the weight dynamic allocation engine is implemented under the condition of high-frequency ship body shaking, the ship attitude sensor is first used to monitor the roll angular velocity and pitch angular velocity in real time. When the average roll angular velocity exceeds 8 degrees per second and the fluctuation amplitude is greater than 3 degrees within 5 seconds, it is determined that the high-frequency shaking state is reached.

[0030] The engine then starts the weight dynamic adjustment program, which includes: for the liquid level sensor installed on the side wall, the centrifugal force interference coefficient caused by shaking is calculated according to the relative distance between its installation position and the ship's gravity axis. Combined with the current surge direction, the confidence weight is gradually reduced by 0.12 for every 0.1g acceleration from the initial value of 1.0, and finally limited to the interval of 0.2-0.3. At the same time, based on the data stability index of the drain pipe pressure sensor within the last 3 minutes, the weight of the drain pipe pressure sensor is increased from the base value of 0.5 to the interval of 0.7-0.8. The data consistency verification result of at least two pressure sensors in the adjacent monitoring area is also associated. If the verification is passed, an additional weight gain of 0.05 is added.

[0031] After the weight distribution is completed, the engine calls the historical data of the liquid level sensor, pressure sensor and flow rate sensor cluster in the same monitoring area, calculates the deviation of the current output value of each sensor from the cluster mean, and if the liquid level sensor data deviation exceeds 25% and the pressure sensor cluster data consistency is above 85%, a majority voting mechanism is triggered, including: selecting two of the three or more same type sensors with the most concentrated data as valid input, forcibly overriding the output value of the abnormal sensor, and re-normalizing the corrected dynamic confidence weight.

[0032] Preferably, the multi-modal cross-validation module comprises:

[0033] The first-level screening unit verifies the logical consistency between the liquid level sensor data and the ship draft, surge direction based on a fluid mechanics model;

[0034] The second-level screening unit analyzes the historical behavior patterns of the sensor data through a long short-term memory network to mark frequently abnormal working condition sensitive sensors;

[0035] When the liquid level sensor reports a sudden increase in water level and the flow rate sensor data does not change, the first-level screening unit triggers a delayed response protection mechanism, suspends the drainage instruction output and starts a data review process for three consecutive sampling periods.

[0036] Preferably, the second-level screening unit performs the following operations after detecting the working condition sensitive sensor:

[0037] Call the historical data of other sensors in the same monitoring area to calculate the current true value;

[0038] Use the calculated value to override the abnormal output data of the working condition sensitive sensor;

[0039] Generate a sensor fault diagnosis signal and upload it to the central controller.

[0040] Preferably, the fault-tolerant execution and reverse verification module comprises:

[0041] The virtual sandbox unit simulates the ship body posture change and cabin water level fluctuation curve within 10 seconds after the execution of the drainage instruction through a digital twin model;

[0042] The reverse comparison unit real-time collects the actual flow data of the drainage pipe pressure sensor and calculates the difference with the predicted value of the virtual sandbox unit;

[0043] If the difference exceeds 5% for three consecutive times, it is determined that the actuator is abnormal, the standby drainage pipeline is switched to and an alarm signal is triggered.

[0044] Preferably, the central controller performs the following logic before generating the drainage instruction:

[0045] receiving a dynamic confidence weight distribution result of the dynamic trust chain fusion decision module;

[0046] receiving a two-stage screening pass identification of the multi-modal cross-validation module;

[0047] receiving a rehearsal deviation detection result of the virtual sandbox unit;

[0048] When the above three inputs all meet the preset conditions, a drainage pump start instruction and a valve opening control signal are generated.

[0049] Preferably, the triggering condition of the majority voting mechanism includes:

[0050] The sum of the dynamic confidence weights of the sensors in the same monitoring area is lower than a preset reliability threshold;

[0051] The difference between the data reported by at least two sensors exceeds a maximum allowable error range;

[0052] The system automatically selects the majority voting result as the final decision basis, and marks the low-weight sensors for offline calibration.

[0053] (Three) beneficial effects

[0054] The present application provides a ship bulwark drainage intelligent management system combined with multi-sensor data monitoring. It has the following beneficial effects:

[0055] (I) The ship bulwark drainage intelligent management system combined with multi-sensor data monitoring improves the accuracy and timeliness of multi-source sensor data fusion in complex sea conditions by constructing a dynamic trust chain fusion decision model and a multi-modal cross-validation mechanism. Based on the dynamic weight distribution algorithm of ship attitude and environmental parameters, the sensor misjudgment caused by mechanical vibration, bubble interference, etc. is effectively suppressed. Combined with digital twin rehearsal and reverse verification closed loop, the drainage instruction false trigger rate is reduced, the response delay is shortened, and through two-stage screening and majority voting mechanism, the sensor cluster collaborative fault tolerance is realized, ensuring the continuous and stable operation of the system under single point failure or abnormal interference.

[0056] (II) The ship bulwark drainage intelligent management system combined with multi-sensor data monitoring solves the adaptability defects of traditional drainage management scheme in dynamic sea conditions, improves the ship navigation safety and energy efficiency management level, and its adaptive learning ability can optimize the sensor evaluation model and decision threshold through historical data, reducing the frequency of manual maintenance. Modular design is compatible with various ship types and drainage system architecture, providing technical support for the automatic upgrade of intelligent ships. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 It is the overall framework diagram of the present application;

[0058] Figure 2 The control logic timing diagram of the present application. DETAILED DESCRIPTION

[0059] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0060] Please refer to Figure 1 and Figure 2 The present application provides a technical solution: a ship bulwark drainage intelligent management system combined with multi-sensor data monitoring, comprising:

[0061] a sensor group, a dynamic trust chain fusion decision module, a multi-modal cross-validation module, a fault-tolerant execution and reverse verification module, and a central controller;

