A method, device, equipment, and storage medium for intelligent inspection of drones based on edge computing.

Abstract

This invention constructs an intelligent inspection system comprising a drone inspection terminal, an edge computing controller, a 5G communication module, and a cloud-integrated management and control platform. The drone inspection terminal collects multi-dimensional physical parameters from the site. The edge computing controller constructs an initial probability density function based on the pre-processed physical parameters, determines the probability density transition kernel function, and obtains the probability density function at each time step through time-series iterative calculation. This allows for the calculation of equipment failure probabilities. Based on the equipment failure probability, the severity of the equipment failure consequences, and the distance between the drone and the equipment, the device priority weight is calculated, an inspection task sequence is generated, and its feasibility is verified. A factor graph model is constructed, defining the potential functions of observation factors, transition factors, time factors, and location factors. The factor graph is dynamically updated based on the equipment failure probability and time complexity. Based on the dynamically updated factor graph and the real-time location of the drone, a total cost function and a heuristic function are constructed. An improved A^* algorithm is used for dynamic path planning. Finally, the edge computing controller transmits the key processing results to the cloud platform via the 5G communication module. The cloud platform performs data storage, global scheduling, and model parameter optimization.

Description

Technical Field

[0001] This invention relates to the field of edge computing and drone intelligent inspection integration, specifically to a drone intelligent inspection method, apparatus, equipment and storage medium based on edge computing. Background Technology

[0002] With the widespread application of drone technology in the industrial field, drone inspection is gradually becoming a replacement for traditional methods. However, existing technical solutions still have significant theoretical shortcomings. Current drone inspection systems generally adopt the traditional architecture of "terminal data acquisition - centralized cloud processing." This architecture centralizes data processing tasks in the cloud, inevitably leading to high latency and high bandwidth consumption during the transmission of massive amounts of raw data. Especially in scenarios with incomplete network coverage or high network load, the stability and real-time performance of data transmission are difficult to guarantee theoretically. Furthermore, existing drone inspection task scheduling and path planning lack deep theoretical integration with the physical attributes of on-site equipment. In terms of task scheduling, traditional solutions are mostly based on preset paths or static equipment location information, failing to fully consider the dynamic evolution of equipment fault states and the temporal changes in physical parameters, resulting in scheduling strategies lacking theoretical specificity and adaptability. In terms of path planning, environmental physical factors (such as the impact of rain and snow on flight stability, and the threat of gas leak concentration to flight safety) are not included in the theoretical modeling scope, making it theoretically impossible to balance inspection efficiency and flight safety in the planned path. Meanwhile, a comprehensive theoretical framework for the deep integration of edge computing and drone inspection has yet to be established. There is a lack of systematic theoretical methods for constructing probability density transfer models based on the physical attributes of field equipment, dynamically updating factor graphs, and implementing dynamic path planning. This fails to meet the theoretical and practical demands of industrial sites for high precision, high real-time performance, and high security in intelligent drone inspection. Therefore, constructing a theoretical and technical solution for intelligent drone inspection based on edge computing has become an important research direction in the current field of industrial inspection. Summary of the Invention

[0003] To address the above problems, this invention provides an intelligent inspection method for unmanned aerial vehicles (UAVs) based on edge computing, comprising the following steps:

[0004] Step 1: Construct a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module, wherein the edge computing controller adopts a heterogeneous computing architecture;

[0005] Step 2: The drone inspection terminal collects multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data;

[0006] Step 3: The edge computing controller constructs an initial probability density function based on the preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ;

[0007] Step 4: Based on equipment failure probability Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility;

[0008] Step 5: Construct a factor graph model, defining the potential functions for the observation factor, transition factor, time factor, and location factor, based on... With time complexity Dynamically update factor graph;

[0009] Step 6: Based on the dynamically updated factor graph and the real-time location of the UAV, construct the total cost function and heuristic function, and adopt an improved... The algorithm performs dynamic path planning;

[0010] Step 7: The edge computing controller transmits key processing results through... The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization.

[0011] Specifically, the data preprocessing in step (2) includes: processing the equipment temperature using a moving average filter. With gas concentration The specific calculation process is as follows:

[0012] ;

[0013] in, For the first The temperature of the filtered equipment at any given time. To adjust the sliding window size, For the first The raw device temperature is collected at all times;

[0014] Median filtering was used to process the ambient temperature, humidity, and rainfall / snowfall. The specific calculation process is as follows:

[0015] ;

[0016] Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half the width. This is a median calculation function that effectively removes impulse noise by selecting the median of the data within a window, and is based on... Criteria for removing outliers.

[0017] Specifically, step (3) includes:

[0018] Step 3.1: Construct the initial probability density function; Step 3.2: Determine the probability density transition kernel function, based on the physical evolution law of equipment failure, define the transition kernel function. ,describe Time-based physical parameter vector Transfer to time Step 3.3: Calculate the temporal probability density transition. Based on the initial probability density function and the transition kernel function, obtain the probability density function at each time step through iterative calculation. Step 3.4: Evaluate the fault probability: based on the joint probability density function. Calculate the probability of equipment failure.

[0019] Specifically, in step 3.1, the initial probability density function is a multidimensional normal distribution, i.e. ;

[0020] Among them, the Initial time The joint probability density function reflects the distribution characteristics of each physical parameter in the initial state; the The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters; For vectors transpose, Covariance matrix The inverse matrix.

[0021] In addition, this invention also provides an edge computing-based intelligent inspection device for unmanned aerial vehicles (UAVs), specifically including the following modules: a system construction module, a data acquisition module, a probability calculation module, a task scheduling module, a factor graph model construction module, a path planning module, and a storage and update module, wherein:

[0022] The system building module is used to construct a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module, wherein the edge computing controller adopts a heterogeneous computing architecture;

[0023] The data acquisition module is used by the drone inspection terminal to collect multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data;

[0024] The probability calculation module is used by the edge computing controller to construct an initial probability density function based on preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ;

[0025] The task scheduling module is used to schedule tasks based on the probability of device failure. Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility;

[0026] The factor graph model building module is used to construct factor graph models, defining the potential functions for observation factors, transition factors, time factors, and location factors, based on... With time complexity Dynamically update factor graph;

[0027] The path planning module is used to construct the total cost function and heuristic function based on the dynamically updated factor graph and the real-time location of the UAV, employing an improved... The algorithm performs dynamic path planning;

[0028] The storage and update module is used by the edge computing controller to transmit key processing results. The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization.

