Intelligent calculation node health state monitoring and fault early warning system for unmanned aerial vehicle cluster

Abstract

The invention is suitable for the technical field of unmanned aerial vehicle cluster monitoring, and provides an unmanned aerial vehicle cluster-oriented intelligent calculation node health state monitoring and fault early warning system, which comprises the following steps of: acquiring multi-dimensional operation state data and spatial position information of each node in real time through an unmanned aerial vehicle cluster node group; a standardized node data set is generated after the data is processed by the ground data aggregation unit, a dynamic weighted topological graph is constructed by a dynamic topological partition module in the cloud intelligent analysis platform, and a cluster is divided into a plurality of collaborative unit sub-graphs; the multi-source symptom extraction module is used for extracting running state data change characteristics of multiple nodes in each sub-graph to generate structured symptom description data, and the fault mode matching module is used for matching the description data with a historical symptom characteristic sequence in a knowledge base; and determining a target fault mode according to a matching result, generating fault early warning information when an early warning condition is met, and finally displaying the fault early warning information through the user terminal. And the accuracy and timeliness of fault early warning are effectively improved.

Description

A Health Status Monitoring and Fault Early Warning System for Intelligent Computing Nodes in Unmanned Aerial Vehicle (UAV) Swarms Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) swarm monitoring technology, and in particular to an intelligent computing node health status monitoring and fault early warning system for UAV swarms. Background Technology

[0002] Unmanned aerial vehicle (UAV) swarms consist of multiple collaborative, intelligent computing nodes and are widely used in reconnaissance, material delivery, and environmental monitoring. During mission execution, each node collects its own operational status data and spatial location information in real time and exchanges data via communication links. Existing health monitoring methods for UAV swarms primarily rely on ground stations monitoring each node independently. Each node periodically uploads data on indicators such as battery level, temperature, and communication signal strength. The ground station uses preset threshold rules to identify anomalies in individual indicators, triggering an alarm when an indicator exceeds the threshold. Simultaneously, some monitoring systems use log recording to store the operational data of each node in a database, allowing maintenance personnel to perform retrospective analysis after a failure.

[0003] Current technologies can only independently assess a single indicator of a single node, failing to detect correlational anomalies arising from collaborative operations among multiple nodes within a cluster. Because nodes in a drone swarm exhibit close communication coupling and spatial coordination during mission execution, an anomaly in a single node's indicator may only be a localized phenomenon. The true cause of mission failure is often a complex fault involving synchronized degradation of communication link quality and coordinated energy consumption anomalies among multiple nodes. The lack of analysis into the internal coupling relationships within the swarm makes it impossible to extract symptom features reflecting the overall operational status of the swarm from multi-source data, resulting in insufficient accuracy and timeliness of fault warnings. Summary of the Invention

[0004] This invention provides a health status monitoring and fault early warning system for intelligent computing nodes in unmanned aerial vehicle (UAV) swarms, which addresses the issues of insufficient accuracy and timeliness in fault early warning.

[0005] This invention provides a health status monitoring and fault early warning system for intelligent computing nodes in a drone swarm, comprising: a drone swarm node group, composed of multiple intelligent computing node drones, used to collect and transmit multi-dimensional operational status data and spatial location information of each node in real time; a ground data aggregation unit, communicatively connected to the drone swarm node group, used to receive multi-dimensional operational status data and spatial location information from multiple intelligent computing node drones, and process the received multi-dimensional operational status data and spatial location information to generate a standardized node dataset; and a cloud-based intelligent analysis platform, communicatively connected to the ground data aggregation unit, the cloud-based intelligent analysis platform including a dynamic topology partitioning module, a multi-source symptom extraction module, and a fault mode matching module; the dynamic topology partitioning module is used to construct a fault mode matching module based on the spatial location relationships of nodes and the quality parameters of inter-node communication links in the standardized node dataset. A dynamically weighted topology graph representing the coupling strength within the cluster is constructed, and the dynamically weighted topology graph is divided into multiple collaborative unit subgraphs based on an adaptive graph segmentation algorithm. The multi-source symptom extraction module is connected to the dynamic topology partitioning module and is used to extract the change characteristics of multi-dimensional operating status data of multiple intelligent computing node UAVs within each collaborative unit subgraph within a preset time window, generating structured symptom description data representing the current operating status of the subgraph. The fault mode matching module is connected to the multi-source symptom extraction module and a pre-built fault mode knowledge base, respectively, and is used to match the structured symptom description data with historical symptom feature sequences in the fault mode knowledge base, determine the target fault mode based on the matching result, and generate fault warning information when the warning conditions are met. The user terminal is communicatively connected to the cloud-based intelligent analysis platform and is used to receive and display the fault warning information.

[0006] Furthermore, each intelligent computing node drone in the drone cluster node group includes: a node status acquisition unit for real-time acquisition of multi-dimensional operational status data of the node; a node positioning unit for real-time acquisition of the node's spatial location information; and a node communication unit connected to the node status acquisition unit and the node positioning unit respectively, for sending the multi-dimensional operational status data and spatial location information to the ground data aggregation unit.

[0007] Furthermore, the ground data aggregation unit includes: a multi-source data receiving module, communicatively connected to the UAV cluster node group, for receiving multi-dimensional operational status data and spatial location information from multiple intelligent computing node UAVs in parallel; a time synchronization and alignment module, connected to the multi-source data receiving module, for time alignment processing of timestamp inconsistencies caused by clock differences and transmission delays among different intelligent computing node UAVs, generating multi-source time-series data with a unified time reference; a heterogeneous data aggregation module, connected to the time synchronization and alignment module, for structurally aggregating multi-dimensional operational status data from different intelligent computing node UAVs with different acquisition frequencies and data formats according to node identifiers and time series; and a standardized output module, connected to the heterogeneous data aggregation module, for converting the aggregated data into a standardized node dataset in a preset format and sending it to the cloud-based intelligent analysis platform.

[0008] Furthermore, in the dynamically weighted topology graph, each node corresponds to a smart computing node drone, and the node weight is determined based on at least one of the node's computing power, remaining battery power, and task role; each edge corresponds to a communication link between two nodes, and the edge weight is determined based on at least one of the communication link's signal strength, transmission delay, and packet loss rate.

