Fault monitoring method and system for rocket engine

Abstract

The invention discloses a fault monitoring method and system for a rocket engine, relates to the technical field of engine testing, and aims to solve the problem that comprehensive fault monitoring cannot be performed on a test part in the prior art. The method comprises the steps of obtaining multiple pieces of state data of a target test piece; the multiple pieces of state data are operation test full state data of a target test piece in the physical test process of the rocket engine; extracting a plurality of target features from the plurality of pieces of state data; the plurality of target features at least comprise spatial features and time sequence features; inputting the plurality of target features into a pre-trained target fault monitoring model, and predicting to obtain a fault monitoring result; based on this, the fault monitoring of the rocket engine with advanced time sequence can be realized, and the fault monitoring range of the test part is widened.

Description

Technical Field

[0001] The present invention relates to the technical field of engine testing, and in particular to a method and system for fault monitoring of a rocket engine. Background Art

[0002] To improve the operation stability and safety of a liquid rocket engine, the fault monitoring link in the physical test run of the liquid rocket engine is particularly important. How to obtain more comprehensive test data is an urgent problem to be solved currently.

[0003] The fault monitoring technology in the prior art mainly relies on the data collected by sensors. Due to the precision of the liquid rocket engine, there are limited places where sensors can be placed, resulting in insufficient monitoring coverage and inability to comprehensively monitor the test components for faults; for example: the sensors of the flow regulator can only measure the pressures at the inlet and outlet, the amount of data collected by the sensors is limited, and the fault monitoring coverage is insufficient; and it can only monitor the state data when an actual fault occurs, and cannot achieve advanced prediction or early warning of the pressure; therefore, the technology that only relies on sensors for fault monitoring cannot achieve comprehensive fault monitoring of the test components.

[0004] Therefore, there is an urgent need to establish a monitoring technology with a larger and wider fault monitoring range for rocket engines to solve the problem in the prior art that comprehensive fault monitoring of test components cannot be achieved. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and system for fault monitoring of a rocket engine, which can perform time series advanced prediction based on the state data collected by existing sensors to obtain more fault monitoring data and broaden the monitoring range; thus solving the problem in the prior art that comprehensive fault monitoring of test components cannot be achieved.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for fault monitoring of a rocket engine, which is applied to a digital twin system of a rocket engine and may include:

[0008] Obtain a plurality of state data of a target test piece; the plurality of state data are the full state data of the operation test of the target test piece during the physical test run of the rocket engine;

[0009] Extract a plurality of target features from the plurality of state data; the plurality of target features at least include spatial features and time series features;

[0010] Input the plurality of target features into a pre-trained target fault monitoring model to predict and obtain a fault monitoring result.

[0011] Preferably, before inputting the multiple target feature data into the pre-trained target fault monitoring model, it may include: obtaining a training sample set from a sample database; extracting training sample feature data from the training sample set; the training sample feature data at least includes spatial feature data and temporal feature data; using the training sample feature data to perform model training on a network model to obtain an initial fault monitoring model; inputting a test sample set into the initial fault monitoring model to obtain an initial fault perception result; the test sample set is a test sample set obtained from the sample database; comparing the initial fault perception result with the actual result, and adjusting the initial fault monitoring model based on an error function until the error function converges to obtain the target fault monitoring model.

[0012] Preferably, before obtaining the training sample set from the sample database, it may include:

[0013] Establishing a sample database; the establishment of the sample database may include: obtaining the historical test full-state data of the target test piece; performing data preprocessing on the historical test full-state data based on a preset data-driven fault prediction method to obtain multiple target data; the results represented by the multiple target data are the fault data of the target test piece at different time sequences; establishing the sample database based on the multiple target data; the multiple target data at least includes spatial features and temporal features.

[0014] Preferably, performing data preprocessing on the historical test full-state data based on a preset data-driven fault prediction method to obtain multiple target data may include:

[0015] Performing signal dimensionality reduction processing on the historical test full-state data to obtain multiple target feature vectors; the multiple target feature vectors are low-dimensional feature vectors including state identifiers obtained by performing dimensionality reduction processing on the high-dimensional feature vectors in the historical test full-state data; the dimensionality reduction processing at least includes signal processing, feature extraction, feature dimensionality reduction, and pattern recognition; wherein, the method of signal processing and feature extraction is a method combining the EMD method and the fuzzy entropy theory; performing state classification and recognition on the multiple target feature vectors to obtain the multiple target data.

