Low-altitude equipment adaptive health management system based on multi-modal perception and collaborative decision-making
The low-altitude equipment health management system, which integrates multi-modal perception and collaborative decision-making, integrates multi-source data to achieve full-dimensional health status perception and personalized prediction of equipment, and dynamically generates maintenance strategies. This solves the problems of multi-source data fusion, prediction model accuracy, and maintenance strategy adaptability of low-altitude equipment, thereby improving operational safety and mission reliability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-27
- Publication Date
- 2026-07-31
AI Technical Summary
The existing low-altitude equipment health management system suffers from problems such as insufficient multi-source data fusion capability, limited accuracy and adaptability of prediction models, lack of dynamic adaptability of maintenance strategies, lack of decision-making architecture and closed-loop verification, and weak cluster collaborative management capability, resulting in low operational safety and maintenance efficiency.
The low-altitude equipment adaptive health management system, which adopts multimodal perception and collaborative decision-making, achieves multi-dimensional parameter integration, millisecond-level fault circuit interruption, personalized prediction, and dynamic maintenance strategy generation through multi-source heterogeneous data acquisition modules, edge-cloud collaborative decision-making architecture, and adaptive maintenance strategy generator.
It enables full-dimensional health status perception of low-altitude equipment, improves the accuracy of fault prediction and scenario adaptability, reduces maintenance costs, enhances operational safety and mission reliability, and adapts to complex low-altitude environments.
Smart Images
Figure CN122492155A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of adaptive health management and intelligent maintenance technology for low-altitude equipment, and in particular to an adaptive health management system for low-altitude equipment based on multimodal perception and collaborative decision-making. Background Technology
[0002] Low-altitude equipment (such as urban logistics drones and electric vertical takeoff and landing aircraft) are emerging air transportation and operational tools. Their working environment is highly dynamic (urban building complexes, complex weather), has high safety requirements (near-ground flight risks), and has multiple mission differences (logistics, inspection, emergency response, etc.). Existing technologies have the following limitations:
[0003] (1) Insufficient multi-source data fusion capability
[0004] Existing health management systems mostly monitor single components (such as power batteries) or single parameters (such as vibration signals), without integrating the unique multi-source heterogeneous data of low-altitude equipment (such as power battery thermal maps, rotor aerodynamic loads, and recognition confidence of sensing systems). This results in a one-sided assessment of the overall health status of the equipment and makes it easy to miss cross-system coupled faults (such as the correlation between battery overheating and flight control accuracy degradation).
[0005] (2) The accuracy and adaptability of the prediction model are limited.
[0006] Battery degradation prediction often relies on a single electrochemical model or data-driven model, without integrating the electrochemical-thermodynamic coupling mechanism. This makes it difficult to accurately capture the dynamic relationship between changes in the cell's internal resistance gradient and the temperature field, resulting in a large error in the remaining life prediction.
[0007] The remaining life prediction of components does not take into account the differences between different models and mission scenarios, lacks transfer learning capabilities, and has poor model generalization.
[0008] The performance degradation of perception systems (such as vision and radar) in complex scenarios such as rain, fog, and glass curtain wall reflections (such as the increase in the false negative rate caused by lens damage) lacks quantitative verification methods, which can easily lead to obstacle avoidance or target recognition failure.
[0009] (3) The maintenance strategy lacks dynamic adaptability.
[0010] Traditional maintenance often adopts a fixed-cycle model without dynamically adjusting based on real-time equipment health status and mission window safety redundancy thresholds. This can lead to either "over-maintenance" (increasing costs) or "under-maintenance" (causing safety incidents). For example, there are no parameter optimization plans to extend the maintenance cycle for components whose degradation can be delayed (such as electric drive systems), and no immediate replacement instructions are triggered for critically failing components.