[0009] Furthermore, the dynamic topology partitioning module is specifically used to: calculate the spatial proximity and communication link quality between each pair of nodes based on the standardized node dataset, and comprehensively obtain the coupling strength between nodes; construct the dynamic weighted topology graph with nodes as vertices and the coupling strength as edge weights; and divide the dynamic weighted topology graph into multiple cooperative unit subgraphs by adopting an adaptive graph partitioning algorithm with the goal of minimizing the coupling strength between subgraphs and maximizing the coupling strength within subgraphs.

[0010] Furthermore, the adaptive graph partitioning algorithm aims to minimize the coupling strength between subgraphs and maximize the coupling strength within subgraphs. It divides the dynamically weighted topology graph into multiple collaborative unit subgraphs, including: initializing each node as an independent subgraph; calculating the merging benefit between adjacent subgraphs, where the merging benefit is a weighted sum of the increase in coupling strength within a subgraph and the decrease in coupling strength between subgraphs after merging two adjacent subgraphs; setting an initial threshold, determined based on a preset cluster size range; merging the corresponding two subgraphs into a new subgraph when the merging benefit exceeds the current threshold; recalculating and updating the current threshold after each round of merging based on the current number of remaining subgraphs and the current node density; and repeating the steps of calculating the merging benefit between adjacent subgraphs, merging, and recalculating and updating the current threshold until the merging benefit between all adjacent subgraphs is less than or equal to the current threshold, resulting in the final multiple collaborative unit subgraphs.

[0011] Furthermore, the multi-source symptom extraction module is specifically used to: acquire multi-dimensional operational status data of all intelligent computing node UAVs within each collaborative unit subgraph within a preset time window; extract statistical features and temporal variation features of the operational status data of each dimension within the preset time window; and generate structured symptom description data corresponding to the current collaborative unit subgraph based on the extracted statistical features and temporal variation features. The structured symptom description data includes subgraph identifier, timestamp, and feature values ​​of each dimension.

[0012] Furthermore, the statistical features include at least one of the mean, variance, maximum value, and minimum value; the time-series variation features include at least one of the linear fitting slope, fluctuation amplitude, and period intensity.

[0013] Furthermore, the fault mode matching module is specifically used for: acquiring structured symptom description data corresponding to each collaborative unit subgraph; calculating the temporal similarity and feature similarity between the structured symptom description data and each historical symptom feature sequence in the fault mode knowledge base; performing weighted fusion on the temporal similarity and feature similarity to obtain the comprehensive matching degree between the structured symptom description data and each historical symptom feature sequence; determining the fault type label corresponding to the historical symptom feature sequence with the highest comprehensive matching degree as a candidate target fault mode; when the comprehensive matching degree of the candidate target fault mode exceeds a preset matching degree threshold, determining that the warning condition is met, and generating fault warning information containing the candidate target fault mode.

[0014] Furthermore, the fault mode data stored in the fault mode knowledge base includes fault type labels, historical symptom feature sequences, and fault propagation patterns; the candidate target fault modes include the fault type labels and the corresponding fault propagation patterns; and the fault warning information includes a list of affected nodes predicted based on the fault propagation patterns and the expected failure time.

[0015] As can be seen from the above technical solution, the present invention has the following advantages: The present invention collects multi-dimensional operational status data and spatial location information of each node in real time through the node group of the UAV swarm. After processing by the ground data aggregation unit, a standardized node dataset is generated. Then, the dynamic topology partitioning module in the cloud intelligent analysis platform constructs a dynamic weighted topology graph based on the spatial location relationship of the nodes and the quality parameters of the communication link. Based on the adaptive graph segmentation algorithm, the swarm is divided into multiple cooperative unit subgraphs. The multi-source symptom extraction module extracts the operational status data change features of multiple nodes in each subgraph within a preset time window to generate structured symptom description data. The fault mode matching module matches this description data with the historical symptom feature sequences in the pre-built fault mode knowledge base. Based on the matching results, the target fault mode is determined, and fault warning information is generated when the warning conditions are met. Finally, it is displayed through the user terminal. The present invention realizes dynamic modeling and refined monitoring of the cooperative relationship of multiple nodes within the UAV swarm. It can automatically extract composite symptom features reflecting the overall operational status of the swarm and realize intelligent warning through knowledge matching with historical fault modes, effectively improving the accuracy and timeliness of fault warning, and realizing the improvement from single-point independent monitoring to swarm collaborative perception. Attached Figure Description

