Fault mode recognition method for airborne navigation functional equipment based on improved graph convolutional neural network

Abstract

The present invention provides an airborne navigation function equipment fault pattern recognition method based on an improved graph convolutional neural network, comprising: collecting key parameters of airborne DME navigation function equipment and constructing a data set, dividing the data set into a training set and a test set; collecting monitoring data and simulated flight data of airborne DME navigation function equipment to obtain source domain data; establishing an airborne navigation function equipment fault pattern recognition model based on an improved graph convolutional neural network; using the source domain data and training set to train the airborne navigation function equipment fault pattern recognition model; using the test set to test the performance of the airborne navigation function equipment fault pattern recognition model. The present invention enables multiple types of electronic system faults that occur during the flight operation of an aircraft to be more quickly identified and located, improves the accuracy of aircraft fault diagnosis and troubleshooting efficiency, reduces maintenance costs, extends the service life of the aircraft, and at the same time improves the safety and reliability of aircraft operation.

Description

Technical Field

[0001] The present invention relates to the technical field of fault identification of airborne navigation function equipment, and in particular to a method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network. Background Art

[0002] The aviation integrated electronic system is a typical representative of the application of the electronic information industry in the aviation field. It is a complex that interconnects multiple airborne electronic subsystems. It coordinates the scattered systems with single functions, such as communication radios, radars, and navigation equipment, to form a multifunctional integrated system, thereby realizing the functions of information measurement, collection, transmission, processing, and monitoring, and completing tasks such as flight control, engine control, navigation, and performance management. With the performance requirements and improvements of large-scale aviation equipment, the complexity and integration of integrated electronic systems continue to increase. At the same time, due to the complex and changeable service environment, the probability of failure of new integrated electronic systems gradually increases, which may lead to system performance degradation or even damage, or even cause serious economic losses and social impacts. Several serious air crashes in aviation history have highlighted the urgency of the safety and reliability of integrated electronic systems.

[0003] As a core component of the aviation integrated electronic system, the airborne navigation function equipment is responsible for providing accurate positioning and navigation information for various carriers. During the flight of the aircraft, the stability and accuracy of the navigation function equipment are directly related to the navigation safety of the aircraft. However, the security threats it faces are also increasing. Factors such as signal interference and system failure may interfere with or damage the navigation function equipment, resulting in a decrease in positioning accuracy or even complete failure. Among them, the distance measuring equipment (DME) navigation function equipment plays an indispensable role in the field of aircraft navigation. The equipment uses radio signals transmitted by ground stations or satellites and combines with the receiving equipment on the aircraft to provide accurate position information of the aircraft. This information can help relevant personnel accurately determine the position of the aircraft during the flight and make adjustments when necessary to ensure safe flight. Due to mountain obstruction, large aircraft attitude, beyond visual range and other reasons, the DME navigation function equipment may have faults including insufficient airborne transmission power, local oscillator lock loss and lockout, thereby affecting the smooth execution of the flight mission. Therefore, the reliability of the DME navigation function equipment is crucial to flight safety.

[0004] As aviation integrated electronic systems become increasingly complex and intelligent, traditional fault diagnosis methods are difficult to meet actual needs, while fault diagnosis methods based on deep learning have more advantages when facing massive data. With the deepening of scholars' research, its feature extraction and classification capabilities for fault data are constantly improving, and the accuracy and time required for fault diagnosis are also constantly being optimized. Therefore, deep learning plays an increasingly important role in the field of fault diagnosis of integrated electronic systems. However, the use of deep learning fault diagnosis methods requires a large amount of labeled data as training samples for the network, and the fault types in the samples need to be rich and sufficient. However, in actual working environments, the monitoring samples generated by the system are difficult to meet the above requirements.

