Fault diagnosis and early warning method based on electric thruster data time sequence characteristics

Abstract

The invention relates to the technical field of space electric propulsion, in particular to a fault diagnosis and early warning method based on electric thruster data time sequence characteristics, which is characterized in that working parameters of an electric thruster are collected at high sampling frequency, a normalization mode is adopted, influences of different parameters are balanced, and a double-label mode is adopted. Future fault state information is embedded in the training data set; performing time sequence feature extraction and fusion on the working feature parameters of the electric thruster by using a long-short memory network, and performing real-time fault diagnosis and early warning through loop iteration training of the network; the method is simple in implementation process and high in engineering applicability, fully excavates the time sequence characteristic connotation information of the operation data of the electric thruster product, and can intelligently realize the operation state monitoring and fault early warning prompt of the electric thruster product.

Description

Technical Field

[0001] This application relates to the field of space electric propulsion technology, and more specifically, to a fault diagnosis and early warning method based on the timing characteristics of electric thruster data. Background Technology

[0002] Electric propulsion technology has advantages such as high specific impulse, long lifespan, and no scale effects, and is widely used in the aerospace field to provide power for aircraft, undertaking key tasks such as attitude control, orbit control, and momentum unloading. The electric thruster is the core single unit for thrust output in the electric propulsion system. However, the electric thruster has a complex structure and operates in environments such as plasma, electromagnetic fields, and thermal flow fields for a long time. There are many factors that can induce failure, and its failure rate cannot be ignored. Fault diagnosis of electric propulsion products plays a very important role in the normal flight of aircraft.

[0003] Currently, some data-based methods for fault diagnosis and early warning of electric propulsion products have emerged, but in practical applications, there are technical problems such as insufficient data utilization, inadequate mining of data connotation information, and weak engineering applicability.

[0004] Traditional diagnostic methods based on feature parameter thresholds struggle to integrate information from multiple feature parameters, resulting in a lack of diagnostic flexibility, difficulty adapting to various operating conditions and environments, and a lack of fault early warning capabilities. Diagnostic methods based on fully connected neural networks input fault data into the neural network module all at once to analyze the relationship between the current data segment and the fault. However, this method disrupts the temporal relationship of each data set within a segment, failing to uncover the causal correlation between the fault and the product's state at different times, and losing crucial feature information, thus limiting the network's diagnostic and early warning performance. Furthermore, inputting all data segments into the neural network at once leads to excessively large network size, consuming significant computational resources and failing to meet the high-efficiency and reliable application requirements of aerospace products, thus limiting its engineering applicability. Therefore, effectively integrating electric propulsion product data information, extracting temporal features, understanding the causal relationship between product state and fault occurrence, and simultaneously reducing computational resource overhead are of great significance for fault diagnosis and early warning of space electric propulsion products. Summary of the Invention

[0005] This application provides a fault diagnosis and early warning method based on the time-series characteristics of electric thruster data. By adopting time series analysis, the method improves the automation and intelligence level of fault diagnosis and early warning for space electric thrusters.

[0006] To achieve the above objectives, this application provides a fault diagnosis and early warning method based on the time-series characteristics of electric thruster data, comprising the following steps: Step 1: Based on the working physical characteristics of the electric thruster, analyze and determine the characteristic parameters reflecting the operating performance of the electric thruster, and accordingly collect the operating parameters of the electric thruster at high frequency, while simultaneously recording the fault states that occur during the process; Step 2: Preprocess the collected data, select data segments with fixed time spans, extract characteristic parameters and quantize them, label the corresponding fault states for the characteristic parameters at different times, and divide the dataset into training dataset and test dataset according to the proportion; Step 3: Construct an electric thruster fault diagnosis and early warning module based on a long short-term memory network; Step 4: Input the preprocessed training dataset into the electric thruster fault diagnosis and early warning module, and continuously iterate and update the network weight parameters until convergence with the goal of minimizing the fault diagnosis and early warning error; Step 5: Input real-time test data into the fault diagnosis and early warning module to monitor the operating status of the electric thruster in real time and provide fault warnings.

[0007] Furthermore, in step 1, based on the working physical characteristics of the electric thruster, a correlation analysis method is used to analyze and determine the characteristic parameters that reflect the operating performance of the electric thruster.

[0008] Furthermore, in step 1, the characteristic parameters reflecting the operating performance of the electric thruster include one or more of the following: ion optics component voltage and current, discharge chamber voltage and current, cathode contact voltage and current, propellant working fluid flow rate, thruster casing voltage, and temperature.

[0009] Furthermore, in step 1, the frequency of collecting the operating parameters of the electric thruster is ≥100MHz.

[0010] Furthermore, in step 2, during data segment preprocessing, the time average and standard deviation of each feature parameter within the segment are obtained, and the state vector is normalized using these values. The normalized parameters are then used as the final feature parameters.

