A method and system for monitoring hydrogen fuel cells of a drone
By employing a multi-model collaborative monitoring method, flight attitude and vibration interference in the UAV hydrogen fuel cell system are eliminated, faults are accurately identified, and hydrogen consumption rate is dynamically corrected. This solves the problems of pressure measurement distortion and misjudgment in existing monitoring methods, thereby improving the operational reliability and flight safety of UAV hydrogen fuel cells.
Patent Information
- Application Number
- CN202610639524.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-11
- Publication Date
- 2026-06-30
AI Technical Summary
Existing drone-based hydrogen fuel cell monitoring methods cannot effectively distinguish the disturbance of flight attitude to the supply airflow field, resulting in distorted pressure measurements, inaccurate identification of vibration-induced hidden micro-leakage, high misjudgment rate, inability to perform dynamic response compensation and timing synchronization, distorted power matching judgment, inability to accurately identify faults, static endurance estimation, and easy overestimation of safe flight time.
By using multi-model collaborative monitoring, flight attitude data, airframe vibration data, flight control status data, and hydrogen fuel cell system parameters are collected. An airflow disturbance model, a vibration attenuation separation model, and a dynamic response compensation model are constructed to isolate flight attitude disturbances and vibration interference, separate micro-leakage attenuation, perform threshold discrimination and real-time power compensation, and calculate the hydrogen consumption rate and remaining safe flight time by combining the comprehensive state assessment model.
It achieves accurate fault identification, reduces the fault misjudgment rate, ensures the purity of the fuel cell stack inlet pressure data, adapts to the entire flight envelope, dynamically corrects the hydrogen consumption rate and remaining safe flight time, and improves the operational reliability of hydrogen fuel cells and the flight safety of UAVs.
Smart Images

Figure CN122314958A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of unmanned aerial vehicle (UAV) battery control technology, and in particular to a method and system for monitoring hydrogen fuel cells in UAVs. Background Technology
[0002] As the application scenarios of drones continue to expand, the demand for long endurance and high reliability is becoming increasingly urgent. Hydrogen fuel cells, due to their advantages such as high energy density and long endurance, have become the mainstream power choice for large and medium-sized drones. However, hydrogen fuel cell systems are complex to operate and are susceptible to various failures due to changes in drone flight attitude, continuous vibration of the airframe, airflow disturbances, and load fluctuations, which seriously threaten flight safety.
[0003] Existing monitoring methods have significant shortcomings. First, they fail to consider the disturbance of the airflow field caused by flight attitude, leading to distorted pressure measurements. Furthermore, they cannot effectively distinguish between normal hydrogen consumption and vibration-induced latent micro-leaks, resulting in high rates of false positives and false negatives. They also lack effective dynamic response compensation and timing synchronization, leading to distorted power matching judgments; they can only issue alarms but cannot locate faults, and their endurance calculations are static, easily overestimating safe flight time.
[0004] To address the aforementioned shortcomings, there is an urgent need for a comprehensive monitoring method that is adaptable to the entire flight envelope, accurate, reliable, and engineering-featured, in order to promote the widespread application of hydrogen fuel cells in the field of unmanned aerial vehicles (UAVs). Summary of the Invention
[0005] This invention addresses the high misjudgment rate of existing methods through multi-model collaborative monitoring, thereby accurately identifying faults, dynamically calculating flight range, and improving operational reliability and flight safety.
[0006] The technical solution proposed in this invention is: a method for monitoring hydrogen fuel cells using unmanned aerial vehicles (UAVs), the method comprising: Collect flight attitude data, airframe vibration data, flight control status data, and hydrogen fuel cell system parameters; Based on flight attitude data and hydrogen fuel cell system parameters, an airflow field disturbance model is constructed, and the hydrogen pressure signal is processed by feedforward noise reduction to obtain the stack inlet pressure. Based on the body vibration data and fuel cell stack inlet pressure, a vibration attenuation separation model is constructed to separate the attenuation of micro-leakage caused by vibration and perform threshold discrimination to obtain the latent leakage status. Based on flight control status data and latent leakage status, a dynamic response compensation model is constructed to compensate for the real-time power of the hydrogen fuel cell stack. The fault partitioning results are obtained based on the matching relationship between the compensated power and the load demand. By inputting the pressure supply status, latent leakage status, and fault zoning results into the integrated state assessment model, the hydrogen consumption rate and remaining safe flight time are calculated.
[0007] Preferably, the specific process for collecting the flight attitude data, the airframe vibration data, the flight control status data, and the hydrogen fuel cell system parameters is as follows: The drone uses an inertial measurement unit to collect flight attitude data in real time. The drone's flight control system uses built-in vibration sensors to collect real-time vibration data of the aircraft and flight control status data. The parameters of the hydrogen fuel cell system are collected by multiple sensors mounted on the system. The collected data is denoised and filtered to remove invalid and interfering data, resulting in valid data that can be used for subsequent modeling and analysis.
[0008] Preferably, the process of obtaining the airflow field disturbance model is as follows: Based on flight attitude data and hydrogen fuel cell system parameters, attitude disturbance characteristics, hydrogen supply pressure fluctuation characteristics and temperature influence characteristics are extracted respectively. The flow field coupling correlation algorithm is used to perform correlation analysis on the extracted attitude disturbance features, pressure fluctuation features and temperature influence features to explore the inherent correspondence between flight attitude and hydrogen supply flow field disturbance. Based on the results of the regularity analysis, multiple features are integrated and fitted to optimize the model, resulting in an airflow field disturbance model for pressure signal denoising.
[0009] Preferably, the process for obtaining the fuel cell stack inlet pressure is as follows: The raw hydrogen pressure data in the parameters of the hydrogen fuel cell system are processed by feedforward noise reduction through the airflow field disturbance model, which removes the flow field disturbance component caused by flight attitude disturbance and retains the real supply pressure change component. The noise-reduced pressure data is calibrated and verified to eliminate spurious fluctuations and output a clean and reliable fuel cell stack inlet pressure.
[0010] Preferably, the construction process of the vibration attenuation separation model is as follows: Based on the body vibration data and the fuel cell stack inlet pressure, vibration amplitude characteristics, vibration frequency characteristics and pressure attenuation change characteristics were extracted respectively. The attenuation component separation algorithm is used to perform coupled analysis on vibration amplitude characteristics, vibration frequency characteristics and pressure attenuation characteristics, distinguish the contribution ratio of normal hydrogen consumption attenuation and leakage-induced attenuation, and obtain the component division results. Based on the coupling analysis and component partitioning results, a vibration attenuation separation model that can identify vibration-induced micro-leakage is constructed.
[0011] Preferably, the process for obtaining the latent leakage state is as follows: The fuel cell stack inlet pressure is input into the vibration attenuation separation model, and the vibration-induced micro-leakage attenuation is obtained by decoupling. Based on the preset sealing safety threshold, the attenuation of micro-leakage is determined by threshold discrimination and status classification. The classification results are validated for rationality, and interference from operating conditions is eliminated to form a latent leakage state that characterizes whether the sealing surface is abnormal.
[0012] Preferably, the construction process of the dynamic response compensation model is as follows: Based on flight control status data and latent leakage status, characteristics of load mutation, flight mode, and the impact of leakage on response performance are extracted. A dynamic hysteresis fitting algorithm is used to analyze the hysteresis law between load demand and stack power output, and to quantify the response hysteresis time under different operating conditions. By integrating and modeling the hysteresis law and its impact characteristics, a dynamic response compensation model for power deviation correction is formed.