[0005] Airborne navigation function equipment will generate a variety of time series data during operation. Therefore, when using deep learning for fault diagnosis, it is necessary to consider not only the temporal features of a single type of data, but also the spatial features between different types of data. Different extracted features will result in different diagnostic results to a certain extent. It is worth discussing how to solve the problem of feature selection and feature combination. Graph Convolutional Network (GCN) is different from traditional neural networks. It is specially designed to process graph structure data and can deeply mine the association between node features and edge features in the graph. Graph Convolutional Neural Network uses information transmission and aggregation mechanisms to allow information exchange between nodes. Nodes transmit and receive information through connections with their neighbor nodes, which enables graph convolutional neural networks to capture contextual information between nodes. Traditional neural networks usually do not have this information transmission mechanism. At the same time, compared with traditional neural networks, GCN has better interpretability in some cases. They can visualize the importance of nodes and information transmission paths to help understand the decision-making process of the model.

[0006] Transfer learning can solve the problem of deep learning fault diagnosis methods' need for a large amount of labeled data and rich fault type samples in actual airborne working environments. Transfer learning improves the classification performance of the model in the new data domain by combining the data distribution obtained in the existing data domain with the data distribution in the new data domain. Transfer learning applies rich labeled data from other similar fields to the fault diagnosis of the target system. By learning useful features from the source domain and then transferring them to the target domain, the amount of labeled data required in the target domain can be significantly reduced. For the fault diagnosis of integrated electronic systems, the model is initially trained using source domain data on the fault diagnosis model architecture of the graph convolutional neural network to enable it to have general feature extraction capabilities. Subsequently, the target domain data is input to retrain and fine-tune the model so that the model can better adapt to the characteristics of the target domain, thus realizing transfer learning from the source domain to the target domain.

[0007] In summary, the existing methods for fault diagnosis of airborne navigation functional equipment are deficient in terms of model training label requirements, effective feature extraction capabilities, and generalization capabilities. There is a large room for research to further improve the accuracy of aircraft fault diagnosis and troubleshooting efficiency, thereby reducing maintenance costs throughout the life cycle of the aircraft, extending the service life of the aircraft, and improving the safety and reliability of aircraft operations, which has far-reaching practical significance for the aviation field. Summary of the invention

[0008] The present invention aims to provide a method for fault pattern recognition of airborne navigation function equipment based on an improved graph convolutional neural network, so as to solve the problems that the existing methods are insufficient in terms of model training label requirements, effective feature extraction capability and generalization capability.

[0009] The present invention provides a method for identifying fault modes of airborne navigation function equipment based on an improved graph convolutional neural network, comprising:

[0010] Step 1: Collect the key parameters of the airborne DME navigation function equipment and construct a data set, which is divided into a training set and a test set;

[0011] Step 2: Obtain source domain data by collecting monitoring data and simulated flight data of airborne DME navigation function equipment of different types of aircraft;

[0012] Step 3: Establish a fault mode recognition model for airborne navigation functional equipment based on the improved graph convolutional neural network;

[0013] Step 4: Using the source domain data and training set, training an airborne navigation function equipment fault mode recognition model based on an improved graph convolutional neural network;

[0014] Step 5: Use the test set to test the performance of the airborne navigation function equipment fault mode recognition model based on the improved graph convolutional neural network.

[0015] In some embodiments, step 1 includes the following sub-steps:

[0016] Step 1.1: Based on multiple flight missions of the aircraft, the monitoring data of the onboard DME navigation function equipment is collected;

[0017] Step 1.2: Select key parameters in the monitoring data according to the failure mode to form a data set;

[0018] Step 1.3: Divide the dataset into training set and test set.

[0019] In some embodiments, there are four types of failure modes, including insufficient airborne transmission power, insufficient airborne receiving link gain, local oscillator loss of lock and equipment being locked; there are six types of key parameters, including the distance from the aircraft to the ground station, the number of reply signals received by the ground station digital signal processor DSP, the signal strength received by the ground station, the locking time, the number of received pulses 1 and the number of received pulses 2.