[0011] (4) Lack of decision-making framework and closed-loop verification
[0012] Existing systems mostly rely on single terminals or cloud-based decision-making, and the processing latency of real-time data such as high-frequency vibration signals is high, making it impossible to achieve high-speed fault circuit interruption.
[0013] The lack of a closed-loop verification mechanism for digital twins means that the effectiveness of maintenance strategies can only be verified through actual operation, making it difficult to verify in advance through fault injection simulation. Furthermore, it is impossible to optimize the prediction model parameters based on feedback data, resulting in low model iteration efficiency.
[0014] (5) Weak cluster collaborative management capabilities
[0015] When multiple low-altitude equipment units form a cluster to perform missions, the existing system lacks a dynamic mission redistribution mechanism based on real-time health scores. If a single piece of equipment fails, it is difficult to quickly schedule backup equipment to take over, which can easily lead to mission interruption. Summary of the Invention
[0016] The technical problem to be solved by the present invention is to provide an adaptive health management system for low-altitude equipment based on multimodal perception and collaborative decision-making, so as to improve the operational safety and maintenance efficiency of low-altitude equipment.
[0017] The technical solution adopted by this invention to solve its technical problem is: to provide an adaptive health management system for low-altitude equipment based on multimodal perception and collaborative decision-making, comprising:
[0018] A multi-source heterogeneous data acquisition module is used to acquire multi-dimensional parameters in real time and integrate the acquired multi-dimensional parameters; the multi-dimensional parameters include power battery pack parameters, electric drive system parameters, flight control and sensing system parameters, and mission payload spectrum;
[0019] The edge-cloud collaborative decision-making architecture includes an equipment-side high-frequency feature extraction unit and a cloud-based intelligent hub. The equipment-side high-frequency feature extraction module is used to analyze the parameters of the electric drive system through variational mode analysis and trigger millisecond-level fault circuit interruption based on the analysis results. The cloud-based intelligent hub is used to predict the failure time and real-time failure probability based on the multi-dimensional parameters.
[0020] An adaptive maintenance strategy generator is used to dynamically generate differentiated maintenance plans based on the prediction results output by the cloud-based intelligent hub.
[0021] The multi-source heterogeneous data acquisition module includes:
[0022] The power battery pack parameter acquisition unit is used to collect the internal resistance of each cell of the power battery pack using an internal resistance tester, generate a spatial gradient distribution matrix, obtain a thermal map of the charging process through an array infrared thermal imager, extract the temperature field distribution on the cell surface, and store it synchronously with the charging current and voltage data.
[0023] The electric drive system parameter acquisition unit is used to acquire the vibration signal of the rotor through a piezoelectric accelerometer, generate the aerodynamic load spectrum through fast Fourier transform, calculate the instantaneous power in real time through the current sensor built into the electronic speed controller, and obtain the response deviation by comparing it with the nominal power curve.
[0024] The flight control and perception system parameter acquisition unit is used to output a positioning error value every preset time period through cross-verification of the positioning system, IMU gyroscope and visual SLAM, record dynamic obstacle avoidance events in real time through the perception system, and calculate the standard deviation of the target recognition confidence sequence.
[0025] The task load spectrum acquisition unit is used to preset task stage labels, collect the overload coefficients of each stage through acceleration sensors, and generate an operational load spectrum bound to the task labels.
[0026] The high-frequency feature extraction module for the equipment includes:
[0027] The extraction unit is used to perform a four-level decomposition of the rotor vibration signal in the electric drive system parameters using the db6 wavelet basis, and extract the characteristic frequencies of the preset frequency band.
[0028] The triggering unit is used to trigger a millisecond-level fault circuit interrupter when the deviation of the characteristic frequency exceeds ±10% of the baseline value.
[0029] The cloud-based intelligent hub includes:
[0030] An electrochemical-thermodynamic coupled degradation model for power batteries is used to obtain the capacity increment curve based on the battery pack parameters, construct a degradation baseline based on the drift of the characteristic points of the capacity increment curve, and then combine a long short-term memory network to predict the entropy productivity change of the future preset number of charge-discharge cycles, and output a personalized charging control matrix containing a time-varying function of the maximum allowable charging current.