[0013] Preferably, the process for obtaining the fault partitioning results is as follows: The real-time power of the hydrogen fuel cell stack is compensated for hysteresis through a dynamic response compensation model. The matching degree between the real-time power of the compensated hydrogen fuel cell stack and the flight control load requirements is compared to distinguish between normal response lag and actual performance abnormality. By combining the degree of matching with the state of hidden leakage, the fault type classification and area location are completed, and the fault partitioning results are obtained.
[0014] Preferably, the process for obtaining the hydrogen consumption rate and the remaining safe flight time is as follows: By integrating the fuel cell stack inlet pressure, latent leakage status, and fault zoning results, a multi-factor comprehensive condition assessment model is constructed. Weighting coefficients are assigned based on the degree of influence of each state on hydrogen consumption, and the actual hydrogen consumption rate is calculated by combining the results. Based on the current hydrogen reserves and real-time flight conditions, and by incorporating these into the range estimation logic, the output hydrogen consumption rate and remaining safe flight time are calculated.
[0015] The present invention also provides a monitoring system for hydrogen fuel cells in unmanned aerial vehicles (UAVs), the system being used to execute the aforementioned method for monitoring hydrogen fuel cells in unmanned aerial vehicles, comprising: The data acquisition module includes an inertial measurement unit, a vibration sensor, and a multi-element sensor. The inertial measurement unit is used to acquire flight attitude data, the vibration sensor is used to acquire airframe vibration data and flight control status data, and the multi-element sensor is used to acquire parameters of the hydrogen fuel cell system. A data processing module, which is connected to the data acquisition module, is used to process the acquired data.
[0016] The beneficial effects of this invention are: 1. Based on flight attitude data and hydrogen fuel cell system parameters, a semi-empirical model containing explicit mapping polynomials is constructed. Combined with gravity compensation and interpolation optimization, feedforward noise reduction is applied to the hydrogen pressure signal. This accurately isolates the interference of flight attitude disturbances and gravity components on the pressure signal, eliminates measurement distortion, ensures the purity of the fuel cell stack inlet pressure data, and adapts to the entire flight envelope of the UAV. This provides reliable data support for subsequent micro-leakage identification and fault diagnosis, avoiding the invalidation of subsequent criteria due to data distortion.
[0017] 2. By combining the attenuation of micro-leakage with the separation of airframe vibration and fuel cell stack inlet pressure, and based on flight control status and leakage level, hysteresis compensation and timing alignment of fuel cell stack power are performed. This effectively decouples normal hydrogen consumption from vibration-induced micro-leakage attenuation, reducing the rate of false and false fault diagnosis. It also corrects deviations caused by electrochemical hysteresis and signal transmission delay, ensuring accurate power matching judgment and enabling precise identification of latent leaks and various faults, laying the foundation for fault zoning.
[0018] 3. Integrating multi-source parameters such as fuel cell stack inlet pressure, latent leakage status, and fault zoning results, the system employs the analytic hierarchy process (AHP) to determine weights and extra flight time by combining fault deterioration trends. This enables precise fault zoning and localization, dynamically corrects hydrogen consumption rates and remaining safe flight time, avoids deviations in static calculations, improves the operational reliability of hydrogen fuel cells and the flight safety of UAVs, and simultaneously adapts to the real-time and lightweight engineering application requirements of airborne platforms. Attached Figure Description
[0019] Figure 1 A flowchart of a method for monitoring hydrogen fuel cells using unmanned aerial vehicles (UAVs); Figure 2 A flowchart for constructing a vibration attenuation separation model for a monitoring method of hydrogen fuel cells using unmanned aerial vehicles (UAVs); Figure 3 A flowchart for obtaining the latent leakage status of a hydrogen fuel cell monitoring method using an unmanned aerial vehicle (UAV); Figure 4 A flowchart for constructing a dynamic response compensation model for a monitoring method of hydrogen fuel cells using unmanned aerial vehicles (UAVs). Detailed Implementation
[0020] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.
[0021] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.
[0022] like Figure 1 As shown, a method for monitoring hydrogen fuel cells in unmanned aerial vehicles (UAVs) includes: collecting flight attitude data, airframe vibration data, flight control status data, and hydrogen fuel cell system parameters; constructing an airflow disturbance model based on the flight attitude data and hydrogen fuel cell system parameters, performing feedforward noise reduction processing on the hydrogen pressure signal to obtain the fuel cell stack inlet pressure; constructing a vibration attenuation separation model based on the airframe vibration data and fuel cell stack inlet pressure, separating the attenuation of micro-leakage caused by vibration and performing threshold discrimination to obtain the latent leakage state; constructing a dynamic response compensation model based on the flight control status data and the latent leakage state, performing response lag compensation on the real-time power of the hydrogen fuel cell stack, and obtaining the fault partitioning result based on the matching relationship between the compensated power and the load demand; inputting the pressure supply state, latent leakage state, and fault partitioning result into a comprehensive state assessment model to calculate the hydrogen consumption rate and the remaining safe flight time.
[0023] Flight attitude data includes the UAV's pitch angle, roll angle, flight acceleration, and yaw angle, used to represent the attitude changes during UAV flight; body vibration data includes vibration frequency, vibration amplitude, and vibration phase, used to reflect the vibration state of the UAV body and fuel cell system during flight; flight control status data includes load demand signals, flight modes (such as hovering, cruise, acceleration, and deceleration), and flight control command parameters, used to reflect the UAV's flight requirements and operating status; hydrogen fuel cell system parameters include hydrogen pressure, hydrogen flow rate, stack terminal voltage, single-cell voltage distribution, stack current, stack temperature, and remaining hydrogen tank capacity, providing basic data for subsequent modeling and monitoring.
[0024] Flight attitude is a core factor affecting the external flow field distribution of a hydrogen fuel cell supply system. Under different attitudes such as climb, dive, turn, and side-flight, the local airflow velocity and static pressure field at the pipeline inlet differ significantly. This difference directly superimposes on the pressure sensor measurements, resulting in non-fault-related pressure fluctuations. Airframe vibration primarily originates from the operation of the power system, aerodynamic excitation, and flight control maneuvers. Vibrations are transmitted to the hydrogen supply pipeline, sealing joints, and fuel cell stack at different frequencies and amplitudes. This can induce fatigue damage to the sealing structure and cause high-frequency jitter in the pressure signal, masking actual leakage characteristics. Flight control status directly determines the stack load level and hydrogen consumption rate. The operating point of the fuel cell varies significantly under different flight modes, and the pressure decay pattern corresponding to normal hydrogen consumption also changes accordingly. Without dynamic baseline judgment based on flight control status, normal operating conditions can easily be misjudged as faults. The parameters of a hydrogen fuel cell system comprehensively reflect the internal operating status of the system. Among them, pressure and flow rate are directly related to the safety of hydrogen supply, the voltage distribution of a single cell is the most sensitive characterization of the health status of the stack, and temperature parameters affect electrochemical efficiency and material aging rate. All kinds of parameters are coupled with each other and together constitute a complete information chain for system status assessment.
[0025] The pressure supply status is determined by combining the fuel cell stack inlet pressure with a preset pressure threshold. When the fuel cell stack inlet pressure is within the preset normal threshold range, the pressure supply status is considered normal; when the fuel cell stack inlet pressure exceeds the preset threshold range, the pressure supply status is considered abnormal. The pressure supply status is one of the core basic parameters for comprehensive evaluation. The flight acceleration change rate refers to the degree of change in flight acceleration per unit time. It is obtained by calculating the acceleration difference between two adjacent sampling points from continuously collected flight acceleration data and then dividing by the sampling time interval. The frequency offset refers to the degree of deviation of the actual dominant frequency of the airframe vibration signal relative to a preset reference dominant frequency. It is obtained by performing spectral analysis on the vibration signal to obtain the actual dominant frequency and then subtracting it from the preset reference dominant frequency.
