Satellite attitude control system fault diagnosis method and device based on time-frequency domain feature fusion

Abstract

The invention provides a satellite attitude control system fault diagnosis method and device based on time-frequency domain feature fusion, and belongs to the technical field of satellite fault diagnosis, and the method can comprise the steps: obtaining target satellite data, and converting the target satellite data into satellite image data through a Gramer angle field; performing feature extraction on the satellite image data to obtain a satellite feature image; global time sequence features of the satellite feature image are extracted, and the global time sequence features refer to image sequence features which are extracted from the whole time sequence of the satellite and can represent the whole sequence features; inputting the global time sequence features into a dense connection model to obtain a fusion time sequence feature graph corresponding to the satellite feature image; and obtaining a target fault type corresponding to the target satellite data according to the fused time sequence feature map. According to the technical scheme, the efficiency and accuracy of satellite fault diagnosis can be improved.

Description

Technical Field

[0001] The present disclosure relates to the technical field of satellite fault diagnosis, and in particular to a method and device for satellite attitude control system fault diagnosis based on time-frequency domain feature fusion. Background Art

[0002] Satellite fault diagnosis can ensure the success of key space missions, improve system safety and reliability, and protect the low-Earth orbit environment. It is a key link in ensuring the long-term and stable operation of satellites in the aerospace field.

[0003] At present, fault diagnosis methods for satellites usually include fault diagnosis methods based on analytical models, fault diagnosis methods based on expert knowledge, and fault diagnosis methods based on data-driven. Due to the uncertainty of the on-orbit environment, it is difficult for fault diagnosis methods based on analytical models to provide a mathematical model with simple structure and high precision. Fault diagnosis methods based on expert knowledge require a large number of knowledge rules, cannot form a complete rule base, and are not portable, so the fault diagnosis efficiency is low. Fault diagnosis methods based on data-driven methods will cause the loss of detailed features in the original fault signal during feature extraction, affecting the message of fault diagnosis, and thus resulting in low accuracy of fault diagnosis. Summary of the invention

[0004] In view of this, the present disclosure hopes to provide a satellite attitude control system fault diagnosis method and device based on time-frequency domain feature fusion, which can improve the efficiency and accuracy of satellite fault diagnosis.

[0005] The technical solution of the present disclosure is achieved as follows: In a first aspect, the present disclosure provides a satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion, comprising: Acquire target satellite data, and convert the target satellite data into satellite image data using the Gram angle field; Extracting features from the satellite image data to obtain a satellite feature image; Extracting global time series features of satellite feature images, wherein the global time series features refer to image sequence features extracted from the entire time series of the satellite and capable of representing the characteristics of the entire sequence; Inputting the global time series feature into a dense connection model to obtain a fused time series feature graph corresponding to the satellite feature image; The target fault type corresponding to the target satellite data is obtained according to the fused time series feature diagram.

[0006] In a second aspect, the present disclosure provides a satellite attitude control system fault diagnosis device based on time-frequency domain feature fusion, the device comprising: A data conversion module, used for acquiring target satellite data and converting the target satellite data into satellite image data using the Gram angle field; A feature extraction module extracts features from the satellite image data to obtain a satellite feature image; A time series extraction module extracts global time series features of satellite feature images, wherein the global time series features refer to image sequence features extracted from the entire time series of the satellite and capable of representing the characteristics of the entire sequence; A feature fusion module, inputting the global time series feature into a dense connection model to obtain a fused time series feature graph corresponding to the satellite feature image; A fault classification module obtains a target fault type corresponding to the target satellite data according to the fused time series feature graph.

[0007] In a third aspect, the present disclosure provides a computing device, the computing device comprising: a memory and a processor; wherein: The memory is used to store a computer program that can be run on the processor; The processor is used to execute the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion described in the first aspect when running the computer program.

[0008] In a fourth aspect, the present disclosure provides a computer-readable storage medium storing at least one instruction, wherein the at least one instruction is used to be executed by a processor to implement the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion as described in the first aspect.

[0009] The present disclosure provides a method and device for fault diagnosis of a satellite attitude control system based on time-frequency domain feature fusion. The method converts target satellite data into satellite image data, extracts features, and then automatically extracts global time series features from the target satellite data to obtain global feature information that characterizes the overall trend and pattern of the target satellite's operating state. The satellite image data is represented by image sequence features, which can improve the performance of the features in expressing satellite data, thereby improving the accuracy of fault diagnosis, reducing manual intervention, and improving the efficiency of fault diagnosis. The design of the dense connection model reduces the number of parameters, thereby reducing the consumption of computing resources and improving the training and reasoning efficiency of the model. The dense connection structure of the dense connection model is used to further extract deep features from the global time series features and fuse features at different levels, providing a more comprehensive description of the satellite state. In addition, feature fusion reduces redundant calculations, improves the efficiency of data processing, makes the fault diagnosis process faster, and can identify more complex fault modes, and can adapt to different fault conditions and satellite environments. The health and fault states of the satellite can be accurately diagnosed according to the target fault type, thereby improving the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flow chart of a satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion provided by the present invention.

[0011] Figure 2 A schematic diagram of a network architecture based on feature fusion provided in the present invention.

