Unmanned aerial vehicle maintenance task scheduling and resource management platform

Abstract

The invention belongs to the technical field of unmanned aerial vehicle technology and intelligent operation and maintenance management, and discloses an unmanned aerial vehicle maintenance task scheduling and resource management platform and method, and the method comprises the steps that a data collection and fusion module obtains and fuses multi-source data; the fault diagnosis and prediction analysis module uses a deep learning model and a CNN-LSTM hybrid model to diagnose faults, predict life and generate a maintenance task list; an optional task dynamic priority evaluation module performs priority ranking on the maintenance task list; the intelligent scheduling decision module uses an improved NSGA-II and VNS mixed algorithm to generate an optimal allocation scheme according to priority and resource constraints; the resource dynamic management and cooperation module realizes dynamic allocation and cooperation of personnel, spare parts, tools and sites, and can integrate AR guidance; and the optional visual monitoring and feedback optimization module displays the state, feeds back data and realizes closed-loop optimization. The maintenance efficiency of the unmanned aerial vehicle can be remarkably improved, the operation cost is reduced, and the flight safety is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of unmanned aerial vehicle (UAV) technology and intelligent operation and maintenance management technology, and in particular relates to a UAV maintenance task scheduling and resource management platform and method. Background Art

[0002] Drone technology is developing rapidly, with its applications becoming increasingly widespread in areas such as inspection, logistics, surveying and mapping, and security. This has led to the expansion of drone fleets and the increasing complexity of operational tasks, placing higher demands on drone reliability, availability, and maintenance efficiency. Traditional drone maintenance models rely heavily on manual experience for fault diagnosis, task assignment, and resource coordination, which presents significant shortcomings: First, there is a lack of fault diagnosis and prediction capabilities. Existing maintenance is mostly a passive response to a fault. There is a lack of real-time monitoring of the drone's health status and early warning of faults, making it difficult to plan maintenance tasks in advance. This leads to frequent sudden failures, which affect mission execution and flight safety.

[0003] Second, task scheduling and resource allocation are inefficient. Prioritizing maintenance tasks often lacks an objective and scientific evaluation system. The allocation of maintenance personnel, spare parts, tools, and sites often relies on manual coordination, making global optimization difficult. This leads to wasted resources, delays, and prolonged average downtime for drones.

[0004] Third, resource management is not sufficiently refined. Spare parts inventory management fails to fully integrate predictive demand, leading to shortages or excessive backlogs of critical spare parts. The skill level of maintenance personnel is not precisely matched to task requirements, impacting maintenance quality and efficiency. The utilization rate of maintenance tools and sites needs to be improved.

[0005] Fourth, information silos are common. Drone status data, historical maintenance records, and available resource information are stored in a decentralized manner, making data sharing and comprehensive analysis difficult, making it difficult to achieve effective knowledge accumulation and closed-loop optimization.

[0006] In the prior art, for example, Chinese patent application number CN119293682B discloses a high-precision drone status determination method and system based on a fault search tree. This method primarily focuses on fault identification using sensor data, but does not delve into the intelligent scheduling of maintenance tasks and comprehensive resource optimization management after identification. Another Chinese patent application number, CN117369260B, discloses a drone scheduling method that combines an energy consumption model with a task time window. This method focuses on scheduling in specific scenarios, but lacks a comprehensive, adaptive scheduling and management solution tailored to the full drone maintenance process, particularly one that incorporates fault prediction, dynamic priority assessment, and complex resource constraints.

[0007] Therefore, there is an urgent need to develop a comprehensive platform that can integrate the entire life cycle data of drones, realize intelligent fault diagnosis and remaining service life prediction, and on this basis, conduct efficient maintenance task scheduling and refined resource management to meet the challenges currently faced by drone operation and maintenance. Summary of the Invention

[0008] The primary purpose of this invention is to overcome existing issues in drone maintenance diagnosis, including insufficient accuracy, weak predictive capabilities, inadequate task scheduling optimization, inadequate resource management, and poor information collaboration. This technology provides a drone maintenance task scheduling and resource management platform. This technology aims to significantly enhance the intelligence, automation, and refinement of drone maintenance, improve maintenance efficiency, shorten drone downtime, optimize maintenance resource allocation, and ultimately ensure drone flight safety.

