Unmanned aerial vehicle system fault diagnosis method and equipment based on AE-GCN

Abstract

The invention discloses an AE-GCN-based unmanned aerial vehicle system fault diagnosis method and equipment, belongs to the technical field of unmanned aerial vehicle fault diagnosis, and is used for solving the technical problems that an existing unmanned aerial vehicle fault diagnosis method depends on expert experience and feature engineering and has the defects of low efficiency, poor generalization ability and the like. The method comprises the following steps: constructing a corresponding structure topological graph based on a topological structure of a target unmanned aerial vehicle, and obtaining a topological matrix; performing node feature extraction on the structure topological graph to obtain a node feature vector; inputting the topological matrix and the node feature vector into a graph convolution neural network for graph convolution operation to obtain a node state feature vector; inputting the data into an attention mechanism module to obtain attention features, and mapping the attention features into node state updating parameters; comparing the node state updating parameter with the node state historical parameter, and positioning a fault node and a fault type; and predicting a fault propagation path based on other nodes directly connected with the fault node in the structural topological graph.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle fault diagnosis, and in particular to an unmanned aerial vehicle system fault diagnosis method and device based on AE-GCN. Background Art

[0002] In recent years, drones, as an important part of strategic emerging industries, have demonstrated remarkable advantages in many industry applications with their unparalleled safety, flexibility, economy and efficient operation capabilities. In the field of traffic monitoring, drones can quickly capture traffic conditions, provide real-time traffic flow and accident information, and provide strong support for traffic management. In terms of power inspection, drones can efficiently conduct line inspections, timely detect potential power failures, and ensure the safe operation of the power grid. In the agricultural field, drones can monitor crop growth conditions in real time by carrying high-precision sensors, providing data support for precision agriculture. In addition, in disaster response and emergency rescue, drones can quickly arrive at the scene, provide key on-site information and rescue support, and greatly improve rescue efficiency. By providing instant and comprehensive on-site information, drone technology provides a scientific basis for decision-making in various fields, accelerates the decision-making process, and improves the effectiveness and scientificity of decision-making.

[0003] With the improvement of industrial automation level and the increase of equipment complexity, drone technology is becoming more and more complex. Drone fault diagnosis has become a key link in ensuring the safety and efficiency of drone systems. First of all, there are various types of drone failures, including but not limited to mechanical structure failures, electronic system failures and flight control system failures. Mechanical structure failures may manifest as frame damage, propeller deformation, etc. Electronic system failures may include battery aging, signal transmission interruption, etc. Flight control system failures may manifest as unstable drone flight, inability to fly according to preset routes, etc. Secondly, drone troubleshooting is complicated. The electronic system and flight control system of drones are complex. When troubleshooting, you may need to have a deep understanding of the internal working principles of the system and the interactions between the various components. This may be a daunting task for general users or maintenance personnel.

[0004] The current methods of UAV fault detection mainly include data-based diagnosis methods, model-based diagnosis methods, knowledge-based diagnosis methods, etc. The data-based diagnosis method is to diagnose the fault of the UAV by analyzing the data generated during the operation of the UAV and extracting the fault characteristics. This method includes the comprehensive application of technical means such as sensor detection, image processing and wireless communication. By analyzing the data generated by the UAV, the potential faults of the UAV can be discovered in time, and corresponding maintenance measures can be taken to avoid further development of the fault and cause greater losses. The model-based diagnosis method is to diagnose the fault of the UAV by establishing a mathematical model of the UAV, simulating and analyzing the model. This method requires an in-depth understanding of the structure and principle of the UAV and is suitable for relatively simple UAV systems. By simulating the operating status of the UAV, possible fault conditions can be predicted and corresponding maintenance plans can be formulated. The knowledge-based diagnosis method is to apply the knowledge and experience of experts to UAV fault diagnosis through technologies such as expert systems and fuzzy logic. This method can handle more complex fault conditions, but requires a lot of expert knowledge and experience. By simulating the decision-making process of experts, the type and cause of UAV faults can be judged more accurately, and corresponding maintenance measures can be taken.