[0031] The component remaining life prediction network uses the degradation data of the electric drive system of multiple UAVs of the same model as the source domain dataset, and migrates it to the target model through the domain adaptation layer. It is used to take the electric drive system parameters as input and output the failure time prediction values of the low-altitude equipment and its components.
[0032] The synthetic defect scenario verification engine constructs a typical urban low-altitude threat library, injects performance degradation factors, calculates the target miss rate rise slope under dynamic conditions, and triggers calibration or cleaning commands when the target miss rate rise slope exceeds the aircraft's safety tolerance threshold.
[0033] The Bayesian network model uses the full-dimensional historical fault data of the low-altitude equipment as the prior probability basis and integrates the multi-dimensional parameters as the posterior probability basis to dynamically calculate the real-time fault probability of the entire machine.
[0034] The adaptive maintenance strategy generator includes:
[0035] The critical safety maintenance instruction module is used to trigger a forced replacement mechanism when the predicted failure time is less than the minimum safety redundancy threshold of the mission window, and to link the low-altitude emergency navigation database to match a safe landing point, correct the landing trajectory based on three-dimensional meteorological grid wind field data, and send graded power-control degradation instructions to the flight control system of the low-altitude equipment.
[0036] The performance degradation maintenance instruction module is used to generate parameter adjustment schemes for extending maintenance cycles and online monitoring enhancement strategies for components with controllable degradation. The parameter adjustment schemes for extending maintenance cycles are generated based on the remaining lifespan of the components. The online monitoring enhancement strategies include increasing the vibration signal sampling rate, enabling temperature trend early warning, and generating a health status report every hour.
[0037] The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making further includes: a closed-loop verification digital twin, used to construct a virtual model of the low-altitude equipment based on the parameters of the flight control and perception system and the mission payload spectrum.
[0038] The closed-loop verification digital twin simulates various failure scenarios through fault injection simulation to verify the effectiveness of the maintenance strategy, and inputs the verified measured data into the model of the cloud intelligent hub to dynamically optimize the model parameters.
[0039] The adaptive health management system for low-altitude equipment based on multimodal perception and collaborative decision-making further includes: a cluster collaborative health scheduler, used to score all low-altitude equipment according to the prediction results output by the cloud-based intelligent hub, and to reallocate the tasks of the low-altitude equipment based on the scores.
[0040] When the cluster collaborative health scheduler reallocates tasks for low-altitude equipment based on the score, if the score is below the score threshold and there is backup low-altitude equipment, an improved Hungarian algorithm is used to minimize the total health loss of the cluster and complete the task node reallocation; if the score is below the score threshold and there is no backup low-altitude equipment, emergency delivery tasks with a priority higher than the first priority threshold are retained, ordinary cargo transportation with a priority lower than the second priority threshold is suspended, the task interruption cost is calculated and a compensation scheme is generated; wherein, the first priority threshold is greater than the second priority threshold.