[0011] Furthermore, in step 2, when performing data segment preprocessing, two types of labels need to be added to the state vector at each time step in the data segment: one is the current running state, and the other is the product running state after time step k. The labels are expressed in the form of confidence vectors.

[0012] Furthermore, in step 2, the number of elements in the vector is the same as the number of fault state categories.

[0013] Furthermore, the electric thruster fault diagnosis and early warning module is built based on a long short-term memory network. The module is trained by selecting data segments of a fixed length and performing labeling processing. The network receives the product feature parameter vector as input, and the output at each time step represents the product state estimate at the current time step and the product state prediction after time step k.

[0014] Furthermore, the output loss of the electric thruster fault diagnosis and early warning module is measured using the cross-entropy loss function.

[0015] Furthermore, after the electric thruster fault diagnosis and early warning module has been trained, when inputting data parameters during application, the time average value is taken as the design value of the data parameter, and the standard deviation is taken as the standard deviation of the historical data of the data parameter.

[0016] The fault diagnosis and early warning method based on the timing characteristics of electric thruster data provided in this application has the following beneficial effects: This application collects the operating parameters of the electric thruster at a high sampling frequency, adopts a normalization method to balance the influence of different parameters, and uses a dual-label approach to embed future fault state information into the training dataset. It utilizes a long short-term memory network to extract and fuse the temporal features of the electric thruster's operating characteristic parameters, and achieves real-time fault diagnosis and early warning through iterative training of the network. The execution process is simple and highly applicable to engineering, fully exploring the temporal characteristics of the electric thruster product's operating data, and can intelligently realize the monitoring of the operating status and fault early warning of the electric thruster product. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.

[0018] In the attached diagram: Figure 1 This is a flowchart illustrating a fault diagnosis and early warning method based on the timing characteristics of electric thruster data provided in an embodiment of this application. Figure 2 This is a schematic diagram of the framework of a fault diagnosis and early warning module based on a long short-term memory network provided in the embodiments of this application. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this application described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] In this application, the terms "upper," "lower," "left," "right," "front," "rear," "top," "bottom," "inner," "outer," "middle," "vertical," "horizontal," "lateral," and "longitudinal" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. These terms are primarily for the purpose of better describing this application and its embodiments, and are not intended to limit the indicated device, element, or component to having a specific orientation, or to be constructed and operated in a specific orientation.

[0022] Furthermore, in addition to indicating location or positional relationship, some of the aforementioned terms may also have other meanings. For example, the term "above" may also be used in some cases to indicate a certain dependency or connection relationship. Those skilled in the art can understand the specific meaning of these terms in this application based on the specific circumstances.

[0023] In addition, the term "multiple" should mean two or more.

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0025] like Figure 1 As shown, this application provides a fault diagnosis and early warning method based on the time-series characteristics of electric thruster data. It collects electric thruster operating parameters at a high sampling frequency, employs normalization to balance the influence of different parameters, and integrates time-series information of various characteristic parameters of the electric thruster, such as voltage, current, and temperature. A dual-label approach is used to embed future fault state information into the training dataset. A Long Short-Term Memory (LSTM) network is used to extract and fuse time-series features of the electric thruster's operating characteristic parameters. Through iterative training of the network, real-time fault diagnosis and early warning are achieved. The specific steps include: Step 1: Based on the working physical characteristics of the electric thruster, analyze and determine the characteristic parameters that reflect the operating performance of the electric thruster, and collect the operating parameters of the electric thruster at high frequency, while recording the fault states that occur during the process. Specifically, based on the working physical characteristics of the electric thruster, a correlation analysis method is used to analyze and determine the characteristic parameters that reflect the operating performance of the electric thruster. These characteristic parameters include one or more of the following: voltage and current of the ion optics component, discharge voltage and current of the discharge chamber, cathode contact voltage and current, propellant flow rate, thruster casing voltage, and temperature. Real-time operating data of each characteristic parameter are collected synchronously using a sampling frequency of 100MHz or higher, and fault states that occur during the process are recorded simultaneously.

[0026] Step 2: Preprocess the collected data, select data segments with a fixed time span, extract feature parameters and quantize them, label the corresponding fault states for the feature parameters at different times, and divide the dataset into training dataset and test dataset according to the proportion; Specifically, the collected characteristic parameters of the electric thruster are preprocessed. For each failure case, a fixed-length data segment is selected, and the data at each moment within the segment is vectorized. The elements of the vector, i.e., the characteristic parameters, are then normalized. Simultaneously, the parameter vectors generated from the data segments are labeled. Each vector is assigned two labels: one label indicates the product state at the current moment t (i.e., the type of failure or normal state), and the other label indicates the product failure type at moment t+k. The product state information is encoded in confidence vector form. This embeds future failure information into the parameter information at each moment, and subsequent deep learning algorithms are used to mine the correlation between future failures and the current product state. After preprocessing, the dataset is divided into training and testing datasets according to a certain ratio.