[0012] Figure 3 A schematic diagram of the composition of a GRU network model provided in the present invention.

[0013] Figure 4 A training flowchart of a GRU network model provided in the present invention.

[0014] Figure 5 A schematic diagram of the composition of a densely connected model provided in the present disclosure.

[0015] Figure 6 A schematic diagram of a confusion matrix of a fault detection result provided in the present disclosure.

[0016] Figure 7 A schematic diagram of a change curve of a loss function and training discussion provided in the present invention.

[0017] Figure 8 A schematic diagram of a curve showing changes in accuracy and training discussion provided by the present invention.

[0018] Fig. 9 A schematic diagram of a satellite attitude control system fault diagnosis device based on time-frequency domain feature fusion provided by the present invention.

[0019] Fig.10 A schematic diagram of the hardware structure of a computing device provided by the present invention. DETAILED DESCRIPTION

[0020] The technical solutions in the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings in the present disclosure.

[0021] Based on the above description of the satellite fault diagnosis method, it is difficult for the fault diagnosis method based on the analytical model to provide a mathematical model with a simple structure and high precision. The fault diagnosis method based on expert knowledge requires a large number of knowledge rules, cannot form a complete rule base and is not portable, so the fault diagnosis efficiency is low. The data-driven fault diagnosis method will cause the loss of detailed features in the original fault signal during the feature extraction process, affecting the fault diagnosis information, and thus resulting in low fault diagnosis accuracy. Based on this, the present disclosure hopes to provide a satellite attitude control system fault diagnosis technical solution based on time-frequency domain feature fusion, see Figure 1, which shows a satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion provided by the present disclosure, the method may include: S101: Acquire target satellite data, and convert the target satellite data into satellite image data using the Gram angle field.

[0022] The target satellite data usually comes from various sensors on the satellite, such as momentum wheel speed sensor, torque sensor, motor current and voltage sensor, temperature sensor, etc. These sensors collect various telemetry data of the satellite during operation at a certain sampling frequency, such as the speed of the momentum wheel, control torque, output torque, attitude quaternion of the satellite, angular velocity and other engineering parameters.

[0023] The target satellite data is time series data with a time dimension, and the sampling frequencies of different sensors may be different. The data volume is large and contains rich feature information that characterizes the various working modes, states, and health conditions of components such as satellite momentum wheels. However, there are also problems such as difficulty in sampling alignment and a large number of irrelevant features.

[0024] The target satellite data is data obtained after preprocessing the initial satellite data using a Kalman filter, wherein the preprocessing at least includes a normalization process for normalizing the initial satellite data to a uniform range, a denoising process for reducing noise in the initial satellite data, and a process for dividing the initial satellite data into segments of a fixed length according to a set time step to obtain the target satellite data, so that the target satellite data is suitable for input into a corresponding fault identification model.

[0025] In some examples, first, a certain type of data (such as momentum wheel speed data) in the acquired satellite time series data is taken as a time series with a length of N. Then, the cosine value of the angle between each pair of data points in the time series is calculated to construct an N×N Gram matrix, in which each element in the matrix represents the cosine of the angle between the corresponding two data points.

[0026] Specifically, suppose we have a set of vectors The Gram matrix is ​​derived from The inner product matrix of every pair of vectors. As shown in the following formula, each element in the matrix is a vector and The vector product between .

[0027]

[0028] Where G represents the Gram matrix, and the vector in the Gram matrix is and Represents the normalized scalar values ​​at different time points in the time series. Specifically: Given a time series {x 1 ,x 2 ,...,x N}, first normalize (scale to [0,1] or [-1,1]), each time point x i is converted to an angle in polar coordinates =arccos(x i ). The Gram matrix element G i , j =cos( + ), reflecting the temporal relationship between time points i and j. That is, the vector and It is a normalized scalar time point value instead of the traditional vector inner product.

[0029] The values ​​in the constructed Gram matrix are standardized, usually using the min-max standardization method to map the values ​​to the range [0,1]. The purpose of this is to unify the scale of the data, which is convenient for subsequent image representation and processing. The standardized Gram matrix is ​​converted into a grayscale image. Specifically, each value in the matrix corresponds to the grayscale value of a pixel in the image, and the size of the grayscale value reflects the relationship between the time series data points. Through this conversion, the original one-dimensional time series satellite data is mapped to a two-dimensional image space to form satellite image data.

[0030] S102: Extract features from satellite image data to obtain satellite feature images.

[0031] The information contained in satellite data images may be relatively redundant, and for subsequent classification, recognition and other tasks, directly using these images may cause noise interference, large amount of calculation and other problems. Through feature extraction, irrelevant and redundant information can be removed, and features related to key information such as satellite failures in the image can be highlighted, thereby improving the performance and efficiency of subsequent tasks. The feature extraction process can be carried out in a variety of ways, such as using a convolutional neural network to extract features and obtain satellite feature images. The specific feature extraction method is not described in detail in this example implementation.

[0032] S103: Extracting global temporal features of satellite feature images.