[0005] Although these methods can detect faults to a certain extent, they rely on expert experience and feature engineering, have low diagnostic efficiency, and are difficult to meet the rapidly changing fault diagnosis needs. On the other hand, due to the complexity of UAV systems, traditional fault diagnosis methods often have difficulty accurately identifying and diagnosing faults, resulting in insufficient accuracy and reliability of diagnostic results. In addition, with the continuous expansion of UAV application scenarios and the increasing diversification of failure modes, traditional methods are obviously insufficient in generalization when faced with complex and changing failure modes, which limits the reliability and stability of UAV systems. Summary of the invention

[0006] The embodiment of the present invention provides a UAV system fault diagnosis method and device based on AE-GCN, which is used to solve the following technical problems: the existing UAV fault diagnosis method relies on expert experience and feature engineering, and has the disadvantages of low efficiency and poor generalization ability.

[0007] The embodiment of the present invention adopts the following technical solutions: On the one hand, an embodiment of the present invention provides a method for diagnosing a fault of a UAV system based on AE-GCN, the method comprising: Based on the topological structure of the target UAV, the corresponding structural topological graph is constructed and the topological matrix is ​​obtained; Extracting node features from the structural topology graph to obtain a node feature vector; Inputting the topology matrix and the node feature vector into a graph convolutional neural network to perform a graph convolution operation to obtain a node state feature vector; Input the node state feature vector into the attention mechanism module to obtain the attention feature, and map the attention feature into the node state update parameter; Compare the node status update parameters with the node status history parameters to locate the faulty node and the fault type; The fault propagation path is predicted based on other nodes in the structural topology diagram that are directly connected to the faulty node.

[0008] In a feasible implementation, based on the topological structure of the target UAV, a structural topological map is constructed and a topological matrix is ​​obtained, specifically including: Obtain the sensor types in the target UAV and the dependencies between each type of sensor; wherein the sensor types include at least one or more of the following: inertial measurement unit, GPS module, barometer, magnetic compass, accelerometer, gyroscope, ultrasonic sensor, lidar, optical flow sensor, temperature sensor, visual sensor and infrared sensor; Mapping each sensor as a node, mapping the dependency relationship as an edge, and constructing a structural topology graph of the target UAV; A topology matrix is ​​generated according to the structural topology diagram.

[0009] In a feasible implementation manner, performing node feature extraction on the structural topology graph to obtain a node feature vector specifically includes: The real-time operating parameters of the sensor corresponding to each node in the structural topology diagram are obtained, and input into the skeleton network for feature extraction to obtain the node feature of each node to form the node feature vector.

[0010] In a feasible implementation, the real-time operating parameters of the sensor corresponding to each node in the structural topology diagram are obtained and input into the skeleton network for feature extraction to obtain the node features of each node, specifically including: according to , extract the overall features of the input parameters to obtain the overall feature vector; where y is the output eigenvalue, x is the input parameter, W represents the initial weight matrix, and b is the initial bias matrix; where the dimension of W is (n-1) The dimensions of n and b are ; The overall feature vector is n-dimensional matrix, n is the total number of nodes in the structural topology graph; The overall feature vector is divided into n one-dimensional matrices to obtain the node feature of each node.

[0011] In a feasible implementation, the topology matrix and the node feature vector are input into a graph convolutional neural network to perform a graph convolution operation to obtain a node state feature vector, specifically including: Using the topology matrix as the adjacency matrix of the graph convolutional neural network, using the node feature vector as the input parameter of the graph convolutional neural network, and outputting the node state feature of each node; The node state features of each node are concatenated to obtain the node state feature vector.

[0012] In a feasible implementation, the formula of the graph convolutional neural network is: ; in, is the node state feature output by the graph convolutional neural network, is the input of the graph convolutional neural network, that is, the node feature vector, is the adjacency matrix, i.e., the topological matrix; and are the parameters of the graph convolutional neural network, and ReLU is the activation function.

[0013] In a feasible implementation, the node state feature vector is input into an attention mechanism module to obtain an attention feature, and the attention feature is mapped to a node state update parameter, specifically including: Using the node state feature vector as an input parameter of the attention mechanism module to obtain attention features; The objective function of the attention mechanism module is: ;in, represents the dimensionality reduction process, represents the dimensionality-increasing process, represents the Sigmoid activation function, represents the attention feature, and F is the node state feature vector; The attention feature is input into a dense layer to obtain an output result; and the output result is mapped to the node state update parameter.