[0016] Figure 1 is an architecture diagram of a smart computing node health status monitoring and fault early warning system for UAV clusters according to the present invention; Figure 2 is a schematic diagram of the working principle of the dynamic topology partitioning module according to the present invention; Figure 3 is a schematic diagram of the working principle of the multi-source symptom extraction module according to the present invention; Figure 4 is a schematic diagram of the working principle of the fault mode matching module according to the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 (Referring to Figure 1): The present invention provides an intelligent computing node health status monitoring and fault early warning system for UAV swarms. This system includes a UAV swarm node group, composed of multiple intelligent computing node UAVs, used to collect and transmit multi-dimensional operational status data and spatial location information of each node in real time; a ground data aggregation unit, communicatively connected to the UAV swarm node group, used to receive multi-dimensional operational status data and spatial location information from multiple intelligent computing node UAVs, and process the received multi-dimensional operational status data and spatial location information to generate a standardized node dataset; and a cloud-based intelligent analysis platform, communicatively connected to the ground data aggregation unit. The cloud-based intelligent analysis platform includes a dynamic topology partitioning module, a multi-source symptom extraction module, and a fault mode matching module. The dynamic topology partitioning module is used to determine the spatial location relationships between nodes and the communication links between nodes in the standardized node dataset. The system employs several methods: Quality parameters are used to construct a dynamically weighted topology graph representing the coupling strength within the cluster, and an adaptive graph segmentation algorithm is used to divide the dynamically weighted topology graph into multiple collaborative unit subgraphs. A multi-source symptom extraction module is connected to the dynamic topology partitioning module to extract the changing characteristics of multi-dimensional operational status data of multiple intelligent computing node UAVs within each collaborative unit subgraph within a preset time window, generating structured symptom description data representing the current operational status of the subgraph. A fault mode matching module is connected to both the multi-source symptom extraction module and a pre-built fault mode knowledge base to match the structured symptom description data with historical symptom feature sequences in the fault mode knowledge base, determine the target fault mode based on the matching results, and generate fault warning information when warning conditions are met. A user terminal communicates with the cloud-based intelligent analysis platform to receive and display fault warning information. The following describes the technical solution of this invention in detail using a specific scenario of a UAV cluster performing a reconnaissance and strike mission in a military operation: In a certain combat mission, the command center received intelligence that "abnormal activity appeared near a water pump factory," and immediately dispatched a UAV cluster consisting of multiple intelligent computing node UAVs to the target area to perform reconnaissance and strike missions. The cluster comprises 20 drones of various types, including reconnaissance drones, strike drones, and communication relay drones, forming a collaborative combat unit. Each drone acts as an intelligent computing node within the cluster, collecting real-time multi-dimensional operational status data and spatial location information, which is then transmitted to the ground data aggregation unit via a wireless communication link. Operational status data includes node voltage, current, remaining battery power, core processor temperature, communication module signal strength, and mission payload operating status; spatial location information includes latitude, longitude, altitude, speed, and heading, provided by the onboard GPS / BeiDou satellite navigation system module. The ground data aggregation unit, serving as the ground station for the forward command post, receives data streams from the 20 drones in parallel via a high-gain antenna, and performs time synchronization and format unification processing on the received multi-source data to generate a standardized node dataset.The dataset contains a unique identifier for each node (reconnaissance drone 01, strike drone 05), a unified timestamp, spatial coordinates, and state values ​​for each dimension, organized in a structured table format for further analysis by a cloud-based intelligent analysis platform.

[0019] The cloud-based intelligent analysis platform, deployed in the rear command center, receives standardized node datasets. First, the dynamic topology partitioning module constructs a dynamically weighted topology graph based on node spatial location and the quality of inter-node communication links. Each vertex in the graph corresponds to an intelligent computing node UAV, and each edge corresponds to a communication link between two UAVs. The edge weight comprehensively reflects the communication quality. Then, an adaptive graph segmentation algorithm is used to divide the topology graph into multiple collaborative unit subgraphs, aiming to minimize inter-subgraph coupling strength and maximize intra-subgraph coupling strength. UAVs within each subgraph are spatially adjacent and have high communication quality, forming functionally coupled small groups. For example, three reconnaissance UAVs and two strike UAVs performing the same area reconnaissance mission are grouped into one subgraph, while two UAVs responsible for communication relay are grouped into a separate subgraph. The partitioning results are updated in real-time as the cluster moves. The multi-source symptom extraction module extracts the changing characteristics of multi-dimensional operational status data within a preset time window for each collaborative unit subgraph. The preset time window is dynamically adjusted according to the task type; for example, a 5-minute window is used for slowly changing parameters such as remaining battery power, and a 30-second window is used for rapidly changing parameters such as communication signal strength. Taking the submap where reconnaissance UAV 01 is located as an example, the changes in data such as battery voltage, processor temperature, and communication signal strength of the five UAVs in this submap within 5 minutes are extracted, including statistical features and temporal variation features of each dimension. All features are combined into structured symptom description data, including the submap identifier "Submap A", the timestamp "10:25:30", and feature vectors of each dimension. The fault mode matching module matches the structured symptom description data with historical symptom feature sequences in a pre-built fault mode knowledge base. This knowledge base stores cluster fault cases recorded in historical combat missions. Each case includes a fault type label, a symptom feature sequence for a period of time before the fault occurred, and the fault propagation pattern. The matching process first calculates the temporal similarity and feature similarity between the real-time symptom and each historical sequence, and then weights and fuses them to obtain a comprehensive matching degree. If the highest comprehensive matching degree exceeds the preset warning threshold, the fault type of the historical fault is determined as the target fault mode, and a fault warning message is generated. The warning threshold is preset based on the statistical distribution of historical fault samples. For example, when the overall matching degree of subgraph A matches the historical case "communication link interruption" by 0.92, the system determines that a communication link interruption failure is about to occur in the subgraph, generates a warning message, including the fault type "communication link interruption", the corresponding fault propagation pattern "after the communication relay node fails, surrounding nodes will successively lose contact within 2 minutes", the list of expected affected nodes "reconnaissance drone 02, strike drone 03, strike drone 04" and the failure time "expected in 30 seconds". The user terminal communicates with the cloud platform to display the warning information in a visual way, such as highlighting the risky subgraph A on the map and popping up a warning details window to remind the commander to adjust the task deployment in time.

[0020] In this embodiment, each intelligent computing node drone in the drone cluster node group includes: a node status acquisition unit, used to collect multi-dimensional operating status data of the node in real time; a node positioning unit, used to obtain the real-time spatial location information of the node; and a node communication unit, which is connected to the node status acquisition unit and the node positioning unit respectively, and is used to send the multi-dimensional operating status data and spatial location information to the ground data aggregation unit.

[0021] Taking the reconnaissance UAV 01 as an example, its node status acquisition unit consists of various sensors embedded in the flight control system. The voltage sensor uses a resistor divider and analog-to-digital converter circuit to acquire the power battery voltage in real time, with a range of 0–60V and an accuracy of ±0.1V. The current sensor uses the Hall effect principle to monitor charging and discharging current, with a range of 0–100A and an accuracy of ±0.5A. The temperature sensor uses a surface-mount negative temperature coefficient thermistor, closely attached to the processor heatsink and battery surface, with a measurement range of -40℃ to 125℃ and an accuracy of ±0.5℃. The communication module status is obtained by reading the registers of the wireless transceiver chip to obtain signal strength and bit error rate. All sensors acquire data at a frequency of 5 Hz, and the data is initially filtered and buffered by the microcontroller built into the node. The node positioning unit uses a multi-mode satellite navigation receiver to output NMEA format positioning data, including latitude and longitude, altitude, ground heading, and horizontal velocity. To cope with satellite signal loss, the unit integrates an inertial measurement unit, which uses Kalman filtering to achieve integrated navigation, outputting continuous and smooth position information at a frequency of 20 Hz. The node communication unit employs a fourth-generation / fifth-generation mobile communication module to establish a two-way communication link with the ground data aggregation unit. The communication protocol is based on the message queue telemetry transmission protocol, and data packets are encrypted using a 128-bit advanced encryption key before transmission to ensure operational data security. The unit has a built-in 256MB buffer to temporarily store data in the event of a network interruption and retransmit it upon recovery, ensuring data integrity. Simultaneously, the unit receives instructions from the ground aggregation unit, such as adjusting the acquisition frequency and uploading cached data.