[0062] The sensor group includes a liquid level sensor installed on the inner side of the ship bulwark, a flow rate sensor and a pressure sensor in the drainage pipeline, a ship deck attitude sensor, and an external environment monitoring sensor. Each sensor is connected to the dynamic trust chain fusion decision module through a data bus. In the implementation of the sensor group, the liquid level sensor is vertically installed on the inner side of the bulwark near the drainage port of the bulkhead surface. The probe is packaged with corrosion-resistant materials, which can collect real-time water depth data in the bulwark area and access the data bus through waterproof cables. The flow rate sensor and the pressure sensor are embedded in the upstream and bending parts of the inner wall of the pipeline, respectively. The flow rate sensor measures the water flow rate based on the ultrasonic Doppler effect, and the pressure sensor detects the fluid pressure fluctuation in the pipeline through a pressure-sensitive element. Both signals are transmitted synchronously to the dynamic trust chain fusion decision module after being processed by an anti-electromagnetic interference shielding layer. The ship deck attitude sensor is fixed on the deck structure near the center of gravity of the ship, integrating a three-axis accelerometer and a gyroscope to output roll angle, pitch angle, and angular acceleration data at a sampling frequency of 100 Hz. The external environment monitoring sensor is arranged at the top of the ship mast, including a wind speed and direction instrument and a wave height radar, which continuously collects external wind and wave parameters.

[0063] Each sensor data is transmitted through an industrial-grade CAN bus. The bus controller has a built-in data verification and conflict arbitration mechanism to ensure real-time synchronization of multi-source heterogeneous data. After receiving the data, the dynamic trust chain fusion decision module first performs sliding mean filtering on the liquid level sensor signal to eliminate transient peak noise caused by ship body pitching. Then, the baseline drift correction is performed on the pressure sensor data, and the initial weight coefficient of each sensor is dynamically calculated in combination with the angular acceleration value output by the deck attitude sensor to form the original data preprocessing link.

[0064] The dynamic trust chain fusion decision module receives real-time data of the sensor group, dynamically adjusts the confidence weight of each sensor based on the ship motion attitude parameters and environmental parameters; the dynamic trust chain fusion decision module obtains the roll angular velocity and pitch angle data in real time through the ship deck attitude sensor, simultaneously receives the surge period and wind speed information collected by the external environment monitoring sensor, and establishes a correlation model of ship motion attitude and environmental interference. When the roll angular velocity is detected to exceed the preset threshold, the module automatically triggers the credibility attenuation mechanism of the liquid level sensor, including: calculating the vibration interference intensity based on the angular acceleration value, performing time domain sliding window analysis on the original data of the liquid level sensor, if the signal fluctuation amplitude in the continuous three windows is positively correlated with the ship rolling frequency, it is determined that the false water level signal caused by mechanical vibration, and the confidence weight is gradually reduced from the initial value 1.0 to 0.3 in linear proportion; at the same time, combined with the surge period and the baseline pressure characteristics of the drainage pipeline pressure sensor, if the current surge frequency is close to the inherent resonance frequency of the ship drainage system, the decision priority of the pressure sensor data is improved, and the confidence weight is improved to above 0.7.

[0065] For the flow rate sensor data, the module adopts frequency domain band pass filtering to eliminate the pipeline vortex noise caused by the hull sway, and dynamically corrects the confidence weight coefficient through the spatial consistency verification of the flow rate data of adjacent monitoring points. When multiple sensor data conflict, the module calls the historical behavior database of the same area sensor cluster, analyzes the abnormal occurrence probability of each sensor under the current working condition, preferentially adopts the data of the high confidence weight sensor, and triggers the majority voting mechanism, including: if the data of at least two sensors of the same type are consistent within the preset error range, the abnormal output value of the low weight sensor is covered, and the finally fused drainage state evaluation result is generated.

[0066] The multi-modal cross-validation module performs two-level screening on the data output by the dynamic trust chain fusion decision module through the physical constraint model and the historical behavior database; in the implementation, the multi-modal cross-validation module first performs first-level screening on the data output by the dynamic trust chain fusion decision module through the physical constraint model, including: based on the current draft of the ship, the diameter of the drainage pipeline and the surge direction parameters, a drainage volume-ship attitude balance equation is constructed to real-time check the logical correlation of the liquid level sensor data and the flow rate sensor data. When the liquid level sensor reports a sudden increase in water level but the flow rate sensor shows that the drainage flow has not changed, the theoretical water accumulation volume in the bulwark area is calculated combined with the real-time roll angle of the ship, if the accumulated water volume corresponding to the sudden increase in water level exceeds the stability limit value of the ship body, it is determined that the instantaneous interference data caused by the deck wave, the delay response protection mechanism is triggered, the drainage instruction generation is suspended and the data review process of continuous three sampling periods is started.

[0067] The second level screening is based on the historical behavior database for anomaly tracing, including: analyzing the data fluctuation characteristics of each sensor under the same working condition in the past 30 days through a long short-term memory network, establishing an association atlas of sensor data and ship speed and turning angle, and if the current flow sensor continuously appears data jump exceeding three times the standard deviation of the historical mean under a certain turning angle, it is marked as a working condition sensitive sensor, and the historical data mean of the pressure sensor and the adjacent liquid level sensor in the same monitoring area is automatically called to replace and compensate, and the abnormal data source is marked as a low priority input.

[0068] In the two-level screening process, the physical constraint model verifies the rationality of the data first, and the historical behavior analysis is used for secondary filtering of data that passes the physical verification but has potential anomalies, and finally a fusion data package with verification marks is generated and transmitted to the central controller, while the conflict data characteristics in the screening process are recorded for subsequent sensor health assessment.

[0069] The fault-tolerant execution and reverse verification module preforms the execution effect of the drainage instruction through the digital twin model and performs reverse verification with the actual drainage result; in the implementation of the fault-tolerant execution and reverse verification module, a digital twin model is first constructed based on the ship three-dimensional model, the drainage system structure parameters and the real-time sensor data. The model integrates the ship attitude dynamics equation and the fluid mechanics simulation engine.

[0070] When the central controller generates a drainage pump start instruction, the module inputs the current liquid level, pipeline pressure, valve opening and ship motion parameters into the digital twin model to simulate the water level drop curve in the bulwark area, the drainage pipeline flow change and the ship center of gravity offset in the next 10 seconds.

