Hydrogen energy unmanned aerial vehicle return control method based on fault diagnosis and path re-planning

Abstract

The application relates to the technical field of unmanned aerial vehicle (UAV) return control, and discloses a hydrogen energy UAV return control method based on fault diagnosis and path re-planning, which comprises the following steps: obtaining state information of the UAV, performing hierarchical fault diagnosis based on the state information, performing feature extraction on the state information, obtaining primary fault features, and inputting the primary fault features into a pre-trained fault diagnosis model. The hydrogen energy UAV return control method based on fault diagnosis and path re-planning can accurately identify fault modes and severity levels through hierarchical fault diagnosis, generate dynamic capability constraints containing key parameters such as flight speed, climbing angle and rolling angle, embed the constraints as rigid conditions into the path re-planning process, and ensure that each flight section of the generated return path does not exceed the actual flight capability boundary of the UAV after the fault, so that the problem that path planning and flight capability are separated in the traditional method is fundamentally solved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) return-to-home control technology, and in particular to a hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning. Background Technology

[0002] Hydrogen-powered drones, with their advantages of long endurance, high energy density, and zero emissions, have shown broad application prospects in long-endurance industrial applications such as power transmission line inspection, pipeline inspection, and geographic mapping. Especially in complex terrain environments such as mountainous areas and plateaus, hydrogen-powered drones have a significant endurance advantage over traditional lithium-ion battery drones.

[0003] However, the fuel cell systems of hydrogen-powered drones are highly sensitive to environmental conditions. In low-temperature environments, the incidence of malfunctions such as membrane electrode icing and gas path blockage increases significantly. When the fuel cell malfunctions, its output power drops noticeably, directly impacting the drone's flight performance. Traditional drone return-to-home control strategies typically employ a pre-defined global return-to-home path, triggering a straight-line return when insufficient battery or hydrogen is detected. However, under fault conditions, the drone's actual flight capabilities have changed, and the pre-planned path may not meet the dynamic constraints after the fault, causing the drone to crash during its return journey due to being unable to overcome terrain obstacles or lacking sufficient power.

[0004] Existing technologies offer some control methods for UAVs after malfunctions, but they lack precise diagnosis of malfunction types and struggle to quantify the specific impact of malfunctions on flight capabilities. Furthermore, path planning is disconnected from flight capability constraints, potentially resulting in planned paths exceeding the actual capabilities of the UAV after a malfunction. Additionally, they cannot adjust return-to-home strategies in real-time based on changes in malfunction severity. Therefore, ensuring the safe return of UAVs in the event of fuel cell failure is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] The technical problem to be solved by this invention is that the pre-planned path in the prior art may not meet the dynamic constraints after a failure. To address this, we propose a return-to-home control method for hydrogen-powered UAVs based on fault diagnosis and path replanning.

[0006] To achieve the above objectives, this application adopts the following technical solution: a hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning, comprising the following steps:

[0007] The system acquires the state information of the UAV, performs hierarchical fault diagnosis based on the state information, extracts features from the state information to obtain primary fault features, inputs the primary fault features into a pre-trained fault diagnosis model to obtain the confidence vector of each fault category and the fault severity prediction result, and determines the current fault mode and the corresponding fault severity level according to the preset judgment rules.

[0008] Starting from the current location of the UAV and setting the preset return point as the target point, the path is replanned using dynamic capability constraints as rigid constraints to generate a return path from the starting point to the target point. The return path consists of waypoints, and each segment of the return path satisfies the dynamic capability constraints. The UAV is then controlled to fly along the return path.

[0009] Preferably, the status information includes the fuel cell stack's output voltage, output current, pressure difference between the hydrogen path and the gas path, cell voltage data, flight attitude data, and control surface feedback data.

[0010] Preferably, the rate of change of the ratio of fuel cell output voltage to output current is calculated as a characteristic of power decay rate; the pressure difference between hydrogen path and gas path is calculated as a characteristic of gas path blockage; and the ratio of the standard deviation to the mean of the voltage of all individual cells in the fuel cell stack is calculated as a single cell voltage inconsistency coefficient.