[0016] Preferably, the step of inputting the multiple pieces of target feature data into a pre-trained target fault monitoring model to predict a fault monitoring result may include: inputting the spatial features into the pre-trained target fault monitoring model to obtain spatial feature prediction data; inputting the temporal features into the pre-trained target fault monitoring model to obtain temporal feature prediction data; performing data fusion on the spatial feature prediction data and the temporal feature prediction data to obtain the fault monitoring result; the data fusion includes at least data splicing and data weighting.

[0017] In a second aspect, the present invention provides a fault monitoring system for a rocket engine, which is applied to a digital twin platform of a rocket engine. The fault monitoring system uses the fault monitoring method according to any one of the first aspects to perform fault monitoring. The system may include: a data acquisition module, a feature extraction module, and a target fault monitoring module; the data acquisition module is configured to acquire multiple state data of a target test piece; the multiple state data are all state data of the target test piece during the physical test run of the rocket engine; the feature extraction module is configured to extract multiple target features from the multiple state data; the multiple target features include at least spatial features and temporal features; the target fault monitoring module is configured to input the multiple target features into a pre-trained target fault monitoring model to predict a fault monitoring result.

[0018] Preferably, the target fault monitoring module may include:

[0019] a data dimensionality reduction unit, which is configured to perform data dimensionality reduction processing on the multiple pieces of all state data according to a preset prediction method for data-driven faults to obtain multiple target data; the multiple pieces of all state data are all state data of historical test runs of the rocket engine; the results represented by the multiple target data are fault data of the target test piece at different time sequences; the multiple target data include high-dimensional spatial features and high-dimensional temporal features; a building unit, which is configured to build the target fault monitoring model based on the multiple target data.

[0020] Preferably, the feature extraction module may include a spatial feature unit and a first building unit;

[0021] The spatial feature unit is configured to extract high-dimensional spatial features from the multiple state data; the first building unit is configured to build a spatial feature model based on the high-dimensional spatial features of the multiple target data.

[0022] Preferably, the feature extraction module includes a temporal feature unit and a second building unit;

[0023] The time series feature unit is used to extract high-dimensional time series features from multiple pieces of the state data; the second construction unit is used to construct a time series feature model based on the high-dimensional time series features of multiple pieces of the target data.

[0024] Preferably, the target fault monitoring module may include a fault monitoring fusion module; the fault monitoring fusion module is used to perform fusion processing on the high-dimensional spatial feature data and the high-dimensional time series feature data, and predict a fault monitoring result.

[0025] Compared with the prior art, a fault monitoring method for a rocket engine provided by the present invention is applied to a digital twin system of a rocket engine. By obtaining multiple pieces of state data of a target test piece; the multiple pieces of state data are all-state operation test data of the target test piece during the physical test run of the rocket engine; extracting multiple target features from the multiple pieces of state data; the multiple target features at least include spatial features and time series features; inputting the multiple target features into a pre-trained target fault monitoring model to predict a fault monitoring result; based on this, because the pre-trained target fault monitoring model is a data-driven fault diagnosis method that analyzes and processes engine measurement data, mines the hidden information in the measurement data, and characterizes the normal or fault mode of the engine, thereby establishing an engine state fault perception model based on a long short-term neural network; therefore, during the physical test run of the rocket engine, the target fault monitoring model can be used to perform time series lead prediction on the all-state data with spatial features and time series features, broadening the fault monitoring scope of each test component during the physical test run of the rocket engine. Description of the Drawings

[0026] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention, and do not constitute an improper limitation of the present invention. In the drawings:

[0027] Figure 1 It is a schematic diagram of the fault monitoring effect in the prior art;

[0028] Figure 2 It is a schematic diagram of the main process of a fault monitoring method for a rocket engine provided by the present invention;

[0029] Figure 3 It is a schematic diagram of the visualization effect of fault monitoring of a fault monitoring method for a rocket engine provided by the present invention;

[0030] Figure 4 It is a schematic diagram of the process of establishing a sample database of a fault monitoring method for a rocket engine provided by the present invention;

[0031] Figure 5Schematic diagram of the main structure of a fault monitoring system for a rocket engine provided by the present invention;

[0032] Figure 6 Schematic diagram of the structure for fault monitoring of vibration data in a fault monitoring system for a rocket engine provided by the present invention.