[0071] If the simulation result shows that the water level drop rate is lower than 95% of the expected value, or the ship roll angle increases by more than the safety threshold, it is determined as a potential abnormal working condition, triggering dynamic strategy backtracking, including: suspending the current instruction execution, re-calling the dynamic trust chain fusion decision module to perform secondary fusion calculation on the sensor data, and generating a revised drainage strategy.

[0072] In the actual drainage process, the module collects pipeline actual flow data through high-precision pressure sensors at a frequency of 20 times per second, compares it with the predicted value of the digital twin model in real time, and calculates the root mean square error of the two in the sliding time window.

[0073] When the error value of three consecutive sampling windows exceeds 5%, it is automatically determined as an actuator failure or sensor failure, the current drainage pipeline control signal is immediately cut off, the standby drainage pipeline is switched to and the sound and light alarm device is activated, and the abnormal data package is marked as a fault case and stored in the historical database.

[0074] In the reverse verification process, the module synchronously updates the parameter calibration coefficient of the digital twin model, dynamically adjusts the fluid resistance coefficient and pump efficiency curve according to the actual drainage effect, and ensures that the model prediction accuracy is continuously optimized with the system running. All verification results and operation records are transmitted to the decision log of the central controller through an encrypted link to provide data support for subsequent strategy optimization.

[0075] The central controller generates drainage instructions according to the output signals of the dynamic trust chain fusion decision module, the multi-modal cross-validation module, and the fault-tolerant execution and reverse verification module, and controls the action of the execution mechanism. In implementation, the central controller first receives the dynamic confidence weight distribution table output by the dynamic trust chain fusion decision module, synchronously acquires the two-stage screening pass identification marked by the multi-modal cross-validation module, and the virtual sandbox pre-play deviation value of the fault-tolerant execution and reverse verification module.

[0076] When the weight of the liquid level sensor in the dynamic confidence weight table is higher than 0.5 and the weight of the pressure sensor is lower than 0.3, the controller calls the historical flow database of the drainage pipeline, calculates the theoretical discharge threshold combined with the current ship draft, and generates the drainage pump start instruction and the opening control signal of the corresponding valve if the physical constraint screening of the multi-modal cross-validation module shows that the liquid level data meets the ship stability requirements and the virtual sandbox pre-play deviation value is less than 3%.

[0077] In the instruction execution phase, the controller monitors the feedback current of the execution mechanism and the valve displacement sensor data in real time, and immediately triggers the gradient load reduction strategy when it detects that the motor current of the drainage pump abnormally rises by more than 15% of the rated value: reduce the pump speed by 5% per second, and gradually reduce the valve opening to prevent pipeline overpressure. If the fault-tolerant execution module finds that the actual drainage flow and the predicted value continuously deviate by more than 3 times, the controller interrupts the current control loop, switches to the backup drainage pipeline, and activates the emergency drainage mode, which prioritizes pressure sensor data-driven decisions while encrypting fault information and storing it in the black box system.

[0078] In all decision-making processes, the controller processes the input signals of each module in parallel through a multi-threaded architecture to ensure that the entire process from data reception to instruction issuance is completed within 200 milliseconds, and to display the drainage strategy decision tree map in real time on the human-machine interaction interface, marking the weight distribution of each sensor and the pass status of the verification link. The final drainage instruction is transmitted to the execution mechanism through the CAN bus, and the operation log is sent to the ship central monitoring system synchronously, recording the instruction timestamp, weight allocation parameters, and reverse verification error value to form a closed-loop control evidence chain.

[0079] The dynamic trust chain fusion decision module includes:

[0080] An environment perception unit is connected with the ship attitude sensor and the environment monitoring sensor, and is configured to acquire the ship roll angular velocity, the trim angle and the surge period in real time.

[0081] A trust chain reconstruction unit is connected with the liquid level sensor, the flow rate sensor and the pressure sensor, and is configured to perform time domain decomposition and frequency domain filtering processing on the sensor signals.

[0082] A weight dynamic allocation engine receives the output signals of the environment perception unit and the filtered data of the trust chain reconstruction unit, and generates the dynamic confidence weight of each sensor.

[0083] It should be further explained that, in the specific implementation process, in the implementation of the dynamic trust chain fusion decision module, the environment perception unit acquires the roll angular velocity and the trim angle data in real time through the ship deck attitude sensor, and simultaneously connects the external environment monitoring sensor to acquire the surge period and the wind speed parameter, and establishes a mapping relationship database of the ship motion state and the environmental disturbance.

[0084] When the roll angular velocity exceeds the preset threshold value, the environment perception unit generates the vibration interference coefficient of the liquid level sensor, the coefficient is positively correlated with the angular acceleration value, and is transmitted to the weight dynamic allocation engine in real time; the trust chain reconstruction unit performs time domain decomposition processing on the liquid level sensor signal, adopts a sliding time window to intercept continuous 10-second original data, separates the vibration noise component consistent with the ship rolling frequency through fast Fourier transform, and retains the real liquid level change trend signal.

[0085] For the pressure sensor data, the unit sets a frequency domain band-pass filter to filter out baseline drift below 0.1 Hz and mechanical impact interference above 50 Hz, and extracts the effective pressure fluctuation characteristics. The weight dynamic allocation engine receives the interference coefficient output by the environment perception unit and the feature data processed by the trust chain reconstruction unit, and dynamically calculates the confidence weight of each sensor by using a fuzzy logic algorithm, including: when the difference between the surge period and the inherent frequency of the ship drainage system is less than 0.5 Hz, the weight of the flow rate sensor is reduced by 30%, and according to the consistency verification result of the pressure sensor data and the adjacent area liquid level sensor, the weight of the pressure sensor is increased to above the preset safety threshold.