[0011] Preferably, the step of fusing the primary fault features with the output of the fault diagnosis model to make a judgment, and determining the current fault mode and the corresponding fault severity level according to the preset judgment rules, specifically includes: obtaining the confidence vector output by the fault diagnosis model in each diagnosis cycle;

[0012] If the confidence level of a certain fault category reaches or exceeds the high confidence threshold, it is directly identified as that fault; if the confidence level of all categories is less than the high confidence threshold, but for a certain category its confidence level reaches or exceeds the low confidence threshold and its historical occurrence frequency within the sliding window reaches or exceeds the preset frequency threshold, it is strongly identified as that fault; if none of the above conditions are met, no clear fault category is generated in the current cycle, and the judgment of the previous cycle is maintained or the safety mode is executed.

[0013] The high confidence threshold is greater than the low confidence threshold, and both are between 0 and 1.

[0014] Preferably, the basic capability constraints are obtained by looking up a table based on the current fault mode and fault severity level and then dynamically corrected to obtain the corrected capability constraints; the remaining driving time is estimated by combining the remaining hydrogen quantity and the power consumption model under the current fault mode, and the remaining driving time is incorporated into the dynamic capability constraints.

[0015] Preferably, the process involves using the current location of the UAV as the starting point and a preset return point as the target point, performing path replanning with dynamic capability constraints as rigid constraints to generate a return path from the starting point to the target point. The feasibility of the newly generated flight segment is then checked based on the current dynamic capability constraints. If the check fails, the current sampling point is discarded and resampled. The process iteratively searches until a feasible path from the starting point to the target point is found, and the path cost is optimized to generate a return path consisting of several waypoints.

[0016] Preferably, the required flight speed, climb angle, and roll angle for the flight segment from the current node to the new sampling point are calculated; the calculated flight speed, climb angle, and roll angle are compared with the maximum flight speed, maximum climb angle, and maximum roll angle in the dynamic capability constraints; if any indicator exceeds the corresponding maximum allowable value, the flight segment is determined to be infeasible.

[0017] Preferably, after generating the return path, a simplified UAV dynamics model is invoked to perform a forward simulation along the planned path, checking whether the angular velocity and acceleration at each step exceed the limits; the remaining hydrogen, the power consumption model under the current fault mode, and the estimated total energy consumption of the path are compared; if the estimated energy consumption exceeds a preset proportion of the remaining available energy, the remaining range is determined to be insufficient; if the verification passes, the generated waypoint sequence is sent to the underlying flight controller to execute the flight; if the verification fails, a secondary replanning is triggered or a preset emergency landing procedure is executed.

[0018] Preferably, if the severity of the fault is further deteriorated during the flight of the UAV along the return path, a second replanning is triggered to regenerate the return path according to the dynamic capability constraints; if the second replanning fails or an extreme fault of complete loss of power is diagnosed, a preset emergency landing procedure is invoked to find a flat area in the current airspace and perform a vertical forced landing.

[0019] The technical effects and advantages of this invention are as follows:

[0020] This invention accurately identifies fault modes and severity levels through hierarchical fault diagnosis, thereby generating dynamic capability constraints that include key parameters such as flight speed, climb angle, and roll angle. These constraints are then embedded as rigid conditions into the path replanning process, ensuring that each segment of the generated return path does not exceed the actual flight capability boundary of the UAV after the fault. This fundamentally solves the problem of the separation between path planning and flight capability in traditional methods, and significantly improves the autonomous return success rate and mission preservation capability of hydrogen-powered UAVs under complex fault conditions such as low-temperature mountainous areas. Attached Figure Description

[0021] The disclosure of this invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. In the drawings, the same reference numerals are used to refer to the same parts:

[0022] Figure 1 This is a diagram of the hydrogen-powered drone return-to-home control architecture of the present invention;

[0023] Figure 2 This is a flowchart of the fault diagnosis process for hydrogen-powered drones according to the present invention;

[0024] Figure 3 This is a logic block diagram of the fault diagnosis and path replanning of the hydrogen-powered drone of the present invention. Detailed Implementation

[0025] It is readily understood that, based on the technical solution of this invention, those skilled in the art can propose various interchangeable structural methods and implementations without altering the essential spirit of the invention. Therefore, the following detailed embodiments and accompanying drawings are merely illustrative examples of the technical solution of this invention and should not be considered as the entirety of the invention or as limitations or restrictions on the technical solution of this invention.

[0026] When hydrogen-powered drones perform tasks such as logistics and inspection, their flight control systems must possess high reliability. In the event of a malfunction, accurately assessing the severity of the failure and planning a feasible return path for the drone within its current capabilities is crucial for ensuring flight safety. Traditional fixed-path or simple straight-line return methods are insufficient in this regard.

