A spacecraft attitude control system fault diagnosis method based on time-frequency coding

Abstract

This invention discloses a fault diagnosis method for a spacecraft attitude control system based on time-frequency coding, comprising the following steps: Step S1: Acquire multi-channel attitude measurement signals of the spacecraft attitude control system within a preset time window, and divide the multi-channel attitude measurement signals into attitude angle channel groups and angular velocity channel groups according to their physical meaning; Step S2: Perform synchronous compressed wavelet time-frequency transformation on each channel signal in the attitude angle channel group and angular velocity channel group; Step S3: Generate RGB time-frequency sub-images of attitude angle and RGB time-frequency sub-images of angular velocity; Step S4: Stitch the RGB time-frequency sub-images of attitude angle and RGB time-frequency sub-images of angular velocity according to a preset spatial layout to obtain a joint time-frequency diagnostic image; Step S5: Output the fault diagnosis result of the spacecraft attitude control system. This invention can generate a joint time-frequency diagnostic image and utilize a visual state space fault diagnosis network to achieve efficient, fine-grained fault diagnosis and axial positioning.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for spacecraft attitude control systems, and specifically to a fault diagnosis method for spacecraft attitude control systems based on time-frequency coding. Background Technology

[0002] As spacecraft missions become increasingly complex and on-orbit operations lengthen, the attitude control system, as a core subsystem ensuring spacecraft attitude stability, precise pointing, and mission execution capabilities, is directly related to spacecraft operational safety. If an attitude sensor, gyroscope, or related measurement channel malfunctions, the impact can propagate through the closed-loop control process to multiple state variables, causing abnormal changes in multi-channel signals such as attitude angle and angular velocity. In severe cases, this can affect spacecraft attitude stability and mission performance.

[0003] Currently, fault diagnosis methods for spacecraft attitude control systems mainly include model-based methods and data-driven methods. Model-based methods typically rely on attitude dynamics models, observers, residual generation, and threshold discrimination techniques. However, these methods are sensitive to model accuracy, disturbance assumptions, and threshold settings, and are easily affected by measurement noise, external disturbances, and parameter uncertainties in complex on-orbit environments. Data-driven methods can learn fault characteristics from historical data, reducing reliance on accurate models. However, spacecraft attitude control system signals typically have characteristics such as multi-channel, strong coupling, non-stationarity, and weak fault characteristics. Directly using one-dimensional time-series signals or processing different channels independently makes it difficult to fully characterize the time-frequency characteristics of faults and inter-axis coupling relationships.

[0004] To enhance the feature representation of non-stationary signals, existing methods often convert the original time-series signal into a time-frequency image before inputting it into a deep learning model for diagnosis. However, methods such as the short-time Fourier transform are limited by a fixed window, making it difficult to simultaneously capture both low-frequency slow-varying features and local abrupt changes. While the ordinary wavelet transform offers multi-resolution analysis capabilities, the time-frequency energy distribution may exhibit diffusion. Furthermore, existing multi-channel time-frequency image construction methods often employ channel-by-channel normalization or simple stacking, which easily weakens the relative energy relationships between different axes. Particularly for spacecraft attitude control systems, the three-axis signals of the same physical quantity not only reflect local anomalies on each axis but also include relative energy changes generated by the coupling of closed-loop control and attitude dynamics. If channel-by-channel normalization is used, the time-frequency energy maps of each axis will be stretched to similar dynamic ranges, thus weakening the amplitude differences between faulty and non-faulty axes. If simple channel stacking is used, it is difficult to reflect the physical grouping relationship between attitude angle signals and angular velocity signals. Therefore, there is an urgent need for a fault diagnosis method for spacecraft attitude control systems that can simultaneously enhance the clustering of non-stationary fault features, maintain multi-axis relative energy relationships, and be applicable to visual diagnostic networks. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the present invention aims to provide a fault diagnosis method for spacecraft attitude control systems based on time-frequency coding. This method enhances the time-frequency aggregation of non-stationary fault features through synchronous compressed wavelet time-frequency transform, and maintains the relative energy relationship between different axes based on physical meaning grouping and three-axis shared normalization within the channel. This generates a joint time-frequency diagnostic image and utilizes a visual state space fault diagnosis network to achieve efficient and fine-grained fault diagnosis and axis positioning.