[0086] Under the condition of severe ship body shaking, the engine starts a majority voting mechanism, compares the data dispersion of at least three sensors in the same monitoring area, if the variance of the liquid level sensor data exceeds the allowed range and the data consistency of the pressure sensor cluster is above 90%, the pressure sensor data is forced to be used as the decision basis, and the abnormal liquid level sensor is marked as a low-confidence device. All weight allocation results and modified decisions are transmitted to the central controller in real time through a high-speed data channel, and are updated to the dynamic trust chain database for subsequent sensor health evaluation and self-learning model optimization.

[0087] The environmental perception unit generates a dynamic attenuation coefficient of the liquid level sensor according to the correlation between the ship roll angular velocity and the surge period, specifically:

[0088] When the ship roll angular velocity exceeds the preset threshold, the dynamic attenuation coefficient of the liquid level sensor decreases according to a linear function;

[0089] When the surge period coincides with the ship's inherent rolling frequency, the reliability threshold of the flow rate sensor is increased to a preset safety value.

[0090] It should be further pointed out that in the specific implementation process, the environmental perception unit in the implementation, through the ship deck attitude sensor real-time monitoring roll angular velocity data, when the detection angle speed value exceeds the preset 5 degrees / second threshold, the unit automatically activates the dynamic attenuation mechanism of the liquid level sensor.

[0091] The mechanism generates an attenuation coefficient positively correlated with vibration intensity based on the integral calculation result of the roll angular acceleration, including: if the angular acceleration continuously increases within 3 seconds, the attenuation coefficient decreases linearly by 0.15 for every 0.1g acceleration, and the minimum is 0.4; At the same time, the surge period data collected by the environmental monitoring sensor is matched with the ship's inherent rolling frequency library, when the difference between the two is less than 0.5 seconds, the unit will increase the reliability threshold of the flow rate sensor from the default value of 0.6 to the safety value of 0.8, and correlate the baseline stability index of the drain pipe pressure sensor, if the standard deviation of the pressure data is continuously less than 10kPa for 5 consecutive sampling periods, the flow rate sensor threshold is further increased to 1.2 times the preset value.

[0092] In the application process of the dynamic attenuation coefficient, the unit performs sliding mean filtering processing on the original data of the liquid level sensor, and the window width is dynamically adjusted according to the attenuation coefficient: when the coefficient is reduced to 0.5 or less, the window is expanded to 15 seconds to suppress high-frequency noise, and the filtered data is compared with the equivalent liquid level value calculated by the pressure sensor in the same area, if the difference exceeds 10cm, the secondary correction is triggered, and the attenuation coefficient is additionally reduced by 0.1.

[0093] All adjusted parameters are written into the ship motion-sensor error correlation matrix in real time, and are synchronized to the decision database of the central controller through the data bus, providing historical reference basis for subsequent weight allocation. When the ship enters a steady sailing state, the unit automatically restores the attenuation coefficient of the liquid level sensor to the initial value 1.0, and updates the adaptive learning model of the flow rate sensor reliability threshold based on the surge period distribution characteristics in the past 30 minutes.

[0094] The trust chain reconstruction unit includes:

[0095] The liquid level signal processing subunit separates the mechanical vibration interference components in the liquid level sensor signal using a wavelet packet decomposition algorithm;

[0096] The pressure signal processing subunit adopts a sliding window entropy value calculation method to identify transient abnormal pulses in the pressure sensor data, which are caused by the rupture of bubbles in the drainage pipeline.

[0097] It should be further explained that, in the implementation process, the liquid level signal processing subunit performs multi-scale wavelet packet decomposition on the original signal of the liquid level sensor installed on the inboard side of the bulwark. First, the collected time domain signal is divided into 8 sub-bands, and the interference components overlapping with the ship mechanical vibration frequency are identified through energy spectrum analysis. The signals of the third to fifth sub-bands are extracted as the vibration noise base, and the corresponding frequency band components are removed from the original data to retain the low-frequency trend signal reflecting the true liquid level. The ship mechanical vibration frequency range is 2-15 Hz. For the pressure sensor data processing subunit, a sliding time window is used to intercept 50 consecutive sampling points of pressure data sequence, and the Shannon entropy value of the data in each window is calculated. When the window entropy value increases by more than twice the standard deviation of the historical mean, it is determined to be a transient abnormal pulse caused by bubble rupture, and the window data is automatically removed and replaced by linear interpolation of the adjacent windows.

[0098] During processing, if the difference between the filtered data of the liquid level sensor and the calculated liquid level value of the pressure sensor exceeds 15 cm for more than 10 seconds, the sensor cooperative calibration mechanism is triggered, including: calling the flow integration data of the same area flow sensor to reverse the theoretical liquid level change, if the reverse result matches the liquid level sensor filtered value by more than 90%, the current window data of the pressure sensor is marked as valid; otherwise, freeze the pressure sensor output and start the data replacement process of the backup sensor. The processed liquid level and pressure signals are compensated for transmission delay differences through time domain alignment algorithm, and finally generate synchronized feature data set to transmit to the weight dynamic allocation engine, while recording the frequency and duration of abnormal pulses for updating the adaptive parameters of the sensor health assessment model.

[0099] The weight dynamic allocation engine performs the following operations under the condition of high frequency ship body oscillation:

[0100] Adjust the dynamic confidence weight of the side wall liquid level sensor to the interval of 0.2-0.3;

[0101] Increase the dynamic confidence weight of the drainage pipe pressure sensor to the interval of 0.7-0.8;

[0102] Based on the majority voting result of the majority voting mechanism of at least three sensors in the same monitoring area, the final output value of the dynamic confidence weight is corrected.

[0103] It needs to be further explained that in the specific implementation process, when the weight dynamic allocation engine is implemented under the high-frequency swing working condition of the ship body, first of all, the roll angular velocity and the change rate of the pitch angle are monitored in real time through the ship attitude sensor, and when the average value of the roll angular velocity in the continuous 5 seconds exceeds 8 degrees per second and the fluctuation amplitude is greater than 3 degrees, it is determined that it is in a high-frequency swing state.