[0020] In some embodiments, step 2 includes the following sub-steps:

[0021] Step 2.1: Collect monitoring data of the onboard DME navigation function equipment generated by the aircraft during multiple flight missions;

[0022] Step 2.2: Generate simulated flight data using a flight simulator, wherein the common positive and abnormal conditions of the onboard DME navigation function equipment in flight are simulated by injecting the fault mode described in step 1, the time point and type of fault injection are recorded during each simulated flight, and labels of the fault mode and normal mode are added to each simulated flight data;

[0023] Step 2.3: Convert the monitoring data of the airborne DME navigation function equipment collected in step 2.1 and the simulated flight data generated in step 2.2 into a unified data format and fuse them to obtain source domain data.

[0024] In some embodiments, in step 3, the process of constructing the fault mode recognition model of the airborne navigation function equipment includes:

[0025] Step 3.1: Extract spatiotemporal correlation features based on self-attention;

[0026] Step 3.2: Build a multi-classification fault diagnosis model.

[0027] In some embodiments, the spatiotemporal correlation feature extraction based on self-attention includes:

[0028] For the input data, multiply it by three randomly initialized coefficient matrices that need to be learned during the training process to obtain the query matrix Q, key matrix K and value matrix V;

[0029] Calculate the attention score A through the dot product of Q and K;

[0030] Use the Softmax function to normalize the attention score A to get the attention weight A softmax , and use the attention weight A softmax Perform a weighted summation on the value matrix V to get the final output Z.

[0031] In some embodiments, the multi-classification fault diagnosis model is constructed, including:

[0032] The output Z is used to analyze the correlation between nodes of features using a graph convolutional neural network: the feature transfer matrix of each node is calculated based on the attention score; after the information transfer is completed, the feature transfer matrix H is convolved through the graph convolution layer to obtain the high-level correlation features of each node;

[0033] High-level correlation features are classified and fault diagnosed through two fully connected layers: the high-level correlation features are input into the first fully connected layer, the ReLU activation function is applied in the first fully connected layer, and then after a Dropout layer, the output is passed to the second fully connected layer; the ReLU activation function is also used in the second fully connected layer, and finally a Softmax layer is used to classify the features, including 4 fault modes and 1 normal mode, to obtain the probability distribution of each situation, and select the mode with the largest probability as the diagnosis result, thereby realizing the construction of a multi-classification fault diagnosis model for airborne DME navigation function equipment.

[0034] In some embodiments, the graph convolution layer includes three graph convolution layers, and the output of each graph convolution layer will be used as the input of the next graph convolution layer. The matrix obtained after passing through all graph convolution layers is the high-level correlation feature.

[0035] In some embodiments, step 4 includes the following sub-steps:

[0036] Step 4.1: Input the source domain data of the airborne DME navigation function equipment obtained in step 2 into the airborne navigation function equipment fault diagnosis model for preliminary training; wherein, the mean square error loss function is used to calculate the loss between the predicted output and the true target value, the training optimizer selects the Adam optimizer, and sets the initial value of the learning rate; the above iterative process is repeated until the specified training round is reached and the preliminary training is completed;

[0037] Step 4.2: Retrain the model trained on the source domain data in step 4.1 using the target domain data collected in step 1, i.e., the training set, and freeze the network parameters of the model except the last fully connected layer; after the training of the frozen model is completed, unfreeze all layers and continue training to finally achieve transfer learning from the source domain to the target domain; after the training is completed, save the model and its weights.

[0038] In some embodiments, step 5 includes the following sub-steps:

[0039] Step 5.1: Load the trained airborne navigation function equipment fault diagnosis model and its weights, input the test set into the trained airborne navigation function equipment fault diagnosis model, and output the diagnosis result;

[0040] Step 5.2: Compare the diagnosis results of the onboard navigation function equipment fault diagnosis model with the actual fault situation, calculate the various performance indicators of the onboard navigation function equipment fault diagnosis model in actual application, analyze the diagnosis results of the onboard navigation function equipment fault diagnosis model in detail, and record the cases of correct diagnosis and incorrect diagnosis;

[0041] Step 5.3: Based on the cases of misdiagnosis, optimize and adjust the fault diagnosis model of the airborne navigation function equipment.