[0009] To achieve the above object, the present invention provides the following technical solutions: A UAV maintenance task scheduling and resource management platform, including: A data acquisition and fusion module is configured to obtain drone operation data, maintenance resource data, and external environment data from the drone system, maintenance resource management system, and external environment information sources, and perform standardization processing and fusion on the multi-source heterogeneous data to form a standardized data set; a fault diagnosis and prediction analysis module configured to diagnose a current fault of the UAV based on the standardized data set using a preset deep learning-based fault diagnosis model, and to predict the remaining useful life of key components using a deep learning-based life prediction model, thereby generating a maintenance task list; The lifespan prediction model based on deep learning is specifically a hybrid prediction model that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM); an intelligent scheduling decision module configured to generate a maintenance task allocation plan that considers multiple constraints and optimization objectives based on the maintenance task list, the dynamic priority of the tasks, and the real-time availability of maintenance resources by using a hybrid optimization scheduling algorithm that integrates an improved non-dominated sorting genetic algorithm (NSGA-II) with an elite retention strategy and a variable neighborhood search (VNS); The resource dynamic management and coordination module is configured to realize the dynamic deployment and operation coordination of maintenance personnel, maintenance spare parts, maintenance tools and maintenance sites according to the maintenance task allocation plan.

[0010] Preferably, the platform further includes a task dynamic priority assessment module, located between the fault diagnosis and prediction analysis module and the intelligent scheduling decision module, configured to: Construct a multi-dimensional mission evaluation indicator system, including evaluation indicators such as fault severity, mission impact, UAV criticality, spare parts acquisition difficulty, maintenance window, and predicted time urgency of fault occurrence; The method combining analytic hierarchy process (AHP) and fuzzy comprehensive evaluation is used to calculate the dynamic priority score of each maintenance task in the maintenance task list, and the maintenance task list is sorted according to the score.

[0011] Preferably, the platform further includes a visual monitoring and feedback optimization module configured to: The graphical user interface displays the drone's health status, maintenance task distribution and progress, and maintenance resource scheduling in real time; and records maintenance process data to form a maintenance knowledge base. Machine learning algorithms are used to analyze historical maintenance data and scheduling execution effects, and the fault diagnosis model, the life prediction model, the weights in the priority evaluation method, and the strategy of the hybrid optimization scheduling algorithm are adaptively adjusted and optimized to form a closed-loop feedback optimization mechanism.

[0012] Compared with the prior art, the present invention has the following beneficial effects: First, by constructing a data acquisition and fusion module and a fault diagnosis and predictive analysis module, specifically employing a lifespan prediction model that combines a convolutional neural network (CNN) with a long short-term memory network (LSTM), this invention enables comprehensive and in-depth analysis of UAV operational data and component health status. This not only effectively improves the accuracy and coverage of fault diagnosis but, more importantly, enables precise prediction of potential failure trends and the remaining useful life of key components. This enables a shift from post-operative maintenance to predictive and proactive maintenance, significantly reducing unplanned groundings and mission interruptions caused by sudden failures and significantly enhancing the reliability and safety of UAV operations.

[0013] Secondly, this invention overcomes the limitations of traditional manual scheduling and simple rule-based scheduling by establishing a dynamic task priority assessment module and employing an intelligent scheduling decision module that integrates an improved non-dominated sorting genetic algorithm (NSGA-II) with an elite retention strategy and a variable neighborhood search (VNS) hybrid optimization scheduling algorithm. The platform scientifically calculates the dynamic priority of tasks based on multi-dimensional evaluation indicators. Under the premise of meeting multiple complex constraints such as maintenance personnel skill requirements, spare parts inventory, and tool availability, it rapidly generates globally optimal or near-optimal maintenance task allocation plans with the optimization objectives of minimizing the total downtime of drones, minimizing the total maintenance cost, and maximizing resource utilization balance. This intelligent scheduling decision ensures the most appropriate allocation of maintenance resources, significantly shortens the average maintenance cycle of drones, and improves the overall fleet availability and mission execution efficiency.

[0014] Furthermore, through a dynamic resource management and collaboration module, this invention enables refined, dynamic management and efficient collaboration of core maintenance resources, including maintenance personnel, spare parts, tools, and repair sites. Intelligent spare parts scheduling and early warning replenishment mechanisms, full lifecycle tracking and management of tools and equipment, and remote maintenance guidance integrated with augmented reality (AR) not only improve the efficiency and turnover of various resources, while reducing delays and waste caused by resource mismatches or shortages, but also enhance the first-time success rate of complex maintenance tasks through technology empowerment, thereby effectively controlling overall maintenance operating costs.