[0104] The engine then starts the weight dynamic adjustment program: for the liquid level sensor installed on the side wall, the centrifugal force disturbance coefficient caused by the swing is calculated according to the relative distance between its installation position and the ship's gravity center axis, combined with the current surge direction, the confidence weight is gradually reduced from the initial value 1.0 by 0.12 per 0.1g acceleration, and finally limited in the interval of 0.2-0.3; at the same time, based on the data stability index (standard deviation less than 15kPa) of the drain pipe pressure sensor in the last 3 minutes, its weight is increased from the basic value 0.5 to the interval of 0.7-0.8, and the data consistency verification result of at least two adjacent monitoring area pressure sensors is associated, if the verification is passed, an additional 0.05 weight gain is added.

[0105] After the weight allocation is completed, the engine calls the historical data of the liquid level sensor, pressure sensor and flow rate sensor cluster in the same monitoring area, calculates the deviation degree of the current output value of each sensor from the cluster mean, and if the liquid level sensor data deviation degree exceeds 25% and the pressure sensor cluster data consistency is above 85%, the majority voting mechanism is triggered, including: selecting two of more than three same type sensors with the most concentrated data as effective input, forcibly covering the output value of the abnormal sensor, and re-normalizing the corrected dynamic confidence weight.

[0106] During the correction process, the engine compares the discharge demand calculation value before and after adjustment in real time, and if the difference exceeds 10% of the preset threshold, a secondary arbitration process is started, the weight is fine-tuned combined with the screening result of the multi-modal cross-validation module, and finally a fusion data package with credibility identification is generated and transmitted to the central controller. All weight adjustment parameters and voting records are updated to the dynamic trust chain database for subsequent sensor performance degradation analysis and self-learning model training, to ensure that the system gradually optimizes the weight allocation strategy in continuous operation.

[0107] The multi-modal cross-validation module includes:

[0108] The first screening unit verifies the logical consistency between the liquid level sensor data and the ship draft depth and the surge direction based on the fluid mechanics model;

[0109] The second screening unit marks the frequently abnormal working condition sensitive sensors by analyzing the historical behavior patterns of the sensor data through the long short-term memory network;

[0110] When the liquid level sensor reports a sudden increase in water level and the flow rate sensor data does not change, the first level screening unit triggers the delayed response protection mechanism, suspends the drainage instruction output and starts the data review process for three consecutive sampling periods.

[0111] It needs to be further explained that in the specific implementation process, the multi-modal cross-validation module, in implementation, the first level physical constraint screening unit constructs a dynamic fluid mechanics model based on the real-time draft of the ship and the surge direction parameters. When the liquid level sensor reports a sudden increase in water level, combined with the design flux calculation theory of the drainage pipeline, the theoretical drainage efficiency is calculated, including: if the current flow rate sensor data shows that the drainage flow rate is not synchronized to increase, and the ship roll angle causes the actual volume of the bulwark area to be less than the theoretical water accumulation corresponding to the sudden increase in water level, it is determined that it is a transient disturbance caused by external surge impact, triggering the delayed response protection mechanism, freezing the drainage instruction generation and starting the data review process for three consecutive sampling periods. During this period, the module synchronously calls the data of the adjacent area pressure sensor cluster, and if at least two pressure sensors show that the pressure fluctuation in the pipeline matches the normal drainage mode, the frozen state is released and the corrected drainage instruction is generated.

[0112] The second level screening unit analyzes the historical behavior pattern of the sensor through the long short-term memory network, establishes a data fluctuation baseline library under different speed, turning angle and surge intensity combinations, and when the flow rate sensor continuously appears data jump exceeding three times the standard deviation of the historical mean at a certain turning angle (such as right full rudder 35 degrees), it is automatically marked as a working condition sensitive sensor, the historical data mean of the liquid level sensor and the pressure sensor in the same monitoring area is called for replacement compensation, and the data weight of the abnormal sensor is forced to zero.

[0113] In the two-level screening process, if the liquid level sensor data passes the physical constraint verification but the historical behavior analysis shows that its false positive rate exceeds 15% in similar working conditions in the past 24 hours, the weight redistribution process of the dynamic trust chain fusion decision module is triggered, the confidence weight of the sensor is reduced and the redundant verification of the backup sensor data is started.

[0114] All screening results and compensation operations are recorded in the verification log and real-time synchronized to the decision tree database of the central controller to ensure that the drainage strategy generation integrates the double verification results of physical laws and historical behavior characteristics.

[0115] The second level screening unit performs the following operations after detecting the working condition sensitive sensor:

[0116] Call the historical data of other sensors in the same monitoring area to calculate the current true value;

[0117] Use the calculated value to cover the abnormal output data of the working condition sensitive sensor;

[0118] A sensor fault diagnosis signal is generated and uploaded to the central controller.

[0119] It should be further explained that in the specific implementation process, after detecting the working condition sensitive sensor, the second level screening unit first calls the historical data of the liquid level sensor, pressure sensor and flow rate sensor in the same monitoring area in the past 30 days, matches similar working condition segments according to the current ship speed, turning angle and surge intensity, and extracts the mean value and standard deviation of each sensor in the corresponding time period as the reference.

[0120] If the current output value of the marked flow rate sensor exceeds the range of ±3 times the standard deviation of the historical mean value, the data replacement compensation algorithm is started, including: based on the pressure sensor data to back-propagate the theoretical value of the pipeline flow, combined with the water level change rate of the adjacent liquid level sensor to calculate the equivalent flow rate, to generate a compensation value to cover the original output of the abnormal sensor, and a sliding time window is used for smooth transition during the covering process to avoid data jumping.

[0121] At the same time, the system automatically triggers the sensor redundancy verification process, including: activating the redundant pressure sensor group in the standby drainage pipeline, comparing its data with the consistency of the main sensor cluster, if the redundancy group data deviates less than 5% from the compensation value, it is confirmed that the replacement is effective and a sensor fault diagnosis signal is generated, including the time of abnormal occurrence, working condition parameters and data comparison curve before and after compensation, and uploaded to the maintenance terminal of the central controller through an encryption protocol. If the redundancy verification fails, further call the simulation data of the digital twin model, combined with the virtual sandbox pre-play results to generate a dynamic compensation coefficient, and mark the sensor as a high-risk device, forcing it to only participate in decision-making as an auxiliary data source in the next 10 minutes.