[0041] Beneficial effects
[0042] By adopting the above-mentioned technical solutions, this invention has the following advantages and positive effects compared with existing technologies: This invention integrates all-dimensional operating parameters of low-altitude equipment through a multi-source heterogeneous data acquisition module, breaking the limitations of single-parameter monitoring and achieving comprehensive perception of the overall health status of the equipment, effectively avoiding the problem of missing cross-system coupled faults. The edge-cloud collaborative decision-making architecture balances real-time performance and intelligence. Millisecond-level fault circuit interruption at the equipment end can quickly respond to emergency anomalies, while the cloud-based intelligent hub accurately predicts failure time and failure probability, improving the accuracy of fault prediction and scenario adaptability. The adaptive maintenance strategy generator dynamically generates differentiated solutions based on cloud prediction results, replacing the traditional fixed-cycle maintenance mode. This avoids increased costs due to over-maintenance and prevents safety accidents caused by insufficient maintenance, achieving on-demand maintenance. This invention significantly improves the safety and mission reliability of low-altitude equipment operation, reduces the total life-cycle maintenance cost, and adapts to cluster operation scenarios, enhancing the equipment's adaptability and risk resistance in complex low-altitude environments. Attached Figure Description
[0043] Figure 1 This is an architecture diagram of the low-altitude equipment adaptive health management system according to an embodiment of the present invention. Detailed Implementation
[0044] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0045] The embodiments of the present invention relate to an adaptive health management system for low-altitude equipment based on multimodal perception and collaborative decision-making, such as... Figure 1 As shown, it includes:
[0046] A multi-source heterogeneous data acquisition module is used to acquire multi-dimensional parameters in real time and integrate the acquired multi-dimensional parameters; the multi-dimensional parameters include power battery pack parameters, electric drive system parameters, flight control and sensing system parameters, and mission payload spectrum;
[0047] The edge-cloud collaborative decision-making architecture includes an equipment-side high-frequency feature extraction unit and a cloud-based intelligent hub. The equipment-side high-frequency feature extraction module is used to analyze the parameters of the electric drive system through variational mode analysis and trigger millisecond-level fault circuit interruption based on the analysis results. The cloud-based intelligent hub is used to predict the failure time and real-time failure probability based on the multi-dimensional parameters.
[0048] An adaptive maintenance strategy generator is used to dynamically generate differentiated maintenance plans based on the prediction results output by the cloud-based intelligent hub.
[0049] The multi-source heterogeneous data acquisition module in this embodiment includes:
[0050] The power battery pack parameter acquisition unit is used to collect the internal resistance of each cell of the 6S1P power battery pack using a high-precision internal resistance tester (measurement range 0-200mΩ, accuracy ±1%), generate a spatial gradient distribution matrix (sampling interval 100ms), acquire the thermal map of the charging process using an 8×8 array infrared thermal imager (temperature measurement range -20℃~120℃, resolution 320×240), extract the temperature field distribution on the cell surface (temperature resolution 0.05℃), and store it synchronously with the charging current and voltage data (sampling frequency 1kHz).
[0051] The electric drive system parameter acquisition unit is used to acquire the vibration signal of the rotor by a piezoelectric accelerometer (range ±50g, sensitivity 100mV / g) installed at the rotor shaft end. The signal is then converted into an aerodynamic load spectrum of 0-2kHz by fast Fourier transform. The instantaneous power is calculated in real time by a current sensor (accuracy ±2A) built into the electronic speed controller. The response deviation is obtained by comparing the result with the nominal power curve (calculation period 10ms).
[0052] The flight control and perception system parameter acquisition unit is used to cross-validate the positioning system (positioning accuracy 1m), IMU gyroscope (gyroscope drift rate 0.1° / h) and visual SLAM, outputting a positioning error value (unit m) every 50ms. It records dynamic obstacle avoidance events (number of successful obstacle avoidances / total number of obstacles) in real time through the perception system (including lidar and visual camera), and calculates the standard deviation of the target recognition confidence sequence (output range 0-1) (calculation window 10s).
[0053] The mission load spectrum acquisition unit is used to preset mission phase labels (takeoff / 3min, cruise / 15min, hover / 2min, landing / 3min), and collects the overload coefficient of each phase through an acceleration sensor (range ±16g) to generate an operational load spectrum (time resolution 100ms) bound to the mission label.