[0015] In addition, the present invention constructs a transparent and efficient information interaction and decision-making support environment by setting up a visual monitoring and feedback optimization module. This module can display task progress, resource status and key performance indicators in real time, so that managers can fully grasp the dynamics of operation and maintenance, and promptly discover and handle abnormal situations. More importantly, through the continuous collection and intelligent analysis of maintenance process data and scheduling execution effects, the module can feed back and optimize the parameters of the diagnostic prediction model, the weight of the priority evaluation system and the strategy parameters of the scheduling algorithm, forming a closed-loop optimization system that continuously learns and improves itself. This ensures that the platform can adapt to the ever-changing drone models, operating scenarios and operation and maintenance needs, and maintain long-term advancement and applicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive work, among which: Figure 1 This is a schematic diagram of the system architecture of a UAV maintenance task scheduling and resource management platform of the present invention; Figure 2 Schematic diagram of the initialization and data preparation process of the hybrid optimization scheduling algorithm in the intelligent scheduling decision module of the present invention; Figure 3 This is a schematic diagram of the NSGA-II core algorithm flow of the hybrid optimization scheduling algorithm in the intelligent scheduling decision module of the present invention; Figure 4 Schematic diagram of the VNS local search enhancement process of the hybrid optimization scheduling algorithm in the intelligent scheduling decision module of the present invention. DETAILED DESCRIPTION

[0017] The following, in conjunction with specific embodiments, describes in detail the specific steps and operating procedures of a two-stage potential diffusion system for high-fidelity virtual try-on of the present invention, so that ordinary technicians in the relevant technical field can implement the present invention. Through the following specific embodiments, technicians can fully understand and implement the technical solutions of the present invention.

[0018] Example 1 See also Figure 1 The core architecture of the drone maintenance task scheduling and resource management platform disclosed in this embodiment is a tightly integrated system comprised of key modules: a data acquisition and fusion module 100, a fault diagnosis and prediction analysis module 200, a dynamic task priority assessment module 300, an intelligent scheduling decision module 400, a dynamic resource management and coordination module 500, and a visual monitoring and feedback optimization module 600. These modules exchange data and collaborate via an internal bus or network, forming the foundation for the platform's intelligent operations and maintenance management.

[0019] The core task of the data acquisition and fusion module 100 is to actively collect original information from multiple data sources, and perform in-depth processing and integration to provide a data basis for subsequent analysis and decision-making.

[0020] During the drone's operational data collection process, the platform communicates with the drone's flight control system, various onboard sensors, the Component Health Management (PHM), and the mission payload via dedicated interfaces. This connection ensures real-time or near-real-time access to detailed flight parameters, the current operating status of key components, battery charge and discharge characteristics, drive motor operating parameters, image transmission link quality data, and any fault alarms detected by system self-tests or sensors. Historical flight logs and mission execution records are also fully imported into the platform's central database via corresponding standardized interfaces for subsequent trend analysis and fault tracing.

[0021] To access maintenance resource data, the platform connects to existing maintenance personnel management systems to accurately capture technician identification numbers, professional skill levels, certification information, current workload, real-time geographic location, and shift schedules. Deeply integrated with spare parts inventory management systems, the platform comprehensively captures each maintenance spare part's unique identification code, name, specifications, current inventory quantity, safety stock threshold, in-transit quantity, storage location information, supplier information, and procurement cycle.

[0022] The platform also connects with the tool and equipment management system to obtain the identification number, type and model, calibration status, borrowing history, and maintenance plan of special maintenance tools, general maintenance tools, and testing equipment. Regarding maintenance site resources, the platform connects to the maintenance site management system to clarify the immediate availability status, capacity limit parameters, specific environmental conditions, and configuration of special maintenance equipment for each maintenance station or site. The immediate availability status can be specifically whether it is idle, occupied, or reserved; the capacity limit parameters can be specifically the size and number of drones that can be accommodated; the specific environmental conditions can be specifically temperature, humidity, and cleanliness; and the special maintenance equipment can be specifically hoisting equipment and test benches.

[0023] The platform also proactively acquires necessary data from external environmental information sources. Using standard application programming interfaces (APIs), it obtains real-time weather forecasts from professional meteorological service platforms, relevant airspace restrictions or temporary flight control notices from air traffic management authorities, and future drone mission plans from mission planning systems, providing key input for forward-looking planning and scheduling.