[0122] All replacement operations are recorded to the sensor health record for subsequent self-learning model updating, and the abnormal sensor position and compensation state are highlighted on the human-machine interface to prompt the crew to perform manual review or maintenance. After the generation of the fault diagnosis signal, the system automatically reduces the initial weight of the sensor in the dynamic trust chain fusion decision-making module to below 0.2, and improves the priority of the standby sensor in the adjacent area to ensure the continuous and stable execution of the drainage strategy.

[0123] The fault-tolerant execution and reverse verification module includes:

[0124] The virtual sandbox unit simulates the ship attitude change and cabin water level fluctuation curve within 10 seconds after the execution of the drainage instruction through the digital twin model;

[0125] The reverse comparison unit real-time collects the actual flow data of the drainage pipe pressure sensor, and calculates the difference value with the predicted value of the virtual sandbox unit;

[0126] If the difference value exceeds 5% for three consecutive times, it is determined that the actuator is abnormal, the standby drainage pipeline is switched to and an alarm signal is triggered.

[0127] It needs to be further explained that in the specific implementation process, when the central controller generates the drainage pump starting instruction, the module first constructs a dynamic simulation scene based on the three-dimensional digital model of the ship and real-time sensor data, simulates the gradual adjustment process of the drainage valve opening degree from 10% to the target value, and predicts the water level drop rate in the bulwark area, the pipeline pressure fluctuation and the ship's center of gravity offset in the next 10 seconds.

[0128] If the simulation results show that the water level drop rate is lower than 95% of the theoretical value or the ship's roll angle increases by more than 2 degrees of the safety threshold, the module immediately triggers dynamic strategy backtracking, including: suspending the current instruction execution, sending a data re-verification request to the dynamic trust chain fusion decision module, and calling the multi-modal cross-validation module to perform secondary screening on the liquid level and pressure sensor data.

[0129] In the actual drainage stage, the module obtains the actual flow value through the pressure sensor in the drainage pipeline at a sampling frequency of 20 times per second, compares it with the predicted curve of the digital twin model in a sliding window, and calculates the average absolute error in each 3-second time window.

[0130] When the error of three consecutive windows exceeds 5%, the module determines that the actuator is abnormal, cuts off the current main drainage pipeline control signal, activates the backup pipeline electromagnetic valve and increases the redundant pump power to 120% of the rated value, and at the same time sends a three-level alarm signal through the deck sound and light alarm.

[0131] During the reverse verification process, the module synchronously analyzes the long-term deviation trend of the actual drainage data and the predicted value, and if it finds that the cumulative deviation of the fluid resistance coefficient exceeds 8% of the initial parameter, it automatically updates the pipeline friction coefficient calibration value of the digital twin model and re-trains the pump efficiency curve fitting model.

[0132] All operation records and verification results are transmitted to the secure storage unit of the central controller through the AES-256 encryption protocol, generating a complete evidence chain containing time stamps, simulation parameters, actual error values and fault codes for subsequent accident tracing and system optimization. When switching to the backup pipeline, the module continuously monitors the pressure stability of the new pipeline, and if the error alarm is not triggered again within 30 seconds, it sends a recovery ready signal to the central controller, otherwise it starts the full-system safety locking mechanism and uploads an emergency shutdown request to the ship control center.

[0133] The central controller performs the following logic before generating the drainage instruction:

[0134] Receive the dynamic confidence weight allocation result of the dynamic trust chain fusion decision module;

[0135] Receive the two-level screening pass identifier of the multi-modal cross-validation module;

[0136] receiving the rehearsal deviation detection result of the virtual sandbox unit;

[0137] When the above three inputs all meet the preset conditions, a drainage pump starting instruction and a valve opening degree control signal are generated.

[0138] It should be further explained that in the specific implementation process, when the central controller receives the dynamic confidence weight distribution table of the dynamic trust chain fusion decision module, it first checks whether the liquid level sensor weight is higher than 0.5 and the pressure sensor weight is lower than 0.3, and if the conditions are met, it calls the water depth and displacement relationship curve in the ship stability database, and calculates the theoretical drainage flow threshold value combined with the current actual water depth value.

[0139] Synchronously reading the two-stage screening pass identification of the multi-modal cross-validation module, including: the physical constraint screening needs to verify the logical consistency of the sudden increase in liquid level data and the surge direction, the ship body roll angle, and the historical behavior screening needs to confirm that there is no high-frequency false alarm record under similar working conditions.

[0140] At the same time, the virtual sandbox rehearsal result of the fault-tolerant execution module is obtained, if the deviation value is less than 3% and the predicted water level drop rate reaches more than 90% of the rated value, the drainage instruction generation process is started. The controller uses a weighted voting mechanism to integrate three types of input signals, including: dynamic confidence weight accounts for 50% of decision weight, two-stage screening result accounts for 30%, and virtual sandbox deviation value accounts for 20%, when the comprehensive score exceeds the preset threshold 0.8, the graded control instruction is generated, that is: the electric valve of the main drainage pipeline is opened to 60% opening degree, and the centrifugal pump is started to run at 80% rated speed.

[0141] During the execution of the instruction, the feedback signal of the valve displacement sensor and the pump motor current value are collected in real time, if the deviation between the actual opening degree of the valve and the instruction value exceeds 5% for 5 seconds, or the pump motor current fluctuation amplitude exceeds the rated value by 10%, the gradient adjustment strategy is triggered, including: gradually increasing the valve opening degree by 2% per second, at the same time, the pump motor speed is increased to 85%, until the actual drainage flow matches the theoretical threshold.

[0142] When the fault-tolerant execution module finds that the flow is continuously abnormal, the controller interrupts the current control signal, switches to the standby pipeline and activates the emergency mode, in which mode the dynamic weight distribution function is closed, and the mean value of the pressure sensor and the redundant liquid level sensor is directly used to drive the decision.