[0054] The high-frequency feature extraction module of the equipment end in this embodiment includes: an extraction unit, which is used to deploy a wavelet packet algorithm using an embedded AI chip (such as NVIDIA Jetson Nano), and use the db6 wavelet basis to decompose the rotor vibration signal in the electric drive system parameters into 4 layers to extract the feature frequencies in the 100-500Hz frequency band (corresponding to the first and second harmonics of the rotor); and a triggering unit, which is used to trigger a millisecond-level fault circuit interruption (response delay ≤8ms) when the deviation of the feature frequency exceeds ±10% of the baseline value (the baseline is calibrated by 100 normal flight data): immediately cut off the power output of the faulty motor, and at the same time start the dual-motor redundant control mode to maintain attitude stability.
[0055] The cloud-based intelligent hub in this embodiment includes:
[0056] An electrochemical-thermodynamic coupled degradation model for power batteries is used to obtain a capacity increment curve based on the battery pack parameters (the capacity increment curve is collected once every 50 charge-discharge cycles). A degradation baseline is constructed based on the drift of the characteristic peaks at 3.4V and 3.7V of the capacity increment curve (accuracy ±0.02V). This is then combined with a long short-term memory network to predict the entropy yield change over a predetermined number of charge-discharge cycles (the input layer of the long short-term memory network contains the entropy yield, core temperature, and depth of charge / discharge (DOD) of the past 30 cycles, and the hidden layer has 128 neurons). The output is a personalized charging control matrix containing a time-varying function of the maximum allowable charging current. The maximum allowable charging current time-varying function... Represented as: ,in, The value is 0.9 (calibrated according to the characteristics of NCM811 cells). For power batteries Constant health status (Corresponding to the safety threshold of 65% SOH). , Set to 60℃ (critical phase transition temperature of ternary lithium materials), and adjust in real time according to the core temperature. (Measured by built-in thermocouple, accuracy ±0.5℃) Dynamically adjust charging current.
[0057] The component remaining life prediction network uses the degradation data of the electric drive system of 100 UAVs of the same model (including vibration RMS values, bearing temperature, and power fluctuations) as the source domain dataset. It is migrated to the target model through a domain adaptive layer (using the CORAL loss function to align the distribution). It takes the electric drive system parameters as input and outputs the failure time prediction value of the low-altitude equipment and its components, that is, the remaining life (RUL) prediction value (error ≤ 6%).
[0058] The synthetic defect scenario verification engine simulates the degradation of perception hardware performance by constructing a typical urban low-altitude threat library (containing 120 semantic scenarios, such as "light rain + reflection from a 30-story glass curtain wall" and "dense fog + cross-shaped high-voltage cables," with each scenario associated with parameters such as visibility (50-1000m) and reflectivity (0-800cd / m²)) and injecting performance degradation factors (such as transmittance reduction (0-25%) corresponding to lens smudges and angle measurement error (0-1.5°) corresponding to radar baseline offset). It calculates the slope of target miss rate increase under dynamic conditions. When the slope of target miss rate increase exceeds the model's safety tolerance threshold, it triggers lidar baseline calibration (accuracy correction to ±0.1°) and optical filter cleaning alarm (activated when the filter image is captured by the airborne camera and the AI identifies a smudge level ≥30%).
[0059] The Bayesian network model uses the comprehensive historical fault data of the low-altitude equipment as the prior probability basis and integrates the multi-dimensional parameters (including battery, electric drive, flight control / sensing, operating conditions, etc.) as the posterior probability basis to dynamically calculate the real-time fault probability of the entire system. .