[0022] In this embodiment, the ground data aggregation unit includes: a multi-source data receiving module, which is communicatively connected to the UAV cluster node group, for receiving multi-dimensional operational status data and spatial location information from multiple intelligent computing node UAVs in parallel; a time synchronization and alignment module, which is connected to the multi-source data receiving module, for processing the time stamp inconsistency caused by clock differences and transmission delays of different intelligent computing node UAVs, and generating multi-source time-series data with a unified time reference; a heterogeneous data aggregation module, which is connected to the time synchronization and alignment module, for structurally aggregating multi-dimensional operational status data from different intelligent computing node UAVs with different acquisition frequencies and data formats according to node identifiers and time series; and a standardized output module, which is connected to the heterogeneous data aggregation module, for converting the aggregated data into a standardized node dataset in a preset format and sending it to the cloud-based intelligent analysis platform.

[0023] Specifically, the multi-source data receiving module is deployed on the ground station server of the forward command post. It employs a high-performance network interface card and a multi-threaded concurrency model, simultaneously monitoring thousands of User Datagram Protocol (UDP) ports and receiving data packets from 20 UAVs in parallel. An internal circular buffer is configured to prevent packet loss due to instantaneous high concurrency. For lost or out-of-order data packets, the module records the data and triggers a retransmission request. The time synchronization and alignment module addresses the inconsistency in timestamps caused by differences in clocks among UAVs and transmission delays. The ground station synchronizes with a standard time source via the BeiDou time synchronization module to maintain a high-precision clock. When a node data packet is received, the module converts the data timestamp to the ground station's unified time reference based on the node timestamp and network transmission delay estimate in the data packet. The specific algorithm uses round-trip delay measurement combined with linear interpolation to time-align non-uniformly arriving data, generating multi-source time-series data with a unified timestamp. The heterogeneous data aggregation module structurally aggregates multi-dimensional state data from different UAVs with different sampling frequencies and formats according to node identifiers and time order. The module first builds an index based on node identifiers, then arranges the data for each dimension of each node by time to form a time series matrix. For dimensions with inconsistent sampling frequencies, linear interpolation is used to align the data to a unified sampling time. Finally, a dense data table is generated, with timestamps as rows and node identifiers and dimension names as columns. The standardization output module converts the aggregated data table into a predefined standardized format to reduce size and improve transmission efficiency. The module adds a metadata header containing data collection start and end times, number of nodes, dimension list, etc., and then sends it to the cloud-based intelligent analysis platform via a military-grade secure fiber optic network. The transmission strategy supports both real-time streaming and batch transmission modes, which can be dynamically switched based on the cloud's processing capacity.

[0024] Example 2: The working principle of each module in the cloud-based intelligent analysis platform is explained below: Please refer to Figure 2. The dynamic topology partitioning module is specifically used for: 1. Calculating the spatial proximity and communication link quality between each pair of nodes based on the standardized node dataset, and comprehensively obtaining the coupling strength between nodes; specifically, spatial proximity is used to measure the closeness of two intelligent computing node drones in spatial location, using Euclidean distance calculation, assuming nodes... and nodes The three-dimensional spatial coordinates are respectively and Then the Euclidean distance between the two The calculation formula is: in: , , They are nodes The longitude, latitude, and altitude coordinates; , , They are nodes The longitude, latitude, and altitude coordinates.

[0025] Defined as a normalized function of distance, the calculation formula is: in: For reference distance, it can be preset according to the cluster task type. It can be set to 500 meters when performing regional reconnaissance tasks and 1000 meters when performing wide-area surveillance tasks. The value range is (0,1], and the larger the value, the closer the two nodes are in space.

[0026] Communication link quality is used to measure the reliability of wireless communication between two nodes, combining three indicators: signal strength, transmission delay, and packet loss rate. Let node... With nodes The signal strength between Transmission delay is Packet loss rate The value range is [0,1]. Normalization is performed on each indicator: normalized signal strength value. The calculation formula is: in: and These are the preset maximum and minimum signal strength values, set according to the performance specifications of the communication equipment. Here, the preferred values ​​are... , .

[0027] Normalized transmission delay The calculation formula is: in: and These are the preset maximum and minimum transmission delay values, set according to communication protocol requirements; the preferred values ​​here are... , .

[0028] Normalized packet loss rate The calculation formula is: in: The value range is [0,1].

[0029] Communication link quality The calculation formula is: in: , , Let be the weight coefficient, and satisfy... The weighting can be dynamically adjusted based on the actual communication environment. For example, in environments with strong electromagnetic interference, the weighting of packet loss rate can be increased. , , .

[0030] Coupling strength between nodes The formula for calculating the combined spatial proximity and communication link quality is as follows: in: , Let be the weight coefficient, and satisfy... Adjustments can be made based on mission requirements; for example, for coordinated strike missions that heavily rely on communication, the intensity can be increased. Get value, get , .

[0031] II. Constructing a Dynamically Weighted Topology Graph with Nodes as Vertices and Coupling Strength as Edge Weights: After calculating the coupling strength between nodes, the dynamic topology partitioning module constructs a dynamically weighted topology graph to describe the coupling relationships within the cluster, based on the location information and communication link quality parameters of all intelligent computing node drones in the standardized node dataset. The construction process of this topology graph is as follows: First, each intelligent computing node drone recorded in the standardized node dataset is treated as an independent vertex, and all vertices form a vertex set. Each vertex has a unique identifier, corresponding to the drone's node number. The number of vertices is equal to the total number of drones in the current cluster. For example, in the 20 drone cluster of Example 1, the vertex set... It contains 20 vertices, corresponding to various drones such as reconnaissance drone 01 and strike drone 03. Secondly, based on the inter-node communication link quality parameters centrally recorded in the standardized node dataset, the connection relationships between vertices are determined. For any two vertices... and If there is a communication link between the two corresponding intelligent computing nodes (drones), then an undirected edge is established between the two vertices, and all edges constitute an edge set. The existence of a communication link can be determined by the link quality parameter. For example, if the quality of the communication link between nodes is greater than zero, a communication link is considered to exist, and an edge is established; if two nodes cannot communicate, no edge is established.