[0029] Specifically, the data preprocessing in the data acquisition module includes: processing the device temperature using a moving average filter. With gas concentration The specific calculation process is as follows:

[0030] ;

[0031] in, For the first The temperature of the filtered equipment at any given time. To adjust the sliding window size, For the first The raw device temperature is collected at all times;

[0032] Median filtering was used to process the ambient temperature, humidity, and rainfall / snowfall. The specific calculation process is as follows:

[0033] ;

[0034] Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half the width. This is a median calculation function that effectively removes impulse noise by selecting the median of the data within a window, and is based on... Criteria for removing outliers.

[0035] Specifically, the probability calculation module includes:

[0036] Construct the initial probability density function; determine the probability density transition kernel function, and define the transition kernel function based on the physical evolution law of equipment failure. ,describe Time-based physical parameter vector Transfer to time The probability density is calculated; the temporal probability density transition is calculated, based on the initial probability density function and the transition kernel function, the probability density function at each time step is obtained through iterative calculation; the failure probability is evaluated based on the joint probability density function. Calculate the probability of equipment failure.

[0037] Specifically, in determining the probability density transition kernel function, the initial probability density function is a multidimensional normal distribution, i.e. ;

[0038] Among them, the Initial time The joint probability density function reflects the distribution characteristics of each physical parameter in the initial state; the The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters; For vectors transpose, Covariance matrix The inverse matrix.

[0039] The present invention also provides a computer device, characterized in that the computer device includes a processor and a memory: the memory is used to store program code and transmit the program code to the processor, the processor is used to execute the above-described edge computing-based UAV intelligent inspection method according to the instructions in the program code.

[0040] The present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium is used to store program code and transmit the program code to a processor, the processor being used to execute the above-described edge computing-based UAV intelligent inspection method according to the instructions in the program code.

[0041] The core objective of this invention is to: construct a localized real-time data processing theoretical model by deeply integrating an edge computing controller with a UAV inspection terminal; establish a probability density transfer calculation theoretical framework based on multi-dimensional physical attribute data of field equipment to accurately characterize the dynamic evolution of equipment fault states; propose a factor graph dynamic update theoretical method based on probability density transfer results and time complexity analysis to achieve the organic integration of multiple state variables in the inspection system; and construct a path dynamic planning theoretical model based on factor graphs and UAV location information to achieve intelligent scheduling of inspection tasks and optimal path planning, ultimately improving inspection efficiency, fault response speed, and flight safety, and providing comprehensive theoretical support and technical solutions for intelligent inspection in industrial sites. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of a drone-based intelligent inspection method based on edge computing.

[0044] Figure 2 This is a schematic diagram of a module for an edge computing-based drone intelligent inspection device. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Example 1

[0047] like Figure 1 As shown, the present invention provides an intelligent inspection method for unmanned aerial vehicles (UAVs) based on edge computing, which specifically includes the following steps:

[0048] Step 1: Construct a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module and the edge computing controller adopt a heterogeneous computing architecture.

[0049] Specifically, in the process of constructing the intelligent inspection system, the UAV inspection terminal, as the execution unit of on-site inspection tasks, serves as the carrier for collecting multi-dimensional physical data and the main body for executing inspection paths. It is equipped with a multi-module acquisition system for collecting equipment operating status and environmental conditions, including a continuous physical parameter acquisition module (equipment temperature acquisition module, gas concentration acquisition module, ambient temperature and humidity acquisition module, rain and snow volume acquisition module) and a discrete physical parameter acquisition module (audio and visual alarm signal acquisition module), while also integrating... The positioning module obtains its own real-time spatial location information.

[0050] Specifically, the edge computing controller, as the core computing unit of the system, is theoretically positioned as a localized data processing and decision-making scheduling center, and adopts a heterogeneous computing architecture. The theoretical design of the system enables high-concurrency data processing and real-time computing. Its core theoretical function lies in real-time preprocessing, probability density transfer calculation, task scheduling decision-making, dynamic updating of factor graphs, and path planning of multi-dimensional physical data collected by UAVs, thereby avoiding latency issues caused by transmitting massive amounts of raw data to the cloud.

[0051] Specifically, the The communication module serves as a low-latency data interaction channel between the edge computing controller and the cloud-integrated management and control platform, based on... The high bandwidth and low latency characteristics of the technology enable bidirectional real-time transmission of edge computing processing results and global instructions in the cloud, theoretically ensuring the coordination and consistency between local processing and global control.

[0052] Specifically, the cloud-based integrated management and control platform serves as the optimization and configuration center for global inspection resources and the storage and analysis center for historical data. It is responsible for receiving key processing results transmitted from edge computing, issuing global scheduling instructions, storing and analyzing historical data, and optimizing and updating inspection model parameters to achieve the optimal global configuration for multi-UAV collaborative inspection.

[0053] Understandably, this involves extending edge computing capabilities to drone terminals to construct a localized closed-loop theoretical model of "collection-processing-decision," while simultaneously... The communication system enables collaboration with the cloud, ensuring the real-time nature of on-site inspections and optimizing the allocation of global resources, thereby alleviating the core contradiction between data transmission latency and cloud computing pressure.

[0054] Step 2: The drone inspection terminal collects multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data.

