A hydrogen energy unmanned aerial vehicle fuel cell water management fault identification method and system based on flight condition correction

Abstract

The application discloses a hydrogen energy unmanned aerial vehicle fuel cell water management fault identification method and system based on flight condition correction, relates to the field of hydrogen energy unmanned aerial vehicle power system state monitoring, and solves the problems of the existing hydrogen energy unmanned aerial vehicle, such as the fact that the voltage fluctuation of a fuel cell stack is easily misjudged as a water management fault when the flight load rapidly changes. The method comprises the following steps: collecting multiple-source state quantities of a fuel cell system, extracting flight conditions and load disturbance auxiliary quantities; pre-processing the multiple-source data to obtain window samples, and constructing a condition normalization feature vector; performing parallel double-path diagnosis, calculating mechanism residual diagnosis evidence according to a polarization, ohmic internal resistance and gas mass transfer water management mechanism model, outputting three types of state candidate probabilities, namely, normal, water flooding and membrane dry, by using a trained data classification model; fusing multiple types of information through a water management fault coupling matrix to cooperatively judge, outputting a water management state, an identification confidence and a fault level, and issuing an alarm, a return, a shutdown protection grading strategy through a safety control interface.

Description

Technical Field

[0001] This invention relates to the field of hydrogen-powered unmanned aerial vehicle (UAV) power system condition monitoring and fault diagnosis technology, specifically to a method and system for identifying water management faults in hydrogen-powered UAV fuel cells based on flight condition correction. Background Technology

[0002] Hydrogen-powered drones typically employ a power system combining fuel cells and battery buffers, offering advantages such as long endurance, rapid refueling, and suitability for long-distance missions. They can be used in scenarios such as inspection, emergency communication, maritime patrol, and logistics transportation. Compared to ground-based fixed platforms, drones experience various conditions during flight, including climb, hovering, turning, gusts of wind, and load variations, causing significant transient fluctuations in the fuel cell output voltage and stack current.

[0003] The water management status of a fuel cell directly affects the water content of the proton exchange membrane, gas mass transfer capacity, and stack output stability. Both flooding and membrane dryness can manifest as a decrease in stack voltage, increased voltage dispersion per cell, or reduced output power, but their mechanisms are opposite: flooding is mainly caused by the accumulation of liquid water leading to limited gas mass transfer, while membrane dryness is mainly caused by insufficient membrane water content leading to increased ohmic impedance. Therefore, existing single-threshold alarms or general fuel cell fault diagnosis methods struggle to distinguish between genuine water management faults and secondary voltage fluctuations caused by propulsion load disturbances or electrical bus disturbances during rapid changes in UAV flight loads. Summary of the Invention

[0004] This invention addresses the problem in existing technologies where rapid changes in fuel cell stack voltage during rapid load changes in hydrogen-powered drones are easily misinterpreted as water management malfunctions. To this end, this invention proposes a method and system for identifying water management malfunctions in hydrogen-powered drone fuel cells based on flight condition correction. The invention achieves the above-mentioned technical problems through the following technical solution: Option 1: This invention proposes a method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles (UAVs) based on flight condition correction. The method includes the following steps: Step 1: Collect multi-source state variables of the hydrogen energy UAV fuel cell system. The multi-source state variables include at least the fuel cell stack state, water management supply and environmental state, and electric bus and propulsion load state. Use any one of the multi-source state variables as an auxiliary variable for flight conditions and load disturbances. Perform time synchronization, filtering and sliding window segmentation on the multi-source state variables to obtain window samples. Step 2: Construct a normalized feature vector for the working condition based on the window samples described in Step 1; Step 3: Calculate the theoretical residual diagnostic evidence based on at least one of the water management mechanism models among the fuel cell polarization model, ohmic internal resistance estimation model, and gas mass transfer model; calculate the candidate probabilities of normal state, flooded state, and membrane dry state based on the trained data classification model; Step 4: Use the water management fault coupling matrix to collaboratively fuse the mechanism residual diagnostic evidence, candidate probabilities, and auxiliary quantities of flight conditions and load disturbances, output the water management status identification results, confidence level and fault level, and output alarm, power limit, return to base or protection control strategies through the safety control interface.