[0032] The inter-node coupling strength calculated above As edge weights, they are assigned to edges between corresponding vertices. Coupling strength. The value range is [0,1]. A larger value indicates a higher spatial proximity between the two nodes and a better communication link quality. For node pairs without a communication link, no edge is established or the edge weight is set to zero. Node weights are determined based on at least one of the following: computing power, remaining battery power, and task role, and assigned to the corresponding vertices. Node Weights Used to characterize nodes The importance within the cluster is calculated using the following formula: in: The normalized value for computing power is determined based on the processor's clock speed. For example, if the processor of a certain type of drone has a clock speed of 2.0 GHz, the normalized value is 0.8. This represents the remaining battery percentage, expressed as the percentage of the battery currently at full capacity, such as 80% being 0.8. The task role weight coefficient can be preset, such as 0.7 for reconnaissance nodes, 0.8 for strike nodes, and 0.9 for relay nodes. The maximum value of these three values ​​is taken as the node weight, reflecting the node's most prominent attribute. The above vertex set... Edge set Node weights and edge weight Together they form a dynamically weighted topology graph, denoted as ,in For the set of node weights, Let be the set of edge weights.

[0033] Third, an adaptive graph segmentation algorithm is adopted to divide the dynamically weighted topology graph into multiple cooperative unit subgraphs with the goal of minimizing the coupling strength between subgraphs and maximizing the coupling strength within subgraphs.

[0034] Adaptive graph segmentation is a graph clustering method that dynamically adjusts the segmentation threshold. It divides the entire topological graph into several subgraphs with tightly connected internal connections and sparse external connections, i.e., cooperative unit subgraphs. Minimizing the coupling strength between subgraphs means minimizing the sum of edge weights between different subgraphs, while maximizing the coupling strength within a subgraph means maximizing the sum of edge weights between nodes within each subgraph. This algorithm achieves this goal by iteratively merging subgraphs. The specific steps are as follows: 1. Initialize each node as an independent subgraph; initially, there are a total of Sub-image, The total number of nodes in the cluster is represented by the number of nodes in each subgraph, which contains only one intelligent computing node drone.

[0035] 2. Calculate the merging benefit between adjacent subgraphs. The merging benefit is the weighted sum of the increase in coupling strength within the subgraph and the decrease in coupling strength between the subgraphs after merging two adjacent subgraphs. Adjacent subgraphs are defined as subgraphs where at least one edge connects the nodes in both subgraphs. Let the subgraphs be... Hezi Diagram are two adjacent subgraphs. Before merging, the internal coupling strength of subgraph is , that is, the sum of the weights of all edges inside subgraph ; the internal coupling strength of subgraph is ; the sum of the edge weights between subgraph and subgraph is . After merging, the new subgraph is obtained, and its internal coupling strength is calculated by the formula: The increase value of the internal coupling strength of the merged subgraph is , and the calculation formula is: Before merging, the sum of the edge weights between subgraph and other adjacent subgraphs (except ) is denoted as , and the sum of the edge weights between subgraph and other adjacent subgraphs (except ) is denoted as . After merging, the sum of the edge weights between the new subgraph and other adjacent subgraphs is denoted as . The decrease value of the coupling strength between subgraphs after merging is calculated by the formula: The merging benefit is defined as the weighted sum of and , and the calculation formula is: Where: is the balance coefficient, which is set according to the task requirements. Taking means that the increase in the internal coupling strength of the subgraph is as important as the decrease in the coupling strength between subgraphs.

[0036] 3. Set an initial threshold. The initial threshold is determined according to the preset scale range of the cluster; the preset scale range refers to the range of the number of nodes expected to be included in the collaborative unit subgraph. For example, it is expected that each subgraph contains 3 to 8 nodes. The initial threshold is calculated by the formula: Where is the total number of cluster nodes; is the adjustment factor, which is determined through simulation experiments. The preferred value is the expected number of subgraphs, which is calculated according to the preset scale range. For example, if the expected subgraph scale is 5, then In military missions, adjustments should be made based on the mission type: when there is a high requirement for reconnaissance area coverage, the sub-map size should be small, and the initial threshold should be set low to make the merging conditions easier to meet, thus increasing the number of sub-maps; when there is a high requirement for coordinated strikes, the sub-map size should be large, the initial threshold should be set high, the merging conditions should be strict, and the number of sub-maps should be reduced.

[0037] 4. When the merged revenue exceeds the current threshold, merge the two corresponding subgraphs into a new subgraph; iterate through all adjacent subgraph pairs and calculate the merged revenue. ,like ,in If the current threshold is met, then a merge operation is performed, and the subgraphs are merged. Hezi Diagram Merge into a new subgraph After merging, the new subgraph It includes all nodes and edges of the original two subgraphs. The node weights can be recalculated as the average or maximum value of the node weights within the subgraph, while the edge weights maintain the original connection relationships.

[0038] 5. After each round of merging, recalculate and update the current threshold based on the current number of remaining subgraphs and the current node density. The current number of remaining subgraphs is denoted as... Current node density Defined as the ratio of the total number of cluster nodes to the area covered by the cluster, the calculation formula is: in The cluster coverage area is estimated using the minimum bounding box or convex hull area of ​​all nodes. Updated threshold. The calculation formula is: in: and The adjustment coefficient is dimensionless and can be obtained through training based on historical data. , This formula causes the threshold to decrease as the number of remaining subgraphs decreases (allowing for further merging) and increase as node density increases (preventing subgraphs from becoming too large).