[0143] All operation data are timestamped and encrypted to generate decision logs, including weight distribution parameters, verification identification status, and actuator response curves, which are transmitted to the ship central monitoring system through a dual-channel redundant bus for synchronous transmission to form a traceable closed-loop control evidence chain, and a dynamic display of the drainage strategy decision tree is displayed on the human-machine interface, highlighting the current effective sensor data source and the verification link status.

[0144] The triggering conditions of the majority voting mechanism include:

[0145] The sum of the dynamic confidence weights of the sensors in the same monitoring area is below the preset reliability threshold;

[0146] The difference between the data reported by at least two sensors exceeds the maximum allowed error range;

[0147] The system automatically selects the majority voting result as the final decision basis and marks the low-weight sensor for offline calibration.

[0148] It needs to be further explained that in the implementation process of the majority voting mechanism of the ship bulwark drainage intelligent management system, when the sum of the dynamic confidence weights of the sensors in the same monitoring area is below the preset reliability threshold 0.7, the system automatically triggers the majority voting process, including: first, selecting at least three effective sensor data from the liquid level sensor, pressure sensor and flow rate sensor cluster in the area, calculating the mean and dispersion, if the data difference of at least two sensors is less than the maximum allowed error range, such as the liquid level difference is not more than ±10 cm, the pressure difference is within ±20 kPa, then the two sensor data are taken as effective input set, and the weighted average value is calculated as the final decision basis. When it is detected that the side wall liquid level sensor data and the effective input set differ by more than 150% of the maximum allowed error, the system marks it as a low-weight sensor, freezes its data output and generates an offline calibration instruction, which contains the abnormal time point, working condition parameters and deviation history record, and is transmitted to the maintenance terminal through an encrypted link. During calibration, the system automatically increases the initial weight of the adjacent area backup sensor to 1.2 times the original value, and calls the digital twin model simulation data to fill in the missing values, ensuring the continuity of decision-making.

[0149] After the voting is completed, the system will reverse verify the corrected data with the original weight distribution result, including: if the correction value causes the drainage strategy adjustment amplitude to exceed 20% of the initial decision, the weight redistribution process of the dynamic trust chain fusion decision module is triggered, the health status of each sensor is re-evaluated and the attenuation coefficient is updated. All voting records and calibration operations are written into the sensor health file for threshold setting optimization of the subsequent self-learning model, and the sensor reliability distribution in each area is displayed in the form of a heat map on the ship human-machine interface, marking the location of the low-weight sensor and the calibration countdown.

[0150] When the ship enters a steady sailing state, the system automatically releases the frozen state of the low-weight sensor, and dynamically restores its initial weight to 80% of the preset value based on the voting results in the past 2 hours. If the abnormality is not triggered for 3 consecutive times, the normal decision-making right is fully restored.

[0151] A ship bulwark drainage intelligent management method combined with multi-sensor data monitoring, comprising the following steps:

[0152] Step S1: Real-time data acquisition through the bulwark liquid level sensor, drainage pipeline flow rate / pressure sensor, ship attitude sensor and environmental monitoring sensor, anti-interference data bus transmission to dynamic trust chain fusion decision module, complete the sliding mean filtering and baseline drift correction of raw data.

[0153] Step S2: Based on the ship roll angular velocity, surge period and environmental parameters, generate sensor dynamic attenuation coefficient, separate signal interference components by wavelet packet decomposition and entropy calculation, dynamically allocate confidence weight of each sensor, trigger majority voting mechanism to eliminate conflicting data.

[0154] Step S3: Verify the physical reasonableness of the liquid level / flow rate data through the fluid mechanics model, analyze the sensor abnormal mode combined with the historical behavior database, and start the data replacement compensation algorithm for the sensor with high frequency false alarm.

[0155] Step S4: Use the digital twin model to simulate the ship body attitude and water level change after the drainage instruction is executed. If the prediction deviation exceeds 5% or the ship stability risk increases, trigger the dynamic strategy backtracking and regenerate the instruction.

[0156] Step S5: The central controller generates a hierarchical drainage instruction by comprehensively considering the dynamic weight, verification identifier and pre-play results, controls the valve opening degree and pump rotating speed, and monitors the execution mechanism current and displacement feedback signal in real time.

[0157] Step S6: Collect the actual drainage flow through the pressure sensor, and compare it with the digital twin prediction value in the sliding window. If the error exceeds the limit for three consecutive times, it is determined that the execution is abnormal and the standby pipeline is switched.

[0158] Step S7: Mark the low-weight state of the faulty sensor and start the redundant data compensation. After switching to the standby pipeline, continuously monitor the pressure stability. If there is no secondary alarm within 30 seconds, the system returns to normal mode.

[0159] Step S8: Encrypt the weight allocation parameters, verification results and execution response curve, update the friction coefficient and pump efficiency curve of the digital twin model, and synchronize the threshold setting of the self-learning algorithm.

[0160] Step S9: Based on the majority voting result and historical calibration records, generate a sensor health degree heat map, mark low-weight device locations and offline calibration countdown, and trigger manual maintenance prompts.

[0161] Step S10: Use actual drainage data to train the dynamic trust chain fusion model in reverse, optimize the weight distribution strategy and majority voting trigger conditions, and improve the system's adaptive ability in complex sea conditions.

[0162] By constructing a dynamic trust chain fusion decision model and a multi-modal cross-validation mechanism, the accuracy and timeliness of multi-source sensor data fusion in complex sea conditions are improved. Based on the dynamic weight distribution algorithm of ship attitude and environmental parameters, the sensor misjudgment caused by mechanical vibration, bubble interference, etc. is effectively suppressed. Combined with digital twin pre-play and reverse verification closed loop, the drainage command mis-triggering rate is reduced, the response delay is shortened, and through two-level screening and majority voting mechanism, the sensor cluster's cooperative fault tolerance is realized, ensuring the continuous and stable operation of the system under single-point failure or abnormal interference.