[0005] Furthermore, a preferred embodiment is provided, wherein the normalized feature vector of the operating condition mentioned in step 2 includes single-chip voltage dispersion, stack voltage drop slope, stack voltage recovery slope, temperature gradient, humidity change, gas pressure fluctuation characteristics, load normalized stack current, and water management comprehensive characteristics composed of voltage dispersion, humidity change and gas pressure fluctuation, as well as bus ripple amplitude, battery compensation current ratio and propulsion load mutation as flight load disturbance exclusion characteristics.

[0006] Furthermore, a preferred embodiment is provided, wherein the mechanistic residual diagnostic evidence mentioned in step 3 consists of the residuals output by at least two of the following models: fuel cell polarization model, ohmic internal resistance estimation model, gas mass transfer model, and thermal-humidity balance model. In this embodiment, the flooded state corresponds to limited mass transfer, gas pressure fluctuations, and enhanced voltage recovery hysteresis, while the dry membrane state corresponds to enhanced ohmic polarization residuals, temperature and humidity deviations, and voltage drop slope. The bus dynamic model can be combined to help distinguish between water management faults and external load disturbances.

[0007] Furthermore, a preferred embodiment is provided, wherein the data classification model in step 3 is any one or a combination of a lightweight gradient boosting model, a support vector machine, a random forest, a temporal convolutional network, or a gated recurrent network. The data classification model outputs candidate probabilities for normal state, flooded state, and membrane dry state, and is deployed in the edge computing unit of the UAV.

[0008] Furthermore, a preferred embodiment is provided in which the data classification model and mechanism residual threshold are obtained through offline training and calibration using multi-condition water management data from a PEMFC fuel cell test bench. The data includes normal state, flooded state, and membrane dry state, and covers different load current conditions. During the training process, the voltage, current, air flow, air pressure, temperature, and humidity data are segmented by sliding window and feature extracted. After training, the model parameters, fusion weights, and threshold parameters are solidified into the edge computing unit of the UAV for online inference during flight.

[0009] Furthermore, a preferred embodiment is provided in which the water management fault coupling matrix described in step 4 is used to record the primary evidence relationship, secondary evidence relationship, and mutual exclusion relationship between the normal state, flooded state, and membrane dry state and the observed characteristics, and to record the secondary exclusion relationship between flight load disturbance and voltage transient change, in order to distinguish the root cause state of fuel cell water management and the secondary phenomenon caused by UAV load mutation.

[0010] Furthermore, a preferred embodiment is provided in which the collaborative fusion described in step 4 is achieved by at least one of Bayesian fusion, DS evidence theory, weighted voting, or rule-constrained fusion, and the warning state is maintained when the water flooding state, membrane dry state, and flight load disturbance conflict degree are greater than a preset threshold.

[0011] Option 2: A fault identification system for water management in hydrogen-powered unmanned aerial vehicles (UAVs) based on flight condition correction, the system comprising: The state acquisition module is used to acquire multi-source state quantities of the hydrogen energy UAV fuel cell system. The multi-source state quantities include at least the fuel cell stack state, water management supply and environmental state, and electric bus and propulsion load state. Any one of the multi-source state quantities is used as an auxiliary quantity for flight conditions and load disturbances. The multi-source state quantities are time-synchronized, filtered, and segmented into sliding windows to obtain window samples. The preprocessing and feature module is used to construct a normalized feature vector of working conditions based on the window samples described in the data acquisition and synchronization module. The diagnostic reasoning module is used to calculate the theoretical residual diagnostic evidence based on at least one water management mechanism model among the fuel cell polarization model, Ohmic internal resistance estimation model and gas mass transfer model; and to calculate the candidate probabilities of normal state, flooded state and membrane dry state based on the trained data classification model. The output and control module is used to collaboratively fuse the mechanism residual diagnostic evidence, candidate probabilities, and auxiliary quantities of flight conditions and load disturbances using the water management fault coupling matrix, output the water management status identification results, confidence level and fault level, and output alarm, power limit, return to base or protection control strategies through the safety control interface.

[0012] Option 3: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in Option 1.

[0013] Option 4: A computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the method described in Option 1.