[0033] Reference numerals: 510 - data acquisition module, 520 - feature extraction module, 530 - target fault monitoring module, 521 - spatial feature unit, 522 - first construction unit, 523 - temporal feature unit, 524 - second construction unit, 531 - dimensionality reduction unit, 532 - establishment unit, 533 - fault monitoring fusion unit. Detailed implementation manners

[0034] For the convenience of clearly describing the technical solutions of the embodiments of the present invention, in the embodiments of the present invention, terms such as "first" and "second" are used to distinguish the same items or similar items with basically the same functions and effects. For example, the first threshold and the second threshold are only used to distinguish different thresholds, and do not limit their chronological order. Those skilled in the art can understand that terms such as "first" and "second" do not limit the quantity and execution order, and terms such as "first" and "second" do not necessarily mean different.

[0035] It should be noted that in the present invention, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the present invention should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Exactly speaking, using words such as "exemplary" or "for example" aims to present relevant concepts in a specific manner.

[0036] In the present invention, "at least one" means one or more, and "a plurality" means two or more. "And / or" describes the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone, where A and B can be singular or plural. The character " / " generally represents an "or" relationship between the previous associated objects. "At least one (piece)" or its similar expression below refers to any combination of these items, including any combination of single item (piece) or plural items (pieces). For example, at least one (piece) of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b and c can be single or multiple.

[0037] The fault monitoring technology in the prior art mainly arranges multiple sensors on the device under test, uses the sensors to collect the status data during the operation of the device, and displays the status data. With the rapid development of liquid rocket engine technology, such a data collection and display method can no longer meet the requirements. Due to the precision of the liquid rocket engine, the places where sensors can be placed are limited, resulting in insufficient monitoring surfaces; as Figure 1 shown Figure 1 is a schematic diagram of the fault monitoring effect in the prior art; in Figure 1 it, the displayed curve graph is a one-dimensional graph displayed on the monitoring platform depending on the data collected by the sensors, which has restricted the data analysis during the physical test run of the liquid rocket engine and cannot meet the technical requirements for improving the operation stability and safety of the liquid rocket engine.

[0038] In view of this, the present invention provides a fault monitoring method for a rocket engine, which is applied to a digital twin system of a rocket engine; uses a pre-trained target fault monitoring model to perform fault prediction on multiple status data of a test piece, and obtains a fault monitoring result with time series lead, broadening the fault monitoring range of each test component during the physical test run of the rocket engine; thus solving the problem in the prior art that comprehensive fault monitoring of test components cannot be achieved; and three-dimensionally visualizes the fault data in the digital twin platform to meet the technical requirements for improving the operation stability and safety of the liquid rocket engine.

[0039] It should be noted that the fault monitoring method for a rocket engine provided by the present invention is based on deep learning, mines and fuses the data collected by sensors, and realizes fast and accurate judgment of potential faults existing in the target component through technical means of fusing multiple pieces of information. As an important branch of artificial intelligence technology, deep learning has been widely applied in many engineering fields in recent years.

[0040] The research on related foreign technologies started earlier. Surajit Roy et al. from Stanford University first based on unsupervised machine learning algorithms and achieved the integrity diagnosis and temperature compensation of structures by using guided ultrasonic waves as the seismic source. René-Vinicio Sánchez Loja et al. from the Warsaw University of Technology proposed a gearbox vibration fault diagnosis method based on the gated recurrent unit (MGRU). By using MGRU to learn and classify the time-frequency domain vibration signals extracted from the test bench, the signal expression and fault type recognition capabilities were improved. Jürgen Pfeffer et al. from the Technical University of Munich first solved the problem of how to use a large amount of unbalanced data in the "structural noise" signal related to the internal excitation of the engine for engine damage prediction by using a convolutional neural network on the time series input signal. Mehrdad Heydarzadeh et al. from the University of Texas used a deep neural network to conduct five-category fault diagnosis on three common monitoring signals of vibration, sound, and torque of the gearbox. The discrete wavelet transform was used to provide initial features as the input of the network, achieving accurate fault diagnosis under different modes, signal specificities, and load conditions.