[0033] Global time series features refer to image sequence features extracted from the entire time series of the satellite that can represent the characteristics of the entire sequence. Usually, long time series data are collected from satellites, that is, satellite feature images are used to identify and extract global sequence image features that can represent the trends and patterns of the entire data set. The image sequence features usually include statistical parameters, frequency domain components, and time domain characteristics, which can capture the overall changes in the satellite's operating status and provide key information for subsequent fault detection, health monitoring, and behavior prediction.

[0034] Exemplarily, the process of extracting global sequence features can be based on the satellite's momentum wheel, star sensor and gyro components, using a gated recurrent unit (GRU) algorithm, Fourier transform and autocorrelation analysis to extract global time series features from satellite feature images, and the global time series features are screened through principal component analysis and recursive feature elimination. The most representative features, namely, star sensor quaternions, gyro rotation speed and system rotation speed, are selected from all global time series features to capture the overall pattern and trend information of satellite operation that is useful for fault detection and diagnosis.

[0035] S104: Input the global temporal features into the dense connection model to obtain a fused temporal feature map corresponding to the satellite feature image.

[0036] The global time series features can be input into the dense connection model 250 (DenseNet), in which each layer is connected to all previous layers, and the input of each layer is the integration of the outputs of all previous layers, so that each layer can use the global feature information of all previous layers. Through the dense connection between layers and the splicing of global feature information in the channel dimension, the fusion of features is achieved to obtain a fused time series feature map. The fused time series feature map includes not only the statistical information of the time series, but also the frequency domain features, time domain features and nonlinear relationships, which together constitute a multi-dimensional data representation.

[0037] S105: Obtain a target fault type corresponding to the target satellite data according to the fused time series feature graph.

[0038] In some example embodiments of the present disclosure, the fused time series feature map may be input into a Kalman filter for smoothing and correction processing to reduce the impact of noise.

[0039] Exemplarily, key features are extracted from the output of the Kalman filter and mapped to a pre-trained fault identification model to obtain target fault types and their locations corresponding to the probability distribution of target satellite data on different fault types, and a corresponding fault report is generated.

[0040] According to the description of the above scheme, by converting the target satellite data into satellite image data, and performing feature extraction, and then automatically extracting the global time series features in the target satellite data, the global feature information that characterizes the overall trend and pattern of the target satellite's operating state is obtained, and the satellite image data is represented by image sequence features, which can improve the performance of the features in expressing satellite data, thereby improving the accuracy of fault diagnosis, reducing manual intervention, and thus improving the efficiency of fault diagnosis. The design of the dense connection model 250 reduces the number of parameters, thereby reducing the consumption of computing resources and improving the training and reasoning efficiency of the model. The dense connection structure of the dense connection model 250 is used to further extract deep features from the global time series features and fuse features at different levels, providing a more comprehensive description of the satellite state. In addition, feature fusion reduces redundant calculations, improves the efficiency of data processing, makes the fault diagnosis process faster, and can identify more complex fault modes, and can adapt to different fault conditions and satellite environments. The health and fault states of the satellite can be accurately diagnosed according to the target fault type, thereby improving the accuracy of fault diagnosis.

[0041] against Figure 1 The technical solution shown in Figure 2 , which shows a schematic diagram of a network architecture composition based on feature fusion provided by the present disclosure, such as Figure 2 As shown, the network architecture based on feature fusion may include a data preprocessing module 210 , a signal conversion module 220 , a feature extraction module 230 , a GRU network model 240 and a dense connection model 250 .

[0042] In the present disclosure, the target satellite data is obtained by preprocessing the initial satellite data. Specifically, the data preprocessing module 210 can be used for processing, specifically, the initial satellite data is preprocessed to obtain the target satellite data.

[0043] For example, initial satellite data may be first acquired, and then preprocessed using a Kalman filter to obtain target satellite data; wherein the preprocessing at least includes normalizing, denoising, and segmenting the initial satellite data.

[0044] For the above example, the initial satellite data is preprocessed using a Kalman filter to obtain the target satellite data. Specifically, the preprocessing includes at least normalizing, denoising, and segmenting the initial satellite data. For example, data cleaning is performed to remove missing values ​​and outliers to ensure data quality; feature selection is performed to select the features most relevant to the fault diagnosis task; normalization is performed to ensure that the data is within a uniform range, thereby improving the convergence speed and stability of model training; denoising is performed to reduce noise in the data, improve the accuracy of the model, and time window division is performed to divide the long time series data of the satellite into multiple time windows.

[0045] In some examples, a fixed-length time step, for example, 24,000, can be used as input to segment the long time series data of the satellite into fixed-length sequences.

[0046] Based on the above example, after obtaining the target satellite data, the above signal conversion module 220 can be used to perform data conversion on the target satellite data to obtain satellite image data. Specifically, the above Gram matrix can be used to obtain the satellite image data of the target satellite data. The acquisition process has been described in detail above, so it will not be repeated here.

[0047] After the satellite image data is acquired, the feature extraction module 230 may be used to extract features from the satellite image data to obtain a satellite feature image. The specific feature extraction method has been described above and will not be repeated here.

[0048] After obtaining the satellite feature image, the GRU network model 240 may be used to extract features from the satellite feature image to obtain global temporal features.