[0014] In a feasible implementation manner, comparing the node status update parameter with the node status history parameter to locate the faulty node and the fault type specifically includes: Compare the node status update parameters of each node with the corresponding node status history parameters, screen out nodes with deviations in status parameters, and preliminarily determine them as faulty nodes; A possible fault type is determined according to the sensor type corresponding to the faulty node.

[0015] In a feasible implementation manner, predicting the fault propagation path based on other nodes directly connected to the faulty node in the structure topology graph specifically includes: Based on the initially determined faulty node, obtaining, in the structure topology map, adjacent nodes directly connected to the faulty node; Based on expert experience, the impact of the faulty node on the adjacent nodes is analyzed, the propagation path and potential impact of the fault are predicted, and the fault diagnosis analysis results are obtained.

[0016] On the other hand, an embodiment of the present invention further provides a UAV system fault diagnosis device based on AE-GCN, the device comprising: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor so that the at least one processor can execute the AE-GCN-based UAV system fault diagnosis method.

[0017] Compared with the prior art, the UAV system fault diagnosis method and device based on AE-GCN provided by the embodiment of the present invention have the following beneficial effects: This paper proposes a UAV system fault diagnosis method based on Attention-Enhanced Graph Convolutional Neural Network (AE-GCN). The graph structure is constructed through the dependency relationship between the various sensors of the UAV to accurately capture the interaction and influence between these sensors. On this basis, the attention mechanism is introduced to guide the algorithm to focus on the key causes of the fault, thereby achieving accurate identification and rapid diagnosis of UAV faults, significantly improving the diagnostic efficiency and accuracy of the algorithm, and enabling the algorithm to quickly lock the source of the problem in a complex fault environment.

[0018] The graph convolutional neural network AE-GCN can automatically learn deep features in drone fault data without manually designing features. This capability effectively overcomes the limitation of traditional methods that require manual feature extraction and improves the accuracy and efficiency of fault diagnosis. Drone fault data often has complex structural relationships. Graph convolutional neural networks can process such non-Euclidean space data, capture the topological structure and relationship information in the data, and thus diagnose faults more accurately. Graph convolutional neural networks can directly input drone fault data into the network and output diagnostic results without complex feature engineering. This end-to-end learning method simplifies the model training process and improves the automation level of fault diagnosis. The attention mechanism can automatically focus on the key features in drone fault data, that is, the parts most relevant to the fault. This enables the model to use data more effectively and improve the accuracy of fault diagnosis. At the same time, the attention mechanism can automatically assign weights to reduce the impact of noise and irrelevant information on the model. This enables the model to maintain stable performance when facing complex environments and noisy data, and improves the robustness of fault diagnosis.

[0019] When graph convolutional neural networks are combined with attention mechanisms and applied to drone fault diagnosis, the advantages of both can be fully utilized. Graph convolutional neural networks can process complex structure data and capture the topological structure and relationship information in the data; while the attention mechanism can focus on the key features in the data. The combination of the two can efficiently capture the key features in drone fault data and improve the accuracy of fault diagnosis. The end-to-end learning method and the ability to automatically focus on key information enable the model to process drone fault data faster and output diagnostic results. This improves the efficiency of fault diagnosis and helps to detect problems in a timely manner and take corresponding measures. The combination of graph convolutional neural networks and attention mechanisms enables the model to better adapt to different drone fault data and scenarios. This generalization ability enables the model to maintain stable performance under new data and scenarios, improving the reliability and practicality of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings: Figure 1 A flow chart of a method for diagnosing a fault in an unmanned aerial vehicle system based on AE-GCN provided in an embodiment of the present invention; Figure 2 A schematic diagram of an AE-GCN-based UAV system fault diagnosis model architecture provided in an embodiment of the present invention; Figure 3 A schematic diagram of the structure of an attention module provided by an embodiment of the present invention; Figure 4 The present invention provides a schematic diagram of the structure of an AE-GCN-based UAV system fault diagnosis device. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0022] The embodiment of the present invention provides a method for diagnosing a fault in an unmanned aerial vehicle system based on AE-GCN. Figure 1 As shown, the UAV system fault diagnosis method based on AE-GCN specifically includes steps S101-S105: S101. Based on the topological structure of the target UAV, a corresponding structural topological map is constructed and a topological matrix is ​​obtained.