[0027] Reference Figure 1-3 As shown, this invention provides a technical solution: a hydrogen-powered unmanned aerial vehicle (UAV) return-to-home control method based on fault diagnosis and path replanning, the method comprising:

[0028] Hierarchical fault diagnosis is performed based on the acquired UAV status information to determine the current fault mode and the corresponding dynamic capability constraints.

[0029] By taking dynamic capability constraints as input, real-time path replanning is performed based on preset global waypoints to generate a return path that meets the dynamic capability constraints.

[0030] Control the drone to fly along the return path.

[0031] The three steps described above are executed sequentially. A hierarchical diagnostic mechanism first quantifies the impact of the fault on flight capability, obtaining dynamic capability constraints. These constraints are then embedded as rigid conditions into the path planning algorithm, thereby generating a return path. This approach fundamentally solves the problem of mismatch between post-fault path planning and real-time flight capability, ensuring a high return success rate.

[0032] Feature extraction is performed on the state information to obtain primary fault features. The specific operations are as follows:

[0033] The processor reads raw state information from the fuel cell voltage and current sensors, IMU, GPS, and servo feedback sensors via the data bus, and then calls a feature extraction algorithm preset in memory to process the raw data. Specifically, this includes: calculating the rate of change of the ratio of fuel cell output voltage to output current as a power decay rate feature; calculating the residual between the gyroscope reading and the expected model output as an attitude sensor anomaly feature; and calculating the difference between the control surface deflection command and the actual feedback position as an actuator deviation feature.

[0034] The specific rules for quantifying primary fault characteristics are as follows: The processor reads raw state information from the fuel cell voltage and current sensors, IMU, GPS, and servo feedback sensors via the data bus. Then, the processor calls a feature extraction algorithm pre-installed in memory to process the raw data.

[0035] The ratio of fuel cell output voltage to output current is calculated as the power decay rate characteristic; the residual between gyroscope readings and expected model output is calculated as the attitude sensor anomaly characteristic; and the difference between the control surface deflection command and the actual feedback position is calculated as the actuator deviation characteristic.

[0036] The specific rules for the quantitative calculation of primary fault characteristics are as follows:

[0037] The rate of change of the real-time internal resistance of a fuel cell relative to its rated operating condition internal resistance is quantified, and the calculation formula is as follows: ,in For the real-time output internal resistance of the fuel cell, To output the total voltage of the fuel cell stack in real time. To output the total current in real time, The nominal internal resistance of the fuel cell under rated operating conditions is used. The rate of change is calculated using a sliding window filter with a window length of 10 sampling periods and a sampling period of 10ms. The average value of the rate of change within the window is taken as the final power decay rate characteristic value to eliminate the influence of instantaneous sampling noise.

[0038] The nominal model of the UAV's rigid body dynamics with degrees of freedom is used as the expected model. The root mean square residual of the gyroscope's three-axis angular velocity measurements and the model's expected output values ​​is used as an anomaly feature, calculated using the following formula: ,in For real-time measurement of angular velocity along the xyz axis of a gyroscope. The rigid body dynamics model is based on the expected three-axis angular velocities output by the current control surface commands and dynamic state. The residual calculation uses a 200ms sliding window, and the maximum value of the root mean square of the residual within the window is taken as the abnormal characteristic value of the attitude sensor.

[0039] The calculation is performed using the absolute value of the relative deviation between the control surface deflection command and the actual feedback position, combined with the dead zone threshold. The calculation formula is as follows: ,in The control surface deflection command angle output by the flight controller. This refers to the actual deflection angle of the control surface as fed back by the servo motor. The preset dead zone threshold for the corresponding servo motor is set to ±0.5° using the servo motor's factory calibration data. When the calculated result is less than 0, 0 is taken as the final actuator deviation characteristic value to eliminate the misjudgment effect of the servo motor's inherent dead zone. These calculated indicators constitute the primary fault feature vector.

[0040] The initial fault features are input into a pre-trained fault diagnosis model, which outputs a preliminary fault classification and confidence score. The resulting feature vector is then input into a Support Vector Machine (SVM) stored in memory. This model, trained on a large amount of historical fault data, can map the feature vectors to different fault categories, including slight power loss in the power system, severe power loss in the power system, left aileron jamming, and GPS signal loss, and outputs a confidence score representing the degree of confidence in the classification.