[0039] 6. Repeat the steps of calculating the merging benefit between adjacent subgraphs, merging, and recalculating and updating the current threshold until the merging benefit between all adjacent subgraphs is less than or equal to the current threshold, thus obtaining the final multiple collaborative unit subgraphs.

[0040] Repeat steps 2 through 5: calculate the merging benefit between adjacent subgraphs, perform a merge when the merging benefit exceeds the current threshold, and update the current threshold after each round of merging, until the merging benefit between all adjacent subgraphs is less than or equal to the current threshold, resulting in multiple collaborative unit subgraphs. At this point, merging any two adjacent subgraphs will not bring sufficient benefit, and the partitioning reaches a local optimum.

[0041] Please refer to Figure 3. The multi-source symptom extraction module is specifically used for: 1. Obtaining multi-dimensional operational status data of all intelligent computing node UAVs within each collaborative unit subgraph within a preset time window; 2. Extracting statistical features and temporal variation features of the operational status data of each dimension within the preset time window; wherein, the statistical features include at least one of mean, variance, maximum value, and minimum value; the temporal variation features include at least one of linear fitting slope, fluctuation amplitude, and periodic intensity; 3. Generating structured symptom description data corresponding to the current collaborative unit subgraph based on the extracted statistical features and temporal variation features. The structured symptom description data includes subgraph identifier, timestamp, and feature values ​​of each dimension.

[0042] Specifically, multidimensional operational status data refers to various status parameters collected in real time by each intelligent computing node UAV, including node voltage, node power, remaining battery power, core processor temperature, communication module signal strength, communication module bit error rate, mission payload operating status, and three-axis acceleration output by the inertial measurement unit. This data is collected by each node at a fixed frequency and processed by the ground data aggregation unit to form a standardized node dataset, which is then used by the multi-source symptom extraction module. The preset time window refers to the length of the time interval used for feature extraction; its value is dynamically adjusted based on the physical characteristics of the operational status data and mission requirements. For slowly changing parameters, the rate of change is slow, requiring a longer time window to capture effective trends; therefore, a larger preset time window value is used. For rapidly changing parameters, fluctuations are rapid due to environmental and load influences, requiring a shorter time window to reflect the current state in real time; therefore, a smaller preset time window value is used. The specific value of the preset time window can be determined through historical data statistical analysis or pre-configured according to the mission type during system initialization and dynamically adjusted based on real-time data during mission execution.

[0043] Statistical features are used to describe the overall distribution characteristics of operational status data within a preset time window. This is based on data from a specific dimension. In the time window sampling sequence within For example, among which Number of sampling points. Mean. The formula that reflects the average level of the data is: variance The formula reflects the degree of dispersion of the data and is as follows: The maximum and minimum values ​​reflect the extreme values ​​of the data, and are taken as the maximum and minimum values ​​in the sequence, respectively. These statistical characteristics can reveal the overall level of node operation, the degree of fluctuation, and whether extreme situations occur.

[0044] Time-series variation characteristics are used to describe the dynamic changes in operational status data within a preset time window. Linear fitting slope. It reflects the overall trend of data change and is obtained by performing least-squares linear fitting on the time series. Let time points be... The corresponding sample value is linear model ,in For time variables, The intercept of the linear fit. The slope; The calculation formula is: in: and These are the time mean and the data mean, respectively. For the first The time value corresponding to each sampling point; slope A positive value indicates an upward trend, a negative value indicates a downward trend, and the larger the absolute value, the more drastic the change.

[0045] Fluctuation range The intensity of data oscillations can be obtained by calculating the root mean square value after detrending. The formula is as follows: in: These are the values ​​obtained through linear fitting or smoothing after moving average. The greater the fluctuation, the more severe the data oscillation.

[0046] Periodic intensity The presence of periodic patterns in data is determined through autocorrelation function analysis. The autocorrelation function of the data is calculated. Then, the maximum peak value (excluding zero delay) is taken as the measure of periodic intensity. The normalized periodic intensity calculation formula is as follows: in: This is the value of the autocorrelation function at zero delay. The closer the value is to 1, the more significant the periodicity.

[0047] When generating structured symptom description data, the extracted statistical features and temporal variation features are organized according to a preset format. Let a certain collaborative unit subgraph be labeled as... The timestamp is ,Include There are nodes, each node has Given operational status data in multiple dimensions, the structured symptom description data can be represented in the form of feature vectors: Subscript Indicates the first The node of the first In practical applications, some features can be selected and combined as needed to reduce data dimensionality and highlight key signs.

[0048] Please refer to Figure 4. The fault mode matching module is specifically used for: 1. Obtaining structured symptom description data corresponding to each collaborative unit subgraph; 2. Calculating the temporal similarity and feature similarity between the structured symptom description data and each historical symptom feature sequence in the fault mode knowledge base; 3. Weighted fusion of temporal similarity and feature similarity to obtain the comprehensive matching degree between the structured symptom description data and each historical symptom feature sequence; 4. Determining the fault type label corresponding to the historical symptom feature sequence with the highest comprehensive matching degree as the candidate target fault mode; 5. When the comprehensive matching degree of the candidate target fault mode exceeds the preset matching degree threshold, it is determined that the warning condition is met, and a fault warning information containing the candidate target fault mode is generated; wherein, the fault mode data stored in the fault mode knowledge base includes fault type labels, historical symptom feature sequences, and fault propagation rules; the candidate target fault mode includes fault type labels and corresponding fault propagation rules; the fault warning information includes a list of affected nodes predicted according to the fault propagation rules and the expected failure time.

[0049] Specifically, a fault mode knowledge base is pre-built, storing cluster fault cases recorded in historical combat missions. Each fault mode data contains three parts: a fault type label, used to uniquely identify the fault category; a historical symptom feature sequence, which is a feature sequence obtained by extracting symptoms from multi-node, multi-dimensional operational status data collected within a preset time period before the fault occurs, stored in vector form; and a fault propagation pattern, used to describe the diffusion path and impact range between nodes within the subgraph after the fault occurs. For each collaborative unit subgraph, let its structured symptom description data be... This data is a multi-dimensional feature vector, containing the statistical features and temporal variation features of each node within the subgraph across various dimensions. The Fault Mode and Effects Knowledge Base contains... A sequence of historical symptom characteristics, denoted as Each This corresponds to a fault mode.