[0049] For the above example, specifically, various signal processing techniques can be used, such as the GRU network model 240, Fourier transform, wavelet transform, and autocorrelation analysis, to extract global time series features from the target satellite data, screen the global time series features through principal component analysis and recursive feature elimination, and select the most representative features from all extracted global time series features as the global time series features corresponding to the target satellite data.

[0050] For some examples, see Figure 3, extracting the global temporal features of the satellite feature image can be achieved by using the GRU network model 240, wherein the GRU network model 240 at least includes at least one feature extraction layer, at least one data processing layer, and a fully connected layer for output connected to the data processing layer. The process of extracting the global temporal features of the satellite feature image, first, input the satellite feature image obtained by preprocessing into the input layer of the GRU network model 240, superimpose the input and output of the feature extraction layer into the data processing layer, and then superimpose the input data and output data of the data processing layer into the next feature extraction layer or fully connected layer to obtain the global temporal features corresponding to the satellite feature image.

[0051] For the above example, specifically, the input layer of the GRU network model 240 receives the preprocessed multi-dimensional satellite feature image, inputs the set batch size batch_size, time step input_time_steps (the length of the sliding window), and feature number input_feature_number into the input layer, and uses a set of cascaded feature extraction layers and at least one feature extraction layer in the data processing layer to extract key global temporal features from the satellite feature image input by the input layer, wherein the feature extraction layer can be a convolutional layer or other types of layers for capturing patterns and trends in the satellite feature image.

[0052] In the present disclosure, the feature extraction layer is a feature extraction layer in a recurrent neural network based on the GRU algorithm. The input and output of the feature extraction layer in the group are superimposed to form an enhanced feature representation, which is input to the data processing layer. At least one data processing layer further processes the output of the feature extraction layer to enhance the nonlinear modeling capability of the feature. The input data and output data of the data processing layer are superimposed and input to the next feature extraction layer or directly to the fully connected layer. In the present disclosure, when the input data and output data of the data processing layer are superimposed and input to the next feature extraction layer, it is also necessary to reduce the feature dimension, extract key information, and reduce the computational complexity. The output result of the data processing layer is then input to the fully connected layer to obtain global time series features.

[0053] Based on the above technical solution, in the specific implementation process, before extracting the global temporal features of the satellite feature image, it is also necessary to train the initial network model based on the GRU algorithm based on the simulated sample feature image and the real sample feature image to obtain the GRU network model 240. Figure 4 , which shows a training flow chart of a GRU network model 240 provided by the present disclosure. In some examples, the training process may include: S401: Acquire a sample feature image and a global temporal feature corresponding to the sample feature image; S402: extracting features from the sample feature image using a feature extraction layer to obtain sample time series features; S403: Input the sample time series features into the GRU-based initial network model to obtain reference time series features; S404: Update the initial network model and feature extraction layer according to the real global time series features and the reference time series features to obtain a GRU network model.

[0054] For the above example, the corresponding global time series features are annotated based on the sample feature images of the satellite, and a unique label or code is assigned to each time series feature for subsequent model training. In an exemplary embodiment of the present disclosure, the GRU network model 240 is mainly a neural network model based on deep learning. For example, the GRU network model 240 can be based on a feedforward neural network. The feedforward network can be implemented as an acyclic graph, in which the nodes are arranged in layers. Generally, the feedforward network topology includes an input layer and an output layer, and the input layer and the output layer are separated by at least one hidden layer. The hidden layer transforms the input received by the input layer into a representation useful for generating output in the output layer. The network nodes are fully connected to the nodes in the adjacent layers via edges, but there are no edges between the nodes in each layer. The data received at the nodes of the input layer of the feedforward network is propagated (ie, "feedforward") to the nodes of the output layer via an activation function, and the activation function calculates the state of the nodes of each continuous layer in the network based on coefficients ("weights"), and the coefficients are respectively associated with each of the edges connecting these layers. The output of the GRU network model 240 can take various forms, and the present disclosure is not limited to this. The GRU network model 240 may also be other neural network models, such as a convolutional neural network (CNN) model, a recurrent neural network (RNN) model, a generative adversarial network (GAN) model, but is not limited thereto, and other neural network models known to those skilled in the art may also be used.

[0055] For the above training process, specifically, first, obtain the corresponding sample feature image, and then use the above sample feature image to train the above initial network model, which may specifically include the following steps: select a network topology; use a set of training data representing the problem modeled by the network; adjust the weights until the network model exhibits the minimum error for all instances of the training data set. For example, during a supervised learning training process for a neural network, the output generated by the network in response to an input representing an instance in the training data set is compared with the "correct" labeled output of the instance; an error signal representing the difference between the output and the labeled output is calculated; and when the error signal is propagated backward through the layers of the network, the weights associated with the connection are adjusted to minimize the error. The model when the error of each output generated from an instance of the training data set is minimized is defined as a GRU network model 240.

[0056] Based on the foregoing description, after obtaining the global time series features, it is necessary to perform feature fusion through the dense connection model 250. Exemplarily, according to the importance of the features or a specific fusion strategy, in some possible implementations, at least two intermediate features can be selected for feature fusion, and the global time series features and at least one of the multiple intermediate features can also be fused, and these intermediate features can come from different layers to capture multi-scale information. The specific feature fusion method can select a suitable fusion method based on prior knowledge, feature importance assessment or experimental results, such as feature splicing, weighted summation or more complex fusion strategies to integrate the information of the selected features, wherein feature splicing can be to splice the selected features in the channel dimension to achieve feature fusion, thereby increasing the dimension and information content of the features.