[0006] To achieve the objective of this invention, the following solution is adopted: A fault diagnosis method for a spacecraft attitude control system based on time-frequency coding includes the following steps: Step S1: Acquire the multi-channel attitude measurement signals of the spacecraft attitude control system within a preset time window, and divide the multi-channel attitude measurement signals into attitude angle channel group and angular velocity channel group according to their physical meaning; Step S2: Perform synchronous compressed wavelet time-frequency transformation on the signals of each channel in the attitude angle channel group and the angular velocity channel group to obtain the synchronous compressed wavelet time-frequency energy map corresponding to each channel; Step S3: For the attitude angle channel group and the angular velocity channel group respectively, the three-axis synchronous compressed wavelet time-frequency energy maps in the same channel group are sequentially subjected to energy compression, shared normalization and RGB mapping to generate attitude angle RGB time-frequency sub-maps and angular velocity RGB time-frequency sub-maps; wherein, the parameters used for the shared normalization are determined by the merged pixel distribution of the three-axis compressed synchronous compressed wavelet time-frequency energy maps in the corresponding channel group; Step S4: The attitude angle RGB time-frequency sub-image and the angular velocity RGB time-frequency sub-image are stitched together according to a preset spatial layout to obtain a joint time-frequency diagnostic image; Step S5: Input the joint time-frequency diagnostic image into the visual state space fault diagnosis network and output the fault diagnosis results of the spacecraft attitude control system; wherein, the fault diagnosis results include fault state, fault sensor type and fault mode; when the fault involves the axial channel, the fault diagnosis results also include the fault axial position.

[0007] Furthermore, the multi-channel attitude measurement signal includes a three-axis attitude angle signal and a three-axis angular velocity signal; The three-axis attitude angle signals include roll angle, pitch angle, and yaw angle; The three-axis angular velocity signals include the spacecraft's own X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity.

[0008] Further, in step S1, the preset time window is either a fixed-length sampling window or a sliding sampling window; when a fixed-length sampling window is used, each sample consists of a fixed number of continuous sampling points; when a sliding sampling window is used, multi-channel attitude measurement signals are continuously intercepted according to a preset step size to achieve online fault diagnosis of the spacecraft attitude control system. Further, the attitude angle channel group consists of roll angle, pitch angle, and yaw angle, used to characterize changes in the spacecraft's attitude state. The angular velocity channel group consists of X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity, and is used to characterize the changes in the attitude motion of the spacecraft. The attitude angle channel group and the angular velocity channel group can maintain the inter-axis coupling relationship between the three-axis signals with the same physical meaning in the subsequent time-frequency coding process.

[0009] Furthermore, prior to step S2, the signal from each channel is standardized using Z-score standardization, the expression of which is: in, Represents the original channel signal. This represents the average value of the channel signal within a preset time window. This represents the standard deviation of the channel signal within a preset time window. This represents the standardized channel signal.

[0010] Further, in step S2, the scale set of the synchronous compressed wavelet time-frequency transform is determined based on the sampling frequency, the lowest frequency of interest, the highest frequency of interest, and the wavelet center frequency; the scale set is expressed as: in, Represents a scale set, Indicates the first One scale, Indicates the wavelet center frequency. Indicates the sampling time interval. and These represent the lowest and highest frequencies of attention, respectively. Indicates the number of scales.

[0011] Further, in step S2, the synchronous compressed wavelet time-frequency transform includes continuous wavelet transform, instantaneous frequency estimation, and frequency reallocation; wherein, the continuous wavelet transform is used to convert the single-channel attitude measurement signal into wavelet coefficients on the scale-time plane, and the continuous wavelet transform is expressed as: in, This represents the single-channel signal to be transformed. Indicates the scale parameter. Indicates the time shift parameter. This represents the conjugate function of the wavelet basis functions. This represents the continuous wavelet transform coefficients.

[0012] Furthermore, the continuous wavelet transform is performed using complex-valued wavelet basis functions; the complex-valued wavelet basis functions include any one of Morlet wavelet, complex Morlet wavelet, or generalized Morse wavelet; using complex-valued wavelet basis functions makes it easier to estimate the instantaneous frequency of the signal based on the phase changes of the wavelet coefficients.

[0013] Furthermore, the instantaneous frequency is estimated based on the phase change of the continuous wavelet transform coefficients, and the instantaneous frequency is expressed as: in, Represents the continuous wavelet transform coefficients relative to the time shift parameter. The partial derivatives, This indicates the operation of taking the imaginary part. This represents the estimated instantaneous frequency.

[0014] Furthermore, the continuous wavelet transform coefficients are synchronously compressed and redistributed in the frequency direction according to the instantaneous frequency to obtain synchronously compressed wavelet coefficients, which are expressed as follows: in, This represents the synchronous compressed wavelet coefficients. Represents frequency variables. This represents the set of scales that satisfy the preset amplitude conditions. Represents the redistribution function; In step S2, the time-frequency energy map of the synchronous compressed wavelet is obtained from the modulus or the square of the modulus of the synchronous compressed wavelet coefficients. The time-frequency energy map of the synchronous compressed wavelet is expressed as follows: in, This represents the time-frequency energy diagram of the synchronous compressed wavelet.

[0015] Furthermore, in step S3, the energy compression employs logarithmic compression, expressed as: in, This represents the energy value in the synchronous compressed wavelet time-frequency energy map. This represents the compressed time-frequency energy value. This represents the energy scaling factor.