[0026] Stable supply pressure is a prerequisite for the continuous and reliable output power of hydrogen fuel cells. Excessive inlet pressure can lead to increased pipeline sealing load and accelerated material fatigue, while insufficient pressure can cause insufficient stack reaction, limited output power, or even shutdown. Therefore, using supply pressure as a fundamental indicator can quickly identify macroscopic anomalies in the gas supply system, providing a basis for subsequent refined fault isolation. The rate of change of flight acceleration reflects the impact intensity of UAV maneuvers. Under severe acceleration, deceleration, and rapid turns, the rate of change of acceleration increases significantly. The resulting inertial effect has an additional impact on the static pressure of hydrogen in the pipeline, which, if not compensated, will cause pressure measurement distortion. Frequency offset characterizes the degree of vibration anomaly. When the sealing structure becomes loose, cracked, or components interfere, the dominant vibration frequency shifts. Monitoring the frequency offset can identify abnormal trends in the mechanical structure in advance, providing an auxiliary criterion for early warning of vibration-induced micro-leakage.
[0027] Furthermore, the specific process for collecting flight attitude data, airframe vibration data, flight control status data, and hydrogen fuel cell system parameters is as follows: The inertial measurement unit on the UAV is used to collect flight attitude data in real time; the vibration sensor built into the UAV flight control system is used to collect body vibration data and flight control status data in real time; the multi-sensor on the hydrogen fuel cell system is used to collect parameters of the hydrogen fuel cell system; the collected data is processed by noise reduction and filtering to remove invalid interference data and obtain effective data that can be used for subsequent modeling and analysis.
[0028] Multi-source data acquisition is the foundation of the entire monitoring method, and its synchronization, accuracy, and anti-interference capability directly determine the accuracy of subsequent models and the reliability of fault diagnosis. In actual UAV platforms, various sensors are distributed in different physical locations, with different signal transmission paths and delay characteristics. Therefore, a unified acquisition architecture and timing mechanism must be adopted to ensure data consistency in time and space. The acquisition process consists of four parallel links: attitude acquisition, vibration acquisition, flight control command acquisition, and fuel cell system parameter acquisition. Each link works synchronously under the constraints of a hardware synchronization mechanism to avoid feature misalignment and logical misjudgment caused by asynchronous acquisition.
[0029] The inertial measurement unit (IMU) sampling frequency is set to 100-200Hz to ensure real-time capture of dynamic changes in the UAV's attitude. The acquired flight attitude data undergoes preliminary noise reduction using a Kalman filter algorithm to ensure data stability. The vibration sensor built into the flight control system is a piezoelectric vibration sensor, with a sampling frequency consistent with the IMU, synchronously acquiring airframe vibration data and flight control status data. The flight control status data is read in real-time via the flight control system's CAN bus, ensuring real-time and accurate data transmission.
[0030] As a core component for flight attitude perception, the IMU (Independent Monitoring Unit) needs to maintain a high refresh rate under high-dynamic flight conditions. A sampling frequency of 100-200Hz can cover the dynamic response range of most UAV maneuvers, avoiding the loss of attitude information due to insufficient sampling rate. Kalman filtering can effectively suppress internal noise and external electromagnetic interference of the IMU, smoothing raw signals such as pitch angle, roll angle, and acceleration, improving the reliability of attitude data. Piezoelectric vibration sensors are characterized by high sensitivity, good high-frequency characteristics, and strong shock resistance, making them suitable for the complex vibration environment of UAVs. Sampling at the same frequency as the IMU ensures strict alignment of attitude and vibration data, providing a consistent data source for subsequent flow field disturbance and vibration coupling analysis. The CAN bus, as a high-reliability communication bus commonly used in the UAV industry, has advantages such as strong anti-interference, low transmission latency, and multi-node collaborative operation. It can stably and in real-time upload key information such as flight control commands, load requirements, and flight modes, avoiding data packet loss and latency distortion.
[0031] Data from the IMU, vibration sensors, and CAN bus are all tagged with a unified timestamp based on the UAV main controller's hardware timer, and time axis alignment is achieved using cubic spline interpolation to ensure strict synchronization of multi-source data in both spatial and temporal dimensions. The multi-sensor system includes a pressure sensor, a miniature hydrogen flow meter, a hydrogen concentration sensor, a voltage acquisition module, a current sensor, a temperature sensor, and a hydrogen quantity sensor, which respectively collect data on hydrogen pressure, real-time hydrogen flow rate in the pipeline, hydrogen concentration around the pipeline, overall and individual cell voltages of the fuel cell stack, fuel cell stack current, fuel cell stack temperature, and remaining hydrogen tank capacity.
[0032] Different sensors and buses have inherent delays, making it difficult to achieve high-precision timing alignment solely through software synchronization. Therefore, a hardware timer on the main controller is used to generate a unified timestamp, ensuring synchronization of multi-source data from the hardware level. For a small amount of non-equal-interval sampling data, time axis reconstruction is performed through cubic spline interpolation, mapping all data to a unified time reference and eliminating phase shifts and feature misalignments. The multi-sensor array covers three dimensions: hydrogen supply safety, fuel cell stack operation, and environmental status. Among them, the pressure sensor monitors the stability of the supply pressure, the hydrogen flow meter accurately measures hydrogen consumption, the hydrogen concentration sensor is used for direct leakage verification, the voltage acquisition module obtains the core health indicator of single-chip voltage consistency, the current and temperature sensors reflect the electrochemical operating conditions, and the hydrogen quantity sensor provides the basis for calculating the remaining range. The entire sensor configuration comprehensively covers the key parameters required for fault characterization.
[0033] The denoising process uses a wavelet threshold denoising algorithm, and the screening process uses the 3σ criterion to remove outlier data points. The resulting valid data all meet the accuracy requirements of subsequent modeling, and the data error is significantly reduced compared to traditional methods.
[0034] Raw sensor data is inevitably affected by factors such as airborne electromagnetic interference, transient airflow impacts, and voltage fluctuations during actual acquisition, resulting in spikes, random noise, and outliers. Directly using this data for modeling can lead to model fitting bias, distorted feature extraction, and even misdiagnosis of faults. Wavelet thresholding denoising algorithms can effectively suppress high-frequency random noise while preserving the signal's abrupt change characteristics, making them particularly suitable for processing non-stationary signals such as pressure and vibration. The 3σ criterion identifies and removes outliers that significantly deviate from the normal distribution based on statistical laws, avoiding interference from extreme single-point data with the overall analysis results. The effective data after dual processing is characterized by low noise, good consistency, and high reliability, providing high-quality input for core algorithms such as airflow field disturbance models and vibration attenuation separation models.
[0035] Furthermore, the process of obtaining the airflow field perturbation model is as follows: Based on flight attitude data and hydrogen fuel cell system parameters, attitude disturbance features, hydrogen supply pressure fluctuation features, and temperature influence features are extracted respectively. A flow field coupling correlation algorithm is used to perform correlation analysis on the extracted attitude disturbance features, pressure fluctuation features, and temperature influence features to uncover the inherent correspondence between flight attitude and hydrogen supply airflow field disturbance. Based on the results of the pattern analysis, multiple features are integrated and fitted to optimize, resulting in an airflow field disturbance model for pressure signal denoising.
[0036] When a drone flies in an atmospheric environment, its pitch and roll attitude changes directly alter the local airflow direction and static pressure distribution at the hydrogen supply pipeline inlet, resulting in significant aerodynamic disturbance components superimposed on the pressure sensor measurements. Simultaneously, the inertial force generated by flight acceleration creates additional static pressure within the pipeline, further causing pressure measurement deviations. Both of these types of interference are unrelated to faults, and directly using them to determine leaks would lead to numerous false alarms. Therefore, this solution constructs an airflow field disturbance model to separate aerodynamic and gravitational disturbances from the total pressure signal, obtaining the true fuel cell stack inlet pressure.