[0023] Specifically, firstly, the sensor types in the target UAV and the dependencies between each type of sensors are obtained; wherein the sensor types include at least one or more of the following: inertial measurement unit, GPS module, barometer, magnetic compass, accelerometer, gyroscope, ultrasonic sensor, lidar, optical flow sensor, temperature sensor, visual sensor and infrared sensor.

[0024] Furthermore, each sensor is mapped as a node, and the dependency relationship is mapped as an edge to construct a structural topology graph of the target UAV. Based on the structural topology graph, a topology matrix is ​​generated.

[0025] As a feasible implementation method, there are many types of sensors in drones, which play a vital role in the flight control and navigation of drones. The main sensor types include inertial measurement unit (IMU), GPS module, barometer, magnetic compass, accelerometer, gyroscope, ultrasonic sensor, laser radar (LiDAR), optical flow sensor, temperature sensor, visual sensor, infrared sensor and other 12 types.

[0026] These sensors do not work in isolation in the drone system, but work together to support the flight control and navigation of the drone. For example, the interaction between them is reflected in the following aspects: 1. UAV flight control systems usually fuse data from different sensors to improve positioning accuracy and stability. For example, fusing data from IMU, magnetic compass, barometer, and GPS can achieve accurate estimation of the UAV's attitude, position, and speed.

[0027] 2. Different sensors have different advantages and scopes of application. For example, GPS provides global location information in open areas, but may not work properly indoors or in areas where the signal is interfered; while ultrasonic sensors and barometers can provide accurate altitude and distance information in these localized ranges. By combining these sensors, drones can achieve stable flight in various environments.

[0028] 3. During the flight of the drone, various sensors need to work together to achieve stable flight control. For example, when the drone needs to perform an obstacle avoidance mission, the lidar and ultrasonic sensors will jointly provide obstacle information, and the flight control system will adjust the flight trajectory of the drone based on this information.

[0029] By sorting out the dependencies between the interactions between the sensors, we can obtain the topological structure between the sensors, and use a topological matrix A to represent the topological structure between the sensors. This topological matrix A will be used as the adjacency matrix of the graph convolutional neural network.

[0030] S102: Extract node features from the structural topology graph to obtain a node feature vector.

[0031] Specifically, Figure 2 A schematic diagram of an AE-GCN-based UAV system fault diagnosis model architecture is provided in an embodiment of the present invention, such as Figure 2 As shown in the figure, the model includes a skeleton network backbone for feature extraction, a graph convolution module, an attention mechanism Attention module, and a fully connected layer Dense module.

[0032] The feature extraction module obtains the real-time operating parameters of the sensor corresponding to each node in the structural topology diagram, and inputs them into the skeleton network for feature extraction to obtain the node features of each node to form a node feature vector.

[0033] As a feasible implementation method, the specific method of obtaining the characteristics of each node is: according to , extract the overall features of the input parameters to obtain the overall feature vector; where y is the output eigenvalue, x is the input parameter, W represents the initial weight matrix, and b is the initial bias matrix; where the dimension of W is (n-1) The dimensions of n and b are The overall eigenvector is n-dimensional matrix, n is the total number of nodes in the structural topology graph. Then the overall feature vector is divided into n one-dimensional matrices to obtain the node features of each node.

[0034] In one embodiment, if the current initial weight matrix is , the size is 12×11, and the current initial bias matrix is , size 1×12. After flipping, its feature size becomes 11×12, and After concatenation, a concatenated vector S of size 12×12 is obtained. Then S is divided into S1, S2…S12, each with a feature size of 12×1, which are used to correspond to the node features of the drone's sensors, inertial measurement unit (IMU), GPS module, barometer, magnetic compass, accelerometer, gyroscope, ultrasonic sensor, laser radar (LiDAR), optical flow sensor, temperature sensor, visual sensor, and infrared sensor.

[0035] S103: Input the topology matrix and the node feature vector into the graph convolutional neural network to perform graph convolution operation to obtain the node state feature vector.

[0036] Specifically, the topological matrix is ​​used as the adjacency matrix of the graph convolutional neural network, the node feature vector is used as the input parameter of the graph convolutional neural network, and the node state features of each node are output.

[0037] Furthermore, the node state features of each node are concatenated to obtain a node state feature vector.