[0016] Further, in step S3, the shared normalization includes: The pixel values ​​of the synchronous compressed wavelet time-frequency energy maps after triaxial compression within the same channel group are merged and statistically analyzed to obtain shared low percentile values ​​and shared high percentile values. The shared low percentile and shared high percentile values ​​are used to uniformly normalize the three-axis synchronous compressed wavelet time-frequency energy map within the same channel group; The unified normalization is expressed as follows: in, This represents the pixel value in the compressed synchronous wavelet time-frequency energy map. This indicates that the lower percentile value is shared. This indicates that the high percentile value is shared. This represents the pixel value after uniform normalization. This represents a truncation function that restricts pixel values ​​to a preset range. Further, in step S3, the shared normalization uses any value from the 1st to the 5th percentile for the shared low percentile and any value from the 95th to the 99th percentile for the shared high percentile.

[0017] Preferably, the shared lower percentile value is the 2nd percentile value, and the shared higher percentile value is the 98th percentile value.

[0018] Furthermore, in step S3, for the attitude angle channel group, the red channel, green channel, and blue channel correspond to the roll angle, pitch angle, and yaw angle, respectively; For the angular velocity channel group, the red channel, green channel, and blue channel correspond to the X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity, respectively.

[0019] Through the RGB mapping method described above, the synchronous compressed wavelet time-frequency energy maps of the three axes within the same physical channel group are organized in the same RGB image, enabling the subsequent visual diagnostic network to learn the coupling relationship between the three axes in a unified visual space.

[0020] Furthermore, in step S4, the attitude angle RGB time-frequency sub-map and the angular velocity RGB time-frequency sub-map are combined into a joint time-frequency diagnostic image by horizontal splicing, vertical splicing, or a preset block region arrangement. The attitude angle RGB time-frequency sub-graph is used to characterize the abnormal characteristics of the spacecraft's attitude state. The angular velocity RGB time-frequency sub-graph is used to characterize the abnormal characteristics of spacecraft attitude motion.

[0021] Furthermore, in step S5, the visual state space fault diagnosis network includes an image patch embedding module, a position encoding module, a bidirectional state space feature extraction module, and a classification output module; The image block embedding module is used to divide the joint time-frequency diagnostic image into an image block sequence; The location encoding module is used to add spatial location information to the image block sequence; The bidirectional state space feature extraction module is built based on a selective state space model and is used to model long-range dependencies in image patch sequences. The bidirectional state space feature extraction module includes a forward state space modeling branch, a reverse state space modeling branch, and a feature fusion unit. The forward state space modeling branch and the reverse state space modeling branch perform state space modeling in opposite directions on the image patch sequence, respectively. The feature fusion unit is used to fuse bidirectional modeling features. The classification output module is used to output fault diagnosis results.

[0022] Furthermore, the visual state space fault diagnosis network adopts a visual backbone network based on a selective state space model.

[0023] Preferably, the visual state space fault diagnosis network adopts the Vision Mamba network.

[0024] Further, in step S5), the fault diagnosis result includes a normal state and a fault state; when the diagnosis result is a fault state, the fault sensor type and fault mode are further output; when the fault category corresponds to a specific axial channel, the fault axial position is further output. The fault sensor type includes a star sensor and a gyroscope; the fault mode includes a jamming fault and a random noise fault; the fault axial position includes at least one of the X-axis, Y-axis, and Z-axis.

[0025] Furthermore, the visual state space fault diagnosis network outputs probability values ​​corresponding to multiple fault categories; when the maximum probability value is greater than or equal to a preset confidence threshold, it outputs the fault diagnosis result corresponding to the maximum probability value; when the maximum probability value is less than the preset confidence threshold, it outputs an abnormal state to be confirmed.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention performs time-frequency analysis on each channel signal by employing synchronous compressed wavelet transform, and introduces instantaneous frequency estimation and frequency redistribution operations on the basis of continuous wavelet transform, so as to refocus the diffused time-frequency energy back to the true frequency position, significantly improving the time-frequency resolution of weak fault features and non-stationary signal components, thereby enhancing the time-frequency aggregation of non-stationary fault features.

[0027] 2. This invention processes the attitude angle channel group and the angular velocity channel group independently, and adopts a shared normalization strategy within each channel group. That is, it uses the merged pixel distribution of the three-axis compressed time-frequency energy map of the group to determine the same set of normalization parameters, and performs unified normalization on the three axes. This avoids the drawback of traditional channel-by-channel normalization, which weakens the amplitude difference between faulty axes and non-faulty axes. It effectively preserves the relative energy relationship between axes, which is more conducive to identifying specific faulty axes and multi-axis coupling anomalies.

[0028] 3. This invention divides multi-channel signals into attitude angle channel groups and angular velocity channel groups according to their physical meaning, generates corresponding RGB time-frequency sub-images, and then stitches them together to form a joint time-frequency diagnostic image. This avoids the simple mixing of signals with different physical meanings and enables the visual diagnostic network to utilize both attitude state abnormality features and attitude motion abnormality features simultaneously, thus conforming to the dynamic coupling mechanism of the spacecraft attitude control system.