[0014] The advantages of this invention are: The hydrogen-powered UAV fuel cell water management fault identification method based on flight condition correction described in this invention can improve the stability of water management status identification in scenarios with rapid changes in flight load of hydrogen-powered UAVs; this invention can distinguish between flooding and membrane drying by combining mechanistic residuals and data classification evidence; this invention can reduce false alarms caused by propulsion load disturbances by utilizing flight condition correction features; and it can convert the identification results into handling suggestions such as early warning, de-rated, return to base, or protection.

[0015] The hydrogen-powered UAV fuel cell water management fault identification method based on flight condition correction described in this invention sets up a mechanism residual diagnosis module and a data classification diagnosis module in parallel. These modules generate mechanism residual evidence based on the physical laws of water transport in the fuel cell stack and data classification evidence based on massive flight data mining, respectively. Then, the two diagnostic evidences are fused and judged through a water management fault coupling matrix. This method relies on the fuel cell water management mechanism to ensure the physical interpretability of the diagnostic results, and uses a data classification model to adapt to the complex and ever-changing UAV flight conditions. It overcomes the shortcomings of poor adaptability of single mechanism models and lack of mechanism support in pure data models, and significantly improves the identification accuracy of two typical water management faults: flooding and membrane drying.

[0016] The fault identification method for fuel cell water management of hydrogen-powered UAVs based on flight condition correction described in this invention, after fusing evidence, can accurately distinguish between three working states: normal equipment, fuel cell stack flooding, and fuel cell stack membrane dryness, and simultaneously outputs the fault identification confidence level and corresponding fault level. The fault level output module connects to the UAV flight control system and airborne energy management system through a safety control interface, and can issue differentiated handling instructions such as early warning prompts, power derating, and emergency shutdown protection at different levels, realizing closed-loop control of fault identification and airborne safety, effectively avoiding the risk of power loss, shutdown and crash caused by fuel cell water management failure, and improving the flight safety of hydrogen-powered UAVs.

[0017] The hydrogen-powered UAV fuel cell water management fault identification method based on flight condition correction described in this invention uses a sliding window to dynamically construct state features and condition correction features. It can dynamically update feature vectors along the entire flight profile of the UAV, including takeoff, climb, hovering, dive, and maneuvering. It can adapt to extreme flight conditions such as short-term heavy load, variable load, vibration and shock of UAV. Compared with fixed feature threshold diagnosis methods, it can still maintain stable and reliable fault identification capability under variable flight loads. It is suitable for various multi-rotor and fixed-wing hydrogen-powered UAV usage scenarios and has strong versatility. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process for identifying water management faults in hydrogen-powered drone fuel cells based on flight condition correction, as described in Implementation Method 1.

[0019] Figure 2This is a schematic diagram of the hydrogen-powered UAV fuel cell water management fault identification system based on flight condition correction as described in Implementation Method 1.

[0020] Figure 3 This is a schematic diagram of the water management fault coupling matrix described in Implementation Method 1.

[0021] Figure 4 This is a schematic diagram of the multi-source feature structure of water management as described in Implementation Method 1.

[0022] Figure 5 This is a schematic diagram of the fault levels and safety handling strategies described in Implementation Method 1.

[0023] Figure 6 This is a schematic diagram of the arrangement of the sensors described in Embodiment 1 on the structure of a hydrogen-powered drone.

[0024] Figure 7 This is a schematic diagram of the model training and data validation process described in Implementation Method 1.

[0025] Figure 8 This is a schematic diagram of typical characteristic curves for the three types of water management states described in Implementation Method 1.

[0026] Figure 9 This is a schematic diagram of the cross-load verification obfuscation matrix described in Implementation Method 1.

[0027] Figure 10 This is a schematic diagram comparing the cross-load recognition performance described in Implementation Method 1.

[0028] Among them, there are hydrogen pressure sensor 1, hydrogen flow sensor 2, fuel cell stack temperature sensor 3, voltage acquisition module 4, current sensor 5, air pressure sensor 6, ambient temperature and humidity sensor 7, and edge computing diagnostic unit 8. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0030] Implementation Method 1, see [link] Figures 1 to 10This embodiment describes a method and system for identifying water management faults in hydrogen-powered unmanned aerial vehicles (UAVs) fuel cells based on flight condition correction. The method specifically includes: collecting multi-source state variables related to the fuel cell stack, water management supply and environmental conditions, electrical bus, and propulsion load; performing synchronization, filtering, and sliding window processing on the multi-source state variables; constructing a water management normalized feature vector related to the current flight condition; generating candidate evidence for normal operation, flooding, and membrane dryness through mechanistic residual diagnosis and data classification diagnosis, respectively; and fusing the candidate evidence and flight condition correction features using a water management fault coupling matrix to output the water management state type, confidence level, and fault level.