[0041] The model employs the C-SVC classification mode, using the radial basis function (RBF) as the kernel function. The kernel parameter gamma is set to the reciprocal of the dimension of the input feature vector, i.e., gamma = 1 / 3, and the penalty coefficient C is set to 100. The input feature vector is a 3D vector composed of the aforementioned power attenuation rate feature, attitude sensor anomaly feature, and actuator deviation feature. Before input, it undergoes min-max normalization processing, with a normalization interval of [0,1]. The upper and lower limits of normalization are pre-calibrated based on the full range of each feature. The model training dataset contains historical fault data throughout the entire lifecycle of the hydrogen-powered UAV, covering four major categories and 12 subcategories of fault scenarios: power attenuation of the power system, ailerons, elevators, rudder jamming and performance reduction, GPS signal loss, jumps, and IMU anomalies. The sample size for each fault scenario is no less than 1000 sets, and the dataset is divided into training and test sets in a 7:3 ratio before training. The model confidence score is calculated by using the Platt scaling method to convert the SVM classification output into a probability value in the range of 0-1, which is used as the confidence score for the corresponding fault category. The closer the confidence score is to 1, the higher the reliability of the classification result.

[0042] The processor performs forward propagation calculations on the model to obtain preliminary diagnostic results.

[0043] Based on the preliminary fault classification and confidence level, a fusion judgment is performed using a preset fault decision logic to determine the final fault mode and level. A decision logic module runs in the processor: if the confidence level of a fault category is higher than a first threshold, it is directly identified as that fault; if the confidence levels of multiple categories are close and all higher than a second threshold, a comprehensive judgment is made based on the temporal characteristics of the state information, or it is identified as a more severe composite fault mode. The final output is the determined fault mode and the corresponding dynamic capability constraints. Through serial processing, a progressive analysis from raw data to accurate diagnosis is achieved, effectively reducing the false positive rate.

[0044] Let the first threshold be The second threshold is ,satisfy Typical value is , In each diagnostic cycle The fault diagnosis model outputs a confidence vector. ,in The total number of preset fault categories, This represents the confidence level that the fault belongs to type i at the current moment.

[0045] Define the sliding window length Used to store the most recent The historical judgment results of each diagnostic cycle. Let the number of occurrences of the i-th type of fault in the historical window be denoted as . Then its frequency of occurrence is The historical judgment results are obtained as follows: for each historical period, if a certain confidence level reaches... Then it is recorded as that class; if the confidence of all classes is lower than that... Then record it as unknown or ignore it; if there is a confidence level of... For classes that meet the conditions at that time, the enhanced judgment rule is used: if the conditions are met, the record is made; otherwise, it is ignored.

[0046] In addition to the above process, key state parameters can also be judged based on a preset single threshold.

[0047] The processor directly determines whether the fuel cell output voltage is below a threshold; if it is, a power failure is triggered. Based on the type of failure, the corresponding preset capability constraints are directly retrieved from a table. Through multi-level feature extraction and intelligent model reasoning, more complex and subtle failure modes can be identified, resulting in more accurate diagnostic results. Consequently, the dynamic capability constraints provided for subsequent planning are more closely aligned with the actual state of the UAV, avoiding potential issues of insufficient or excessive response due to a single threshold judgment.

[0048] In this embodiment, the dynamic capability constraints include at least one of the maximum flight speed constraints, maximum climb or dive angle constraints, and maximum roll angle constraints; the fault modes include at least one of the following: power system power attenuation, actuator jamming or reduced efficiency, and navigation sensor data anomalies. Different fault modes correspond to different sets of dynamic capability constraints.

[0049] When the diagnostic result is "Power system power attenuation level 1", the corresponding dynamic capability constraints are that the maximum flight speed, maximum climb angle, and maximum roll angle are all limited within the corresponding safe ranges. When the diagnostic result is "Right elevator effectiveness reduced by 50%", the corresponding constraints may be that the climb angle is limited and the roll to the left is limited, while the maximum level flight speed may remain unchanged. These constraint parameter values ​​are obtained from a large number of simulations and experimental calibrations and are stored in the configuration file in memory for the diagnostic module to call.