[0055] Specifically, the UAV inspection terminal flies along a preset initial inspection path and, based on the physical mechanism of equipment fault diagnosis, collects multi-dimensional physical parameters reflecting the equipment's operating status and environmental conditions. The theoretical definitions of various parameters are as follows:

[0056] Equipment surface temperature This reflects the equipment's operating load and potential for failure. For example, excessively high motor temperature indicates an internal fault; it is a continuous physical parameter. Gas leakage concentration... This reflects the sealing condition of the gas equipment; an excessive concentration indicates a leak. It is a continuous physical parameter. Ambient temperature. Ambient humidity : Affects the aging rate of equipment and the flight stability of drones; these are continuous physical parameters. Ambient humidity and ambient temperature jointly affect the equipment condition and the operational stability of drone electronic components; these are also continuous physical parameters. Rain and snowfall. Directly affects the flight safety and data acquisition accuracy of drones, as it is a continuous physical parameter; audible and visual alarm signals. : Directly reflects whether the equipment has triggered a fault alarm, with a value of 0 (normal) or 1 (alarm), and is a discrete physical parameter; UAV position The above refers to Indicates longitude, the stated For latitude, the Here, represents altitude, and represents spatial location parameters.

[0057] Understandably, continuous physical parameters are sampled continuously based on time series to ensure the capture of dynamic trends in parameter changes; discrete physical parameters are sampled based on event-triggered sampling, with the sampling frequency automatically increased when an alarm signal is triggered; spatial location parameters are timestamped synchronously with physical parameter acquisition to ensure consistency in the time dimension.

[0058] Furthermore, based on data noise reduction and outlier removal, the edge computing controller performs real-time preprocessing on the collected multi-dimensional physical data. The specific theoretical methods are as follows:

[0059] For equipment temperature Gas concentration The moving average filtering method is used, and the exemplary calculation process is as follows:

[0060] ;

[0061] in, For the first The temperature of the filtered equipment at any given time. The size of the sliding window is understood to be based on the sampling frequency and noise characteristics. For the first The raw device temperature collected at all times can be used to calculate the average value of the data within a window to observe and smooth data fluctuations while preserving the trend of change.

[0062] Regarding ambient temperature Ambient humidity Rain and snow volume This parameter, which is easily affected by impulse noise, is calculated using the median filtering method, as exemplarily shown below:

[0063]

[0064] Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half the width. This is a median calculation function that can effectively remove impulse noise by selecting the median of the data within a window.

[0065] Furthermore, in the outlier removal process, based on The criterion performs outlier detection on all continuous physical parameters, and its calculation process is as follows:

[0066]

[0067] Among them, the For the first The specific physical parameter values ​​of the device at any given time, exemplarily as shown in the example below. Device surface temperature at any time , No. Gas leakage concentration at any given time The The historical sampling average of specific physical parameter values, the The historical sampling standard deviation of this specific physical parameter. For the first The effective parameter values ​​at time are obtained through The criteria can identify data that exceeds the mean by ±3 standard deviations as outliers and replace them with valid data from the previous time point.

[0068] Furthermore, after extracting key target regions from the video data collected by the drone using the inter-frame difference method, a compression algorithm is used to reduce the data volume; simultaneously, all preprocessed data is timestamped and uniformly uses high-precision timestamps. Time ensures the consistency of data across different dimensions in the time dimension, providing a time reference for subsequent time series calculations.

[0069] Understandably, in this step, a complete physical information model of equipment fault status and environmental conditions is constructed through the comprehensive collection of multi-dimensional physical data. The preprocessing process is based on data processing theory and methods to ensure data reliability and provide high-quality input for subsequent probability density transfer calculations.

[0070] Step 3: The edge computing controller constructs an initial probability density function based on the preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ;

[0071] Specifically, the edge computing controller constructs a probability density transition theoretical model of equipment fault states based on multi-dimensional physical data of field devices. Its purpose is to quantitatively describe the dynamic evolution of equipment fault states, providing a theoretical basis for task scheduling and factor graph updates. The specific steps are as follows:

[0072] Step 3.1: Construct the initial probability density function

[0073] Based on the physical mechanism of equipment failure and historical inspection data, the multi-dimensional physical parameters are abstracted into a vector of random variables and set as follows:

[0074]

[0075] in The filtered gas concentration, The ambient temperature and humidity are determined by median filtering. Then, an initial probability density function is constructed using a multidimensional normal distribution. The calculation process is as follows:

[0076]

[0077] Wherein, the Initial time The joint probability density function reflects the distribution characteristics of each physical parameter under the initial state;

[0078] The The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters, such as the gas concentration. With audible and visual alarm signals The covariance is positive, indicating a positive correlation between the two; the aforementioned For vectors transpose, Covariance matrix The inverse matrix.

[0079] Step 3.2: Determine the probability density transition kernel function. Based on the physical evolution law of equipment failure, define the transition kernel function. ,describe Time-based physical parameter vector Transfer to time The probability density is calculated as follows:

[0080]

[0081] Among them, the For the first The physical parameters of the device are The value at time, For the first The transfer kernel function for the physical parameters of a device. It can be understood that for continuous parameters acquired by the device, i.e. =1 to 5 correspond to In this case, a normal transition kernel is required, that is:

[0082]

[0083] Wherein, the For the first The parameters are in The rate of change over time reflects the dynamic trend of equipment parameters. For the first The transfer standard deviation of each equipment parameter is obtained based on historical data statistics and reflects the uncertainty of changes in equipment parameters.

[0084] Furthermore, for discrete parameters, i.e. =6 corresponds to an audible and visual alarm signal. At that time, the Bernoulli transfer kernel is used, that is:

[0085]

[0086] Among them, the for Caused by the state at a given moment The probability of triggering an alarm at any given time is, for example, based on the gas concentration. Equipment temperature Parameter threshold determination.

[0087] Step 3.3: Calculate the temporal probability density transition. Based on the initial probability density function and the transition kernel function, the probability density function at each time step is obtained through iterative calculation. The calculation process is as follows:

[0088]

[0089] Among them, the for The joint probability density function at time t, with an integral range of . Time device physical parameter vector All possible values ​​of .

[0090] Furthermore, to achieve real-time computation, the Monte Carlo sampling method is used for numerical solution, with the following steps:

[0091] from Medium sampling One sample point; , , ... ;based on generate Time sample points ; Constructed through kernel density estimation The specific process is as follows:

[0092]

[0093] Among them, the For kernel density estimation bandwidth, It is a multidimensional Gaussian kernel function.