[0060] The adaptive maintenance strategy generator in this embodiment includes:
[0061] The critical safety maintenance instruction module is used to trigger a forced replacement mechanism when the predicted failure time is less than the minimum safety redundancy threshold of the mission window (3 minutes for emergency missions and 15 minutes for regular missions). It also links with the low-altitude emergency navigation database (including safe landing points such as rooftop platforms and playgrounds within a 500m range, updated once per hour) to match safe landing points (prioritizing candidate points with an altitude ≥ 10m of the current altitude). Based on three-dimensional meteorological grid data (500m × 500m × 50m resolution), it corrects the landing trajectory (lateral deviation compensation ≤ 3m). Simultaneously, it sends a graded power-control degradation instruction to the flight control system of the low-altitude equipment. This graded power-control degradation instruction is as follows: (the design thresholds for failure probabilities are 0.3 and 0.6 respectively): Level 1 ( <0.3: Power off mission payloads (such as cargo hold temperature control and redundant communication modules) to reduce power consumption by 15%; Level 2 (0.3≤ <0.6): Software locks the rotor's maximum tilt angle to 50% (mechanical limit combined with PID closed-loop control); Level 3 ( (≥0.6): Activate the blended wing-body aerodynamic compensation (by fine-tuning the flaps, the lift-to-drag ratio is increased by 8%) to ensure attitude stability.
[0062] The performance degradation maintenance instruction module is used to generate parameter adjustment schemes for extending maintenance cycles and online monitoring enhancement strategies for components with controllable degradation. The parameter adjustment schemes for extending maintenance cycles are generated based on the remaining life of the component, for example, based on RUL prediction, extending the bearing lubrication cycle from 150 flight hours to 180 hours (remaining life must be >200 hours). The online monitoring enhancement strategies refer to increasing the vibration signal sampling rate (from 2kHz to 8kHz), enabling temperature trend early warning (setting a temperature rise rate threshold of 0.8℃ / min), and generating a health status report every hour.
[0063] The low-altitude equipment adaptive health management system of this embodiment also includes a closed-loop verification digital twin, used to construct a virtual model of the low-altitude equipment based on the parameters of the flight control and sensing system and the mission payload spectrum. This closed-loop verification digital twin simulates various failure scenarios through fault injection simulation (supporting 30 typical fault injections, such as "single cell short circuit," "right front rotor dynamic balance failure," and "visual algorithm drift"), verifying the effectiveness of maintenance strategies (e.g., mission completion rate increases to 98% after forced replacement). The verified measured data (e.g., battery cycle life after replacement, sensing accuracy after calibration) is input into the model of the cloud-based intelligent hub to dynamically optimize model parameters and continuously improve prediction accuracy (error is reduced by 15-20% after model iteration).
[0064] The low-altitude equipment adaptive health management system of this embodiment further includes: a cluster collaborative health scheduler, used to score all low-altitude equipment based on the prediction results output by the cloud-based intelligent hub, wherein the score is calculated as follows: ,in, (RUL weight) (Failure probability weight) This is the predicted failure time for low-altitude equipment. This represents the real-time failure probability of the entire system. Then, based on the score, tasks for the low-altitude equipment are redistributed. Specifically, when the score is below a threshold (0.4, corresponding to high risk) and there is backup low-altitude equipment, an improved Hungarian algorithm (time complexity O(n^2)) is used. 3 (n is the number of cluster equipment), with the goal of minimizing the total health loss of the cluster, the task nodes are redistributed; when the score is lower than the score threshold and there is no spare low-altitude equipment, emergency delivery tasks with a priority of more than 0.8 (such as medical supplies) are retained, and ordinary cargo transportation with a priority of less than 0.5 is suspended. The task interruption cost (including delay penalty and customer satisfaction loss) is calculated and a compensation plan is generated.
[0065] The following is an example of a single 30-minute city delivery task, and its specific process is as follows:
[0066] S1 Data Acquisition: Before takeoff, the initial internal resistance distribution and temperature field of the power battery are collected. During flight, parameters such as rotor vibration signal and flight control positioning error are recorded every 10ms and stored synchronously to the edge node.
[0067] S2 Perception Performance Quantization: Through the synthetic defect scene verification engine, the slope of the missed detection rate Δη / Δt=0.03 / h for the current scene (such as "sunny + building cluster occlusion") is calculated in real time (less than β=0.08 / h, so no alarm is triggered for the time being).
[0068] S3 Degradation Prediction: The power battery model outputs the next degradation inflection point as 120 cycles later, and the predicted remaining lifespan of the electric drive system is 85 hours.