[0050] An optimization problem that takes the current position of the UAV as the starting point, global path points as intermediate or ending points, and satisfies the aforementioned dynamic capability constraints; the objective function of the optimization problem includes at least one of path length, flight time, and energy consumption, as detailed below:

[0051] The processor determines the drone's current position based on the information obtained from the navigation system. and read preset intermediate path points from memory , and the destination of the return flight ; By pre-setting a global return path sequence Construct a path optimization problem. The path consists of waypoints. The path points and the connecting segments constitute the decision variables of the optimization problem. The constraints include:

[0052] Dynamic constraints: The required flight speed, climb angle, and roll angle for a flight segment must not exceed the output dynamic capability constraints, namely the maximum permissible speed, maximum permissible climb angle, and maximum permissible roll angle.

[0053] Start and end point constraints: , Or some intermediate path point; where The first waypoint for replanning the route. , For intermediate waypoints in the replanned route, This is the last path point in the replanned path.

[0054] Obstacles constrain the path, which must avoid known no-fly zones or obstacles.

[0055] The objective function can be to minimize the total path length, or a weighted sum of the path length and estimated flight time, with the estimated flight time estimated by the path length and the suggested cruise speed under the current fault mode. Solving the optimization problem generates a return path consisting of a series of waypoints and feasible segments between these waypoints that satisfy the dynamic capability constraints. The processor calls the optimization algorithm stored in memory to solve the problem.

[0056] The solver uses iterative calculations to find a sequence of waypoints that satisfies all constraints and minimizes the objective function. This sequence is the replanned return path. Because the constraints include embedded dynamic capability constraints, each segment of the solved path is one that the UAV can actually execute under the current fault condition.

[0057] Considering dynamic capability constraints, the environmental space is discretized into grids, and a passage cost is assigned to each grid. The minimum cost path from the starting point to the target point is searched in the discrete state space. Employing a continuous optimization-based method, smoother paths that better reflect the continuous dynamics of the UAV are generated, typically with higher spatial resolution and superior path quality.

[0058] The method also includes, after generating the return path, further ensuring the robustness of the return and providing backup safety strategies for possible planning failures;

[0059] The processor performs simulation verification on the generated return path. It calls a simplified UAV dynamics model, inputs the capability parameters under the current fault mode and the planned path, and performs a forward simulation to check whether the UAV will violate angular velocity limits, acceleration limits, or collide with obstacles while tracking the path. At the same time, it also verifies whether the endpoint of the path is within the current remaining range capability.

[0060] If the verification passes, the processor executes the steps to control the UAV to fly along the return path; if the verification fails, it triggers a secondary replanning of the return path or executes a preset emergency landing procedure. If the verification passes, the processor sends the generated waypoint sequence to the underlying flight controller, which generates specific control surface and throttle commands to control the UAV to fly along that path. If the verification fails, such as a collision occurring in the simulation or the required range exceeding the capability, the processor will trigger a re-execution of the above steps, i.e., a secondary replanning. If the secondary replanning still fails or an extreme fault such as complete loss of power is diagnosed, the processor retrieves the preset emergency landing procedure from memory, finds a flat area in the current airspace, and performs a vertical forced landing.

[0061] In the above method steps, the memory is used to store the computer program. The memory can be random access memory or read-only memory.

[0062] The processor executes the computer program stored in the memory to implement the steps in the aforementioned hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning. The processor is an integrated circuit chip with signal processing capabilities; it can be a general-purpose processor, such as a CPU, digital signal processor, or application-specific integrated circuit (ASIC). The processor and memory are connected via an internal bus, which serves as an interface module for communication with the UAV's sensors and actuators.

[0063] Example 1: Hydrogen-powered drones perform power transmission line inspection tasks.

[0064] During a certain stage of the inspection, the fuel cell management system reported an anomaly: the stack output voltage continued to drop, but the output current remained unchanged, resulting in a significant decrease in output power; the pressure difference between the hydrogen inlet and outlet increased significantly; the air compressor speed increased, but the response became slower. The attitude data recorded by the flight control system showed no anomalies, and the control surface feedback was normal. The onboard processor periodically collected the aforementioned sensor data via the CAN bus, with a pre-set fixed sampling frequency.