[0094] Step 3.4: Evaluate the failure probability: based on the joint probability density function. The calculation process for the equipment failure probability is as follows:

[0095]

[0096] Among them, the for Probability of equipment failure at any given time. Given the range of physical parameters for the fault, this process can quantify the severity and probability of equipment failure through integration. Understandably, this step transforms multi-dimensional physical data into quantified fault probabilities, accurately depicting the dynamic evolution of the fault state, and achieving localized real-time computation through edge computing to ensure the real-time nature of fault assessment.

[0097] Step 4: Based on equipment failure probability Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility;

[0098] Specifically, the edge computing controller constructs a theoretical model for inspection task scheduling based on device failure probability, real-time UAV location, and device location information. The goal is to achieve dynamic optimization scheduling of inspection tasks. The specific steps are as follows:

[0099] Step 4.1: Theoretical Model for Equipment Priority Ranking: Constructing an equipment priority weight formula, integrating failure probability, severity of failure consequences, and distance factors, the equipment priority weight... The calculation process is as follows:

[0100]

[0101] Among them, the For equipment The priority weight is determined by the value; a larger value indicates a higher priority. For equipment exist The failure probability at time t is specifically calculated using probability density transition; the... The severity of the consequences of a device failure (eq) is graded based on the device type and location importance; for Inspect drones and equipment regularly The straight-line distance, wherein The calculation process is as follows:

[0102]

[0103] in, Indicates that the inspection drone is in The position coordinates at that time, ( ) for equipment The pre-stored position coordinates; The weighting coefficient is set based on on-site safety requirements, for example.

[0104] Step 4.2: Generate inspection tasks based on the equipment priority weights. The sorting results generate an inspection task sequence. High-priority devices increase their sampling frequency, for example, doubling the sampling frequency of gas concentration and increasing the video frame rate; in harsh environmental areas, for example... In such cases, the priority of equipment inspection should be reduced to decrease flight risks.

[0105] Step 4.3: Verify mission feasibility based on the remaining battery power of the inspection drone. Total energy consumption To verify feasibility, the total energy consumption calculation process is as follows:

[0106]

[0107] Among them, the To inspect the flight speed of the drone, the For flight power, the For equipment The sampling time, the This represents the total power of the acquisition module; it is understandable that if That is, reserve 20% emergency power if the mission is feasible; otherwise, retain the core high-priority mission and upload the remaining missions to the cloud to coordinate with other inspection drones to complete them.

[0108] This step, based on fault probability and physical environment data, enables intelligent optimization of inspection tasks, ensuring optimal resource allocation and improving the targeting and efficiency of inspections.

[0109] Step 5: Construct a factor graph model, defining the potential functions for the observation factor, transition factor, time factor, and location factor, based on... With time complexity Dynamically update factor graph;

[0110] Specifically, the edge computing controller constructs a factor graph theoretical model, abstracting elements such as UAV location, equipment fault status, and inspection time into nodes and factors. Dynamic updates are achieved based on probability density transfer results and time complexity. The specific steps are as follows:

[0111] Step 5.1: Factor graph construction process. The factor graph consists of variable nodes and factor nodes, where variable nodes are set. Indicates the location of the inspection drone , Indicates equipment fault status ,in , Indicates inspection time Further setting of factor nodes, Represents the observed factor, and Related, Indicates the transfer factor, and and Related, Represents the time factor, and and Related, Represents the position factor, and and Related.

[0112] Step 5.2: Calculate the factor potential energy function. The potential energy function quantifies the rationality of the variable node values. Therefore, the potential energy calculation process for the above factors is as follows:

[0113] Observation factor potential function:

[0114]

[0115] Among them, the The fault probability observation value, i.e., the probability density transition result, is the... For factor graph The estimated value, the The standard deviation of the noise observed for the failure probability.

[0116] Transfer factor potential function:

[0117]

[0118] Among them, the The rate of change of the failure probability, the To transfer the standard deviation of noise.

[0119] Time factor potential function:

[0120]

[0121] Among them, the For inspection drones from arrive The distance, the The estimated inspection time is... This represents the standard deviation of time noise.

[0122] Position factor potential function:

[0123]

[0124] Among them, the For inspecting drones and equipment The distance, the The standard deviation of the location noise indicates that the higher the probability of failure, the more stringent the distance requirements.

[0125] Step 5.3: Dynamically update the factor graph, for example, every interval ,Will Input observation factors This updates the potential function of the observation factor. Updated via belief propagation algorithm ,based on Update the transfer factor potential function It reflects the failure evolution trend and calculates the time complexity.

[0126] ;

[0127] Update time factor potential function Compared with the estimated inspection time Then, updating the position factor based on the change in the drone's position is an energy function. This ensures the real-time nature of location constraints. Understandably, this step involves integrating multi-dimensional information and dynamically updating the factor graph to provide comprehensive and real-time constraints for path planning.

[0128] Step 6: Based on the dynamically updated factor graph and the real-time location of the UAV, construct the total cost function and heuristic function, and adopt an improved... The algorithm performs dynamic path planning;

[0129] Specifically, the edge computing controller constructs an improved [system / mechanism] based on a dynamically updated factor graph and the real-time location of the UAV. The theoretical model of algorithmic path planning specifically includes:

[0130] Step 6.1: Construct the total cost function for multi-objective optimization: ;

[0131] in, , , The weighting coefficients are defined as follows:

[0132] The time cost is Reflecting time costs; the security cost is

[0133] ;

[0134] Among them, the The maximum amount of rain and snow required for safe flight of the drone, the The lower explosive limit of the gas, the For optimal flight temperature, To allow for extreme flight temperatures; fault response cost is This reflects the timeliness of fault response.

[0135] Step 6.2: Improvement Algorithm heuristic function ,Right now

[0136] ;

[0137] in, ;

[0138] This indicates the maximum flight speed of the inspection drone;

[0139] , , For device nodes Environmental parameters;

[0140] The The probability of failure of the target device.