[0050] Temporal similarity is used to measure the degree of similarity between real-time symptom description data and historical symptom feature sequences in terms of their evolutionary trends over time. Because... It is the feature vector within the current time window, while These are feature vectors within a certain time window prior to the occurrence of historical faults. Both can be considered as fixed-length feature vectors because, after symptom extraction, the data within the time window has been compressed into statistical features and temporal variation features, no longer retaining the original time series. Therefore, temporal similarity actually refers to the similarity of feature vectors in terms of temporal variation features. Dynamic time warping distance is used to calculate the similarity between two sequences, but here... and Since it's already in vector form, we calculate the cosine similarity as the feature similarity. However, if... and Since the data itself contains temporal variation characteristics, these characteristics already implicitly contain time information. For more accurate matching, a portion of the multi-node time-series data before the original symptom extraction can be retained to form a short time series, and then dynamic time-normalized distance matching can be performed. However, for simplicity, this embodiment uses similarity calculation between feature vectors.

[0051] Feature similarity is calculated using cosine similarity. Let's assume real-time symptom description data... The corresponding feature vector is ,in Total number of features; historical symptom feature sequence The corresponding feature vector is Then feature similarity The calculation formula is: in: Data describing real-time symptoms The Each feature component Historical symptom characteristic sequence The There are 1 feature components; the cosine similarity ranges from [-1, 1]. The closer the value is to 1, the more consistent the directions of the two vectors are, that is, the more similar the features are.

[0052] Temporal similarity is used to measure real-time symptom description data. With historical symptom characteristic sequence The degree of similarity in temporal variation patterns. To highlight the temporal characteristics, this embodiment uses a subset of temporal variation features from the feature vector for calculation. Let's assume that from... The extracted time-series change feature vector is ,in For the first The slope of each node, For the first The fluctuation range of each node, For the first The periodic intensity of each node; from The extracted time-series change feature vector is Then the temporal similarity The calculation formula is: in: The total dimension of the time-varying feature sub-vectors. The first of the real-time time-series changing feature sub-vectors One portion, The first sub-vector of historical time series change features Each component.

[0053] We perform a weighted fusion of temporal similarity and feature similarity to obtain the sum of matching degrees between the structured symptom description data and each historical symptom feature sequence. Weighted fusion formula: in: This is a weighting coefficient, ranging from [0,1], adjusted according to the sensitivity of the fault type to time-series changes. For faults involving battery depletion, the time-series trend is more important, so [the weighting coefficient is not specified]. For communication quality faults, statistical characteristics and timing characteristics are equally important, and one can be chosen over the other. The weighting coefficients can be determined through training on historical data.

[0054] After calculating the comprehensive matching degree between real-time symptoms and all historical symptom feature sequences, the fault mode corresponding to the maximum value is selected as the candidate target fault mode. A preset matching degree threshold is used to determine whether the warning is reliable enough. If the comprehensive matching degree is greater than or equal to the threshold, the warning condition is met, and a fault warning message is generated. This threshold is set according to actual needs, preferably 0.85, meaning that a warning is triggered only when the matching degree exceeds 85%, in order to reduce false alarms. The warning message includes the fault type label of the candidate target fault mode and the corresponding fault propagation law, and predicts the list of affected nodes and the expected failure time in the current subgraph based on this law. The fault propagation law is stored in the form of conditional probability or time sequence rules. For example, when the fault type is communication link interruption, the first node to fail is the communication relay node, and the probability of the nodes directly connected to it losing contact within 30 seconds is 80%, and the probability of the entire subgraph losing contact within 60 seconds is 95%. Based on the specific topology and node roles of the real-time subgraph, the specific list of affected nodes and the expected failure time can be inferred.

[0055] Using the sub-diagram of Example 1 For example, this sub-graph contains reconnaissance drone 01, strike drone 03, and strike drone 04. Assume real-time extracted structured symptom description data. During the process, the battery power at each node showed a rapid downward trend (with a large absolute slope), and the communication signal strength fluctuated drastically. The fault mode knowledge base contains a historical fault mode, "Sudden Drop in Battery Power," with its historical symptom sequence. The signal is characterized by low average power consumption, large variance, negative slope, and large absolute value; another historical symptom sequence of "communication link interruption" is also present. The signal strength has a low mean, high variance, and large fluctuation range. Calculation and The feature similarity is 0.82, and the temporal similarity is 0.90 (due to the obvious decreasing trend in battery power). We take... The score is 0.6, resulting in a comprehensive matching degree of 0.868. The feature similarity is 0.75, the temporal similarity is 0.70, and the overall matching degree is 0.72. The maximum value of 0.868 corresponds to... The candidate target failure mode is "sudden battery drain". A preset threshold of 0.85 is set; 0.868 > 0.85, satisfying the warning condition. The failure propagation pattern of the "sudden battery drain" failure mode is as follows: after this failure occurs, the node with the lowest battery level will fail within 10 minutes. Subsequently, the nodes cooperating with it will experience accelerated battery drain due to task redistribution, and the entire sub-map may lose combat capability within 20 minutes. Based on the current battery values ​​of each node in the sub-map (Reconnaissance UAV 01 has the lowest battery level), the predicted list of affected nodes is [Reconnaissance UAV 01, Strike UAV 03, Strike UAV 04], with expected failure times of 8 minutes for Reconnaissance UAV 01, 15 minutes for Strike UAV 03, and 18 minutes for Strike UAV 04. This information is integrated into the failure warning message and displayed to the commander through the user terminal.

[0056] The above embodiments transform massive amounts of isolated data into structured feature information that can characterize the collaborative status of the drone swarm through dynamic topology partitioning and multi-source symptom extraction. Based on intelligent matching of historical fault mode database, the accuracy and timeliness of fault early warning are effectively improved.