[0057] In some examples, to perform feature fusion through the dense connection model 250, it is first necessary to build a network structure of the dense connection model 250, such as Figure 5 As shown, the densely connected model 250 is composed of multiple layers of feature fusion layers connected in sequence, and each layer is connected to the previous layer to ensure the flow of information between layers and the accumulation of features.

[0058] In some examples, such as Figure 5 As shown, the feature fusion layer includes multiple layers of convolution modules; inputting the global temporal features into the first layer of feature fusion layer to obtain the first layer of fused temporal feature map can include: using the multiple layers of convolution modules to perform convolution processing on the global temporal features to obtain multiple layers of intermediate features; wherein the input of the P-th layer of convolution module is the output of the P-1-th layer of convolution module, and P is greater than or equal to 2; and fusing at least two of the multiple layers of intermediate features to obtain the first layer of fused temporal feature map.

[0059] For the above example, specifically, the feature fusion layer is composed of multiple convolution modules, each of which is responsible for extracting features at different levels. Each convolution module can include settings for parameters such as convolution kernel size, step size, and padding to meet the extraction requirements of different features. The global time series features are transmitted to the first layer of convolution modules as input data. Optionally, the first layer of convolution modules performs a convolution operation on the input global time series features to generate convolution time series features of the first layer, wherein the convolution operation is to extract local features using convolution kernels in each layer of feature fusion layer to obtain the intermediate features of the corresponding layer. Optionally, the first layer of convolution modules performs batch normalization and nonlinear activation functions on the input global time series features, wherein batch normalization is to apply batch normalization to reduce internal covariate shift; nonlinear activation is to introduce nonlinearity through nonlinear activation functions such as ReLU to enhance the expression ability of the model. The output of each layer of convolution modules is used as the input of the next layer of convolution modules to form multiple layers of intermediate features. For the P-th layer of convolution modules, where P≥2, its input is the output of the P-1-th layer of convolution modules to ensure the flow of information between layers. Among the multiple layers of intermediate features, before fusion, feature selection techniques can be applied, such as coefficient of variation-based and information gain-based methods, to select at least two intermediate features with the most information for fusion.

[0060] After the fusion operation, the first layer of fused time series feature graph, i.e., the fused time series feature graph of feature fusion layer 1, is obtained, which integrates the information extracted by the multi-layer convolution module. The first layer of fused time series feature graph will be used as the input of the second layer of fused feature layer, and accordingly, the output of the second layer of fused feature layer, i.e., feature fusion layer 2, will be used as the input of the subsequent layer, or for further processing of the fault diagnosis model.

[0061] Through the above multi-layer convolution processing and feature fusion, the first feature fusion layer can effectively extract and integrate key information from the global time series features, providing strong data support for fault diagnosis. Fusion of intermediate features at different levels can capture multi-scale information and improve the accuracy of fault diagnosis.

[0062] For example, see Figure 5 Assuming that the dense connection model 250 includes M layers of feature fusion layers connected in sequence, inputting the global time series feature into the dense connection model 250 to obtain the fused time series feature map corresponding to the satellite feature image can include: inputting the global time series feature into the first feature fusion layer to obtain the first fused time series feature map; using the output of the feature fusion layer of the Nth layer as the input of the N+1th feature fusion layer to obtain the fused time series feature map of the N+1th layer; and using the fused time series feature map of the Mth layer as the target fused time series feature map.

[0063] For the above example, specifically, the extracted global temporal features are transmitted as input data to the first feature fusion layer of the densely connected model 250. In the first feature fusion layer, the input global temporal features are preliminarily processed, such as the convolution operation shown in the foregoing example content. The first fused temporal feature map is used as the input for the next layer for feature transfer. Each subsequent feature fusion layer receives the outputs from all previous layers as input to achieve feature accumulation and feature fusion. For the feature fusion layer of the Nth layer (1 < N < M), the operations of convolution, normalization, and activation function are repeatedly performed, and the output is passed to the (N + 1)th layer, which will not be elaborated in detail here. After the Mth layer, that is, the last feature fusion layer, completes the processing, the target fused temporal feature map is output. The target fused temporal feature map contains all the information accumulated from the input to the last layer. Through the method of layer-by-layer feature fusion, the densely connected model 250 can effectively extract and integrate key information from the global temporal features, providing strong data support for fault diagnosis.

[0064] In some examples, when obtaining the target fused temporal feature map, at least one of the global temporal features and multiple intermediate features can be fused to obtain the fused temporal feature map corresponding to the satellite feature image.

[0065] Exemplarily, through the above satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion, the global temporal features can also be fused with at least one intermediate feature. First, they are concatenated in the channel dimension to increase the dimension of the features. Second, weighted summation is performed, and the weights can be dynamically adjusted according to the importance of the features. Then, element-wise multiplication or division operations are performed to combine the global temporal features and the intermediate features to achieve feature fusion. The features after the fusion operation and processing are used as the fused temporal feature map corresponding to the satellite feature image, and this fused temporal feature map synthesizes the information of the global temporal features and the intermediate features.