[0141] Step 6.3: Theoretical execution of path dynamic programming: Initialize the open list (nodes to be searched) and the closed list (nodes already searched), and set the starting node as... ;Calculate the total cost of nodes The From the starting node to The actual cost; choice The smallest node is used to generate neighboring nodes and calculate costs until the target device node is found; This process is repeated. The spline curve smooths the path, ensuring curvature continuity and avoiding high-risk areas.

[0142] The goal of this step is to integrate multi-dimensional costs to achieve dynamic path optimization for inspection drones, balancing efficiency and safety.

[0143] Step 7: The edge computing controller transmits key processing results through... The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization.

[0144] Specifically, the steps for constructing collaborative management and control between edge computing and cloud platforms are as follows: The theoretical mechanism for transmitting edge computing results to the cloud is based on data importance-based hierarchical transmission: high-frequency transmission of data with high real-time requirements, and real-time transmission of high-risk fault data, with a latency of ≤100ms. Cloud data storage and display adopts a distributed storage architecture. Store historical data. Real-time data is cached, and a visual interface displays information such as drone location and fault status. Global scheduling is implemented in the cloud; during multi-drone inspections, the cloud uses edge computing results from each drone to optimize global resource allocation through algorithms, avoiding collisions and balancing the load. In case of drone failure, other drones are scheduled to take over the task. A theoretical closed loop of data feedback and model optimization is established: the cloud periodically analyzes historical data, optimizes model parameters, and distributes the optimization results to the edge computing controller, enabling continuous system optimization.

[0145] This step enables the synergy between localized processing and global control, improving system stability and scalability.

[0146] Example 2

[0147] like Figure 2 As shown, the present invention provides an intelligent inspection device for unmanned aerial vehicles based on edge computing, which specifically includes the following modules: system construction module, data acquisition module, probability calculation module, task scheduling module, factor graph model construction module, path planning module, and storage and update module.

[0148] The system building module is used to construct a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module and the edge computing controller adopt a heterogeneous computing architecture.

[0149] Specifically, in the process of constructing the intelligent inspection system, the UAV inspection terminal, as the execution unit of on-site inspection tasks, serves as the carrier for collecting multi-dimensional physical data and the main body for executing inspection paths. It is equipped with a multi-module acquisition system for collecting equipment operating status and environmental conditions, including a continuous physical parameter acquisition module (equipment temperature acquisition module, gas concentration acquisition module, ambient temperature and humidity acquisition module, rain and snow volume acquisition module) and a discrete physical parameter acquisition module (audio and visual alarm signal acquisition module), while also integrating... The positioning module obtains its own real-time spatial location information.

[0150] Specifically, the edge computing controller, as the core computing unit of the system, is theoretically positioned as a localized data processing and decision-making scheduling center, and adopts a heterogeneous computing architecture. The theoretical design of the system enables high-concurrency data processing and real-time computing. Its core theoretical function lies in real-time preprocessing, probability density transfer calculation, task scheduling decision-making, dynamic updating of factor graphs, and path planning of multi-dimensional physical data collected by UAVs, thereby avoiding latency issues caused by transmitting massive amounts of raw data to the cloud.

[0151] Specifically, the The communication module serves as a low-latency data interaction channel between the edge computing controller and the cloud-integrated management and control platform, based on... The high bandwidth and low latency characteristics of the technology enable bidirectional real-time transmission of edge computing processing results and global instructions in the cloud, theoretically ensuring the coordination and consistency between local processing and global control.

[0152] Specifically, the cloud-based integrated management and control platform serves as the optimization and configuration center for global inspection resources and the storage and analysis center for historical data. It is responsible for receiving key processing results transmitted from edge computing, issuing global scheduling instructions, storing and analyzing historical data, and optimizing and updating inspection model parameters to achieve the optimal global configuration for multi-UAV collaborative inspection.

[0153] Understandably, this involves extending edge computing capabilities to drone terminals to construct a localized closed-loop theoretical model of "collection-processing-decision," while simultaneously... The communication system enables collaboration with the cloud, ensuring the real-time nature of on-site inspections and optimizing the allocation of global resources, thereby alleviating the core contradiction between data transmission latency and cloud computing pressure.

[0154] The data acquisition module is used by the drone inspection terminal to collect multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data.

[0155] Specifically, the UAV inspection terminal flies along a preset initial inspection path and, based on the physical mechanism of equipment fault diagnosis, collects multi-dimensional physical parameters reflecting the equipment's operating status and environmental conditions. The theoretical definitions of various parameters are as follows:

[0156] Equipment surface temperature This reflects the equipment's operating load and potential for failure. For example, excessively high motor temperature indicates an internal fault; it is a continuous physical parameter. Gas leakage concentration... This reflects the sealing condition of the gas equipment; an excessive concentration indicates a leak. It is a continuous physical parameter. Ambient temperature. Ambient humidity : Affects the aging rate of equipment and the flight stability of drones; these are continuous physical parameters. Ambient humidity and ambient temperature jointly affect the equipment condition and the operational stability of drone electronic components; these are also continuous physical parameters. Rain and snowfall. Directly affects the flight safety and data acquisition accuracy of drones, as it is a continuous physical parameter; audible and visual alarm signals. : Directly reflects whether the equipment has triggered a fault alarm, with a value of 0 (normal) or 1 (alarm), and is a discrete physical parameter; UAV position The above refers to Indicates longitude, the stated For latitude, the Here, altitude is denoted as , and spatial location is denoted as . It can be understood that continuous physical parameters are sampled continuously over time to ensure the capture of dynamic trends; discrete physical parameters are sampled based on event-triggered sampling, with the sampling frequency automatically increasing when an alarm signal is triggered; spatial location parameters are timestamped synchronously with physical parameter acquisition to ensure consistency in the time dimension.

[0157] Furthermore, based on data noise reduction and outlier removal, the edge computing controller performs real-time preprocessing on the collected multi-dimensional physical data. The specific theoretical methods are as follows:

[0158] For equipment temperature Gas concentration The moving average filtering method is used, and the exemplary calculation process is as follows:

[0159]

[0160] in, For the first The temperature of the filtered equipment at any given time. The size of the sliding window is understood to be based on the sampling frequency and noise characteristics. For the first The raw device temperature collected at all times can be used to calculate the average value of the data within a window to observe and smooth data fluctuations while preserving the trend of change.