[0027] More specifically, taking a data segment of length p as an example, a fixed duration p is set, and a high-frequency acquired fault data segment is selected. The feature parameters of each time step of the segment are converted into vector form and saved, i.e. , where v i If (t) is the i-th feature at time t, then for a data segment of duration p, its storage format is as follows: Because the feature parameters include voltage parameters (high voltage can exceed 1000V), current parameters (approximately 0.5~3A), temperature parameters, flow rate parameters, etc., and these parameters have different dimensions and different variances for different amplitudes, parameters with larger amplitudes or variances have a greater impact on the network output in fault diagnosis using deep learning methods, leading to slower convergence of network weight parameters during training. Furthermore, these parameters can mask the correlation between other low-value parameters and the fault, interfering with the accuracy of network calculations. Therefore, it is necessary to normalize these parameters. Normalization consists of two steps: centering and normalization. The centering process takes the average value of each parameter in the data segment as the center and normalizes each feature parameter to zero mean, as shown in the following formulas: , where μ i Let r be the average value of the i-th dimension feature within the data segment. i (t) represents the eigenvalues ​​after centering.

[0028] Normalization calculates the variance of each parameter within the data segment after centering, and then uses this variance to weight each parameter, as shown in the following formulas: ,in, Let x be the variance of the eigenvalues ​​after centering. i (t) represents the normalized eigenvalues. These are the normalized eigenvectors.

[0029] Furthermore, the fault diagnosis and early warning module constructed in this embodiment is based on a Long Short-Term Memory (LSTM) network, which is a supervised learning method. The training dataset needs to be labeled in advance for network training and loss function calculation. However, adding only one label to a data segment cannot reflect the in-depth relationship between the product state and the fault at a specific moment. Therefore, this embodiment adds a label to each moment in the data segment, ensuring that the output of each time layer of the LSTM has an accurate label for calculation, thus improving the accuracy of the loss function calculation. Moreover, to uncover the causal relationship between the fault and the product state at previous moments, this embodiment adds two types of labels to the data at each moment: representing the fault state at the current moment t, and the fault state at moment k (t+k), enabling the network to have the ability to predict fault states at future moments after training. The labels are given in a vector form similar to one-hot encoding, that is, the number of elements in the vector is the same as the number of fault categories. The element corresponding to the fault that occurs at time t is marked as 1, and the element that does not occur is marked as 0. This can express the situation where multiple faults occur at the same time. If all elements in the vector are 0, it indicates that the product is working normally. At the same time, multiple variables of 1 are allowed to appear, ensuring that the module can diagnose and warn of multiple faults that occur at the same time.

[0030] Step 3: Construct a fault diagnosis and early warning module for electric thrusters based on Long Short-Term Memory (LSTM) networks. The module receives the current state vector of the product and the estimated output from the previous time step as input, and outputs the current fault diagnosis and early warning confidence vector.

[0031] Specifically, LSTM is an improvement on traditional recurrent neural networks (RNNs), capable of extracting the time-series characteristics of data and capturing causal relationships between time-series data. Simultaneously, it can assign reasonable weights to data at different time intervals, effectively mitigating the gradient vanishing problem during deep learning training and improving training efficiency. After network training, it can output discrimination results in real time during use. These technical characteristics are well-suited to the requirements of efficient and reliable data fault analysis in electric thrusters. For example... Figure 2 As shown, the LSTM network uses computational units as its core, iteratively calculating the input and output. The computational unit structure can be divided into four parts, including the forget gate (f t ), Input gate (i t ), output gate (o t ), cell state (c t ),in: , In the formula, S t c t h t Let c be the input vector (i.e., the feature vector at time t), the cell state, and the output vector, respectively; t-1 h t-1 These represent the unit state and output vector at time t-1, respectively; δ is the nonlinear activation function; W f W i W o These are the forgetting, input, and output weight matrices, respectively, b. f b i b o These are the corresponding biases.

[0032] Step 4: Input the preprocessed training dataset into the electric thruster fault diagnosis and early warning module, and continuously iterate and update the network weight parameters until convergence with the goal of minimizing the fault diagnosis and early warning error; Specifically, the training set is divided into training groups of ten segments each, i.e., a batch. These are input into the fault diagnosis and early warning module for forward computation. After obtaining the cross-entropy loss, the network weight parameters are iteratively updated using the time-series backpropagation algorithm with the goal of minimizing the loss. After one round of training is completed with all training set data, the order of the segments in the training set is shuffled, and the batches are reorganized into groups of ten segments each. A new round of training begins, stopping after a specified number of training rounds. The network performance is then tested using a test set. The network output is the confidence level of each fault at the current time and at future time. If the confidence level exceeds a threshold, the fault is considered to have occurred.