[0031] like Figure 3 As shown, the water management fault coupling matrix establishes primary evidence relationships, secondary evidence relationships, and mutual exclusion relationships between the normal state, flooded state, and membrane dry state and observational features such as voltage dispersion, voltage drop slope, temperature gradient, humidity change, gas pressure fluctuation, and load normalized current. It also uses propulsion load abrupt changes, electrical bus ripple, and airframe vibration as external disturbance sources to distinguish between the root cause state of fuel cell water management and secondary phenomena of flight conditions.

[0032] like Figure 4 As shown, the feature construction unit generates feature vectors based on a sliding window. The feature vectors include single-chip voltage dispersion, stack voltage drop slope, stack voltage recovery slope, temperature gradient, humidity change, gas pressure fluctuation characteristics, and water management comprehensive characteristics composed of voltage dispersion, humidity change, and gas pressure fluctuation. Furthermore, the bus ripple amplitude, load normalized stack current, and propulsion load abrupt change are combined to correct flight conditions and eliminate false alarms.

[0033] like Figure 5 As shown, the fault level output module can classify diagnostic results into normal (L0), warning (L1), derated (L2), return to home (L3), and protection (L4). The safety control interface communicates with the UAV flight control system or energy management system, converting water management status identification results into control commands such as power limiting, power allocation, return to home, protection shutdown, or alarm recording.

[0034] like Figure 7 As shown, in one embodiment of model training and verification, the model training and verification module (300) reads the PEMFC fuel cell bench water management dataset. The bench water management dataset includes normal state, flooded state and membrane dry state, and covers different load current conditions such as 108A, 270A and 405A. Its state quantities include at least one of stack voltage, stack current, air flow rate, air pressure, temperature and humidity.

[0035] In a specific numerical embodiment, the sliding window sample segmentation uses an overlapping window with a length of 600 sampling points and a step size of 60 sampling points. When the data sampling frequency changes, the window length can be from 100 to 1000 sampling points, and the step size can be from 5% to 20% of the window length. When the data classification model uses a random forest model, the number of decision trees can be from 100 to 300, and the maximum depth of a single tree can be from 4 to 12. When using a gradient boosting model, the number of weak learners can be from 100 to 300, and the learning rate can be from 0.03 to 0.10. During cross-load validation, any one of the load conditions of 108A, 270A, and 405A is used as the test condition, and the remaining load conditions are used as training or validation conditions. The model's recognition performance is evaluated using confusion matrix, accuracy, recall, and Macro-F1 metrics.

[0036] The test bench water management data is time-synchronized, outlier-handled, and segmented into sliding window samples. The feature extraction unit constructs features such as voltage drop slope, voltage recovery slope, single-chip voltage dispersion, gas pressure fluctuation characteristics, temperature gradient, humidity change, and load normalized current. The classification model training unit trains a three-classification model using the features as input. The mechanism residual threshold calibration unit determines the flooding evidence threshold and membrane dryness evidence threshold based on the residual distribution under normal operating conditions.

[0037] Furthermore, the risk function early warning unit calculates the risk function based on the cumulative degree of voltage residual, mass transfer limited residual, ohmic polarization residual, and temperature and humidity residual. When the risk function exceeds the early warning threshold, it outputs the early warning level. The verification and evaluation unit evaluates the model performance using indicators such as confusion matrix, accuracy, recall, F1 score, AUC, false alarm rate, and early warning lead time. It also solidifies the verified model parameters, fusion weights, and fault level thresholds into the UAV edge computing unit.

[0038] It should be noted that the PEMFC bench data is mainly used to verify the ability to identify normal, flooded, and dry membrane states. In UAV applications, the water management state identification results can be further fused with flight control, battery, DC / DC, and propulsion load data for flight condition correction, fault level determination, and safety handling strategy output.