[0041] In recent years, domestic researchers have achieved certain results in this application field. The team led by Professor Peng Gaoliang from Harbin Institute of Technology proposed a new deep convolutional neural network (WDCNN) based on a wide first-layer convolutional kernel. This method uses the original vibration signal as the input (using data augmentation methods) to generate more inputs, and uses a wide kernel in the first convolutional layer to extract features and suppress high-frequency noise, improving the fault diagnosis accuracy of the Convolutional Neural Networks (abbreviated as CNN). The team from South China University of Technology constructed a one-dimensional convolutional neural network model by fusing time-domain and frequency-domain data, and using multi-wide kernel layers, attention mechanisms, and global pooling layers according to the characteristics of one-dimensional vibration data to improve the recognition effect of the model on fault data in a noisy environment.

[0042] In summary, it is an advanced technology application to use the processing ability of the recurrent neural network for time series signals and the processing ability of the convolutional neural network for spatial signals, as well as other efficient information fusion methods based on deep learning technologies to achieve fault diagnosis. A rocket engine fault monitoring method provided by the present invention detects the time series and spatial relationship data of different structures of a liquid rocket engine through sensors, designs a relevant neural network structure for supervised learning based on this, extracts and generalizes features from the detection data of each structure of the liquid rocket engine, and can achieve high-fidelity prediction of the operating states of each structure of the liquid rocket engine, broaden the fault monitoring range, and reduce the occurrence of potential faults.

[0043] Next, the technical solution of the present invention will be described in detail with reference to the accompanying drawings:

[0044] Please refer to Figure 2 , Figure 2 , which is a schematic diagram of the main process of a fault monitoring method for a rocket engine provided by the present invention; the execution subject is a platform or terminal device equipped with the technical solution disclosed in the embodiment of the present invention, such as a digital twin platform or an ipad, etc.

[0045] In Figure 2 , the method can be applied to a rocket engine digital twin system, and the method may include the following steps:

[0046] Step 210: Obtain multiple state data of the target test piece; the multiple state data are the full state data of the operation test of the target test piece during the physical test run of the rocket engine.

[0047] Step 220: Extract multiple target features from the multiple state data; the multiple target features at least include spatial features and temporal features.

[0048] Step 230: Input the multiple target features into a pre-trained target fault monitoring model to predict a fault monitoring result.

[0049] In steps 210 to 230, the rocket engine digital twin platform obtains multiple state data of the target test piece; wherein the target test piece is any component on the liquid rocket engine, such as a thrust chamber, a combustion chamber or a cooling chamber, etc.; the state data are the full state data of the operation test of each target test piece monitored by sensors arranged on each target test piece during the physical test run. Further, from the extracted multiple state data, at least spatial features and temporal features are extracted; because the target fault monitoring model is a monitoring model trained based on the spatial features and temporal features in the state data, therefore, data including at least spatial features and temporal features are input into the target fault monitoring model, so that a fault monitoring result including fault prediction can be obtained, broadening the scope of fault monitoring.

[0050] Further, please also refer to Figure 3 , Figure 3 , which is a schematic diagram of the visualization effect of the fault monitoring of a fault monitoring method for a rocket engine provided by the present invention. That is, applying the fault monitoring method for a rocket engine provided by the present invention to a rocket engine digital twin system can form a fault monitoring based on digital twin and obtain a three-dimensional visualization fault monitoring effect.