[0033] The output h(t) of the diagnosis and early warning module is a vector containing different fault confidence levels at the current and future times, as shown in the following formula: , In the formula, ρ i ρ(t) represents the confidence level of different types of product failures occurring at time t, where ρ1(t) ~ ρ m (t) represents the diagnosis result at the current moment, ρ m+1 (t)~ρ 2m (t) represents the early warning information for a future time.

[0034] The computational loss of the LSTM network is obtained by summing the fault diagnosis loss Loss(t) at the current time step and the fault warning loss Loss(t+k) at the future time step. Both losses are calculated using the cross-entropy loss function, as shown in the following equations: In the formula, batch represents the data segment in the smallest training group, which is preferably 10 in this embodiment. j (t) represents the actual fault occurrence at time t. A value of 1 indicates that fault j has occurred, while a value of 0 indicates that fault j has not occurred. L t Let t be the loss value calculated from the data at time t. Then, the loss L calculated by LSTM for the entire data segment of length P is shown in the following formula: , The network weight coefficients are updated using the time-series backpropagation algorithm, as shown in the following formula: , In the formula, Let be the partial derivative of the training loss for the entire data segment with respect to the weight parameters in the network. Let t be the partial derivative of the training loss with respect to the network weight parameters.

[0035] Step 5: Input the real-time test data into the fault diagnosis and early warning module to monitor the operating status of the electric thruster in real time and issue fault warnings.

[0036] Specifically, in the vector h(t) output by the diagnosis and warning module, when ρ i (t) exceeds the threshold, for 1 < i < m, it is considered that the corresponding fault has occurred at the current moment, that is, the fault diagnosis is completed; for m < i < 2m, it is considered that there is a high probability of occurrence of this fault at a future moment, that is, a fault warning is given.

[0037] The above are only the preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A fault diagnosis and early warning method based on the timing characteristics of electric thruster data, characterized in that, Includes the following steps: Step 1: Based on the working physical characteristics of the electric thruster, analyze and determine the characteristic parameters that reflect the operating performance of the electric thruster, and collect the operating parameters of the electric thruster at high frequency, while recording the fault states that occur during the process. Step 2: Preprocess the collected data, select data segments with a fixed time span, extract feature parameters and quantize them, label the corresponding fault states for the feature parameters at different times, and divide the dataset into training dataset and test dataset according to the proportion; Step 3: Construct a fault diagnosis and early warning module for electric thrusters based on long short-term memory networks; Step 4: Input the preprocessed training dataset into the electric thruster fault diagnosis and early warning module, and continuously iterate and update the network weight parameters until convergence with the goal of minimizing the fault diagnosis and early warning error; Step 5: Input the real-time test data into the fault diagnosis and early warning module to monitor the operating status of the electric thruster in real time and issue fault warnings.

2. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 1, characterized in that, In step 1, based on the working physical characteristics of the electric thruster, the correlation analysis method is used to analyze and determine the characteristic parameters that reflect the operating performance of the electric thruster.

3. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 2, characterized in that, In step 1, the characteristic parameters reflecting the operating performance of the electric thruster include one or more of the following: voltage and current of the ion optics component, discharge voltage and current of the discharge chamber, cathode contact voltage and current, propellant flow rate, thruster casing voltage, and temperature.

4. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 3, characterized in that, In step 1, the frequency of collecting the operating parameters of the electric thruster is ≥100MHz.

5. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 4, characterized in that, In step 2, during data segment preprocessing, the time average and standard deviation of each feature parameter within the segment are obtained, and the state vector is normalized using these values. The normalized parameters are then used as the final feature parameters.

6. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 5, characterized in that, In step 2, when performing data segment preprocessing, two types of labels need to be added to the state vector at each time step in the data segment: one is the current running state, and the other is the product running state after time step k. The labels are expressed in the form of confidence vectors.

7. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 6, characterized in that, In step 2, the number of elements in the vector is the same as the number of fault state categories.

8. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 7, characterized in that, The electric thruster fault diagnosis and early warning module is built based on a long short-term memory network. The module is trained by selecting fixed-length data segments and labeling them. The network receives the product feature parameter vector as input, and the output at each time step represents the product state estimate at the current time step and the product state prediction after time step k.

9. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 8, characterized in that, The output loss of the electric thruster fault diagnosis and early warning module is measured using the cross-entropy loss function.

10. The fault diagnosis and early warning method based on the timing characteristics of electric thruster data according to claim 9, characterized in that, After the electric thruster fault diagnosis and early warning module is trained, when inputting data parameters during application, the time average value is taken as the design value of the data parameter, and the standard deviation is taken as the standard deviation of the historical data of the data parameter.

Images