[0161] Regarding ambient temperature Ambient humidity Rain and snow volume This parameter, which is easily affected by impulse noise, is calculated using the median filtering method, as exemplarily shown below:

[0162]

[0163] Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half the width. This is a median calculation function that can effectively remove impulse noise by selecting the median of the data within a window.

[0164] Furthermore, in the outlier removal process, based on The criterion performs outlier detection on all continuous physical parameters, and its calculation process is as follows:

[0165]

[0166] Among them, the For the first The specific physical parameter values ​​of the device at any given time, exemplarily as shown in the example below. Device surface temperature at any time , No. Gas leakage concentration at any given time The The historical sampling average of specific physical parameter values, the The historical sampling standard deviation of this specific physical parameter. For the first The effective parameter values ​​at time are obtained through The criteria can identify data that exceeds the mean by ±3 standard deviations as outliers and replace them with valid data from the previous time point.

[0167] Furthermore, after extracting key target regions from the video data collected by the drone using the inter-frame difference method, a compression algorithm is used to reduce the data volume; simultaneously, all preprocessed data is timestamped and uniformly uses high-precision timestamps. Time is crucial for ensuring consistency of data across different dimensions, providing a time baseline for subsequent time-series calculations. Understandably, this step involves comprehensively collecting multi-dimensional physical data to construct a complete physical information model of equipment fault states and environmental conditions. The preprocessing process, based on data processing theories and methods, ensures data reliability, providing high-quality input for subsequent probability density transfer calculations.

[0168] The probability calculation module is used by the edge computing controller to construct an initial probability density function based on preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ;

[0169] Specifically, the edge computing controller constructs a probability density transition theoretical model of equipment fault states based on multi-dimensional physical data of field devices. Its purpose is to quantitatively describe the dynamic evolution of equipment fault states, providing a theoretical basis for task scheduling and factor graph updates. The specific steps are as follows:

[0170] The initial probability density function is theoretically constructed based on the physical mechanism of equipment failure and historical inspection data, abstracting multi-dimensional physical parameters into a vector of random variables:

[0171]

[0172] in The filtered gas concentration, The ambient temperature and humidity are determined by median filtering. Then, an initial probability density function is constructed using a multidimensional normal distribution. The calculation process is as follows:

[0173]

[0174] Wherein, the Initial time The joint probability density function reflects the distribution characteristics of each physical parameter under the initial state;

[0175] The The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters, such as the gas concentration. With audible and visual alarm signals The covariance is positive, indicating a positive correlation between the two; the aforementioned For vectors transpose, Covariance matrix The inverse matrix.

[0176] Determine the probability density transition kernel function and define the transition kernel function based on the physical evolution law of equipment failure. ,describe Time-based physical parameter vector Transfer to time The probability density is calculated as follows:

[0177]

[0178] Among them, the For the first The physical parameters of the device are The value at time, For the first The transfer kernel function for the physical parameters of a device. It can be understood that for continuous parameters acquired by the device, i.e. =1 to 5 correspond to In this case, a normal transition kernel is required, that is:

[0179]

[0180] Wherein, the For the first The parameters are in The rate of change over time reflects the dynamic trend of equipment parameters. For the first The transfer standard deviation of each equipment parameter, obtained based on historical data statistics, reflects the uncertainty of parameter changes. Furthermore, for discrete parameters... =6 corresponds to an audible and visual alarm signal. At that time, the Bernoulli transfer kernel is used, that is:

[0181]

[0182] Among them, the for Caused by the state at a given moment The probability of triggering an alarm at any given time is, for example, based on the gas concentration. Equipment temperature Parameter threshold determination.

[0183] The temporal probability density transition is calculated by iteratively obtaining the probability density function at each time step based on the initial probability density function and the transition kernel function. The calculation process is as follows:

[0184]

[0185] Among them, the for The joint probability density function at time t, with an integral range of . Time device physical parameter vector All possible values ​​of .

[0186] Furthermore, to achieve real-time computation, the Monte Carlo sampling method is used for numerical solution, with the following steps:

[0187] from Medium sampling One sample point; , , ... ;

[0188] based on generate Time sample points ;

[0189] Constructed through kernel density estimation The specific process is as follows:

[0190]

[0191] Among them, the For kernel density estimation bandwidth, It is a multidimensional Gaussian kernel function.

[0192] Assess the probability of failure: based on the joint probability density function The calculation process for the equipment failure probability is as follows:

[0193]

[0194] Among them, the for Probability of equipment failure at any given time. Given the range of physical parameters for the fault, this process can quantify the severity and probability of equipment failure through integration. Understandably, this step transforms multi-dimensional physical data into quantified fault probabilities, accurately depicting the dynamic evolution of the fault state, and achieving localized real-time computation through edge computing to ensure the real-time nature of fault assessment.

[0195] The task scheduling module is used to schedule tasks based on the probability of device failure. Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility;

[0196] Specifically, the edge computing controller constructs a theoretical model for inspection task scheduling based on device failure probability, real-time UAV location, and device location information. The goal is to achieve dynamic optimization scheduling of inspection tasks. The specific steps are as follows:

[0197] The theoretical model for equipment priority ranking: Constructing an equipment priority weight formula that integrates failure probability, severity of failure consequences, and distance factors. The equipment priority weights... The calculation process is as follows:

[0198]

[0199] Among them, the For equipment The priority weight is determined by the value; a larger value indicates a higher priority. For equipment exist The failure probability at time t is specifically calculated using probability density transition; the... The severity of the consequences of a device failure (eq) is graded based on the device type and location importance; for Inspect drones and equipment regularly The straight-line distance, wherein The calculation process is as follows:

[0200]

[0201] in, Indicates that the inspection drone is in The position coordinates at that time, ( ) for equipment The pre-stored position coordinates; The weighting coefficient is set based on on-site safety requirements, for example.

[0202] Generate inspection tasks based on the device priority weights. The sorting results generate an inspection task sequence. High-priority devices increase their sampling frequency, for example, doubling the sampling frequency of gas concentration and increasing the video frame rate; in harsh environmental areas, for example... In such cases, the priority of equipment inspection should be reduced to decrease flight risks.