[0037] The specific laws governing airflow disturbances identified in actual flight are as follows: Under the same flight attitude, the inlet static pressure increases in a quadratic curve with increasing pitch angle and linearly with increasing roll and yaw angle; under the same spatial attitude, the greater the flight acceleration, the greater the additional static pressure in the duct, and this relationship is linearly superimposed along the acceleration direction; simultaneously, under the same aerodynamic conditions, for every 10°C increase in the temperature difference between the fuel cell stack and the ambient temperature, the pressure measurement will produce a fixed offset. Based on these clearly defined physical laws, this scheme adopts a semi-empirical model based on CFD simulation and wind tunnel test data fitting to construct an explicit mapping relationship between disturbances and pressure.
[0038] The airflow disturbance model specifically adopts a second-order polynomial form, taking pitch angle, roll angle, three-axis flight acceleration components, and temperature difference as inputs, and disturbance correction coefficients as outputs. The model expression can be described as follows: the disturbance correction coefficient equals the quadratic coefficient of pitch angle multiplied by the square of pitch angle, plus the linear coefficient of pitch angle multiplied by pitch angle, plus the roll angle coefficient multiplied by roll angle, plus the corresponding compensation coefficients for each of the three-axis accelerations multiplied by the three-axis acceleration components, plus the temperature compensation coefficient multiplied by the temperature difference, and finally superimposed with a fixed bias term. All coefficients were determined by fitting multiple sets of experimental data using the least squares method, which can comprehensively characterize the combined influence of attitude, acceleration, and temperature on pressure measurement.
[0039] In model construction, a controlled variable method is used to keep temperature conditions constant, prioritizing the accuracy of attitude and acceleration perturbation fitting. A three-axis flight acceleration is introduced as a gravity compensation factor to directly decouple aerodynamic and inertial perturbations, enabling the model to adapt to any flight attitude, including vertical climb, dive, lateral flight, and hovering, covering the entire flight envelope. During feature processing, features with different dimensions, such as pitch angle, roll angle, acceleration, and temperature difference, are normalized to the same numerical range to eliminate calculation biases caused by differences in magnitude. After fitting and optimization, the model can output perturbation correction coefficients in real time based on the current flight state, performing feedforward correction on the original pressure signal to eliminate spurious fluctuations caused by attitude perturbations and gravity components, outputting pure and reliable fuel cell inlet pressure, providing an accurate data foundation for subsequent micro-leakage identification.
[0040] Furthermore, the process for obtaining the fuel cell stack inlet pressure is as follows: The raw hydrogen pressure data in the hydrogen fuel cell system parameters are processed by feedforward noise reduction using an airflow field disturbance model. This process removes the flow field interference components caused by flight attitude disturbances and retains the true supply pressure change components. The noise-reduced pressure data is then calibrated and verified to eliminate spurious fluctuation interference and output a clean and reliable stack inlet pressure.
[0041] Feedforward noise reduction achieves active compensation based on disturbance prediction, offering a faster response speed compared to traditional feedback filtering, and can complete corrections simultaneously with pressure fluctuations. By eliminating aerodynamic interference and the effects of gravity, only the true changes caused by hydrogen consumption and leakage are retained in the pressure signal, creating a prerequisite for vibration attenuation separation.
[0042] The feedforward noise reduction process adopts an attitude-based lookup table compensation method. It uses the disturbance correction coefficient output by the airflow field disturbance model to calibrate and correct the original hydrogen pressure. The correction follows the calculation relationship: the actual fuel cell stack inlet pressure = the original measured pressure - the aerodynamic disturbance component - the gravitational inertial disturbance component - the temperature drift component.
[0043] The airflow disturbance model adopts the above-mentioned second-order polynomial mapping relationship, and the specific formula is: Disturbance correction coefficient = pitch angle quadratic term coefficient × pitch angle squared + pitch angle linear term coefficient × pitch angle + roll angle coefficient × roll angle + X-axis acceleration coefficient × X-axis acceleration + Y-axis acceleration coefficient × Y-axis acceleration + Z-axis acceleration coefficient × Z-axis acceleration + temperature compensation coefficient × temperature difference + fixed bias term.
[0044] In this embodiment, by fitting multiple sets of wind tunnel and flight test data, the model coefficients were determined as follows: pitch angle quadratic coefficient 0.001, pitch angle linear coefficient 0.002, roll angle coefficient 0.0015, X-axis acceleration coefficient 0.008, Y-axis acceleration coefficient 0.006, Z-axis acceleration coefficient 0.012, temperature compensation coefficient 0.0005, and fixed bias term 0.003. After calculating the disturbance correction coefficient, it was further decomposed into three parts: aerodynamic disturbance component, gravitational inertial disturbance component, and temperature drift component, corresponding to attitude influence, acceleration influence, and temperature influence, respectively. The final actual fuel cell stack inlet pressure was calculated according to the following formula: Actual fuel cell stack inlet pressure = Original measured pressure − Total pressure disturbance value corresponding to the disturbance correction coefficient. Calibration processing uses a standard pressure sensor for periodic comparison to correct zero-point drift and sensitivity attenuation caused by long-term sensor use. The lookup table compensation calculation is highly efficient and suitable for embedded airborne real-time systems, achieving rapid correction through pre-stored multi-condition disturbance coefficients. Subsequently, a sliding window verification algorithm was used for further smoothing. The window length was set to 20 consecutive sampling points. The data within the window was smoothed by mean and outliers were removed to filter out pseudo fluctuations caused by instantaneous shocks, so that the final output fuel cell stack inlet pressure stably reflects the real gas supply status.
[0045] For example, at a certain moment, the drone collects raw hydrogen pressure measurements of 0.92 MPa, pitch angle of 10°, roll angle of 5°, X-axis acceleration of 0.5g, Y-axis acceleration of 0.2g, Z-axis acceleration of 0.3g, and the temperature difference between the fuel cell stack and the environment of 20°C. First, these data are substituted into the airflow field disturbance model to calculate the disturbance correction coefficient, which is 0.001 × 10⁻⁶. 2+0.002×10+0.0015×5+0.008×0.5+0.006×0.2+0.012×0.3+0.0005×20+0.003=0.1493. Then, according to the model structure, the disturbance correction coefficient is decomposed into aerodynamic disturbance component, gravitational inertial disturbance component, and temperature drift component. Among them, the aerodynamic disturbance component is 0.0893MPa, the gravitational inertial disturbance component is 0.04MPa, the temperature drift component is 0.02MPa, and the total pressure disturbance value is 0.1493MPa, which is consistent with the disturbance correction coefficient. Then, according to the feedforward noise reduction logic, the original measured pressure is subtracted from the total pressure disturbance value to obtain the real fuel cell stack inlet pressure of 0.7707MPa. Finally, through sliding window smoothing, the final stable and reliable fuel cell stack inlet pressure is output as 0.7707MPa.
[0046] Furthermore, the construction process of the vibration attenuation separation model is as follows: Based on the body vibration data and the fuel cell stack inlet pressure, vibration amplitude features, vibration frequency features, and pressure attenuation change features are extracted respectively; an attenuation component separation algorithm is used to perform coupled analysis on the vibration amplitude features, vibration frequency features, and pressure attenuation features, distinguish the contribution ratio of normal hydrogen consumption attenuation and leakage-induced attenuation, and obtain the component partitioning results; based on the coupling analysis and component partitioning results, a vibration attenuation separation model that can identify vibration-induced micro-leakage is constructed.