[0024] During the data fusion and processing phase, the platform standardizes the collected multi-source, heterogeneous data. This process includes data format conversion, unit unification, timestamp alignment, outlier cleaning, and missing value filling. Outlier cleaning can be performed using, for example, the 3σ principle or the isolation forest algorithm, while missing value filling can be achieved using mean interpolation, regression interpolation, or machine learning-based predictive interpolation methods. After preprocessing and fusion, the resulting standardized dataset is stored in the platform's central database or distributed data lake for access by other modules.

[0025] The fault diagnosis and prediction analysis module 200 uses analytical techniques and algorithm models based on the standardized data set provided by the module 100 to conduct in-depth analysis to accurately identify faults and predict risks.

[0026] At the fault diagnosis level, the platform uses a combination of strategies: First, rule-based diagnostic methods utilize a built-in expert knowledge base containing FMECA data and preset diagnostic rules for common drone failure modes, causes, and effects. Second, model-based diagnostic methods build digital twin models or mathematical and physical models for key components, detecting and isolating faults by comparing the residuals between model outputs and actual sensor readings. Third, data-driven diagnostic methods use historical fault data and corresponding sensor data to train machine learning classification models, such as support vector machines (SVMs), random forests, gradient boosted decision trees (GBDTs), or deep neural networks (DNNs). These models identify current sensor data patterns and determine the presence and likely type of fault.

[0027] In terms of failure prediction, the module focuses on identifying potential degradation trends and estimating the remaining useful life (RUL) of components: First, a prediction method based on state parameter trends is used to analyze the time series of key component health indicator parameters to predict their future evolution. Secondly, for components with significant wear or fatigue characteristics, a hybrid prediction model combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network is used to estimate the RUL. This CNN-LSTM model takes as input a series of component historical operating data and current state data. The CNN layer extracts local and spatial features from the data sequence, while the LSTM layer captures temporal dependencies and long-term trends in the data sequence. The output layer predicts the component's RUL value or the probability of failure at a specific point in the future.

[0028] Specifically, the network structure of the CNN-LSTM hybrid prediction model includes: Input layer: receives time series data of length T , where each time step Contains n-dimensional feature vectors, representing the key components of the drone in the first The state parameters at a time point.

[0029] CNN feature extraction layer: uses one-dimensional convolution operation to extract local feature patterns, including: the first convolution layer: the convolution kernel size is 3, the number of output channels is 64, and the ReLU activation function is used; the second convolution layer: the convolution kernel size is 5, the number of output channels is 128, and the ReLU activation function is used; the maximum pooling layer: the pooling window size is 2 and the stride is 1; the dropout layer: the dropout rate is set to 0.2 to prevent overfitting LSTM time series modeling layer: A bidirectional LSTM network is used to capture long-term temporal dependencies. The number of LSTM units is 256, and the number of layers is two stacked LSTM layers. The outputs of the forward and backward LSTMs are concatenated to form a 512-dimensional feature vector. Fully connected output layer: First fully connected layer: 512→128, using ReLU activation; second fully connected layer: 128→64, using ReLU activation; output layer: 64→1, output RUL prediction value.

[0030] Loss function: Use the mean square error loss function MSE, combined with the L2 regularization term: ,in is the regularization coefficient, set to 0.001.

[0031] Training strategy: Adam optimizer is used, the initial learning rate is set to 0.001, the learning rate decays to 0.9 times the original every 10 epochs, the batch size is 32, and the maximum number of training rounds is 200 rounds.

[0032] Taking lithium batteries as an example, its RUL model It can be expressed as: , in: is the mapping function obtained through training, is the number of charge and discharge cycles, is the average depth of discharge, is the operating temperature, is the discharge current, is the initial health state, Represents historical sequence data. Model training optimizes network weights through the back-propagation algorithm to minimize the error between the predicted RUL and the actual RUL.

[0033] After completing the diagnostic prediction, the module analyzes the results and automatically generates a detailed maintenance task list. Each task includes the drone ID, task type, fault description or prediction details, involved component information, recommended repair plan, required skill level, estimated work time, and a list of required spare parts and special tools. This list serves as a key input for subsequent evaluation and decision-making.