[0038] As a feasible implementation method, the present invention adopts a two-layer graph convolutional neural network, and the formula is: ;in, is the node state feature output by the graph convolutional neural network, is the input of the graph convolutional neural network, that is, the node feature vector. is the adjacency matrix, i.e. the topological matrix; and is the parameter of the graph convolutional neural network, and ReLU is the activation function. S1, S2 ... S12 are input into the graph convolution module, and T1, T2 ... T12 are output. T1, T2 ... T12 are concatenated to obtain the node state feature vector T with a size of 12 × 12, and then the feature vector T is input into the attention mechanism module.

[0039] S104: Input the node state feature vector into the attention mechanism module to obtain the attention feature, and map the attention feature into the node state update parameter.

[0040] Specifically, Figure 3A schematic diagram of an attention module structure provided by an embodiment of the present invention is shown in FIG. Figure 3 As shown in the figure, the node state feature vector (i.e., the feature matrix in the figure) obtained in the graph convolutional neural network is used as the input parameter of the attention mechanism module, and is processed by the multi-layer perceptron MLP to obtain the attention feature.

[0041] The objective function of the attention mechanism module is: ;in, represents the dimensionality reduction process, represents the dimensionality-increasing process, represents the Sigmoid activation function, represents the attention feature, and F is the node state feature vector.

[0042] Furthermore, the attention features obtained according to the above objective function are input into the dense layer of the model to obtain the output result, and the output result is mapped to the node state update parameter.

[0043] S105: Compare the node status update parameters with the node status history parameters to locate the faulty node and the fault type; and predict the fault propagation path based on other nodes directly connected to the faulty node in the structure topology diagram.

[0044] Specifically, the node state update parameters of each node are compared with the corresponding node state history parameters, and the nodes with deviations in state parameters are screened out and preliminarily determined as faulty nodes. Then, the possible fault type is determined according to the sensor type corresponding to the faulty node.

[0045] Furthermore, based on the initially determined faulty node, the adjacent nodes directly connected to the faulty node are obtained in the structural topology map. Then, based on expert experience, the impact of the faulty node on the adjacent nodes is analyzed, the propagation path and potential impact of the fault are predicted, and the fault diagnosis analysis results are obtained.

[0046] like Figure 2As shown, the present invention integrates the real-time data and topological structure of each sensor of the drone. Each sensor is mapped to a node of the topological graph, and the dependencies between the sensors are mapped to the edges of the graph, thereby generating a complete topological graph. The graph convolutional neural network is used to extract the node features of the currently generated topological graph, and the graph convolution operation is performed according to the topological matrix and the node features to update the node state, and the attention mechanism is used as an auxiliary to further improve the accuracy and speed of fault location. In order to determine the faulty node, the present invention compares the current state and the historical state of the node, and screens the node based on the state deviation. This process can preliminarily locate the faulty node and determine its fault type. On the basis of the preliminary fault location, other nodes directly connected to the faulty node are further analyzed. By analyzing the faulty node and the neighboring nodes affected by the faulty node, the propagation path and potential impact of the fault can be predicted, thereby obtaining the analysis result of the fault diagnosis.

[0047] In addition, the embodiment of the present invention also provides a UAV system fault diagnosis device based on AE-GCN, such as Figure 4 As shown, the UAV system fault diagnosis equipment based on AE-GCN specifically includes: at least one processor; and a memory in communication with the at least one processor; wherein, The memory stores instructions executable by at least one processor to enable the at least one processor to perform: Based on the topological structure of the target UAV, the corresponding structural topological graph is constructed and the topological matrix is ​​obtained; Extracting node features from the structural topology graph to obtain a node feature vector; Inputting the topology matrix and the node feature vector into a graph convolutional neural network to perform a graph convolution operation to obtain a node state feature vector; Input the node state feature vector into the attention mechanism module to obtain the attention feature, and map the attention feature into the node state update parameter; Compare the node status update parameters with the node status history parameters to locate the faulty node and the fault type; The fault propagation path is predicted based on other nodes in the structural topology diagram that are directly connected to the faulty node.

[0048] Each embodiment of the present invention is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0049] The above describes specific embodiments of the present invention. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0050] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the embodiments of the present invention should be included in the protection scope of the present invention.