[0047] like Figure 2 As shown, vibration is one of the main causes of micro-leakage in the hydrogen supply system of unmanned aerial vehicles (UAVs). Vibration can lead to a decrease in the sealing surface fit and the propagation of micro-cracks, resulting in abnormal pressure decay. However, pressure decay also comes from normal hydrogen consumption of the fuel cell stack. Without effective separation, the system cannot determine the source of the decay. The core function of the vibration decay separation model is to decouple the vibration-induced micro-leakage component from the total decay, thereby achieving accurate identification of hidden leaks.
[0048] Vibration amplitude feature extraction includes vibration peak value, RMS value, and peak factor; vibration frequency feature extraction includes fundamental frequency, harmonic frequency, and frequency offset; pressure decay change features include pressure decay rate, decay amplitude, and decay period. Decay component separation is achieved based on frequency domain feature differences. Considering that the relationship between the normal hydrogen consumption rate of a fuel cell and the pressure decay slope is not a linear constant and is affected by multiple factors such as stack temperature, hydrogen purity, and membrane water content, calibrating the baseline using only a single load variable can lead to misjudgments under different operating conditions such as high and low temperatures. Therefore, orthogonal experimental calibration is used to construct a multidimensional baseline slope lookup table with load demand, stack temperature, and hydrogen supply temperature as input parameters.
[0049] Vibration peak value and effective value reflect vibration intensity, and peak factor can identify impact vibration; frequency characteristics are used to match the resonance range of the sealed structure, and pressure decay rate is the core indicator for judging whether there is a leak. Since the normal hydrogen consumption slope is affected by the coupling of multiple physical fields, a single variable baseline cannot adapt to all working conditions. Therefore, a three-maintenance formal baseline is adopted to cover key factors such as load and temperature. The coupling analysis of the attenuation component separation algorithm takes feature correlation, frequency domain decoupling, weight allocation, and error correction as its core process. The specific implementation process is as follows: First, the extracted vibration amplitude features, vibration frequency features, and pressure decay change features are standardized and preprocessed to eliminate dimensional differences. All feature values are mapped to the [0,1] interval to avoid coupling analysis deviations caused by differences in the magnitude of feature values. At the same time, the 3σ criterion is used to remove outliers in the feature data to ensure the reliability of the input features.
[0050] After preprocessing, the Pearson correlation coefficient method is used for feature correlation coupling analysis. The correlation coefficient between each vibration feature and the pressure attenuation feature is calculated, and vibration features with a strong correlation to pressure attenuation are screened out, typically features with an absolute correlation coefficient ≥ 0.6. If the absolute correlation coefficient is ≥ 0.6, the strongly correlated features such as vibration peak value, fundamental frequency, frequency shift, and pressure attenuation rate are directly retained. Then, a feature coupling matrix is constructed for the screened features to quantify the linear and nonlinear correlation between vibration features and pressure attenuation features, clarifying the intrinsic mapping relationship between vibration intensity, vibration frequency, and pressure attenuation. If not, the process returns to the previous step to re-screen features, recalculate feature correlation, and determine thresholds until strongly correlated features that meet the correlation coefficient threshold requirements are selected before proceeding to the subsequent feature coupling matrix construction step.
[0051] Based on this, frequency domain decoupling and separation are performed. The pressure attenuation signal and vibration signal are synchronously Fourier transformed to convert the time domain signal to the frequency domain. Decoupling is achieved by utilizing the difference in frequency domain characteristics between normal hydrogen consumption attenuation and vibration-induced leakage attenuation. The pressure attenuation signal corresponding to normal hydrogen consumption has energy concentrated in the low frequency band of 0.01Hz-0.1Hz, and the frequency domain waveform is smooth with no obvious peak. However, the pressure attenuation signal corresponding to vibration-induced micro-leakage will show obvious energy peak in a specific frequency band (500Hz-2kHz) corresponding to the resonant frequency of the sealing structure. Moreover, this peak has a fixed correlation with the fundamental frequency and harmonic frequency of vibration, that is, the leakage frequency domain peak is 2-3 times the fundamental frequency of vibration.
[0052] After frequency domain decoupling, the contribution ratio of the attenuation component is calculated. The low-frequency signal energy is recorded as the normal hydrogen consumption attenuation energy, and the high-frequency resonance signal energy is recorded as the leakage attenuation energy. The component values are calculated using a weighted summation formula. The calculation formula is as follows: the normal hydrogen consumption attenuation component is equal to the low-frequency energy ratio multiplied by the low-frequency weight coefficient, and the leakage attenuation component is equal to the high-frequency energy ratio multiplied by the high-frequency weight coefficient. The weight coefficient is determined by the correlation coefficient of the characteristic coupling matrix. The higher the correlation, the greater the weight.
[0053] To improve the accuracy of component partitioning, the model introduces an error correction factor to compensate for the calculation results. The error correction factor ranges from 0.95 to 1.05, keeping the overall error of the coupling analysis within 5%.
[0054] During the coupling analysis, a multidimensional baseline slope lookup table is used for auxiliary calculations. A trilinear interpolation algorithm is employed. When real-time parameters exceed the lookup table's defined domain, the nearest neighbor interpolation method is used to truncate boundary values. The truncated baseline slope is then amplified by a safety factor. This amplification factor is determined based on the standard deviation of slope fluctuations at the lookup table boundary, typically set to 1.2-1.5 times, to ensure coverage of boundary uncertainties. Simultaneously, an over-limit alarm is triggered to ensure the system does not crash due to interpolation failure. This trilinear interpolation method ensures smooth baseline output under continuous operating conditions. The safety factor amplification in the boundary extrapolation process conservatively improves the baseline slope, preventing false leak alarms due to underestimating normal hydrogen consumption caused by over-limit operating conditions, and effectively enhancing robustness under extreme operating conditions.
[0055] The final microleakage attenuation is obtained by subtracting the theoretical normal hydrogen consumption attenuation slope from the real-time pressure attenuation rate. The calculation formula is: Vibration-induced microleakage attenuation equals the real-time pressure attenuation rate minus the theoretical normal hydrogen consumption attenuation slope corresponding to the current operating condition. This completely decouples normal hydrogen consumption from vibration-induced attenuation.
[0056] Based on the aforementioned attenuation component separation algorithm, a four-layer vibration attenuation separation model was constructed, consisting of a feature input layer, a coupling analysis layer, a frequency domain decoupling layer, and an output layer. The model inputs include standardized vibration features, pressure attenuation features, and the theoretical normal hydrogen consumption attenuation slope. The model outputs vibration-induced micro-leakage attenuation, normal hydrogen consumption attenuation, and their respective contributions. The model was calibrated using experimental data, with a correlation coefficient threshold set at 0.6. Weight coefficients were dynamically allocated based on correlation, and the error correction factor was adaptively adjusted between 0.95 and 1.05 according to operating conditions. Parameters were iteratively optimized using the least squares method, and cross-validation ensured that the model did not overfit under hovering, cruising, or acceleration conditions. The final model was validated under extreme conditions, achieving a micro-leakage attenuation identification accuracy of no less than 95%, a false positive rate of no more than 3%, and a single-step calculation delay of no more than 10ms, making it stable for airborne real-time monitoring. This model achieves accurate separation of normal hydrogen consumption and vibration-induced micro-leakage through multi-dimensional coupling of vibration, pressure, frequency domain, and baseline, solving the problem that traditional methods cannot distinguish the source of attenuation and providing a reliable basis for subsequent latent leakage detection.