[0066] It should be noted that based on the foregoing training process of the GRU network model 240, correspondingly, the densely connected model 250 also needs to be trained. Comparing with the training process of the GRU network model 240, the input of the GRU network model 240 is the target satellite data, and the output is the global temporal features. The input of the densely connected model 250 is the global temporal features obtained through the GRU network model 240, and the output is the fused temporal feature map. For the specific training process, please refer to the elaboration of the foregoing training process content, which will not be elaborated in detail here.

[0067] Based on the above description, after obtaining the fused time series feature graph, a Kalman filter is constructed, and the parameters of the Kalman filter, such as the state transfer matrix, the observation matrix, the process noise covariance matrix, and the observation noise covariance matrix, are defined. The output results of the GRU network model 240 and the dense connection model 250, i.e., the global time series features and the fused time series feature graph, are input into the Kalman filter, and the output results are smoothed and corrected by the Kalman filter to reduce the influence of random noise and correct the estimation of the system state, thereby providing a more accurate and stable fault diagnosis result.

[0068] In some examples, obtaining a target fault type corresponding to a satellite feature image based on a fused time series feature graph may include: obtaining a fault characterization vector corresponding to the satellite feature image based on the fused time series feature graph; and determining a target fault type corresponding to the satellite feature image based on the fault characterization vector.

[0069] For the above example, specifically, taking the input of the fused time series feature graph of the target satellite data into a pre-trained fault identification model as an example, the pre-trained fault identification model may include a global average pooling layer, a fully connected layer, and an output layer. The hierarchical processing in the pre-trained fault identification model is used to capture the complex patterns and trends in the target satellite data, and the feature dimension is reduced and key information is extracted through the global average pooling layer, and finally a fault characterization vector is output, wherein the fault characterization vector contains a high-level feature representation of the input data. Based on the fault characterization vector, the pre-trained fault identification model determines the fault type corresponding to the target satellite data through the classification decision mechanism of the fully connected layer. This process involves mapping the fault characterization vector to a probability distribution, comparing the probability distribution values ​​with the set thresholds to obtain the output values ​​of the corresponding fault types, and combining the output values ​​of the fault types to obtain the fault characterization vector.

[0070] In some examples, the fault type with the maximum probability distribution value can also be identified as the target fault type corresponding to the satellite feature image. This implementation not only improves the accuracy and efficiency of fault diagnosis, but also reduces the reliance on a large amount of labeled data by using the pre-trained fault identification model, making the fault identification process more efficient and practical.

[0071] Based on the explanation of the above example, for example, obtaining a fault characterization vector corresponding to a satellite feature image according to a fused time series feature graph may include: determining at least one candidate fault type; determining a probability distribution value of target satellite data for each candidate fault type based on the fused time series feature graph; and obtaining a fault characterization vector according to the probability distribution value.

[0072] For the above example, one or more candidate fault types are first determined, and then the corresponding probability distribution value is calculated for each candidate fault type based on the one or more candidate fault types. Finally, the probability distribution values ​​are combined to form a fault characterization vector, thereby providing a comprehensive fault type probability description for the satellite data. The fault characterization vector summarizes the possibility of the satellite data belonging to each fault type.

[0073] For example, taking the momentum wheel as an example, the above method can be used to identify six types of momentum wheel faults (idling fault, stuck fault, excessive friction fault, abnormal noise fault and periodic oscillation fault) and normal conditions, and the following are obtained: Figure 6 As shown in the confusion matrix, the larger the value, the greater the predicted probability. It can be obtained that the maximum value in each row corresponds to the maximum value of the representative probability. The conclusion obtained is consistent with the true value, that is, the method disclosed in the present invention can accurately determine the fault type.

[0074] Reference Figure 7 and Figure 8 ,In the training process of the fault recognition model, as the number of ,training rounds increases, the loss function gradually converges and the ,accuracy gradually improves.

[0075] In some examples, the fault identification model may be a GASF-DenseNet-GRU model, a long short-term memory network model, a convolutional neural network model, a GRU network model, etc. See Table 1 for details: Table 1 Network results comparison

[0076] Referring to Table 1, the GASF-DenseNet-GRU fault recognition model was used to classify five types of faults in the satellite attitude control subsystem, achieving a recognition accuracy of 92%.

[0077] For the momentum wheel jamming fault and excessive friction fault, the fault signal classifier has the best fault identification effect.

[0078] In the present disclosure, the GRU network model is combined with the dense connection model to provide a more comprehensive data analysis. The entire long sequence data of the satellite is processed by the GRU network model to extract the global time series features in the target satellite data, namely the global statistical features, frequency domain features and time domain features, and then the dense connection model is used to fuse features of different scales and types, so that the model has better generalization ability and can adapt to different failure modes and environmental changes. In addition, the dense connection model reduces the number of parameters through feature reuse and improves the efficiency of data processing.