[0065] At this point, the processor initiates a hierarchical fault diagnosis mechanism. First, it performs rapid rule-based judgment, monitoring the pressure difference between the hydrogen and air sides in real time. When the pressure difference exceeds a threshold dynamically calibrated based on the current current and temperature, a potential gas path blockage flag is triggered. Next, it calculates the ratio of the standard deviation to the mean of the voltage of all individual cells in the fuel cell stack, obtaining the individual cell voltage inconsistency coefficient. If this coefficient exceeds the normal upper limit, a poor stack consistency flag is triggered. Finally, it monitors the output voltage change rate; if it exceeds a set value, a rapid power decay flag is triggered. These flags, along with the raw data, constitute a primary fault feature vector.

[0066] Next, the processor constructs time-series features from historical data within a defined time window and inputs them into a pre-trained Long Short-Term Memory (LSTM) network model. This model has been trained in the laboratory using fault-injected data, covering various fault scenarios such as membrane electrode icing, gas path blockage, and normal degradation. The model outputs the probability distribution of various faults and predictions of fault severity.

[0067] The decision logic module integrates the results of the rapid rule judgment with the output of the deep learning model. If the probability of gas path blockage exceeds the set threshold and the differential pressure indicator of the rapid rule judgment also supports it, a preliminary judgment of gas path blockage is made. Then, by combining the temporal characteristics of differential pressure changes and the severity predicted by the deep learning model, the fault mode and severity level are comprehensively determined.

[0068] Specifically, in each diagnostic cycle, the fault diagnosis model outputs a confidence vector with the same number of elements as the total number of fault categories. A high confidence threshold and a low confidence threshold are set, where the low threshold is less than the high threshold and both are between 0 and 1. If the confidence of a fault category reaches or exceeds the high threshold, it is directly identified as that fault. If the confidence of all categories is less than the high threshold, but the confidence of a certain category reaches or exceeds the low threshold, and its historical occurrence frequency within the sliding window reaches or exceeds a preset frequency threshold, it is further identified as that fault. If none of the above conditions are met, no clear fault category is generated in the current cycle, and the judgment from the previous cycle is maintained or a safe mode is executed. Through this hierarchical judgment mechanism, the final output is the determined fault mode and its corresponding severity level.

[0069] The onboard computer retrieves basic capability constraints based on the diagnostic results, including parameters such as maximum flight speed, maximum climb angle, and maximum roll angle. It also dynamically adjusts these constraints based on current environmental conditions such as temperature and humidity, and estimates remaining flight time using a power consumption model based on remaining hydrogen reserves and the current fault condition. These constraints are encapsulated as a dynamic capability constraint set.

[0070] The processor then initiates path replanning. A path optimization problem is constructed, using the UAV's current position as the starting point, the final return point as the target point, and intermediate preset points as reference nodes. The path consists of waypoints and connecting segments, with the decision variable being the three-dimensional position of the waypoints. Constraints include dynamic constraints (the required flight speed, climb angle, and roll angle for each segment must not exceed dynamic capability constraints), origin-end point constraints, and obstacle constraints. The objective function comprehensively considers path length and energy consumption, with weighting coefficients assigned to both based on mission requirements.

[0071] The processor invokes an improved fast search random tree algorithm to solve the problem. When randomly sampling and expanding nodes, this algorithm checks the feasibility of the newly generated flight segment at each step based on the current dynamic capability constraints: it calculates the required speed, climb angle, and roll angle for the flight segment from the current node to the new sampling point. If any of these parameters exceeds the maximum allowable value, the sampling point is discarded and resampling is performed. The same feasibility check is also performed on candidate paths during the reconnection optimization process. The algorithm uses the UAV's current position as the root node and the target point region as the sampling guide, iteratively searching until a feasible path from the starting point to the destination is found and the path cost is gradually optimized, ultimately generating a return path consisting of a series of waypoints.

[0072] After the path is generated, a simplified UAV dynamics model is invoked, and the mass, thrust, and aerodynamic parameters after the fault are input. A forward simulation is performed along the planned path to check whether the angular velocity and acceleration at each step exceed the limits. At the same time, the remaining hydrogen in the fuel cell, the power consumption model under the current fault, and the estimated total energy consumption of the path are compared. If the estimated energy consumption exceeds a certain proportion of the remaining available energy, the remaining range is determined to be insufficient.