[0034] The task dynamic priority evaluation module 300 prioritizes the maintenance task list generated by module 200. The module first establishes a multi-dimensional evaluation index system. The first-level index covers the degree of safety impact, task impact, economic impact, and timeliness impact. The AHP method is used to determine the basic weight of each first-level index. The second-level indicators include the fault severity level, the UAV criticality coefficient, the spare parts acquisition difficulty coefficient, the maintenance window length, the predicted fault occurrence time proximity value, etc., and the fuzzy comprehensive evaluation method is used to deal with its uncertainty and ambiguity. Each maintenance task Dynamic priority Calculated by the following example formula: in, Score the final dynamic priority, For the The weight of the first-level evaluation indicators, For the The task in The comprehensive evaluation value under the first-level indicators is The expected completion deadline for the task, is the current time, The acceptable maintenance window duration for the task, is the time urgency impact coefficient, Estimate the comprehensive resource consumption value for the task, It is the reference value of the maximum possible resource consumption of a single task in the system. is the reverse impact coefficient of resource consumption.

[0035] The specific calculation steps of the combination of the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation are as follows: Step 1: Construct a judgment matrix for evaluation indicators, construct Judgment matrix A, where Indicator Relative to the indicator Importance: ,in =1 / , =1.

[0036] Step 2: Calculate the weight vector The weights are calculated using the eigenvector method: ,in is the weight vector, is the maximum eigenvalue.

[0037] Step 3: Consistency Check Calculating the consistency ratio ,in is a random consistency indicator. When <0.1, the judgment matrix is ​​considered to have satisfactory consistency.

[0038] Step 4: Fuzzy comprehensive evaluation For each secondary indicator, a fuzzy evaluation set V = {excellent, good, average, poor} is established, with corresponding scores {0.9, 0.7, 0.5, 0.3}. Trapezoidal fuzzy numbers are used to represent the uncertainty of the evaluation value.

[0039] Step 5: Final score calculation The final dynamic priority score is calculated by combining the AHP weights and fuzzy evaluation results.

[0040] This module dynamically updates task priorities based on real-time conditions.

[0041] See also Figure 2 、 Figure 3 and Figure 4 After obtaining the sorted maintenance task list and real-time maintenance resource availability, the intelligent scheduling decision module 400 initiates its core scheduling process. It uses a hybrid optimization scheduling algorithm that combines an improved non-dominated sorting genetic algorithm (NSGA-II) with an elite retention strategy and a variable neighborhood search (VNS) to generate a maintenance task allocation plan.

[0042] See also Figure 2,In the algorithm initialization and input phase, the module obtains the sorted ,maintenance task list and real-time maintenance resource status, and sets ,optimization objectives, including minimizing the total UAV downtime, minimizing the total maintenance ,cost, maximizing the resource utilization balance, and maximizing the completion rate of high-priority ,tasks.

[0043] The scheduling scheme is encoded as a chromosome and an initial population is randomly generated. For each scheduling scheme in the population, the fitness is calculated according to the optimization objective and constraints.

[0044] See also Figure 3 The core process of genetic evolution operation, namely NSGA-II, includes: performing non-dominated sorting and crowding calculation, dividing the Pareto frontier and calculating the individual crowding distance; performing elite selection to retain excellent individuals for the next generation; performing adaptive crossover and adaptive mutation operations, and their operators and probabilities are dynamically adjusted according to population diversity and convergence status.

[0045] The adaptive crossover operator dynamically adjusts the crossover probability and mode according to the population convergence state and individual fitness differences: Crossover probability adaptive formula: in, =0.9 is the maximum crossover probability, =0.6 is the minimum crossover probability, is the current algebra, is the maximum algebra, =2 is the convergence control parameter.

[0046] Specific steps of the crossover operation: Parent selection: Tournament selection is used with a tournament size of 3; Crossover point determination: For the maintenance task scheduling chromosome, 1-3 crossover points are randomly selected; Segment exchange: The task allocation segments of the parent individuals are exchanged between the crossover points; Constraint repair: Check whether the crossover offspring violates constraints such as skill matching and resource capacity. If so, the constraint repair strategy is adopted: Skill mismatch: reallocate to someone with corresponding skills; resource conflict: adjust task start time or replace resources; timing dependency violation: reorder based on dependency relationships.