[0079] Based on the same inventive concept as the above technical solution, see Fig. 9, which shows a satellite attitude control system fault diagnosis device 900 based on time-frequency domain feature fusion provided by the present disclosure, the device 900 may include: a data conversion module 901, a feature extraction module 902, a time series extraction module 903, a feature fusion module 904 and a fault classification module 905; wherein, The data conversion module 901 may be used to obtain target satellite data and convert the target satellite data into satellite image data using the Gram angle field; The feature extraction module 902 may be used to extract features from the satellite image data to obtain a satellite feature image; The time series extraction module 903 can be used to extract the global time series features of the satellite feature image, wherein the global time series features refer to image sequence features extracted from the entire time series of the satellite and capable of representing the characteristics of the entire sequence; The feature fusion module 904 may be used to input the global temporal features into a dense connection model to obtain a fused temporal feature graph corresponding to the satellite feature image; The fault classification module 905 may be configured to obtain a target fault type corresponding to the target satellite data according to the fused time series feature graph.

[0080] In some examples, the feature fusion module 904 is configured to: Input the global temporal features into the first-layer feature fusion layer to obtain the first-layer fusion temporal feature map; The output of the feature fusion layer of the Nth layer is used as the input of the N+1th feature fusion layer to obtain the fused temporal feature map of the N+1th layer, where N is greater than 1 and less than M-1; The fused temporal feature map of the Mth layer is used as the target fused temporal feature map.

[0081] In some examples, the feature fusion module 904 is configured to: The global temporal features are convolved using multi-layer convolution modules to obtain multi-layer intermediate features; wherein the input of the P-th layer convolution module is the output of the P-1-th layer convolution module, and P is greater than or equal to 2; At least two of the multiple layers of intermediate features are fused to obtain a first layer of fused temporal feature graph.

[0082] In some examples, the feature fusion module 904 is configured to: The global time series feature and at least one of the multiple intermediate features are fused to obtain a fused time series feature map corresponding to the satellite feature image.

[0083] In some examples, the timing extraction module 903 is configured to: Get initial satellite data; The initial satellite data is preprocessed by a Kalman filter to obtain target satellite data, wherein the preprocessing at least includes normalizing, denoising and segmenting the initial satellite data to obtain the target satellite data.

[0084] In some examples, the timing extraction module 903 is configured to: The GRU network model is used to extract the features of the target satellite data to obtain the global time series features; The GRU network model includes at least one feature extraction layer, at least one data processing layer, and a fully connected layer for output connected to the data processing layer. The input and output of the feature extraction layer are superimposed and input into the data processing layer; The input data and output data of the data processing layer are superimposed and input into the next feature extraction layer or fully connected layer to obtain the global time series features corresponding to the target satellite data.

[0085] In some examples, the fault classification module 905 is configured to: Obtain the fault characterization vector corresponding to the target satellite data according to the fused time series feature graph; The target fault type corresponding to the target satellite data is determined according to the fault characterization vector.

[0086] It can be understood that the exemplary technical solution of the satellite attitude control system fault diagnosis device 900 based on time-frequency domain feature fusion belongs to the same concept as the technical solution of the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion. Therefore, the details not described in detail in the technical solution of the satellite attitude control system fault diagnosis device 900 based on time-frequency domain feature fusion can be referred to the description of the technical solution of the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion. This disclosure will not elaborate on this.

[0087] Please refer to Fig.10 , which shows a schematic diagram of the hardware structure of a computing device provided by an exemplary embodiment of the present disclosure. In some examples, the computing device may be at least one of a smart phone, a smart watch, a desktop computer, a laptop, a virtual reality terminal, an augmented reality terminal, a wireless terminal, and a laptop portable computer. The computing device has a communication function and can access a wired network or a wireless network. The computing device may generally refer to one of a plurality of terminals, and those skilled in the art may know that the number of the above terminals may be more or less. In some examples, the computing device may receive data of a satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion based on the wired network or wireless network to which it is connected. It can be understood that the computing device undertakes the calculation and processing work of the technical solution of the present disclosure, and the present disclosure does not limit this.

[0088] like Fig.10 As shown, the computing device in the present disclosure may include one or more of the following components: a processor 1010 and a memory 1020 .

[0089] Optionally, the processor 1010 uses various interfaces and lines to connect various parts within the entire computing device, and executes various functions of the computing device and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 1020, and calling data stored in the memory 1020. Optionally, the processor 1010 can be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor 1010 can integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), a neural network processor (NPU), and a baseband chip. Among them, the CPU mainly processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content that needs to be displayed on the touch display; the NPU is used to implement artificial intelligence (AI) functions; and the baseband chip is used to process wireless communications. It is understandable that the above-mentioned baseband chip may not be integrated into the processor 1010, but may be implemented by a separate chip.

[0090] The memory 1020 may include a random access memory (RAM) or a read-only memory (ROM). Optionally, the memory 1020 includes a non-transitory computer-readable storage medium. The memory 1020 may be used to store instructions, programs, codes, code sets, or instruction sets. The memory 1020 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above various method embodiments, etc.; the data storage area may store data created according to the use of the computing device, etc.