[0029] 4. This invention uses a visual state-space fault diagnosis network to replace the traditional Transformer model. This network is built based on a selective state-space model, which can achieve linear computational complexity while maintaining global dependency modeling capabilities. It is suitable for processing high-resolution joint time-frequency diagnostic images, thereby reducing computational overhead and meeting the real-time requirements of on-orbit fault diagnosis for spacecraft.

[0030] 5. This invention can output fine-grained fault diagnosis results, including not only the fault state, but also the fault sensor type and fault mode. Furthermore, when the fault involves a specific axial channel, it can output the fault axial position, thereby providing clear and direct information support for fault isolation, fault-tolerant control and on-orbit health management of spacecraft attitude control systems. Attached Figure Description

[0031] Figure 1 This is a flowchart of a spacecraft attitude control system fault diagnosis method based on time-frequency coding in an embodiment of the present invention; Figure 2 This is a schematic diagram of a fault diagnosis method for a spacecraft attitude control system based on time-frequency coding, as described in an embodiment of the present invention. Figure 3 This is a schematic diagram of the synchronous compressed wavelet time-frequency energy map generation process in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the shared normalization, RGB mapping, and joint time-frequency diagnostic image generation of the attitude angle channel group and angular velocity channel group in an embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of the visual state space fault diagnosis network in an embodiment of the present invention. Detailed Implementation

[0032] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments. It should be noted that, without conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

[0033] like Figure 1-5 As shown, this embodiment of the invention provides a fault diagnosis method for a spacecraft attitude control system based on time-frequency coding, including the following steps: Step S1: Acquire the multi-channel attitude measurement signals of the spacecraft attitude control system within a preset time window, and divide the multi-channel attitude measurement signals into attitude angle channel group and angular velocity channel group according to their physical meaning.

[0034] Step S2: Perform synchronous compressed wavelet time-frequency transformation on the signals of each channel in the attitude angle channel group and the angular velocity channel group to obtain the synchronous compressed wavelet time-frequency energy map corresponding to each channel.

[0035] Step S3: For the attitude angle channel group and the angular velocity channel group respectively, the three-axis synchronous compressed wavelet time-frequency energy maps in the same channel group are sequentially subjected to energy compression, shared normalization and RGB mapping to generate attitude angle RGB time-frequency sub-maps and angular velocity RGB time-frequency sub-maps; wherein, the parameters used for shared normalization are determined by the merged pixel distribution of the three-axis compressed synchronous compressed wavelet time-frequency energy maps in the corresponding channel group.

[0036] Step S4: The attitude angle RGB time-frequency sub-image and the angular velocity RGB time-frequency sub-image are stitched together according to a preset spatial layout to obtain a joint time-frequency diagnostic image.

[0037] Step S5: Input the joint time-frequency diagnostic image into the visual state space fault diagnosis network and output the fault diagnosis results of the spacecraft attitude control system; wherein, the fault diagnosis results include fault state, fault sensor type and fault mode; when the fault involves the axial channel, the fault diagnosis results also include the fault axial position.

[0038] The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to the embodiments of the present invention will be further described in detail below.

[0039] To address the problems of insufficient time-frequency clustering of non-stationary fault features, difficulty in maintaining physical coupling due to simple stacking of multi-axis signals, and distortion of inter-axis relative energy relationships caused by channel-by-channel normalization in existing methods, the fault diagnosis method for spacecraft attitude control systems based on time-frequency coding in this invention first generates a time-frequency energy map with strong energy clustering using synchronous compressed wavelet transform. Then, channel groups are constructed according to the physical meaning of attitude angles and angular velocities, and unified normalization is performed within each channel group using shared normalization parameters for three axes, so that the relative energy difference between faulty axes and non-faulty axes is preserved. Subsequently, the three-axis time-frequency maps within the group are mapped to RGB time-frequency sub-maps and stitched together to form a joint time-frequency diagnostic image, which is finally input into a visual state space fault diagnosis network for classification.

[0040] In this embodiment, multi-channel attitude measurement signals are acquired and physically grouped, as detailed below: The spacecraft attitude control system consists of attitude sensors, gyroscopes, controllers, actuators, and attitude dynamics components. During spacecraft operation, attitude sensors are used to obtain the spacecraft's attitude angle information, and gyroscopes are used to obtain the spacecraft's angular velocity information.

[0041] The multi-channel attitude measurement signal acquired in this embodiment includes six channels, namely: in, Indicates the roll angle. Indicates pitch angle, Indicates the yaw angle. , , These represent the angular velocities along the X, Y, and Z axes of the spacecraft's main system, respectively.