[0057] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A health status monitoring and fault early warning system for intelligent computing nodes in a drone swarm, characterized in that, include: The drone swarm node group consists of multiple intelligent computing node drones, which are used to collect and send multi-dimensional operational status data and spatial location information of each node in real time; The ground data aggregation unit is communicatively connected to the UAV cluster node group. It is used to receive multi-dimensional operational status data and spatial location information from multiple intelligent computing node UAVs, and to process the received multi-dimensional operational status data and spatial location information to generate a standardized node dataset. The cloud-based intelligent analysis platform is communicatively connected to the ground data aggregation unit. The cloud-based intelligent analysis platform includes a dynamic topology partitioning module, a multi-source symptom extraction module, and a fault mode matching module. The dynamic topology partitioning module is used to construct a dynamically weighted topology graph representing the coupling strength within the cluster based on the spatial positional relationships of nodes and the quality parameters of inter-node communication links in the standardized node dataset, and to divide the dynamically weighted topology graph into multiple collaborative unit subgraphs based on an adaptive graph segmentation algorithm; the multi-source symptom extraction module is connected to the dynamic topology partitioning module, and is used to extract the change characteristics of multi-dimensional operating status data of multiple intelligent computing node UAVs within each collaborative unit subgraph within a preset time window, and generate structured symptom description data representing the current operating status of the subgraph; the fault mode matching module is connected to the multi-source symptom extraction module and the pre-built fault mode knowledge base, respectively, and is used to match the structured symptom description data with the historical symptom feature sequences in the fault mode knowledge base, determine the target fault mode based on the matching result, and generate fault warning information when the warning conditions are met; The user terminal is connected to the cloud-based intelligent analysis platform to receive and display the fault warning information.

2. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 1, characterized in that, Each intelligent computing node drone in the drone cluster node group includes: a node status acquisition unit for real-time acquisition of multi-dimensional operational status data of the node; a node positioning unit for real-time acquisition of the node's spatial location information; and a node communication unit connected to the node status acquisition unit and the node positioning unit respectively, for sending the multi-dimensional operational status data and spatial location information to the ground data aggregation unit.

3. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 1, characterized in that, The ground data aggregation unit includes: a multi-source data receiving module, communicatively connected to the UAV cluster node group, for receiving multi-dimensional operational status data and spatial location information from multiple intelligent computing node UAVs in parallel; a time synchronization and alignment module, connected to the multi-source data receiving module, for time alignment processing of timestamp inconsistencies caused by clock differences and transmission delays among different intelligent computing node UAVs, generating multi-source time-series data with a unified time reference; a heterogeneous data aggregation module, connected to the time synchronization and alignment module, for structurally aggregating multi-dimensional operational status data from different intelligent computing node UAVs with different acquisition frequencies and data formats according to node identifiers and time series; and a standardized output module, connected to the heterogeneous data aggregation module, for converting the aggregated data into a standardized node dataset in a preset format and sending it to the cloud-based intelligent analysis platform.

4. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 1, characterized in that, In the dynamic weighted topology graph, each node corresponds to a smart computing node drone, and the node weight is determined based on at least one of the node's computing power, remaining battery power, and task role; each edge corresponds to a communication link between two nodes, and the edge weight is determined based on at least one of the communication link's signal strength, transmission delay, and packet loss rate.

5. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 4, characterized in that, The dynamic topology partitioning module is specifically used to: calculate the spatial proximity and communication link quality between each pair of nodes based on the standardized node dataset, and obtain the coupling strength between nodes; construct the dynamic weighted topology graph with nodes as vertices and the coupling strength as edge weights; An adaptive graph segmentation algorithm is adopted to divide the dynamic weighted topology graph into multiple cooperative unit subgraphs with the goal of minimizing the coupling strength between subgraphs and maximizing the coupling strength within subgraphs.

6. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 5, characterized in that, The adaptive graph partitioning algorithm aims to minimize the coupling strength between subgraphs and maximize the coupling strength within subgraphs. It divides the dynamically weighted topology graph into multiple collaborative unit subgraphs, including: initializing each node as an independent subgraph; calculating the merging benefit between adjacent subgraphs, where the merging benefit is a weighted sum of the increase in coupling strength within a subgraph and the decrease in coupling strength between subgraphs after merging two adjacent subgraphs; setting an initial threshold, determined based on a preset cluster size range; merging the corresponding two subgraphs into a new subgraph when the merging benefit exceeds the current threshold; recalculating and updating the current threshold after each round of merging based on the current number of remaining subgraphs and the current node density; and repeating the steps of calculating the merging benefit between adjacent subgraphs, merging, and recalculating and updating the current threshold until the merging benefit between all adjacent subgraphs is less than or equal to the current threshold, resulting in the final multiple collaborative unit subgraphs.

7. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 1, characterized in that, The multi-source symptom extraction module is specifically used to: acquire multi-dimensional operational status data of all intelligent computing node drones within each collaborative unit subgraph within a preset time window; and extract the statistical characteristics and temporal variation characteristics of the operational status data of each dimension within the preset time window. Based on the extracted statistical features and temporal change features, structured symptom description data corresponding to the current collaborative unit subgraph is generated. The structured symptom description data includes subgraph identifier, timestamp, and feature values ​​of each dimension.

8. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 7, characterized in that, The statistical features include at least one of the mean, variance, maximum value, and minimum value; the time series variation features include at least one of the linear fitting slope, fluctuation amplitude, and period intensity.

9. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 1, characterized in that, The fault mode matching module is specifically used for: acquiring structured symptom description data corresponding to each collaborative unit subgraph; calculating the temporal similarity and feature similarity between the structured symptom description data and each historical symptom feature sequence in the fault mode knowledge base; performing weighted fusion on the temporal similarity and feature similarity to obtain the comprehensive matching degree between the structured symptom description data and each historical symptom feature sequence; determining the fault type label corresponding to the historical symptom feature sequence with the highest comprehensive matching degree as a candidate target fault mode; when the comprehensive matching degree of the candidate target fault mode exceeds a preset matching degree threshold, determining that the warning condition is met, and generating fault warning information containing the candidate target fault mode.

10. The intelligent computing node health status monitoring and fault early warning system for UAV swarms according to claim 9, characterized in that, The fault mode knowledge base stores fault mode data including fault type labels, historical symptom feature sequences, and fault propagation patterns; the candidate target fault modes include the fault type labels and corresponding fault propagation patterns; the fault warning information includes a list of affected nodes predicted based on the fault propagation patterns and the expected failure time.

Images