[0047] Local search enhancement is the VNS process (see Figure 4 ) is started after NSGA-II evolves to a certain stage, and a representative solution is selected from the Pareto optimal solution set as the initial solution. Local deep optimization is performed by searching in multiple predefined neighborhood structures (such as task order exchange, resource reallocation, and start time fine-tuning).

[0048] The VNS algorithm defines five neighborhood structures for local search optimization: Neighborhood structure - Task sequence swapping: Randomly select two adjacent tasks and swap their execution order; applicable to optimizing the timing arrangement between tasks.

[0049] Neighborhood structure - Task reassignment: Randomly select one task and reassign it to other maintenance personnel who meet the skill requirements; used to balance the personnel load.

[0050] Neighborhood structure - Time window adjustment: Randomly adjust the start time of the selected task within the range of ±2 hours; used to optimize resource utilization.

[0051] Neighborhood structure - Resource replacement: Replace the spare parts, tools or sites required for the task with equivalent resources; applicable to alleviating resource conflicts.

[0052] Neighborhood structure - Task decomposition and merging: Decompose complex tasks into multiple subtasks, or merge similar subtasks; used to improve maintenance efficiency.

[0053] VNS search process: Step 1, Initial solution s = Select from the Pareto optimal solution set; Step 2, k = 1 (neighborhood structure index); Step 3, While k ≤ 5: Randomly generate a solution s' in the neighborhood N k (s); Improve s' using local search to get s''; Iff(s'') < f(s): s = s'', k = 1; Else: k = k + 1; Step 4, Return the optimal solution s.

[0054] Throughout the process, strictly check various constraint conditions, such as the matching of maintenance personnel skills and task requirements, the consistency of spare part model specifications, the specialization of special tools, the capacity and environmental adaptability of the maintenance site, the continuous operation duration limit, and the timing dependence between tasks. The algorithm terminates when it reaches the preset maximum number of iterations or there is no significant improvement in the Pareto front, and outputs the final Pareto optimal solution set or recommends a comprehensive optimal solution, specifying the planned time, executor, required resources and site for each task.

[0055] Dynamic Resource Management and Collaboration Module 500: Responsible for executing the scheduling plan generated by Module 400 and managing resources. Its functions include: automatically assigning maintenance tasks to currently available maintenance personnel with the appropriate skills and planning the optimal route; upon task confirmation, the system automatically reserves or requests the required spare parts for the task, triggering expedited procurement or allocation if inventory is insufficient, and tracking spare parts; maintenance personnel reserve or request the use of tools and equipment through the platform, and the system displays their status and location in real time; the use of maintenance sites is integrated into scheduling, with the system allocating resources based on task characteristics and avoiding conflicts; and providing remote maintenance guidance based on augmented reality (AR), allowing experts to provide real-time guidance on on-site maintenance, with all guidance processes recorded.

[0056] Visual Monitoring and Feedback Optimization Module 600: This module provides a unified monitoring and management interface and drives system optimization. Its functions include: Real-time display of KPIs such as fleet health, maintenance task queues and progress, resource load, spare parts inventory, and personnel efficiency through customized dashboards; users can query the full lifecycle status of any maintenance task, and the system automatically issues alerts for delayed or abnormal tasks; historical maintenance cases are collected and organized to form a structured maintenance knowledge base to assist in diagnosis and training; and closed-loop feedback and continuous optimization are achieved by comparing actual fault information with model results to update fault diagnosis and prediction model parameters; analyzing the completion of high-priority tasks to adjust priority assessment model weights; and comparing scheduling plans with actual execution to fine-tune scheduling algorithm parameters or strategies.

[0057] Example 2 This embodiment provides a method for scheduling and managing drone maintenance tasks and resources based on the platform described in Example 1. The main steps include: Step S10: Data collection and initialization The fusion platform continuously collects UAV flight data, sensor data, component status data, maintenance resource real-time information and historical data through the data collection and fusion module 100, performs fusion processing and stores them in the database.

[0058] Step S20: Fault Monitoring, Prediction, and Task Generation: The fault diagnosis and prediction analysis module 200 monitors the health status of drones in real time. When a real-time fault alarm is detected or a current fault is identified through model analysis, a repair task is immediately generated. Simultaneously, based on the CNN-LSTM RUL prediction model, this module regularly scans the fleet and generates predictive maintenance tasks for components whose RUL falls below the warning threshold. This is combined with the pre-set regular maintenance plan to generate planned maintenance tasks. All tasks are entered into the initial maintenance task list.