[0042] In one optional implementation, the sampling frequency of the multi-channel attitude measurement signal is: The preset time window length is Then the number of sampling points contained in each channel within each time window is: For example, when the sampling frequency is The time window length is At that time, each channel contains There are 100 sampling points. At this point, each sample can be represented as: in, These correspond to roll angle, pitch angle, and yaw angle, respectively. These correspond to the angular velocities of the X, Y, and Z axes, respectively.

[0043] Since attitude angles and angular velocities in a spacecraft attitude control system have different physical meanings, directly mixing the six channels could weaken the structural relationships between these different physical quantities. Therefore, this embodiment divides the six-channel attitude measurement signals into two channel groups according to their physical meanings: The first group is the attitude angle channel group: The second group is the angular velocity channel group: Among them, the attitude angle channel group is used to characterize the changes in the spacecraft's attitude state, and the angular velocity channel group is used to characterize the changes in the spacecraft's attitude motion.

[0044] By using the above grouping method, the three axial signals with the same physical meaning can be processed together in the subsequent time-frequency image construction process, thereby preserving the coupling relationship and relative change characteristics between the three axes.

[0045] Before performing synchronous compressed wavelet time-frequency transform, this embodiment preferably performs standardization processing on each channel signal to reduce the impact of different dimensions and amplitude ranges on subsequent time-frequency analysis. The standardization processing adopts the Z-score standardization method, the expression of which is: In the formula, Indicates the first The original signal of each channel, This represents the average value of the channel signal within the current time window. This represents the standard deviation of the channel signal within the current time window. This represents the standardized channel signal.

[0046] In this embodiment, synchronous compressed wavelet time-frequency transform is performed on the signals of each channel to obtain the synchronous compressed wavelet time-frequency energy map, the details of which are as follows: like Figure 3 As shown, this embodiment employs synchronous compressed wavelet transform for time-frequency analysis of each channel signal. Synchronous compressed wavelet transform is a time-frequency analysis method that introduces instantaneous frequency estimation and frequency reallocation operations on the basis of continuous wavelet transform, thereby improving the time-frequency energy concentration.

[0047] For any standardized single-channel signal First, the scale set for synchronous compressed wavelet transform is determined. This scale set is determined based on the sampling frequency, the lowest frequency of interest, the highest frequency of interest, and the wavelet center frequency, and is expressed as: In the formula, Represents a scale set, Indicates the first One scale, Indicates the wavelet center frequency. Indicates the sampling time interval. Indicates the minimum frequency of attention. Indicates the highest frequency of attention. Indicates the number of scales.

[0048] in, and The frequency bands can be determined based on the spacecraft's attitude motion frequency bands and fault-sensitive frequency bands. For the attitude angle channel group, the focus should be on the low-frequency slow change components of attitude and the local abrupt changes caused by faults; for the angular velocity channel group, the focus should be on the attitude motion change frequency bands and the frequency change regions caused by gyroscope anomalies.

[0049] Then, for single-channel signals Perform a continuous wavelet transform to obtain the wavelet coefficients: In the formula, Indicates the scale parameter. Indicates the time shift parameter. This represents the conjugate function of the wavelet basis functions. This represents the continuous wavelet transform coefficients.

[0050] In a preferred embodiment, the wavelet basis function is a complex-valued wavelet basis function, such as a Morlet wavelet, a complex Morlet wavelet, or a generalized Morse wavelet. The purpose of using a complex-valued wavelet basis function is to facilitate the estimation of the instantaneous frequency based on the phase changes of the wavelet coefficients.

[0051] Furthermore, the instantaneous frequency is estimated based on the phase variation of the continuous wavelet transform coefficients: In the formula, Represents the wavelet coefficients relative to the time shift parameter The partial derivatives, This indicates the operation of taking the imaginary part. This represents the estimated instantaneous frequency.

[0052] Subsequently, the continuous wavelet transform coefficients are synchronously compressed and redistributed in the frequency direction according to the instantaneous frequency to obtain the synchronously compressed wavelet coefficients: In the formula, This represents the synchronous compressed wavelet coefficients. Represents frequency variables. This represents the set of scales that satisfy the preset amplitude conditions. This represents the redistribution function.

[0053] Through the above synchronous compression and redistribution operation, the energy dispersed in the scale direction in the continuous wavelet transform can be reconcentrated to the corresponding frequency position, thereby obtaining a time-frequency representation with stronger energy concentration.

[0054] Finally, the time-frequency energy map of the synchronous compressed wavelet is calculated based on the synchronous compressed wavelet coefficients: In the formula, This represents the time-frequency energy diagram of the synchronous compressed wavelet.

[0055] By performing the above process on each of the six channels, six synchronous compressed wavelet time-frequency energy maps can be obtained: in, Corresponding attitude angle channel group, Corresponding angular velocity channel group.