[0163] The adaptability of traditional drainage management schemes in dynamic sea conditions is solved, and the ship navigation safety and energy efficiency management level are improved. The adaptive learning ability can optimize the sensor evaluation model and decision threshold through historical data, reducing the frequency of manual maintenance. The modular design is compatible with various ship types and drainage system architectures, providing technical support for the automatic upgrade of intelligent ships.

[0164] It should be noted that in this text, relationship terms such as first and second are only used to distinguish one entity or operation from another, and do not necessarily require or imply any actual relationship or order between the entities or operations. Moreover, the term "includes", "contains" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of another identical element in the process, method, article or device that includes the element.

[0165] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A smart management system for ship bulwark drainage that integrates multi-sensor data monitoring, characterized in that, include: Sensor array, dynamic trust chain fusion decision module, multimodal cross-validation module, fault-tolerant execution and reverse verification module, and central controller; The sensor group includes a liquid level sensor installed inside the ship's bulwark, a flow velocity sensor and a pressure sensor in the drainage pipe, a ship deck attitude sensor, and an external environmental monitoring sensor. Each sensor is connected to the dynamic trust chain fusion decision module via a data bus. The dynamic trust chain fusion decision module receives real-time data from the sensor group and dynamically adjusts the confidence weights of each sensor based on the ship's motion attitude parameters and environmental parameters. The multimodal cross-validation module performs two-level screening on the data output by the dynamic trust chain fusion decision module through a physical constraint model and a historical behavior database. The fault-tolerant execution and reverse verification module uses a digital twin model to pre-simulate the execution effect of drainage commands and performs reverse verification with the actual drainage results. The central controller generates drainage instructions based on the output signals of the dynamic trust chain fusion decision module, the multimodal cross-validation module, and the fault-tolerant execution and reverse verification module, and controls the action of the actuator. The dynamic trust chain fusion decision module includes: The environmental sensing unit, connected to the ship's attitude sensor and environmental monitoring sensor, is used to acquire the ship's roll rate, pitch angle and surge period in real time. The trust chain reconstruction unit is connected to the level sensor, flow rate sensor and pressure sensor, and is used to perform time-domain decomposition and frequency-domain filtering on the sensor signals. The dynamic weight allocation engine receives the output signal from the environmental perception unit and the filtered data from the trust chain reconstruction unit to generate dynamic confidence weights for each sensor. The trust chain reconstruction unit includes: The liquid level signal processing subunit uses a wavelet packet decomposition algorithm to separate the mechanical vibration interference component in the liquid level sensor signal. The pressure signal processing subunit uses a sliding window entropy calculation method to identify transient abnormal pulses in the pressure sensor data. These transient abnormal pulses are caused by the rupture of air bubbles in the drainage pipe. The multimodal cross-validation module includes: The first-level screening unit verifies the logical consistency between liquid level sensor data and ship draft and surge direction based on a fluid dynamics model; The second-level screening unit analyzes the historical behavior patterns of sensor data through long short-term memory networks to identify condition-sensitive sensors that frequently exhibit abnormalities. When the level sensor reports a sudden increase in water level and the flow velocity sensor data remains unchanged, the first-level screening unit triggers a delayed response protection mechanism, suspends the output of drainage instructions, and starts a data verification process for three consecutive sampling cycles. The fault-tolerant execution and reverse verification module includes: The virtual sandbox unit simulates the changes in hull attitude and the fluctuation curve of water level inside the cabin within 10 seconds after the execution of the drainage command through a digital twin model. The reverse comparison unit collects the actual flow data from the drainage pipe pressure sensor in real time and calculates the difference between it and the predicted value from the virtual sandbox unit. If the difference exceeds 5% for three consecutive times, the actuator is deemed abnormal, the system is switched to the backup drainage pipeline, and an alarm signal is triggered.

2. The intelligent management system for ship bulwark drainage combining multi-sensor data monitoring as described in claim 1, characterized in that: The environmental sensing unit generates a dynamic attenuation coefficient for the liquid level sensor based on the correlation between the ship's roll rate and the surge period, specifically: When the ship's roll rate exceeds a preset threshold, the dynamic attenuation coefficient of the liquid level sensor decreases according to a linear function. When the surge cycle coincides with the ship's inherent rolling frequency, the reliability threshold of the flow velocity sensor is increased to a preset safety value.

3. The intelligent management system for ship bulwark drainage combining multi-sensor data monitoring as described in claim 1, characterized in that: The weight dynamic allocation engine performs the following operations under high-frequency hull rolling conditions: The dynamic confidence weight of the sidewall liquid level sensor was adjusted to the range of 0.2-0.3; Increase the dynamic confidence weight of the drain pipe pressure sensor to the range of 0.7-0.8; The final output value of the dynamic confidence weight is corrected based on the majority voting result of a majority voting mechanism involving at least three sensors within the same monitoring area.

4. The intelligent management system for ship bulwark drainage combining multi-sensor data monitoring as described in claim 1, characterized in that: After detecting a condition-sensitive sensor, the second-level screening unit performs the following operations: The current true value is calculated by calling historical data from other sensors within the same monitoring area; The calculated value is used to cover the abnormal output data of the condition-sensitive sensor; Generate sensor fault diagnosis signals and upload them to the central controller.

5. The intelligent management system for ship bulwark drainage combining multi-sensor data monitoring as described in claim 1, characterized in that: The central controller executes the following logic before generating a drainage command: Receive the dynamic confidence weight allocation results from the dynamic trust chain fusion decision module; The two-level screening of the multimodal cross-validation module passes the identification. Receive the pre-simulation deviation detection results from the virtual sandbox unit; When all three inputs meet the preset conditions, a drainage pump start command and a valve opening control signal are generated.

6. The intelligent management system for ship bulwark drainage combining multi-sensor data monitoring according to claim 3, characterized in that: The triggering conditions for the majority voting mechanism include: The sum of the dynamic confidence weights of sensors within the same monitoring area is lower than a preset reliability threshold; The difference between the data reported by at least two sensors exceeds the maximum permissible error range; The system automatically selects the majority vote result as the basis for the final decision and marks low-weight sensors for offline calibration.

Images