[0057] Furthermore, the process of obtaining the latent leakage state is as follows: The fuel cell stack intake pressure is input into the vibration attenuation separation model to decouple and obtain the vibration-induced micro-leakage attenuation amount. According to the preset sealing safety threshold, the micro-leakage attenuation amount is threshold-discriminated and classified into states. The classification results are verified for rationality, and operating condition interference is eliminated to form a latent leakage state characterizing whether the sealing surface is abnormal.
[0058] like Figure 3 As shown, the state of latent leakage is divided into three levels: no leakage, slight leakage, and obvious leakage, which provides a basis for subsequent fault zoning and safety early warning.
[0059] The decoupling process is based on frequency domain energy separation logic, extracting the attenuation component strongly correlated with vibration characteristics from the total pressure attenuation, reducing the cross-influence of normal hydrogen consumption and temperature changes. In this embodiment, the vibration attenuation separation model uses the following calculation formula: real-time pressure attenuation rate = (previous stack inlet pressure - current stack inlet pressure) ÷ sampling time interval; theoretical normal hydrogen consumption attenuation slope = multidimensional baseline lookup table and trilinear interpolation result; vibration-induced micro-leakage attenuation = real-time pressure attenuation rate - theoretical normal hydrogen consumption attenuation slope. The preset sealing safety thresholds are calibrated through system sealing design standards and multi-condition flight tests. The first threshold is set at 0.01 MPa / s, and the second threshold is set at 0.03 MPa / s. Micro-leakage attenuation below 0.01 MPa / s is considered no hidden leakage; between 0.01 MPa / s and 0.03 MPa / s is considered a slight hidden leakage; and above 0.03 MPa / s is considered a significant hidden leakage. The threshold division balances monitoring sensitivity and flight safety, avoiding missed detections and over-warnings. The rationality verification combines the current flight attitude, vibration intensity and load change trend, and uses a sliding window smoothing algorithm to smooth out suspected anomalies caused by instantaneous violent fluctuations, filter out sudden interference caused by single disturbances, reduce the probability of misjudgment caused by extreme working conditions, and improve the overall robustness of leakage status judgment.
[0060] For example, after feedforward noise reduction and calibration, the UAV obtains the current fuel cell stack inlet pressure as 0.7707 MPa and the previous inlet pressure as 0.7827 MPa, with a sampling interval of 0.1 seconds. Based on the model formula, the real-time pressure decay rate is calculated to be 0.125 MPa / s. Through multidimensional baseline slope lookup table and trilinear interpolation, the theoretical normal hydrogen consumption decay slope under the current operating condition is calculated to be 0.10 MPa / s. Further calculation shows that the vibration-induced micro-leakage decay is 0.025 MPa / s, which is within the threshold range of 0.01 MPa / s to 0.03 MPa / s. At the same time, the UAV is currently in cruise mode, with stable flight attitude, stable vibration amplitude, and no drastic load changes. The rationality verification is passed, and the latent leakage state is finally determined to be a minor leak.
[0061] Furthermore, the construction process of the dynamic response compensation model is as follows: Based on flight control status data and latent leakage status, load mutation characteristics, flight mode characteristics, and leakage impact characteristics on response performance are extracted. A dynamic hysteresis fitting algorithm is used to analyze the hysteresis law between load demand and fuel cell power output, and to quantify the response hysteresis time under different operating conditions. The hysteresis law and impact characteristics are fused and modeled to form a dynamic response compensation model for power deviation correction.
[0062] like Figure 4As shown, hydrogen fuel cells have inherent electrochemical hysteresis, and their power response is delayed when the load changes abruptly. If this is not compensated, it can lead to misjudgment of faults. The core function of the dynamic response compensation model is to eliminate the influence of this hysteresis while taking into account the sensor transmission delay, restoring the true power output capability of the fuel cell stack, and providing accurate data support for power matching judgment in subsequent fault partitioning.
[0063] Load mutation characteristics include load mutation amplitude, mutation rate, and duration. Flight modes are classified and identified as hovering, cruise, acceleration, and deceleration. The power response characteristics of the fuel cell stack differ under different flight modes. The impact of leakage on the response is converted into a corresponding influence coefficient according to the latent leakage level. The higher the leakage level, the larger the influence coefficient and the more obvious the suppression effect on the power response. The analysis process of the dynamic hysteresis fitting algorithm is based on multi-condition test data. First, load demand signals and real-time power output signals of the fuel cell stack under different load levels, different flight modes, and different latent leakage levels are collected through experiments. The time axis of the collected dual signals is strictly aligned. During the alignment process, the inherent delay effects of piezoelectric vibration sensors, IMU, and CAN bus transmission mentioned above are eliminated in advance to ensure the accuracy of the analysis object.
[0064] After alignment, a piecewise fitting method is used to conduct dynamic lag fitting analysis. The load demand signal is divided into two categories according to the type of abrupt change: step abrupt change and gradual abrupt change. For each type of abrupt change, different load intervals and leakage level intervals are divided. The time difference between load demand and power output in each interval is calculated, i.e., the initial lag time. Then, the initial lag time of each interval is fitted by the least squares method to obtain the nonlinear correlation equation between lag time and load level, leakage level, and flight mode. At the same time, the compensation coefficient under different operating conditions is calculated. The compensation coefficient is positively correlated with the lag time. The longer the lag time, the larger the compensation coefficient, thereby realizing the dynamic adaptation of the compensation intensity.
[0065] To ensure the reliability of the fitting results, an error calibration term is introduced during the fitting process to correct the fitting deviation caused by operating condition fluctuations and signal noise. The error calibration term is obtained by fitting multiple sets of measured data under standard operating conditions (fixed load, fixed flight mode, known leakage level) to ensure that the fitting error of the lag time is controlled within 1ms, which meets the real-time requirements of the airborne platform and avoids system oscillation caused by high-frequency compensation.
[0066] In this embodiment, the dynamic response compensation model uses the following calculation formula: the lag correction time equals the basic lag time plus the leakage influence coefficient multiplied by the lag correction factor; the real-time compensation coefficient equals the benchmark compensation coefficient plus the lag time multiplied by the compensation gain coefficient. The basic lag time, leakage influence coefficient, lag correction factor, benchmark compensation coefficient, and compensation gain coefficient are all determined through experimental data calibration, and the lag time has been pre-added with the overall transmission delay calibration value of the sensor network. The model input consists of pre-processed load mutation characteristics, flight mode characteristics, leakage influence coefficient, and lag law parameters; the model output is the real-time compensation coefficient and the lag correction time.
[0067] Subsequently, a model structure was built, employing a four-layer structure: feature input layer, hysteresis fitting layer, fusion modeling layer, and output layer. The input layer is responsible for receiving standardized multi-source features and hysteresis law parameters. The hysteresis fitting layer embeds the aforementioned dynamic hysteresis fitting algorithm, outputting the hysteresis duration and basic compensation coefficient under the corresponding operating conditions in real time. The fusion modeling layer deeply integrates the hysteresis law and leakage influence coefficient, dynamically adjusting the compensation coefficient according to the current flight mode, load status, and latent leakage level to ensure compensation accuracy and adaptability to operating conditions. The output layer outputs power deviation correction parameters, providing a reliable basis for subsequent power correction.
[0068] After the model is built, parameter calibration and optimization iterations are performed. Using multiple sets of test data under different operating conditions, the fitting coefficients, error calibration terms, and fusion weights in the model are calibrated to ensure that the compensation parameters output by the model are highly matched with the actual power lag. Then, through cross-validation, it is ensured that the model has good generalization ability under different flight conditions such as hovering, cruise, acceleration, and deceleration, as well as different leakage levels, to avoid overfitting.