[0203] Step 4.3: Verify mission feasibility based on the remaining battery power of the inspection drone. Total energy consumption To verify feasibility, the total energy consumption calculation process is as follows:

[0204]

[0205] Among them, the To inspect the flight speed of the drone, the For flight power, the For equipment The sampling time, the This represents the total power of the acquisition module; it is understandable that if That is, reserve 20% emergency power if the mission is feasible; otherwise, retain the core high-priority mission and upload the remaining missions to the cloud to coordinate with other inspection drones to complete them.

[0206] This step, based on fault probability and physical environment data, enables intelligent optimization of inspection tasks, ensuring optimal resource allocation and improving the targeting and efficiency of inspections.

[0207] The factor graph model building module is used to construct factor graph models, defining the potential functions for observation factors, transition factors, time factors, and location factors, based on... With time complexity Dynamically update factor graph;

[0208] Specifically, the edge computing controller constructs a factor graph theoretical model, abstracting elements such as UAV location, equipment fault status, and inspection time into nodes and factors. Dynamic updates are achieved based on probability density transfer results and time complexity. The specific steps are as follows:

[0209] The factor graph construction process involves setting variable nodes and factor nodes. Indicates the location of the inspection drone , Indicates equipment fault status ,in , Indicates inspection time ;

[0210] Further configure factor nodes, Represents the observed factor, and Related, Indicates the transfer factor, and and Related, Represents the time factor, and and Related, Represents the position factor, and and Related.

[0211] The potential energy function of the factor is calculated, and the rationality of the variable node values ​​is quantified through the potential energy function. Therefore, the potential energy calculation process of the above factor is as follows:

[0212] Observation factor potential function:

[0213]

[0214] Among them, the The fault probability observation value, i.e., the probability density transition result, is the... For factor graph The estimated value, the The standard deviation of the noise observed for the failure probability.

[0215] Transfer factor potential function:

[0216]

[0217] Among them, the The rate of change of the failure probability, the To transfer the standard deviation of noise.

[0218] Time factor potential function:

[0219]

[0220] Among them, the For inspection drones from arrive The distance, the The estimated inspection time is... This represents the standard deviation of time noise.

[0221] Position factor potential function:

[0222]

[0223] Among them, the For inspecting drones and equipment The distance, the The standard deviation of the location noise indicates that the higher the probability of failure, the more stringent the distance requirements.

[0224] Dynamically update the factor graph, for example, every interval ,Will Input observation factors This updates the potential function of the observation factor. Updated via belief propagation algorithm ,based on Update the transfer factor potential function It reflects the failure evolution trend and calculates the time complexity. Update the time factor potential function Compared with the estimated inspection time Then, updating the position factor based on the change in the drone's position is an energy function. This ensures the real-time nature of location constraints. Understandably, this step involves integrating multi-dimensional information and dynamically updating the factor graph to provide comprehensive and real-time constraints for path planning.

[0225] The path planning module is used to construct the total cost function and heuristic function based on the dynamically updated factor graph and the real-time location of the inspection drone, employing an improved... The algorithm performs dynamic path planning;

[0226] Specifically, the edge computing controller constructs an improved [system / mechanism] based on a dynamically updated factor graph and the real-time location of the UAV. The theoretical model of algorithmic path planning specifically includes:

[0227] Construct the total cost function for multi-objective optimization:

[0228] in, , , The weighting coefficients are defined as follows:

[0229] The time cost is Reflecting time costs; the security cost is

[0230] ;

[0231] Among them, the The maximum amount of rain and snow required for safe flight of the drone, the The lower explosive limit of the gas, the For optimal flight temperature, To allow for extreme flight temperatures; fault response cost is This reflects the timeliness of fault response. Improvement Algorithm heuristic function ,Right now

[0232] ;

[0233] in, ,

[0234] This indicates the maximum flight speed of the inspection drone;

[0235] , , For device nodes Environmental parameters;

[0236] The The probability of failure of the target device.

[0237] The theoretical execution of path dynamic programming involves initializing the open list (nodes to be searched) and the closed list (nodes already searched), and setting the starting node as... ;Calculate the total cost of nodes The From the starting node to The actual cost; choice The smallest node is used to generate neighboring nodes and calculate costs until the target device node is found; This process is repeated. The spline curve smooths the path, ensuring curvature continuity and avoiding high-risk areas.

[0238] The goal of this step is to integrate multi-dimensional costs to achieve dynamic path optimization for inspection drones, balancing efficiency and safety.

[0239] The storage and update module is used by the edge computing controller to transmit key processing results. The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization.

[0240] Specifically, the steps for constructing collaborative management and control between edge computing and cloud platforms are as follows: The theoretical mechanism for transmitting edge computing results to the cloud is based on data importance-based hierarchical transmission: high-frequency transmission of data with high real-time requirements, and real-time transmission of high-risk fault data, with a latency of ≤100ms. Cloud data storage and display adopts a distributed storage architecture. Store historical data. Real-time data is cached, and a visual interface displays information such as drone location and fault status. Global scheduling is implemented in the cloud; during multi-drone inspections, the cloud uses edge computing results from each drone to optimize global resource allocation through algorithms, avoiding collisions and balancing the load. In case of drone failure, other drones are scheduled to take over the task. A theoretical closed loop of data feedback and model optimization is established: the cloud periodically analyzes historical data, optimizes model parameters, and distributes the optimization results to the edge computing controller, enabling continuous system optimization.

[0241] This step enables the synergy between localized processing and global control, improving system stability and scalability.

[0242] The present invention also provides a computer device, characterized in that the computer device includes a processor and a memory: the memory is used to store program code and transmit the program code to the processor, the processor is used to execute the above-described edge computing-based UAV intelligent inspection method according to the instructions in the program code.

[0243] The present invention also provides a computer-readable storage medium, characterized in that the computer-readable storage medium is used to store program code and transmit the program code to a processor, the processor being used to execute the above-described edge computing-based UAV intelligent inspection method according to the instructions in the program code.