[0059] Step S30: Dynamic evaluation and sorting of task priorities The task dynamic priority evaluation module 300 calculates the dynamic priority score of each task in the maintenance task list generated in S20 according to a preset evaluation model, and sorts the list according to the score.

[0060] Step S40: Intelligent Scheduling and Resource Allocation: The intelligent scheduling decision module 400 obtains the sorted maintenance task list and the latest resource availability status provided by the dynamic resource management and collaboration module 500. It then initiates the NSGA-II and VNS hybrid optimization scheduling algorithm, iteratively searching for the optimal maintenance task allocation solution based on the preset optimization objective while satisfying multiple constraints.

[0061] Step S50: Task Execution and Resource Collaboration. The dynamic resource management and collaboration module 500 automatically pushes tasks to relevant maintenance personnel based on the selected maintenance task allocation plan and coordinates spare parts delivery, tool preparation, and site reservation. Maintenance personnel execute tasks as instructed, record work hours and consumption, and complete reports. They can also initiate AR remote collaboration as needed.

[0062] Step S60: Monitoring, Feedback, and Optimization. The visual monitoring and feedback optimization module 600 monitors task execution progress and resource usage throughout the entire process. Upon task completion, complete maintenance records are collected and the knowledge base is updated. The system regularly analyzes historical scheduling performance and actual operation and maintenance data, adaptively adjusting fault diagnosis models, priority assessment weights, and scheduling algorithm parameters to achieve closed-loop optimization.

[0063] Although specific embodiments of the present invention have been described above, those skilled in the art will appreciate that these specific embodiments are merely illustrative, and that those skilled in the art may omit, substitute, and modify the details of the methods and systems described above without departing from the principles and spirit of the present invention. For example, combining the above method steps to perform substantially the same functions in substantially the same manner to achieve substantially the same results falls within the scope of the present invention. Accordingly, the scope of the present invention is limited solely by the appended claims.

Claims

1. A UAV maintenance task scheduling and resource management platform, characterized by: include: The data acquisition and fusion module is configured to acquire and fuse UAV operation data, maintenance resource data, and external environment data to form a standardized data set; a fault diagnosis and prediction analysis module configured to diagnose a current fault of the UAV based on the standardized data set using a first preset deep learning-based fault diagnosis model, and predict the remaining useful life of key components using a hybrid prediction model combining a convolutional neural network and a long short-term memory network, thereby generating a maintenance task list; an intelligent scheduling decision module configured to generate a maintenance task allocation plan based on the maintenance task list, the dynamic priority of the tasks, and the real-time availability of maintenance resources by using a hybrid optimization scheduling algorithm that integrates an improved non-dominated sorting genetic algorithm with an elite retention strategy and a variable neighborhood search; The resource dynamic management and coordination module is configured to realize the dynamic deployment and operation coordination of maintenance personnel, maintenance spare parts, maintenance tools and maintenance sites according to the maintenance task allocation plan.

2. The UAV maintenance task scheduling and resource management platform according to claim 1 is characterized in that: The UAV operation data acquired by the data acquisition and fusion module includes: UAV flight parameter data, sensor telemetry data, component health status information, historical maintenance records, and fault code data; In addition, the maintenance resource data acquired by the data collection and fusion module includes: skill level information, qualification certification information, real-time location information, busy and idle status information of maintenance personnel, model parameters, inventory quantity, storage location information, expected replenishment time information of spare parts, type information, available status information, calibration cycle information of maintenance tools, as well as capacity parameters, occupancy information, and environmental condition information of the maintenance site.

3. The UAV maintenance task scheduling and resource management platform according to claim 1 is characterized in that: The fault diagnosis and prediction analysis module also combines a fault propagation path analysis method based on a dynamic Bayesian network to identify potential fault modes and their probability of occurrence.

4. The UAV maintenance task scheduling and resource management platform according to claim 1, characterized in that: The module further includes a task dynamic priority evaluation module, the module being configured to: Construct a multi-dimensional task evaluation indicator system, including the severity of the fault, the impact on subsequent tasks, the criticality of the UAV, the difficulty of obtaining spare parts, the maintenance window, and the urgency of the predicted fault occurrence time; The method of combining analytic hierarchy process and fuzzy comprehensive evaluation is used to calculate the dynamic priority score of each maintenance task; the following formula is used to calculate the priority score of each maintenance task. Dynamic priority of maintenance tasks: in, For the The final dynamic priority score of each maintenance task, For the The weight of the first-level evaluation indicators, For the The task in The comprehensive evaluation value under the first-level indicators is For the The expected completion deadline of each task, is the current time, For the The acceptable maintenance window duration for each task, is the time urgency impact coefficient, For the The estimated comprehensive resource consumption value of each task, It is the reference value of the maximum possible resource consumption of a single task in the system. The estimated comprehensive resource consumption value of each task; Sort the maintenance task list based on the calculated priority score.