[0042] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0043] 1. The present invention constructs an intelligent fault diagnosis method for airborne navigation function equipment, so that various types of electronic system faults occurring during the flight operation of the aircraft can be identified and located more quickly, further improving the accuracy of aircraft fault diagnosis and troubleshooting efficiency, thereby reducing maintenance costs throughout the life cycle of the aircraft, extending the service life of the aircraft, and improving the safety and reliability of aircraft operation.

[0044] 2. The present invention combines the self-attention mechanism with a graph convolutional neural network to build a fault diagnosis model. The self-attention mechanism is first used to focus on capturing the long-range dependencies and key features in the data, and then the graph convolutional neural network is further combined to model the feature correlation information. This allows the model to more accurately infer the fault source and impact path, and maintain high accuracy and robustness when faced with complex system structures and changeable fault modes.

[0045] 3. The present invention performs fault diagnosis on avionics integrated electronic systems based on deep learning. Compared with traditional fault diagnosis algorithms, it can automatically extract fault features, capture complex patterns and relationships in the data, and significantly improve the accuracy and robustness of fault diagnosis.

[0046] 4. The present invention adopts a training method based on transfer learning during training. In the case where the airborne navigation function equipment lacks a large number of training samples with fault labels, it makes full use of the existing relevant data and simulated flight data, thereby reducing the dependence on a large amount of labeled data in the target domain and further improving the generalization ability of the neural network fault diagnosis model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The present invention is a flowchart of a method for identifying fault modes of airborne navigation function equipment based on an improved graph convolutional neural network in an embodiment of the present invention.

[0048] Figure 2 It is a structural schematic diagram of the fault mode recognition model of airborne navigation function equipment based on the improved graph convolutional neural network in an embodiment of the present invention.

[0049] Figure 3 The present invention is a schematic diagram of the training process of the fault mode recognition model of the airborne navigation function equipment based on the improved graph convolutional neural network in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, 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 part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0051] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0052] Example

[0053] like Figure 1 As shown, this embodiment proposes a method for identifying fault modes of airborne navigation function equipment based on an improved graph convolutional neural network, including:

[0054] Step 1: Collect the key parameters of the airborne DME navigation function equipment and construct a data set, which is divided into a training set and a test set;

[0055] In some embodiments, step 1 includes the following sub-steps:

[0056] Step 1.1: Through the antenna, sensor and other equipment of a certain type of aircraft, the monitoring data of the airborne DME navigation function equipment is collected according to the aircraft's multiple flight missions;

[0057] Step 1.2: Select key parameters in the monitoring data according to the fault mode to form a data set; in this embodiment, there are 4 types of fault modes, including insufficient airborne transmission power, insufficient airborne receiving link gain, local oscillator loss of lock and equipment being locked; there are 6 types of key parameters, including the distance from the aircraft to the ground station, the number of reply signals received by the ground station digital signal processor DSP, the signal strength received by the ground station, the locked time, the number of received pulses 1 and the number of received pulses 2.

[0058] Step 1.3: Divide the data set into a training set and a test set; the division ratio can be set as needed. In this embodiment, the training set accounts for 70% and the test set accounts for 30%. The training set is the target domain data required for subsequent transfer learning, retraining and fine-tuning, and labels are added to each sample data according to 4 failure modes and 1 normal mode.

[0059] Step 2: Obtain source domain data by collecting monitoring data and simulated flight data from airborne DME navigation function equipment of different types of aircraft.

[0060] In some embodiments, step 2 includes the following sub-steps:

[0061] Step 2.1: For different models of aircraft with similar navigation systems and sensor layouts, which have been in use for a longer time and have more labeled fault data, regularly collect monitoring data of the onboard DME navigation function equipment generated by the aircraft during multiple flight missions through flight data recorders or data acquisition equipment.

[0062] Step 2.2: Generate simulated flight data using the flight simulator X-Plane, and simulate the common positive and abnormal conditions of the onboard DME navigation function equipment in flight by injecting the fault mode described in step 1. In addition, the time point and type of fault injection are recorded during each simulated flight, and labels are added for each sample data according to 4 fault modes and 1 normal mode.