[0091] In addition, those skilled in the art can understand that the structure of the computing device shown in the above drawings does not constitute a limitation on the computing device, and the computing device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, the computing device also includes a display screen, a camera assembly, a microphone, a speaker, a radio frequency circuit, an input unit, a sensor (such as an acceleration sensor, an angular velocity sensor, an optical fiber sensor, etc.), an audio circuit, a WiFi module, a power supply, a Bluetooth module and other components, which will not be described in detail here.

[0092] The present disclosure also provides a computer-readable storage medium storing at least one instruction, wherein the at least one instruction is used to be executed by a processor to implement the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion as described in the above embodiments.

[0093] The present disclosure also provides a computer program product, which includes computer instructions, which are stored in a computer-readable storage medium; a processor of a computing device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computing device executes to implement the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion of the above-mentioned embodiments.

[0094] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the present disclosure can be implemented with hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer-readable storage media and communication media, wherein the communication media includes any medium that is convenient for transmitting a computer program from one place to another. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0095] The above are only specific implementation methods of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present disclosure, which should be covered by the protection scope of the present disclosure.

Claims

1. A satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion, characterized in that: The method comprises: Acquire target satellite data, and convert the target satellite data into satellite image data using the Gram angle field; Extracting features from the satellite image data to obtain a satellite feature image; Extracting global time series features of satellite feature images, wherein the global time series features refer to image sequence features extracted from the entire time series of the satellite and capable of representing the characteristics of the entire sequence; Inputting the global time series feature into a dense connection model to obtain a fused time series feature graph corresponding to the satellite feature image; The target fault type corresponding to the target satellite data is obtained according to the fused time series feature diagram.

2. The method according to claim 1, characterized in that The dense connection model includes M layers of feature fusion layers connected in sequence; the inputting the global time series feature into the dense connection model to obtain the fused time series feature graph corresponding to the satellite feature image includes: Inputting the global temporal features into the first feature fusion layer to obtain a first fusion temporal feature graph; The output of the feature fusion layer of the Nth layer is used as the input of the N+1th feature fusion layer to obtain the fused temporal feature map of the N+1th layer, where N is greater than 1 and less than M-1; The fused temporal feature map of the Mth layer is used as the target fused temporal feature map.

3. The method according to claim 2, characterized in that The feature fusion layer includes a multi-layer convolution module; the step of inputting the global temporal features into the first-layer feature fusion layer to obtain a first-layer fusion temporal feature graph includes: Using the multi-layer convolution module to perform convolution processing on the global temporal features to obtain multi-layer intermediate features; wherein the input of the P-th layer convolution module is the output of the P-1-th layer convolution module, and P is greater than or equal to 2; At least two of the multiple layers of intermediate features are fused to obtain the first layer of fused temporal feature graph.

4. The method according to claim 3, characterized in that The step of inputting the global temporal features into a densely connected model to obtain a fused temporal feature graph corresponding to the satellite feature image comprises: The global time series feature and at least one of the multiple intermediate features are fused to obtain a fused time series feature map corresponding to the satellite feature image.

5. The method according to claim 1, characterized in that The method further comprises: Get initial satellite data; The initial satellite data is preprocessed by a Kalman filter to obtain the target satellite data, wherein the preprocessing at least includes normalizing, denoising and segmenting the initial satellite data to obtain the target satellite data.

6. The method according to claim 1, characterized in that The step of extracting the global temporal features of the satellite feature image comprises: The satellite feature image is extracted using a GRU network model to obtain global temporal features; wherein, The GRU network model includes at least one feature extraction layer, at least one data processing layer, and a fully connected layer for output connected to the data processing layer. The input and output of the feature extraction layer are superimposed and input into the data processing layer; The input data and output data of the data processing layer are superimposed and input into the next feature extraction layer or the fully connected layer to obtain the global time series features corresponding to the target satellite data.

7. The method according to claim 1, characterized in that The acquiring the target fault type corresponding to the target satellite data according to the fused time series characteristic graph includes: Acquire a fault characterization vector corresponding to the target satellite data according to the fused time series feature graph; A target fault type corresponding to the target satellite data is determined according to the fault characterization vector.

8. A satellite attitude control system fault diagnosis device based on time-frequency domain feature fusion, characterized in that: The device comprises: A data conversion module, used for acquiring target satellite data and converting the target satellite data into satellite image data using the Gram angle field; A feature extraction module, used for extracting features from the satellite image data to obtain a satellite feature image; A time series extraction module is used to extract the global time series features of the satellite feature image, wherein the global time series features refer to image sequence features extracted from the entire time series of the satellite and capable of representing the characteristics of the entire sequence; A feature fusion module, used for inputting the global time series feature into a dense connection model to obtain a fused time series feature map corresponding to the satellite feature image; A fault classification module is used to obtain a target fault type corresponding to the target satellite data according to the fused time series feature diagram.

9. A computing device, characterized in that The computing device comprises: a processor and a memory; wherein, The memory is used to store a computer program that can be run on the processor; The processor is used to execute the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion as described in any one of claims 1 to 7 when running the computer program.

10. A computer-readable storage medium, characterized in that: The readable storage medium stores at least one instruction, and the at least one instruction is used to be executed by a processor to implement the satellite attitude control system fault diagnosis method based on time-frequency domain feature fusion as described in any one of claims 1 to 7.

Images