[0039] like Figure 8 As shown, under different load conditions of 108A, 270A and 405A, the voltage-related characteristics of the normal state, flooded state and membrane dry state show different trends, which can provide a data basis for water management status identification.

[0040] like Figure 9As shown, when the model's recognition capability was evaluated using a cross-load verification method, the recognition rates for normal state, flooded state, and membrane dry state were 96.5%, 93.3%, and 100.0%, respectively, indicating that the model can distinguish the three types of water management states under different load conditions.

[0041] like Figure 10 As shown, different models exhibit differences in Macro-F1 scores during cross-load validation. Among them, the Random Forest model and Gradient Boosting model maintain high recognition performance under the 108A, 270A, and 405A test conditions, and can be used as candidate models for deployment of edge computing units on UAVs.

[0042] This embodiment also proposes a hydrogen-powered UAV fuel cell water management fault identification system based on flight condition correction. The system includes a state acquisition module, a preprocessing and feature module, a diagnostic reasoning module, and an output and control module. The state acquisition module includes a sensor array and data on stack status, water management-related supply and environmental status, and bus / load status. The preprocessing and feature module includes a data acquisition and synchronization module, an operating condition identification module, and a feature construction unit. The diagnostic reasoning module includes a mechanism residual diagnosis module, a data classification diagnosis module, and a water management fault coupling matrix. The output and control module includes a collaborative fusion decision module and a fault level output module.

[0043] like Figure 6 As shown, in one embodiment, the UAV body includes a fuel cell compartment, a hydrogen tank compartment, and a battery compartment. A hydrogen pressure sensor 1 is located near the hydrogen tank compartment or its outlet pressure relief valve; a fuel cell stack temperature sensor 3 and a voltage acquisition module 4 are located in the fuel cell compartment; an air pressure sensor 6 is located in the air supply channel; the voltage acquisition module 4 and a current sensor 5 are located in the battery compartment or at the power bus; the current sensor 5 and a speed sensor are located at the propulsion motor; a vibration sensor is located near the aircraft's center of gravity or near the fuel cell compartment mounting base; and an ambient temperature and humidity sensor 7 is located on the outer surface of the aircraft or near the air inlet.

[0044] In summary, this implementation method constructs water management status characteristics and flight condition correction characteristics based on a sliding window, respectively calculates residual evidence and data classification evidence, and performs collaborative fusion with a water management fault coupling matrix to output the identification results, confidence level, and fault level of normal state, flooded state, or membrane dry state. It solves the problem that rapid changes in fuel cell voltage fluctuations during hydrogen-powered UAV flight loads can easily be misjudged as water management faults. It uses propulsion load abrupt changes, electrical bus disturbances, and airframe vibrations as auxiliary quantities to eliminate false alarms, and outputs graded handling suggestions such as warnings, derating, or protection to the flight control or energy management system.

[0045] Those skilled in the art will understand that the above description is merely a preferred embodiment of the present invention, and the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. This is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0046] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for identifying water management faults in the fuel cell of a hydrogen-powered unmanned aerial vehicle based on flight condition correction, characterized in that, The method includes the following steps: Step 1: Collect multi-source state variables of the hydrogen energy UAV fuel cell system. The multi-source state variables include at least the fuel cell stack state, water management supply and environmental state, and electric bus and propulsion load state. Use any one of the multi-source state variables as an auxiliary variable for flight conditions and load disturbances. Perform time synchronization, filtering and sliding window segmentation on the multi-source state variables to obtain window samples. Step 2: Construct a normalized feature vector for the working condition based on the window samples described in Step 1; Step 3: Calculate the theoretical residual diagnostic evidence based on at least one of the water management mechanism models among the fuel cell polarization model, ohmic internal resistance estimation model, and gas mass transfer model; calculate the candidate probabilities of normal state, flooded state, and membrane dry state based on the trained data classification model; Step 4: Use the water management fault coupling matrix to collaboratively fuse the mechanism residual diagnostic evidence, candidate probabilities, and auxiliary quantities of flight conditions and load disturbances, output the water management status identification results, confidence level and fault level, and output alarm, power limit, return to base or protection control strategies through the safety control interface.