[0063] Step 2.3: Convert the monitoring data of the airborne DME navigation function equipment collected in step 2.1 and the simulated flight data generated in step 2.2 into a unified data format and fuse them to obtain the source domain data required for subsequent transfer learning training.

[0064] Step 3: Establish an airborne navigation function equipment fault pattern recognition model based on an improved graph convolutional neural network; the construction process of the airborne navigation function equipment fault pattern recognition model is divided into spatiotemporal correlation feature extraction based on self-attention and multi-classification fault diagnosis model construction. The overall model structure is as follows: Figure 2 shown.

[0065] Step 3.1: Extract spatiotemporal correlation features based on self-attention;

[0066] For input data X of size n×m, where n is the number of parameters and m is the length of the time series, for the input data X, multiply it by three randomly initialized coefficient matrices W that need to be learned during the training process Q ,W K ,W V, to obtain the query matrix Q, key matrix K and value matrix V. Then, the attention score A is calculated by the dot product of Q and K, which represents the correlation between each time point. Finally, the attention score A is normalized using the Softmax function to obtain the attention weight A softmax , and use the attention weight A softmax Perform weighted summation on the value matrix V to get the final output Z:

[0067] A softmax =Softmax(A)

[0068] Z=A softmax V

[0069] Among them, the output Z is the enhanced feature representation of the input data, that is, the enhanced feature representation, which contains the dependencies between different time points in the input data and integrates the global information. The representation of each time point not only contains its own information, but also contains important information related to other time points and spatial correlation.

[0070] Step 3.2: Building a multi-classification fault diagnosis model;

[0071] For the enhanced feature representation obtained by the self-attention layer, that is, the output Z, we further use the graph convolutional neural network to analyze the correlation between nodes between features. First, we calculate the feature transfer matrix H of each node, that is, the part of the node that transfers information to its neighbors, and the calculation formula is:

[0072]

[0073] In the formula, H is the feature transfer matrix of the node, A is the attention score matrix, that is, the adjacency matrix, I is the identity matrix, is a diagonal matrix where each diagonal element D ii is the reciprocal of the square root of the sum of the i-th row of the adjacency matrix A.

[0074] After the information transfer is completed, the feature transfer matrix H is convolved through the graph convolution layer to obtain the high-level association feature Y of each node:

[0075] Y = ReLU(H·W)

[0076] Where W is a learnable weight matrix used for convolution operation; ReLU is a nonlinear activation function. In this embodiment, three graph convolution layers are designed, and the output of each graph convolution layer will be used as the input of the next graph convolution layer. The matrix obtained after all graph convolution layers is the high-level association feature Y.

[0077] After obtaining the high-level correlation feature Y of the monitoring data of the airborne DME navigation function equipment, the high-level correlation features are classified and fault diagnosed through two fully connected layers, and finally the fault diagnosis results of the airborne DME navigation function equipment are obtained. First, the high-level correlation features are input into the first fully connected layer, and the ReLU activation function is applied to introduce nonlinear characteristics and enhance the expression ability of the model; then, a Dropout layer is used to prevent overfitting and improve the generalization ability of the model; then, the output is passed to the second fully connected layer, and the ReLU activation function is also used in the second fully connected layer to further extract and fuse feature information; finally, the features are classified through a Softmax layer, including 4 fault modes and 1 normal mode, and the probability distribution of each situation is obtained. The mode with the largest probability is selected as the diagnosis result, thereby realizing the construction of a multi-classification fault diagnosis model for airborne DME navigation function equipment.

[0078] Step 4: Train the fault mode recognition model of airborne navigation function equipment based on the improved graph convolutional neural network;

[0079] In some embodiments, transfer learning is used to train the fault diagnosis model of the airborne navigation function equipment. The training process is as follows: Figure 3 The training process is divided into two parts: initial training of source domain data and fine-tuning and retraining of target domain data. Step 4 includes the following sub-steps:

[0080] Step 4.1: Input the source domain data of the airborne DME navigation function equipment obtained in step 2 into the airborne navigation function equipment fault diagnosis model for preliminary training, using the mean square error loss function Calculate the loss between the predicted output Y and the true target value y, select the Adam optimizer for training, and set the initial value of the learning rate to 0.001. Repeat the above iterative process until the specified 40 training rounds are reached, completing the preliminary training.