[0073] If the verification passes, the processor sends the generated waypoint sequence to the underlying flight controller, which then uses a model predictive control algorithm to dynamically generate control surface and throttle commands to control the UAV to fly along that path. If the verification fails, a secondary replanning is triggered or a preset emergency landing procedure is executed. During flight, if the diagnostic module detects that the fault severity has further deteriorated, the processor immediately triggers a secondary replanning, regenerating a smoother, lower-power path based on the new capability constraints. If the secondary replanning still fails or an extreme fault such as complete loss of power is diagnosed, the processor invokes the preset emergency landing procedure, searches for a flat area in the current airspace, and performs a vertical emergency landing.

[0074] This embodiment demonstrates how, when a hydrogen-powered drone encounters typical faults such as gas path blockage, this method accurately identifies the fault type and severity through hierarchical diagnosis, dynamically calculates flight capability boundaries, and replans a matching return path in real time, ultimately achieving a safe return. Compared to the traditional method of triggering a straight-line return based solely on a single threshold, this method significantly improves the drone's autonomous survivability and mission preservation capabilities under complex fault conditions.

[0075] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.

Claims

1. A hydrogen-powered unmanned aerial vehicle (UAV) return-to-home control method based on fault diagnosis and path replanning, characterized in that, Includes the following steps: Obtain the drone's status information; Perform hierarchical fault diagnosis based on status information to determine the current fault mode and the corresponding fault severity level; Dynamic capability constraints are determined based on the current fault mode and fault severity level. These dynamic capability constraints characterize the flight performance boundaries of the UAV under fault conditions. Starting from the current position of the UAV and with a preset return point as the target point, the dynamic capability constraints are used as constraints for path replanning to generate a return path that satisfies the dynamic capability constraints. The UAV is then controlled to fly along the return path.

2. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: The hierarchical fault diagnosis extracts features from the state information to obtain primary fault features, inputs the primary fault features into the fault diagnosis model to obtain fault diagnosis results, and determines the current fault mode and the corresponding fault severity level by fusing the primary fault features with the output of the fault diagnosis model.

3. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 2, characterized in that: The fault diagnosis model is a pre-trained deep learning model used to output the confidence vector of each fault category and the fault severity prediction result.

4. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 2, characterized in that: The process involves fusing the primary fault features with the output of the fault diagnosis model to obtain a confidence vector from the fault diagnosis model output. The current fault mode is determined by comparing the confidence vector with a preset threshold and determining the frequency of occurrence of the confidence vector within the historical diagnosis period.

5. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: The dynamic capability constraints include maximum flight speed, maximum climb angle, maximum roll angle, maximum acceleration, maximum angular velocity, and remaining endurance.

6. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: Dynamic capability constraints are determined based on the current fault mode and fault severity level, and basic capability constraints are determined accordingly. The basic capability constraints are then modified based on environmental conditions or the remaining energy of the UAV to obtain dynamic capability constraints. The environmental conditions include ambient temperature or ambient humidity. The remaining energy of the UAV includes remaining hydrogen or remaining electrical energy.

7. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: The path replanning includes starting from the current location of the UAV and setting a preset return point as the target point, searching for a path consisting of several waypoints. During the search process, the feasibility of each segment of the candidate path is checked. If the flight parameters required for any segment exceed the corresponding boundary of the dynamic capability constraint, the candidate path is discarded.

8. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: After generating the return path, the feasibility of the return path is verified, including dynamic feasibility verification or energy feasibility verification. If the verification is successful, the steps of controlling the UAV to fly along the return path are executed; if the verification fails, the path is replanned or an emergency landing procedure is executed.

9. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 8, characterized in that: The dynamic feasibility verification involves calling the UAV dynamic model to perform a forward simulation along the planned path, checking whether the UAV's angular velocity and acceleration exceed the limits during the simulation, and then estimating the remaining available energy based on the UAV's remaining energy and the power consumption model under the current fault mode. The estimated total energy consumption of the return path is compared with the remaining available energy. If the estimated total energy consumption exceeds the preset proportion of the remaining available energy, the energy feasibility verification is deemed unsuccessful.

10. The hydrogen-powered UAV return-to-home control method based on fault diagnosis and path replanning according to claim 1, characterized in that: If a change in the severity level of a fault is detected during the UAV's flight along the return path, the dynamic capability constraints are redefined based on the changed fault severity level, and the path is replanned based on the redefined dynamic capability constraints. If a return path that meets the dynamic capability constraints cannot be generated, or if a preset extreme fault is detected, an emergency landing procedure is executed.

Images