[0069] S4 maintenance decision: Task planning time Ttask=30min, safety buffer δt=10min, predicted failure time Tf=150min>40min, therefore, the performance degradation maintenance instruction is generated: extend the lubrication cycle of the electric drive system from 150h to 180h and increase the vibration sampling rate to 5kHz.
[0070] S5 closed-loop verification: The digital twin is injected with a "minor lens contamination" fault. Simulation shows that the maintenance strategy can keep the missed detection rate within 5%. The feedback data is used to optimize the attenuation factor weight of the perception system prediction model.
[0071] This implementation method achieves real-time perception, accurate prediction and dynamic maintenance of the health status of low-altitude equipment through refined parameter configuration, algorithm selection and process design, which significantly improves the operational safety (fault detection rate ≥99%) and maintenance economy (overall cost reduction of 25-30%) in complex low-altitude environments.
[0072] It is easy to see that this invention integrates all-dimensional operating parameters of low-altitude equipment through a multi-source heterogeneous data acquisition module, breaking the limitations of single-parameter monitoring and achieving comprehensive perception of the overall health status of the equipment, effectively avoiding the problem of missing cross-system coupled faults. The edge-cloud collaborative decision-making architecture balances real-time performance and intelligence. Millisecond-level fault circuit interruption at the equipment end can quickly respond to emergency anomalies, while the cloud-based intelligent hub accurately predicts failure time and failure probability, improving the accuracy of fault prediction and scenario adaptability. The adaptive maintenance strategy generator dynamically generates differentiated solutions based on cloud prediction results, replacing the traditional fixed-cycle maintenance mode. This avoids the increased costs caused by over-maintenance and prevents safety accidents caused by insufficient maintenance, achieving on-demand maintenance. This invention significantly improves the safety and mission reliability of low-altitude equipment operation, reduces the total life-cycle maintenance cost, and adapts to cluster operation scenarios, enhancing the equipment's adaptability and risk resistance in complex low-altitude environments.
Claims
1. A low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making, characterized in that, include: A multi-source heterogeneous data acquisition module is used to acquire multi-dimensional parameters in real time and integrate the acquired multi-dimensional parameters; the multi-dimensional parameters include power battery pack parameters, electric drive system parameters, flight control and sensing system parameters, and mission payload spectrum; The edge-cloud collaborative decision-making architecture includes an equipment-side high-frequency feature extraction unit and a cloud-based intelligent hub. The equipment-side high-frequency feature extraction module is used to analyze the parameters of the electric drive system through variational mode analysis and trigger millisecond-level fault circuit interruption based on the analysis results. The cloud-based intelligent hub is used to predict the failure time and real-time failure probability based on the multi-dimensional parameters. An adaptive maintenance strategy generator is used to dynamically generate differentiated maintenance plans based on the prediction results output by the cloud-based intelligent hub.
2. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, The multi-source heterogeneous data acquisition module includes: The power battery pack parameter acquisition unit is used to collect the internal resistance of each cell of the power battery pack using an internal resistance tester, generate a spatial gradient distribution matrix, obtain a thermal map of the charging process through an array infrared thermal imager, extract the temperature field distribution on the cell surface, and store it synchronously with the charging current and voltage data. The electric drive system parameter acquisition unit is used to acquire the vibration signal of the rotor through a piezoelectric accelerometer, generate the aerodynamic load spectrum through fast Fourier transform, calculate the instantaneous power in real time through the current sensor built into the electronic speed controller, and obtain the response deviation by comparing it with the nominal power curve. The flight control and perception system parameter acquisition unit is used to output a positioning error value every preset time period through cross-verification of the positioning system, IMU gyroscope and visual SLAM, record dynamic obstacle avoidance events in real time through the perception system, and calculate the standard deviation of the target recognition confidence sequence. The task load spectrum acquisition unit is used to preset task stage labels, collect the overload coefficients of each stage through an accelerometer, and generate an operational load spectrum that is bound to the task labels.
3. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, The high-frequency feature extraction module for the equipment includes: The extraction unit is used to perform a four-level decomposition of the rotor vibration signal in the electric drive system parameters using the db6 wavelet basis, and extract the characteristic frequencies of the preset frequency band. The triggering unit is used to trigger a millisecond-level fault circuit interrupter when the deviation of the characteristic frequency exceeds ±10% of the baseline value.
4. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, The cloud-based intelligent hub includes: An electrochemical-thermodynamic coupled degradation model for power batteries is used to obtain the capacity increment curve based on the battery pack parameters, construct a degradation baseline based on the drift of the characteristic points of the capacity increment curve, and then combine a long short-term memory network to predict the entropy productivity change of the future preset number of charge-discharge cycles, and output a personalized charging control matrix containing a time-varying function of the maximum allowable charging current. The component remaining life prediction network uses the degradation data of the electric drive system of multiple UAVs of the same model as the source domain dataset, and migrates it to the target model through the domain adaptation layer. It is used to take the electric drive system parameters as input and output the failure time prediction values of the low-altitude equipment and its components. The synthetic defect scenario verification engine constructs a typical urban low-altitude threat library, injects performance degradation factors, calculates the target miss rate rise slope under dynamic conditions, and triggers calibration or cleaning commands when the target miss rate rise slope exceeds the aircraft's safety tolerance threshold. The Bayesian network model uses the full-dimensional historical fault data of the low-altitude equipment as the prior probability basis and integrates the multi-dimensional parameters as the posterior probability basis to dynamically calculate the real-time fault probability of the entire machine.
5. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, The adaptive maintenance strategy generator includes: The critical safety maintenance instruction module is used to trigger a forced replacement mechanism when the predicted failure time is less than the minimum safety redundancy threshold of the mission window, and to link the low-altitude emergency navigation database to match a safe landing point, correct the landing trajectory based on three-dimensional meteorological grid wind field data, and send graded power-control degradation instructions to the flight control system of the low-altitude equipment. The performance degradation maintenance instruction module is used to generate parameter adjustment schemes for extending maintenance cycles and online monitoring enhancement strategies for components with controllable degradation. The parameter adjustment schemes for extending maintenance cycles are generated based on the remaining lifespan of the components. The online monitoring enhancement strategies include increasing the vibration signal sampling rate, enabling temperature trend early warning, and generating a health status report every hour.
6. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, Also includes: A closed-loop verification digital twin is used to construct a virtual model of the low-altitude equipment based on the parameters of the flight control and perception system and the mission payload spectrum.
7. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 6, characterized in that, The closed-loop verification digital twin simulates various failure scenarios through fault injection simulation to verify the effectiveness of the maintenance strategy, and inputs the verified measured data into the model of the cloud intelligent hub to dynamically optimize the model parameters.
8. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 1, characterized in that, Also includes: The cluster collaborative health scheduler is used to score all low-altitude equipment based on the prediction results output by the cloud-based intelligent hub, and to reallocate the tasks of the low-altitude equipment based on the scores.
9. The low-altitude equipment adaptive health management system based on multimodal perception and collaborative decision-making according to claim 8, characterized in that, When the cluster collaborative health scheduler reallocates tasks for low-altitude equipment based on the score, if the score is below the score threshold and there is backup low-altitude equipment, an improved Hungarian algorithm is used to minimize the total health loss of the cluster and complete the task node reallocation; if the score is below the score threshold and there is no backup low-altitude equipment, emergency delivery tasks with a priority higher than the first priority threshold are retained, ordinary cargo transportation with a priority lower than the second priority threshold is suspended, the task interruption cost is calculated and a compensation scheme is generated; wherein, the first priority threshold is greater than the second priority threshold.