[0056] In this embodiment, energy compression, shared normalization, and RGB encoding are performed on the three-axis synchronous compressed wavelet time-frequency energy maps within the same channel group. The specific details are as follows: Since the dynamic range of energy values ​​at different locations in the synchronous compressed wavelet time-frequency energy map can be quite large, direct image encoding can easily lead to weak fault features being masked by high-energy regions. Therefore, this embodiment first performs energy compression on the synchronous compressed wavelet time-frequency energy map.

[0057] In a preferred embodiment, the energy compression employs logarithmic compression: In the formula, This represents the energy value in the synchronous compressed wavelet time-frequency energy map. This represents the compressed time-frequency energy value. This represents the energy scaling factor.

[0058] After energy compression is completed, such as Figure 4 As shown, this embodiment performs shared normalization processing on the three-axis time-frequency energy maps within the same channel group, and further maps them into RGB time-frequency sub-maps.

[0059] Taking the attitude angle channel group as an example, , , All pixel values ​​in the three images are merged into a single pixel set: Calculate the shared low percentile value based on the pixel set. and sharing high percentile values .

[0060] Similarly, for the angular velocity channel group, , , All pixel values ​​in the three images are merged into: And calculate the shared lower percentile value and sharing high percentile values .

[0061] In a preferred embodiment, the shared low percentile value is the 2nd percentile value, and the shared high percentile value is the 98th percentile value. Alternatively, the low percentile value can be set to any value from the 1st to the 5th percentile values, and the high percentile value can be set to any value from the 95th to the 99th percentile values, depending on the actual data distribution.

[0062] For any compressed synchronous wavelet time-frequency energy map within the same channel group, its unified normalization formula is: In the formula, This represents the pixel value in the compressed synchronous wavelet time-frequency energy map. This indicates the shared lower percentile value corresponding to this channel group. This indicates the shared high percentile value corresponding to this channel group. This represents the pixel value after uniform normalization. This indicates that the pixel value is limited to A cutoff function within a specified range.

[0063] Through the shared normalization process described above, triaxial signals within the same channel group use the same set of normalization parameters, instead of being normalized independently. This avoids the disruption of the relative energy relationships between the three axes caused by channel-by-channel normalization, which is beneficial for subsequent diagnostic networks to identify specific axial anomalies or multiaxial coupling anomalies.

[0064] After completing the shared normalization, the uniformly normalized three-axis time-frequency energy maps within the attitude angle channel group are mapped to the three RGB color channels respectively, generating attitude angle RGB time-frequency sub-maps: In the formula, This represents the RGB time-frequency subplot of the attitude angle. , , These represent the unified normalized time-frequency energy maps corresponding to roll angle, pitch angle, and yaw angle, respectively. , , These represent the red channel, green channel, and blue channel, respectively.

[0065] For the angular velocity channel group: In the formula, Represents the RGB time-frequency subgraph of angular velocity. , , These represent the unified normalized time-frequency energy diagrams corresponding to the angular velocities along the X, Y, and Z axes, respectively.

[0066] Through the RGB mapping method described above, the time-frequency features of the three axes within the same physical channel group are organized into the same RGB image, enabling the subsequent visual diagnostic network to learn the coupling relationship and relative energy difference between the three axes in a unified visual space.

[0067] In this embodiment, the attitude angle RGB time-frequency sub-image and the angular velocity RGB time-frequency sub-image are stitched together to form a joint time-frequency diagnostic image, as detailed below: like Figure 4 As shown, this embodiment uses the attitude angle RGB time-frequency sub-graph. Angular velocity RGB time-frequency subgraph The images are stitched together according to a preset spatial layout to obtain a joint time-frequency diagnostic image. .

[0068] In a preferred embodiment, a horizontal splicing method is used: In another implementation, a vertical splicing method can also be used: Alternatively, a preset block area arrangement can be used.

[0069] The attitude angle RGB time-frequency sub-image is used to characterize abnormal features of the spacecraft's attitude state, while the angular velocity RGB time-frequency sub-image is used to characterize abnormal features of the spacecraft's attitude motion. By stitching the two together into a joint time-frequency diagnostic image, the diagnostic network can simultaneously utilize fault information from both attitude state changes and attitude motion changes.

[0070] Before inputting the diagnostic network, the joint time-frequency diagnostic image can be adjusted to a preset size, for example... , Or other sizes suitable for visual diagnostic network inputs.

[0071] In this embodiment, the joint time-frequency diagnostic image is input into the visual state-space fault diagnosis network, and the fault diagnosis result is output, as detailed below: like Figure 5 As shown, this embodiment uses a visual state space fault diagnosis network to extract features and classify joint time-frequency diagnostic images.

[0072] The visual state space fault diagnosis network includes an image block embedding module, a position encoding module, a bidirectional state space feature extraction module, and a classification output module.

[0073] First, the image patch embedding module will combine time-frequency diagnostic images. The image is divided into several blocks, and each block is mapped to a feature vector to obtain a sequence of image blocks: in, Indicates the number of image patches. Indicates the first The feature vector corresponding to each image patch.