2. The method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles based on flight condition correction according to claim 1, characterized in that, The normalized feature vector of operating conditions mentioned in step 2 includes single-chip voltage dispersion, stack voltage drop slope, stack voltage recovery slope, temperature gradient, humidity change, gas pressure fluctuation characteristics, load normalized stack current, and water management comprehensive characteristics composed of voltage dispersion, humidity change and gas pressure fluctuation, as well as bus ripple amplitude, battery compensation current ratio and propulsion load mutation as flight load disturbance exclusion characteristics.

3. The method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles based on flight condition correction according to claim 1, characterized in that, The mechanistic residual diagnostic evidence mentioned in step 3 consists of the residuals output by at least two of the following models: fuel cell polarization model, ohmic internal resistance estimation model, gas mass transfer model, and thermal-humidity balance model. Among them, the flooded state corresponds to limited mass transfer, gas pressure fluctuation, and enhanced voltage recovery hysteresis, while the dry membrane state corresponds to enhanced ohmic polarization residuals, temperature and humidity deviations, and voltage drop slope. It can also be combined with the bus dynamic model to help distinguish between water management faults and external load disturbances.

4. The method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles based on flight condition correction according to claim 1, characterized in that, The data classification model mentioned in step 3 is any one or a combination of a lightweight gradient boosting model, support vector machine, random forest, temporal convolutional network or gated recurrent network. The data classification model outputs candidate probabilities of normal state, flooded state and membrane dry state, and is deployed in the edge computing unit of the UAV.

5. The method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles based on flight condition correction according to claim 1, characterized in that, The data classification model and mechanism residual threshold were obtained through offline training and calibration using multi-condition water management data from the PEMFC fuel cell bench. The data included normal state, flooded state, and membrane dry state, and covered different load current conditions. The training process involves sliding window segmentation and feature extraction of voltage, current, airflow, air pressure, temperature, and humidity data. After training, the model parameters, fusion weights, and threshold parameters are embedded into the edge computing unit of the UAV for online inference during flight.

6. The method for identifying fuel cell water management faults in hydrogen-powered unmanned aerial vehicles based on flight condition correction according to claim 1, characterized in that, The water management fault coupling matrix described in step 4 is used to record the primary evidence relationship, secondary evidence relationship, and mutual exclusion relationship between the normal state, flooded state, and membrane dry state and the observed characteristics, and to record the secondary exclusion relationship between flight load disturbance and voltage transient change, in order to distinguish the root cause state of fuel cell water management and the secondary phenomenon caused by UAV load mutation.

7. The method for identifying water management faults in hydrogen-powered UAV fuel cells based on flight condition correction according to claim 1, characterized in that, The collaborative fusion described in step 4 is achieved using at least one of Bayesian fusion, DS evidence theory, weighted voting, or rule-constrained fusion, and maintains an early warning state when the water-flooded state, membrane dry state, and flight load disturbance conflict degree are greater than a preset threshold.

8. A fault identification system for water management in hydrogen-powered unmanned aerial vehicles (UAVs) based on flight condition correction, characterized in that, The system includes: The state acquisition module is used to acquire multi-source state quantities of the hydrogen energy UAV fuel cell system. The multi-source state quantities include at least the fuel cell stack state, water management supply and environmental state, and electric bus and propulsion load state. Any one of the multi-source state quantities is used as an auxiliary quantity for flight conditions and load disturbances. The multi-source state quantities are time-synchronized, filtered, and segmented into sliding windows to obtain window samples. The preprocessing and feature module is used to construct a normalized feature vector of working conditions based on the window samples described in the data acquisition and synchronization module. The diagnostic reasoning module is used to calculate the theoretical residual diagnostic evidence based on at least one water management mechanism model among the fuel cell polarization model, Ohmic internal resistance estimation model and gas mass transfer model; and to calculate the candidate probabilities of normal state, flooded state and membrane dry state based on the trained data classification model. The output and control module is used to collaboratively fuse the mechanism residual diagnostic evidence, candidate probabilities, and auxiliary quantities of flight conditions and load disturbances using the water management fault coupling matrix, output the water management status identification results, confidence level and fault level, and output alarm, power limit, return to base or protection control strategies through the safety control interface.

9. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

10. A computer device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method of any one of claims 1-7.

Images