[0069] Finally, extreme operating condition tests were conducted to verify the model's performance, ensuring that the model's response delay was ≤5ms and the compensation accuracy was ≥98%. This effectively eliminated the dual impact of electrochemical hysteresis and sensor transmission delay, ensured strict timing alignment, and significantly improved the accuracy of power matching judgment. This laid a solid foundation for distinguishing between normal response hysteresis and actual performance anomalies in subsequent fault partitioning, and further improved the reliability and engineering practicality of the entire monitoring system.
[0070] Furthermore, the process of obtaining the fault partitioning results is as follows: The real-time power of the hydrogen fuel cell stack is compensated for by a dynamic response compensation model to restore the theoretically required power level. The real-time power of the compensated hydrogen fuel cell stack is compared with the flight control load requirements to distinguish between normal response lag and actual performance anomalies. Based on the degree of matching and the state of hidden leakage, the fault type classification and area location are completed to obtain the fault zoning results.
[0071] Fault partitioning enables the upgrade from abnormal alarms to locating the source of the abnormality, facilitating rapid response.
[0072] The lag compensation adopts a correction method similar to the lead-lag link in the control loop. The power time series is time-aligned according to the lag duration given by the compensation model. The lag duration has been pre-added with the overall signal transmission delay calibration value of the sensor network to ensure the physical authenticity of the time alignment and avoid system oscillation caused by high-frequency correction. The combination of compensation coefficient and lag duration must satisfy the Nyquist stability criterion to ensure that the phase margin of the compensated system is kept above 30 degrees. The power amplitude is proportionally corrected according to the compensation coefficient to make the compensated power closer to the theoretical output level.
[0073] The matching degree calculation adopts the normalized mean square error method, which is the sum of the squares of the error between the compensated power and the load demand, divided by the sum of the square of the load demand power amplitude and the preset minimum value. The preset minimum value is 0.3% of the rated power. This completes the overflow prevention normalization process, eliminates the influence of the difference in error magnitude under different power levels, and avoids division by zero error when the load is extremely small. The relative error is used as the judgment basis, and the relative error threshold is set to 5%. A relative error within 5% is considered as normal response lag, and a relative error exceeding 5% is considered as having a real performance abnormality.
[0074] Fault partitioning is achieved by combining the hydrogen concentration sensor, pipeline hydrogen flow meter and fuel cell stack single-chip voltage consistency. The fuel cell stack single-chip voltage consistency data is resampled and low-pass filtered. The low-pass filter cutoff frequency is set to 0.5Hz. The resampling uses linear interpolation to reduce the voltage data sampling rate to 100Hz, so that it is consistent with the sampling rate of pressure signal and flow signal, ensuring time synchronization of multi-source data and avoiding the failure of judgment logic under rapid transient conditions caused by electrochemical hysteresis. The specific judgment logic is as follows: When the latent leakage state is normal, the hydrogen concentration and flow rate are not obviously abnormal, but the voltage consistency of the stack cells is poor and the power matching degree is insufficient, it is judged as an internal performance degradation fault of the stack; when the latent leakage state is abnormal, the local hydrogen concentration is too high and the power matching degree is insufficient, it is judged as a micro-leakage fault of the sealing surface or pipeline; when the pressure supply state is abnormal, the pipeline flow rate fluctuates significantly and the power matching degree is insufficient, it is judged as an abnormal pressure fault of the gas supply system; when the latent leakage state is normal, the pressure supply state is normal, there is no local hydrogen concentration abnormality, but the voltage consistency of the stack cells is extremely poor and the power matching degree is insufficient, it is judged as a sudden performance fault of the stack body, and the fault area is located as the membrane electrode or catalyst layer. This forms a logically consistent, comprehensive, and physically realistic fault zoning result.
[0075] In this embodiment, the lag time calculated by the dynamic response compensation model is 35ms, the compensation coefficient is 1.035, and after power timing alignment correction, the compensated power is 1242W, the load demand is 1200W, and the relative error calculated by the normalized mean square error formula is 3.5%, which is less than the 5% judgment threshold, indicating normal power matching. The current fuel cell stack inlet pressure is 0.7707MPa, which is within the normal pressure range, and the pipeline hydrogen flow fluctuation is less than 2%, with no abnormal fluctuations. The hydrogen concentration sensor detects a local hydrogen concentration of 20ppm, which is lower than the safety alarm threshold. The vibration attenuation separation model decouples the vibration-induced micro-leakage attenuation amount to 0.025MPa / s, and the latent leakage state is determined to be a minor leak. The standard deviation of the fuel cell stack single-cell voltage is 15mV, which does not exceed the abnormal threshold of 20mV, and the maximum and minimum cell voltage difference is 8% of the rated single-cell voltage, which does not exceed the critical threshold of 10%, indicating normal voltage consistency. Based on the above multi-dimensional status information and the fault zoning judgment rules, the current system fault type is finally determined to be a sealing surface or pipeline micro-leakage fault, thus achieving accurate fault location and zoning output.
[0076] Furthermore, the process for obtaining the hydrogen consumption rate and remaining safe flight time is as follows: By integrating the results of fuel cell stack inlet pressure, latent leakage status, and fault zoning, a multi-factor comprehensive state assessment model is constructed. Weighting coefficients are assigned according to the degree of influence of each state on hydrogen consumption, and the actual hydrogen consumption rate is calculated. Combining the current hydrogen reserves and real-time flight conditions, the model is substituted into the endurance calculation logic to output the hydrogen consumption rate and the remaining safe flight time.
[0077] The multi-factor integrated state assessment model takes the fuel cell stack inlet pressure, latent leakage status, and fault zoning results as the main inputs, while introducing flight mode and load demand as auxiliary influencing factors to ensure that the model can comprehensively cover various key operating conditions that affect hydrogen consumption. Among them, the fuel cell stack inlet pressure directly determines the basic hydrogen supply rate, the latent leakage status reflects the degree of leakage loss, the fault zoning results reflect the additional hydrogen consumption impact of faults, and the flight mode and load demand determine the basic hydrogen consumption level of the fuel cell stack. The synergistic effect of these factors constitutes a complete hydrogen consumption impact system.
[0078] The model employs the Analytic Hierarchy Process (AHP) to determine the weights of each factor. A judgment matrix is constructed based on the degree of influence and fitted with experimental data. Each weight satisfies a consistency ratio (CR) of less than 0.1. Specifically, the weights are allocated as follows: stack inlet pressure 0.35, latent leakage state 0.3, fault zoning result 0.25, and flight mode and load requirements 0.1. Higher leakage and fault levels correspond to larger weighting coefficients. The actual hydrogen consumption rate is calculated using a weighted summation formula: the actual hydrogen consumption rate equals the base hydrogen consumption multiplied by the stack inlet pressure weight, plus the leakage loss multiplied by the latent leakage state weight, plus the additional hydrogen consumption due to faults multiplied by the fault zoning result weight, plus the baseline hydrogen consumption weight corresponding to the flight mode and load requirements. The base hydrogen consumption is determined by looking up a table based on the current flight mode, load requirements, and stack inlet pressure. Leakage loss is assigned values according to three levels: no leakage, minor leakage, and significant leakage. The additional hydrogen consumption due to faults is determined based on the fault zoning result. The calculation of remaining safe flight time incorporates a degradation correction factor. The specific formula is: remaining safe flight time equals the current remaining hydrogen storage in the hydrogen cylinder divided by the product of the actual hydrogen consumption rate and the degradation coefficient. The degradation coefficient is determined based on the changing trend of micro-leakage attenuation. A linear regression slope is calculated using a sliding time window, and the slope is compared with a preset degradation threshold to determine if there is an accelerating degradation trend. If there is accelerating degradation, the degradation coefficient is increased to shorten the safe flight time; if there is a stable state, the degradation coefficient is set to 1. In this embodiment, the sliding window is 30 seconds, the preset degradation threshold corresponds to the critical rate of crack propagation in the seal, and the low-pass filter cutoff frequency is set to 0.5Hz to filter out instantaneous vibration interference.