[0074] Then, the location encoding module adds spatial location information to the image patch sequence to preserve the two-dimensional spatial structure in the joint time-frequency diagnostic image: In the formula, Indicates position code, This represents the sequence of image blocks after location coding has been added.

[0075] Subsequently, the bidirectional state space feature extraction module performs forward and reverse state space modeling on the image patch sequence to extract global dependencies and cross-regional correlation features in the joint time-frequency diagnostic image.

[0076] Finally, the classification output module outputs the probability of the fault category based on the extracted features: In the formula, Indicates the first Output values ​​corresponding to each fault category Indicates the number of fault categories. Indicates the first The predicted probability corresponding to each fault category.

[0077] The final diagnosis was: in, This indicates the predicted fault category.

[0078] During the training phase, a training sample set is constructed using joint time-frequency diagnostic images labeled with fault categories. The training sample set includes multiple training samples, each consisting of a joint time-frequency diagnostic image and its corresponding fault category label. The fault category label represents the normal state of the spacecraft's attitude control system, the type of faulty sensor, the fault mode, and the fault axial position when the fault category corresponds to a specific axial channel.

[0079] The joint time-frequency diagnostic images from the training sample set are input into the visual state-space fault diagnosis network, and the classification output module outputs the predicted probability corresponding to each fault category. A classification loss function is constructed based on the difference between the predicted probability and the true fault category label, and the parameters of the visual state-space fault diagnosis network are updated through backpropagation.

[0080] In a preferred embodiment, the classification loss function is the cross-entropy loss function, expressed as: In the formula, Indicates the number of fault categories. Indicates the first The true label corresponding to the fault category, when the sample belongs to the first fault category. Class Time ,otherwise ; The output of the visual state space fault diagnosis network represents the first... Predicted probabilities for each fault category This represents classification loss.

[0081] In one implementation, the visual state space fault diagnosis network can be updated with parameters using stochastic gradient descent, Adam optimization, or other gradient optimization algorithms until the classification loss converges or a preset number of training rounds are reached, thus obtaining a trained visual state space fault diagnosis network.

[0082] During the diagnostic phase, the multi-channel attitude measurement signals of the spacecraft's attitude control system are processed according to the steps described above to obtain a joint time-frequency diagnostic image. This joint time-frequency diagnostic image is then input into the trained visual state-space fault diagnosis network to obtain the predicted probability corresponding to each fault category. The fault diagnosis result is determined based on the fault category corresponding to the highest predicted probability.

[0083] In this embodiment, the fault diagnosis result includes fault status, fault sensor type, and fault mode; when the fault category corresponds to a specific axial channel, the fault axial position is further output. The fault status includes normal status and fault status; the fault sensor type includes star sensor and gyroscope; the fault mode includes jamming fault and random noise fault; the fault axial position includes at least one of the X-axis, Y-axis, and Z-axis.

[0084] In one implementation, the fault categories may include: normal state; star sensor stuck fault; star sensor random noise fault; gyroscope X-axis stuck fault; gyroscope Y-axis stuck fault; gyroscope Z-axis stuck fault; gyroscope X-axis random noise fault; gyroscope Y-axis random noise fault; gyroscope Z-axis random noise fault.

[0085] Furthermore, if the maximum category probability output by the visual state space fault diagnosis network is greater than or equal to a preset confidence threshold, the fault diagnosis result corresponding to the maximum probability value is output; if the maximum category probability is less than the preset confidence threshold, the abnormal state to be confirmed is output, so as to reduce the impact of low confidence misjudgment on the spacecraft fault diagnosis result.

[0086] The spacecraft attitude control system fault diagnosis method based on time-frequency coding according to the embodiments of the present invention has the following advantages: This invention employs synchronous compressed wavelet transform to generate time-frequency energy maps, enhancing the time-frequency clustering of non-stationary weak fault features and improving fault feature representation. Physical grouping according to attitude angle and angular velocity channels avoids simple mixing of signals with different physical meanings, facilitating the maintenance of coupling between signals in the spacecraft attitude control system. Shared normalization is used for the three-axis time-frequency energy maps within the same channel group, preserving the relative energy relationships between different axes and avoiding energy ratio distortion caused by channel-by-channel normalization. The invention stitches the attitude angle RGB time-frequency sub-maps and angular velocity RGB time-frequency sub-maps into a joint time-frequency diagnostic image, enabling the diagnostic network to simultaneously utilize attitude state anomaly features and attitude motion anomaly features. The invention utilizes a visual state space fault diagnosis network for feature extraction and classification, reducing the computational overhead of high-resolution time-frequency image diagnosis while maintaining global dependency modeling capabilities. The output of this invention includes not only the fault state but also the fault sensor type and fault mode. When the fault category corresponds to a specific axial channel, the fault axial position is further output, providing more explicit information for fault isolation, fault-tolerant control, and health management of the spacecraft attitude control system.