Claims

1. A UAV system fault diagnosis method based on AE-GCN, characterized in that: The method comprises: Based on the topological structure of the target UAV, the corresponding structural topological graph is constructed and the topological matrix is ​​obtained; Extracting node features from the structural topology graph to obtain a node feature vector; Inputting the topology matrix and the node feature vector into a graph convolutional neural network to perform a graph convolution operation to obtain a node state feature vector; Input the node state feature vector into the attention mechanism module to obtain the attention feature, and map the attention feature into the node state update parameter; Compare the node status update parameters with the node status history parameters to locate the faulty node and the fault type; The fault propagation path is predicted based on other nodes in the structural topology diagram that are directly connected to the faulty node.

2. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 1 is characterized in that: Based on the topological structure of the target UAV, a structural topological map is constructed and a topological matrix is ​​obtained, including: Obtain the sensor types in the target UAV and the dependencies between each type of sensor; wherein the sensor types include at least one or more of the following: inertial measurement unit, GPS module, barometer, magnetic compass, accelerometer, gyroscope, ultrasonic sensor, lidar, optical flow sensor, temperature sensor, visual sensor and infrared sensor; Mapping each sensor as a node, mapping the dependency relationship as an edge, and constructing a structural topology graph of the target UAV; A topology matrix is ​​generated according to the structural topology diagram.

3. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 1 is characterized in that: Extracting node features from the structural topology graph to obtain a node feature vector specifically includes: The real-time operating parameters of the sensor corresponding to each node in the structural topology diagram are obtained, and input into the skeleton network for feature extraction to obtain the node feature of each node to form the node feature vector.

4. The method for AE-GCN-based fault diagnosis of unmanned aerial vehicle system according to claim 3 is characterized in that: The real-time operating parameters of the sensor corresponding to each node in the structural topology diagram are obtained and input into the skeleton network for feature extraction to obtain the node features of each node, specifically including: according to , extract the overall features of the input parameters to obtain the overall feature vector; where y is the output eigenvalue, x is the input parameter, W represents the initial weight matrix, and b is the initial bias matrix; where the dimension of W is (n-1) The dimensions of n and b are ; The overall feature vector is n-dimensional matrix, n is the total number of nodes in the structural topology graph; The overall feature vector is divided into n one-dimensional matrices to obtain the node feature of each node.

5. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 1, characterized in that: The topology matrix and the node feature vector are input into a graph convolutional neural network to perform a graph convolution operation to obtain a node state feature vector, specifically including: Using the topology matrix as the adjacency matrix of the graph convolutional neural network, using the node feature vector as the input parameter of the graph convolutional neural network, and outputting the node state feature of each node; The node state features of each node are concatenated to obtain the node state feature vector.

6. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 5 is characterized in that: The formula of the graph convolutional neural network is: ; in, is the node state feature output by the graph convolutional neural network, is the input of the graph convolutional neural network, that is, the node feature vector, is the adjacency matrix, i.e., the topological matrix; and are the parameters of the graph convolutional neural network, and ReLU is the activation function.

7. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 1, characterized in that: The node state feature vector is input into the attention mechanism module to obtain the attention feature, and the attention feature is mapped into the node state update parameter, specifically including: Using the node state feature vector as an input parameter of the attention mechanism module to obtain attention features; The objective function of the attention mechanism module is: ;in, represents the dimensionality reduction process, represents the dimensionality-increasing process, represents the Sigmoid activation function, represents the attention feature, and F is the node state feature vector; The attention feature is input into a dense layer to obtain an output result; and the output result is mapped to the node state update parameter.

8. The method for fault diagnosis of unmanned aerial vehicle system based on AE-GCN according to claim 1, characterized in that: Comparing the node status update parameters with the node status history parameters to locate the faulty node and the fault type, specifically including: Compare the node status update parameters of each node with the corresponding node status history parameters, screen out nodes with deviations in status parameters, and preliminarily determine them as faulty nodes; A possible fault type is determined according to the sensor type corresponding to the faulty node.

9. The method for AE-GCN-based fault diagnosis of unmanned aerial vehicle system according to claim 8, characterized in that: Predicting a fault propagation path based on other nodes directly connected to the faulty node in the structure topology graph specifically includes: Based on the initially determined faulty node, obtaining, in the structure topology map, adjacent nodes directly connected to the faulty node; Based on expert experience, the impact of the faulty node on the adjacent nodes is analyzed, the propagation path and potential impact of the fault are predicted, and the fault diagnosis analysis results are obtained.

10. A UAV system fault diagnosis device based on AE-GCN, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the AE-GCN-based UAV system fault diagnosis method according to any one of claims 1-9.

Images