[0244] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this disclosure.

Claims

1. A drone-based intelligent inspection method based on edge computing, characterized in that... The steps include: Step 1: Constructing a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module, wherein the edge computing controller adopts a heterogeneous computing architecture; Step 2: The drone inspection terminal collects multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data; Step 3: The edge computing controller constructs an initial probability density function based on the preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ; Step 4: Based on equipment failure probability Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility; Step 5: Construct a factor graph model, defining the potential functions for the observation factor, transition factor, time factor, and location factor, based on... With time complexity Dynamically update factor graph; Step 6: Based on the dynamically updated factor graph and the real-time location of the UAV, construct the total cost function and heuristic function, and adopt an improved... The algorithm performs dynamic path planning; Step 7: The edge computing controller transmits key processing results through... The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization. Step 3 specifically includes: Step 3.1: Construct the initial probability density function; Step 3.2: Determine the probability density transition kernel function, based on the physical evolution law of equipment failure, define the transition kernel function. ,describe Time-based physical parameter vector Transfer to time Step 3.3: Calculate the temporal probability density transition. Based on the initial probability density function and the transition kernel function, obtain the probability density function at each time step through iterative calculation. Step 3.4: Evaluate the failure probability: based on the probability density function... Calculate the probability of equipment failure; In step 3.1, the initial probability density function is a multidimensional normal distribution, i.e. ; Among them, the Initial time The probability density function reflects the distribution characteristics of each physical parameter in the initial state; The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters; For vectors transpose, Covariance matrix The inverse matrix.

2. The UAV intelligent inspection method based on edge computing according to claim 1, characterized in that: Step 2 data preprocessing includes: processing equipment temperature using a moving average filter. With gas concentration The specific calculation process is as follows: ； in, For the first The temperature of the filtered equipment at any given time. To adjust the sliding window size, For the first The raw device temperature is collected at all times; Median filtering was used to process the ambient temperature, humidity, and rainfall / snowfall. The specific calculation process is as follows: ; Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half of the width. This is a median calculation function that effectively removes impulse noise by selecting the median of the data within a window, and is based on... Criteria for removing outliers.

3. A drone-based intelligent inspection device based on edge computing, characterized in that: Specifically, it includes the following modules: system construction module, data acquisition module, probability calculation module, task scheduling module, factor graph model construction module, path planning module, and storage and update module, among which: The system building module is used to construct a system consisting of a drone inspection terminal, an edge computing controller, and... The intelligent inspection system consists of a communication module and a cloud-based integrated management and control platform. The drone inspection terminal is equipped with a multi-dimensional physical parameter acquisition module and... The positioning module, wherein the edge computing controller adopts a heterogeneous computing architecture; The data acquisition module is used by the drone inspection terminal to collect multi-dimensional physical parameters on site, including equipment surface temperature. Gas leak concentration Ambient temperature Ambient humidity Rain and snow volume Audible and visual alarm signals and drone location The edge computing controller preprocesses the collected data; The probability calculation module is used by the edge computing controller to construct an initial probability density function based on preprocessed physical parameters. Determine the probability density transition kernel function The probability density function at each time step is obtained through time-series iterative calculation. Then calculate the probability of equipment failure. ; The task scheduling module is used to schedule tasks based on the probability of device failure. Severity of consequences of equipment failure and the distance between the drone and the equipment Calculate device priority weight Generate inspection task sequences and verify their feasibility; The factor graph model building module is used to construct factor graph models, defining the potential functions for observation factors, transition factors, time factors, and location factors, based on... With time complexity Dynamically update factor graph; The path planning module is used to construct the total cost function and heuristic function based on the dynamically updated factor graph and the real-time location of the UAV, employing an improved... The algorithm performs dynamic path planning; The storage and update module is used by the edge computing controller to transmit key processing results. The communication module transmits data to the cloud platform, which then performs data storage, global scheduling, and model parameter optimization. The probability calculation module specifically includes: Construct the initial probability density function; determine the probability density transition kernel function, and define the transition kernel function based on the physical evolution law of equipment failure. ,describe Time-based physical parameter vector Transfer to time The probability density is calculated; the temporal probability density transition is calculated, based on the initial probability density function and the transition kernel function, the probability density function at each time step is obtained through iterative calculation; the failure probability is evaluated based on the probability density function. Calculate the probability of equipment failure; In determining the probability density transition kernel function, the initial probability density function is a multidimensional normal distribution, i.e. ; Among them, the Initial time The probability density function reflects the distribution characteristics of each physical parameter in the initial state; The initial mean vector is obtained by statistical analysis of physical parameters under historical normal conditions. The average temperature during normal operation of the equipment, the The average concentration of natural gas in a normal environment, the The average amount of rain and snow during rainless or snowless weather is described below. , The average temperature and humidity in a normal environment, the The average value of the audible and visual alarm signal under normal conditions; The initial covariance matrix reflects the correlation between various physical parameters; For vectors transpose, Covariance matrix The inverse matrix.

4. The edge computing-based unmanned aerial vehicle (UAV) intelligent inspection device according to claim 3, characterized in that: The data preprocessing in the data acquisition module includes: processing the device temperature using a moving average filter. With gas concentration The specific calculation process is as follows: ； in, For the first The temperature of the filtered equipment at any given time. To adjust the sliding window size, For the first The raw device temperature is collected at all times; Median filtering was used to process the ambient temperature, humidity, and rainfall / snowfall. The specific calculation process is as follows: ; Among them, the For the first Rain and snowfall amount after median filtering at time points, the The median window width is half of the width. This is a median calculation function that effectively removes impulse noise by selecting the median of the data within a window, and is based on... Criteria for removing outliers.

5. A computer device, characterized in that, The computer device includes a processor and a memory: the memory is used to store program code and transmit the program code to the processor, and the processor is used to execute the UAV intelligent inspection method based on edge computing as described in claim 1 or 2 according to the instructions in the program code.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code and transmit the program code to a processor, which is used to execute the UAV intelligent inspection method based on edge computing as described in claim 1 or 2 according to the instructions in the program code.

Images