[0087] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A fault diagnosis method for a spacecraft attitude control system based on time-frequency coding, characterized in that, Includes the following steps: Step S1: Acquire the multi-channel attitude measurement signals of the spacecraft attitude control system within a preset time window, and divide the multi-channel attitude measurement signals into attitude angle channel group and angular velocity channel group according to their physical meaning; Step S2: Perform synchronous compressed wavelet time-frequency transformation on the signals of each channel in the attitude angle channel group and the angular velocity channel group to obtain the synchronous compressed wavelet time-frequency energy map corresponding to each channel; Step S3: For the attitude angle channel group and the angular velocity channel group respectively, the three-axis synchronous compressed wavelet time-frequency energy maps in the same channel group are sequentially subjected to energy compression, shared normalization and RGB mapping to generate attitude angle RGB time-frequency sub-maps and angular velocity RGB time-frequency sub-maps; wherein, the parameters used for the shared normalization are determined by the merged pixel distribution of the three-axis compressed synchronous compressed wavelet time-frequency energy maps in the corresponding channel group; Step S4: The attitude angle RGB time-frequency sub-image and the angular velocity RGB time-frequency sub-image are stitched together according to a preset spatial layout to obtain a joint time-frequency diagnostic image; Step S5: Input the joint time-frequency diagnostic image into the visual state space fault diagnosis network and output the fault diagnosis results of the spacecraft attitude control system; wherein, the fault diagnosis results include fault state, fault sensor type and fault mode; when the fault involves the axial channel, the fault diagnosis results also include the fault axial position.

2. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, The multi-channel attitude measurement signal includes a three-axis attitude angle signal and a three-axis angular velocity signal; The three-axis attitude angle signals include roll angle, pitch angle, and yaw angle; The three-axis angular velocity signals include the spacecraft's own X-axis angular velocity, Y-axis angular velocity, and Z-axis angular velocity.

3. The fault diagnosis method for spacecraft attitude control systems based on time-frequency coding according to claim 1, characterized in that, The attitude angle channel group consists of roll angle, pitch angle and yaw angle, and is used to characterize changes in the attitude state of the spacecraft. The angular velocity channel group consists of X-axis angular velocity, Y-axis angular velocity and Z-axis angular velocity, and is used to characterize the changes in the attitude motion of the spacecraft. The attitude angle channel group and the angular velocity channel group can maintain the inter-axis coupling relationship between the three-axis signals with the same physical meaning in the subsequent time-frequency coding process.

4. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, Before step S2, the signal of each channel is further normalized. The normalization process uses Z-score normalization, and its expression is: in, Represents the original channel signal. This represents the average value of the channel signal within a preset time window. This represents the standard deviation of the channel signal within a preset time window. This represents the standardized channel signal.

5. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, In step S2, the scale set of the synchronous compressed wavelet time-frequency transform is determined based on the sampling frequency, the lowest frequency of interest, the highest frequency of interest, and the wavelet center frequency; the scale set is expressed as: in, Represents a scale set, Indicates the first One scale, Indicates the wavelet center frequency. Indicates the sampling time interval. and These represent the lowest and highest frequencies of attention, respectively. Indicates the number of scales.

6. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, In step S2, the synchronous compressed wavelet time-frequency transform includes continuous wavelet transform, instantaneous frequency estimation, and frequency reallocation; wherein, the continuous wavelet transform is expressed as: in, This represents the single-channel signal to be transformed. Indicates the scale parameter. Indicates the time shift parameter. This represents the conjugate function of the wavelet basis functions. This represents the continuous wavelet transform coefficients.

7. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 6, characterized in that, The instantaneous frequency is estimated based on the phase change of the continuous wavelet transform coefficients. The instantaneous frequency is expressed as: in, Represents the continuous wavelet transform coefficients relative to the time shift parameter. The partial derivatives, This indicates the operation of taking the imaginary part. This represents the estimated instantaneous frequency.

8. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 7, characterized in that, Synchronous compression wavelet coefficients are obtained by synchronously compressing and redistributing the continuous wavelet transform coefficients in the frequency direction based on the instantaneous frequency. These synchronous compression wavelet coefficients are expressed as follows: in, This represents the synchronous compressed wavelet coefficients. Represents frequency variables. This represents the set of scales that satisfy the preset amplitude conditions. Represents the redistribution function; The time-frequency energy diagram of the synchronous compressed wavelet in step S2 is represented as follows: in, This represents the time-frequency energy diagram of the synchronous compressed wavelet.

9. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, In step S3, the energy compression adopts logarithmic compression, which is expressed as: in, This represents the energy value in the synchronous compressed wavelet time-frequency energy map. This represents the compressed time-frequency energy value. This represents the energy scaling factor.

10. The fault diagnosis method for spacecraft attitude control system based on time-frequency coding according to claim 1, characterized in that, In step S3, the shared normalization includes any value from the 1st to the 5th percentile and any value from the 95th to the 99th percentile.

Images