[0081] Step 4.2: The model trained on the source domain data is retrained using the target domain data collected in step 1. The network parameters of the model except the last fully connected layer are frozen to retain the learned common features. The loss function, optimizer, learning rate, and training rounds are set in the same way as the initial training. After the frozen model training is completed, all layers are unfrozen and training is continued so that the model can further adapt to the characteristics of the target domain data. At this time, the learning rate is appropriately reduced to fine-tune the model weights and finally achieve transfer learning from the source domain to the target domain. After training is completed, save the fine-tuned model and its weights for use in the actual fault diagnosis process.

[0082] Step 5: Test the performance of the fault pattern recognition model of airborne navigation functional equipment based on the improved graph convolutional neural network;

[0083] In some embodiments, step 5 includes the following sub-steps:

[0084] Step 5.1: Apply the airborne navigation function equipment fault diagnosis model based on the improved graph convolutional neural network trained by transfer learning to the airborne DME navigation function equipment in the actual aviation integrated electronic system, load the trained airborne navigation function equipment fault diagnosis model and its weights, and input the test set collected in step 1 into the airborne navigation function equipment fault diagnosis model based on the improved graph convolutional neural network trained by transfer learning, and output the diagnosis results - insufficient airborne transmission power, insufficient airborne receiving link gain, local oscillator loss of lock, equipment locked and normal mode.

[0085] Step 5.2: Compare the diagnosis results of the onboard navigation function equipment fault diagnosis model with the actual fault situation, and calculate the various performance indicators of the onboard navigation function equipment fault diagnosis model in actual application, including accuracy, recall rate and F1 score. The calculation formula is as follows:

[0086]

[0087] Conduct a detailed analysis of the diagnostic results of the fault diagnosis model for airborne navigation function equipment, and record cases of correct and incorrect diagnosis.

[0088] Step 5.3: In view of the poor performance of the airborne navigation function equipment fault diagnosis model on certain fault types, that is, cases of misdiagnosis, the airborne navigation function equipment fault diagnosis model is further optimized and adjusted to ensure the reliability and effectiveness of the airborne navigation function equipment fault diagnosis model based on the improved graph convolutional neural network in the actual fault diagnosis application of airborne DME navigation function equipment.

[0089] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for fault mode recognition of airborne navigation function equipment based on improved graph convolutional neural network, characterized in that: include: Step 1: Collect the key parameters of the airborne DME navigation function equipment and construct a data set, which is divided into a training set and a test set; Step 2: Obtain source domain data by collecting monitoring data and simulated flight data of airborne DME navigation function equipment of different types of aircraft; Step 3: Establish a fault mode recognition model for airborne navigation functional equipment based on the improved graph convolutional neural network; Step 4: Using the source domain data and training set, training an airborne navigation function equipment fault mode recognition model based on an improved graph convolutional neural network; Step 5: Use the test set to test the performance of the airborne navigation function equipment fault mode recognition model based on the improved graph convolutional neural network.

2. According to claim 1, the method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network is characterized in that: Step 1 includes the following sub-steps: Step 1.1: Based on multiple flight missions of the aircraft, the monitoring data of the onboard DME navigation function equipment is collected; Step 1.2: Select key parameters in the monitoring data according to the failure mode to form a data set; Step 1.3: Divide the dataset into training set and test set.

3. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 2 is characterized in that: There are four types of failure modes, including insufficient airborne transmission power, insufficient airborne receiving link gain, local oscillator loss of lock and equipment being locked; there are six types of key parameters, including the distance from the aircraft to the ground station, the number of reply signals received by the ground station's digital signal processor DSP, the signal strength received by the ground station, the locking time, the number of received pulses 1 and the number of received pulses 2.

4. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 2 is characterized in that: Step 2 includes the following sub-steps: Step 2.1: Collect monitoring data of the onboard DME navigation function equipment generated by the aircraft during multiple flight missions; Step 2.2: Generate simulated flight data using a flight simulator, wherein the common positive and abnormal conditions of the onboard DME navigation function equipment in flight are simulated by injecting the fault mode described in step 1, the time point and type of fault injection are recorded during each simulated flight, and labels of the fault mode and normal mode are added to each simulated flight data; Step 2.3: Convert the monitoring data of the airborne DME navigation function equipment collected in step 2.1 and the simulated flight data generated in step 2.2 into a unified data format and fuse them to obtain source domain data.

5. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 1, characterized in that: In step 3, the process of constructing the fault mode recognition model of the airborne navigation function equipment includes: Step 3.1: Extract spatiotemporal correlation features based on self-attention; Step 3.2: Build a multi-classification fault diagnosis model.

6. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 5 is characterized in that: The spatiotemporal correlation feature extraction based on self-attention includes: For the input data, multiply it by three randomly initialized coefficient matrices that need to be learned during the training process to obtain the query matrix Q, key matrix K and value matrix V; Calculate the attention score A through the dot product of Q and K; Use the Softmax function to normalize the attention score A to get the attention weight A softmax , and use the attention weight A softmax Perform a weighted summation on the value matrix V to get the final output Z.

7. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 6 is characterized in that: The multi-classification fault diagnosis model is constructed, including: The output Z is used to analyze the correlation between nodes of features using a graph convolutional neural network: the feature transfer matrix of each node is calculated based on the attention score; after the information transfer is completed, the feature transfer matrix H is convolved through the graph convolution layer to obtain the high-level correlation features of each node; High-level correlation features are classified and fault diagnosed through two fully connected layers: the high-level correlation features are input into the first fully connected layer, the ReLU activation function is applied in the first fully connected layer, and then after a Dropout layer, the output is passed to the second fully connected layer; the ReLU activation function is also used in the second fully connected layer, and finally a Softmax layer is used to classify the features, including 4 fault modes and 1 normal mode, to obtain the probability distribution of each situation, and select the mode with the largest probability as the diagnosis result, thereby realizing the construction of a multi-classification fault diagnosis model for airborne DME navigation function equipment.

8. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 7 is characterized in that: The graph convolution layer includes three graph convolution layers. The output of each graph convolution layer will be used as the input of the next graph convolution layer. The matrix obtained after passing through all graph convolution layers is the high-level correlation feature.

9. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 1, characterized in that: Step 4 includes the following sub-steps: Step 4.1: Input the source domain data of the airborne DME navigation function equipment obtained in step 2 into the airborne navigation function equipment fault diagnosis model for preliminary training; wherein, the mean square error loss function is used to calculate the loss between the predicted output and the true target value, the training optimizer selects the Adam optimizer, and sets the initial value of the learning rate; the above iterative process is repeated until the specified training round is reached and the preliminary training is completed; Step 4.2: Retrain the model trained on the source domain data in step 4.1 using the target domain data collected in step 1, i.e., the training set, and freeze the network parameters of the model except the last fully connected layer; after the training of the frozen model is completed, unfreeze all layers and continue training to finally achieve transfer learning from the source domain to the target domain; after the training is completed, save the model and its weights.

10. The method for identifying fault patterns of airborne navigation function equipment based on an improved graph convolutional neural network according to claim 1, characterized in that: Step 5 includes the following sub-steps: Step 5.1: Load the trained airborne navigation function equipment fault diagnosis model and its weights, input the test set into the trained airborne navigation function equipment fault diagnosis model, and output the diagnosis result; Step 5.2: Compare the diagnosis results of the onboard navigation function equipment fault diagnosis model with the actual fault situation, calculate the various performance indicators of the onboard navigation function equipment fault diagnosis model in actual application, analyze the diagnosis results of the onboard navigation function equipment fault diagnosis model in detail, and record the cases of correct diagnosis and incorrect diagnosis; Step 5.3: Based on the cases of misdiagnosis, optimize and adjust the fault diagnosis model of the airborne navigation function equipment.

Images