[0079] For example, the current fuel cell stack inlet pressure is 0.7707 MPa, the latent leakage status is minor leakage, the fault zoning result is a micro-leakage fault on the sealing surface or pipeline, the flight mode is cruise, and the load requirement is 1200W; the model shows a base hydrogen consumption of 18 L / min, a leakage loss of 1.2 L / min, an additional hydrogen consumption due to faults of 0.8 L / min, and a base hydrogen consumption of 17.5 L / min for flight mode and load requirement; substituting these values into the weighted summation formula, the actual hydrogen consumption rate is 18 multiplied by 0.35 plus 1.2 multiplied by 0.3 plus 0.8 multiplied by 0.25 plus 17.5 multiplied by 0.1, resulting in 8.61 L / min. The current remaining hydrogen storage in the hydrogen cylinder is 380L. The slope of the micro-leakage decay is calculated using a 30-second sliding window. The slope is less than the preset deterioration threshold, indicating no accelerated deterioration trend. The deterioration coefficient is set to 1. The remaining safe flight time is equal to 380L divided by 8.61L / min, which is approximately 44.13 minutes. After rounding and safety redundancy correction, the final output of the remaining safe flight time is 44 minutes.
[0080] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.
[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0082] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.
Claims
1. A method for monitoring hydrogen fuel cells using an unmanned aerial vehicle (UAV), characterized in that, The method includes: Collect flight attitude data, airframe vibration data, flight control status data, and hydrogen fuel cell system parameters; Based on flight attitude data and hydrogen fuel cell system parameters, an airflow field disturbance model is constructed, and the hydrogen pressure signal is processed by feedforward noise reduction to obtain the stack inlet pressure. Based on the body vibration data and fuel cell stack inlet pressure, a vibration attenuation separation model is constructed to separate the attenuation of micro-leakage caused by vibration and perform threshold discrimination to obtain the latent leakage status. Based on flight control status data and latent leakage status, a dynamic response compensation model is constructed to compensate for the real-time power of the hydrogen fuel cell stack. The fault partitioning results are obtained based on the matching relationship between the compensated power and the load demand. By inputting the pressure supply status, latent leakage status, and fault zoning results into the integrated state assessment model, the hydrogen consumption rate and remaining safe flight time are calculated.
2. The method for monitoring hydrogen fuel cells using an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The specific process for collecting the flight attitude data, the airframe vibration data, the flight control status data, and the hydrogen fuel cell system parameters is as follows: The drone uses an inertial measurement unit to collect flight attitude data in real time. The drone's flight control system uses built-in vibration sensors to collect real-time vibration data of the aircraft and flight control status data. The parameters of the hydrogen fuel cell system are collected by multiple sensors mounted on the system. The collected data is denoised and filtered to remove invalid and interfering data, resulting in valid data that can be used for subsequent modeling and analysis.
3. The method for monitoring hydrogen fuel cells using a drone according to claim 2, characterized in that, The process of obtaining the airflow field disturbance model is as follows: Based on flight attitude data and hydrogen fuel cell system parameters, attitude disturbance characteristics, hydrogen supply pressure fluctuation characteristics and temperature influence characteristics are extracted respectively. The flow field coupling correlation algorithm is used to perform correlation analysis on the extracted attitude disturbance features, pressure fluctuation features and temperature influence features to explore the inherent correspondence between flight attitude and hydrogen supply flow field disturbance. Based on the results of the regularity analysis, multiple features are integrated and fitted to optimize the model, resulting in an airflow field disturbance model for pressure signal denoising.
4. The method for monitoring hydrogen fuel cells using a drone according to claim 3, characterized in that, The process of obtaining the fuel cell stack inlet pressure is as follows: The raw hydrogen pressure data in the parameters of the hydrogen fuel cell system are processed by feedforward noise reduction through the airflow field disturbance model, which removes the flow field disturbance component caused by flight attitude disturbance and retains the real supply pressure change component. The noise-reduced pressure data is calibrated and verified to eliminate spurious fluctuations and output a clean and reliable fuel cell stack inlet pressure.
5. The method for monitoring hydrogen fuel cells using a drone according to claim 4, characterized in that, The construction process of the vibration attenuation separation model is as follows: Based on the body vibration data and the fuel cell stack inlet pressure, vibration amplitude characteristics, vibration frequency characteristics and pressure attenuation change characteristics were extracted respectively. The attenuation component separation algorithm is used to perform coupled analysis on vibration amplitude characteristics, vibration frequency characteristics and pressure attenuation characteristics, distinguish the contribution ratio of normal hydrogen consumption attenuation and leakage-induced attenuation, and obtain the component division results. Based on the coupling analysis and component partitioning results, a vibration attenuation separation model that can identify vibration-induced micro-leakage is constructed.
6. The method for monitoring hydrogen fuel cells using a drone according to claim 5, characterized in that, The process of obtaining the latent leakage state is as follows: The fuel cell stack inlet pressure is input into the vibration attenuation separation model, and the vibration-induced micro-leakage attenuation is obtained by decoupling. Based on the preset sealing safety threshold, the attenuation of micro-leakage is determined by threshold discrimination and status classification. The classification results are validated for rationality, and interference from operating conditions is eliminated to form a latent leakage state that characterizes whether the sealing surface is abnormal.
7. A method for monitoring hydrogen fuel cells using an unmanned aerial vehicle (UAV) according to claim 6, characterized in that, The construction process of the dynamic response compensation model is as follows: Based on flight control status data and latent leakage status, characteristics of load mutation, flight mode, and the impact of leakage on response performance are extracted. A dynamic hysteresis fitting algorithm is used to analyze the hysteresis law between load demand and stack power output, and to quantify the response hysteresis time under different operating conditions. By integrating and modeling the hysteresis law and its impact characteristics, a dynamic response compensation model for power deviation correction is formed.
8. The method for monitoring hydrogen fuel cells using a drone according to claim 7, characterized in that, The process of obtaining the fault partitioning results is as follows: The real-time power of the hydrogen fuel cell stack is compensated for hysteresis through a dynamic response compensation model. The matching degree between the real-time power of the compensated hydrogen fuel cell stack and the flight control load requirements is compared to distinguish between normal response lag and actual performance abnormality. By combining the degree of matching with the state of hidden leakage, the fault type classification and area location are completed, and the fault partitioning results are obtained.
9. A method for monitoring hydrogen fuel cells using an unmanned aerial vehicle (UAV) according to claim 8, characterized in that, The process for obtaining the hydrogen consumption rate and the remaining safe flight time is as follows: By integrating the fuel cell stack inlet pressure, latent leakage status, and fault zoning results, a multi-factor comprehensive condition assessment model is constructed. Weighting coefficients are assigned based on the degree of influence of each state on hydrogen consumption, and the actual hydrogen consumption rate is calculated by combining the results. Based on the current hydrogen reserves and real-time flight conditions, and by incorporating these into the range estimation logic, the output hydrogen consumption rate and remaining safe flight time are calculated.
10. A drone hydrogen fuel cell monitoring system, the system being used to execute the drone hydrogen fuel cell monitoring method according to any one of claims 1-9, characterized in that, include: The data acquisition module includes an inertial measurement unit, a vibration sensor, and a multi-element sensor. The inertial measurement unit is used to acquire flight attitude data, the vibration sensor is used to acquire airframe vibration data and flight control status data, and the multi-element sensor is used to acquire parameters of the hydrogen fuel cell system. A data processing module, which is connected to the data acquisition module, is used to process the acquired data.