5. The UAV maintenance task scheduling and resource management platform according to claim 1, characterized in that: The hybrid optimization scheduling algorithm adopted by the intelligent scheduling decision module has multiple optimization goals: minimizing the total downtime of drones, minimizing the total maintenance cost, maximizing the balance of resource utilization, and maximizing the completion rate of high-priority tasks; In the evolution process of the non-dominated sorting genetic algorithm, an adaptive crossover operator and an adaptive mutation operator based on maintenance task characteristics and resource status are introduced; After obtaining the Pareto optimal solution set, the variable neighborhood search is used to perform local deep optimization on the key scheduling solutions in the solution set.

6. The UAV maintenance task scheduling and resource management platform according to claim 1, characterized in that: When generating maintenance task allocation plans, the intelligent scheduling decision module also considers the following constraints: matching maintenance personnel skills with task requirements, consistency between spare part models and specifications, specificity of special tools, maintenance site capacity and environmental adaptability, continuous operation duration limit, and timing dependency between different maintenance tasks.

7. The UAV maintenance task scheduling and resource management platform according to claim 1 is characterized in that: The resource dynamic management and collaboration module performs the following functions: Automatically assign the identified maintenance tasks to currently available maintenance personnel with corresponding skills, and plan the optimal route to the maintenance location or the location of the drone; Automatically reserve or request required spare parts based on task requirements, and trigger intelligent replenishment or emergency procurement processes based on spare parts inventory and in-transit damage risks; Track the real-time location and usage status of maintenance tools to ensure tool availability and schedule maintenance in a timely manner; Reserve and allocate maintenance site resources, optimize site turnover, and coordinate site sharing during multi-task parallel maintenance; It also provides remote maintenance guidance based on augmented reality, allowing senior technical experts to remotely guide on-site maintenance personnel to perform complex maintenance tasks in real time.

8. The UAV maintenance task scheduling and resource management platform according to claim 1 is characterized in that: It also includes a visual monitoring and feedback optimization module, which is configured to: Real-time display of drone location, maintenance task distribution, maintenance personnel dynamics, and resource scheduling through the geographic information system interface; Provides status tracking and early warning functions for the entire life cycle of maintenance tasks; records actual working hours data, spare parts consumption data, and fault resolution data during the maintenance process to form a maintenance knowledge base; A machine learning algorithm is used to analyze historical maintenance data and scheduling execution effects, and the fault diagnosis model parameters, the hybrid prediction model parameters, the evaluation weight of the task dynamic priority evaluation module and the strategy of the hybrid optimization scheduling algorithm are adaptively adjusted and optimized.

9. A method for scheduling and managing UAV maintenance tasks based on the platform of any one of claims 1 to 8, characterized in that: The following steps are involved: Step S1: Acquire and fuse UAV operation data, maintenance resource data, and external environment data through the data acquisition and fusion module to obtain a standardized data set; Step S2: using a fault diagnosis and prediction analysis module to diagnose the current fault of the UAV based on the standardized data set using a first preset deep learning-based fault diagnosis model, and using a hybrid prediction model combining a convolutional neural network and a long short-term memory network to predict the remaining useful life of key components, and generating a maintenance task list. If the platform includes a task dynamic priority assessment module, the maintenance task list is prioritized using the module; Step S3: The intelligent scheduling decision module generates a maintenance task allocation plan based on the sorted maintenance task list and the real-time maintenance resource availability by adopting a hybrid optimization scheduling algorithm that integrates an improved non-dominated sorting genetic algorithm with an elite retention strategy and a variable neighborhood search; Step S4: Dynamically allocate maintenance resources and coordinate operations according to the maintenance task allocation plan through the resource dynamic management and collaboration module; Step S5: If the platform includes a visual monitoring and feedback optimization module, the module is used to monitor task execution and perform feedback optimization on the system.

Images