[0051] In Figure 3In this case, a rocket engine fault monitoring method provided by the present invention is applied to a rocket engine digital twin system. By using finite element simulation technology to train and construct the 3D physical field of the core components, and using the sensor data of finite measurement points to drive the 3D physical field, the full - aspect monitoring and fault prediction of the test are realized, and the scope of fault monitoring is broadened. For example, by using simulation technology plus an engine state fault perception model based on long - short - term neural network, the operation of the test object can be understood more comprehensively. The velocity field and pressure in the flow regulator can be seen, and at the same time, time - series lead prediction can be carried out on the perceived full - state information, broadening the fault monitoring scope of the physical test of liquid rocket engines.

[0052] Based on this, a rocket engine fault monitoring method provided by the present invention is applied to a rocket engine digital twin system. By obtaining multiple state data of the target test piece; extracting multiple target features from the multiple state data; the multiple target features at least include spatial features and time - series features; inputting the multiple target features into a pre - trained target fault monitoring model, the fault monitoring result is predicted; realizing the time - series lead fault monitoring of the rocket engine, broadening the fault monitoring scope of the test components; meeting the stability and safety test requirements of liquid rocket engines.

[0053] Preferably, a sample database needs to be established before step 210. Please refer to Figure 4 , Figure 4 which is a schematic flow chart of establishing a sample database for a rocket engine fault monitoring method provided by the present invention.

[0054] In Figure 4 , the process of establishing a sample database may include:

[0055] Step 410: Obtain the historical test full - state data of the target test piece.

[0056] In step 410, the historical test full - state data generated during the physical test of each test piece can be obtained from the database. These historical data are the data collected by sensors arranged on the finite measurement points of the engine; of course, the test full - state data during the current physical test of the target test piece can also be obtained simultaneously; the full - state data represents the detection data generated by a complete test.

[0057] Step 420: Based on a preset data - driven fault prediction method, perform data pre - processing on the historical test full - state data to obtain multiple target data; the results represented by the multiple target data are the fault data of the target test piece at different time series.

[0058] Step 430: Establish the sample database based on the multiple target data; the multiple target data includes at least spatial features and temporal features.

[0059] In steps 420 to 430, the data can be processed according to the principle of data-driven fault diagnosis to obtain data with fault status identifiers, that is, data representing healthy or faulty states can be obtained; thus, corresponding data samples in different states can be established based on these data; that is, healthy data samples can be established based on the data corresponding to the healthy state, and faulty data samples can be established based on the data corresponding to the faulty state; each sample also includes at least the spatial features and temporal features of the data; thus, a sample database including healthy data samples and faulty data samples is established; so that the target fault monitoring model can be obtained by training the network model with the faulty data samples in the sample database.

[0060] Preferably, in step 420, the data preprocessing of the historical test full-state data based on the preset data-driven fault prediction method to obtain multiple target data may include: performing signal dimensionality reduction processing on the historical test full-state data to obtain multiple target feature vectors; the multiple target feature vectors are low-dimensional feature vectors including state identifiers obtained by performing dimensionality reduction processing on the high-dimensional feature vectors in the historical test full-state data; the dimensionality reduction processing includes at least signal processing, feature extraction, feature dimensionality reduction, and pattern recognition; wherein, the method of signal processing and feature extraction is a method combining the EMD method and the fuzzy entropy theory; performing state classification and recognition on the multiple target feature vectors to obtain the multiple target data.

[0061] Specifically, the goal of data-driven fault diagnosis is to convert the high-dimensional feature vector into a state identifier (low-dimensional feature vector or sensitive feature) with better discriminant performance through four steps of signal processing, feature extraction, feature dimensionality reduction, and pattern recognition, and then input it into the pattern recognition classifier to achieve the identification and classification of fault states, that is, to identify healthy or faulty, and the faults are further divided into multiple failure modes.

[0062] Exemplarily, in the process of performing dimensionality reduction processing on the data, a method combining the EMD (Empirical Mode Decomposition) method and the fuzzy entropy theory can be used for signal processing and feature extraction to obtain the fault information of the system; taking the vibration signal of misaligned rotor as an example, the method of combining the EMD (Empirical Mode Decomposition) method and the fuzzy entropy theory for signal processing and feature extraction may include the following steps to complete the extraction of feature vectors.

[0063] (1) First, the fault signal is decomposed using the EMD method. During the fault decomposition process, the extreme symmetric extension method can be used to improve the endpoint effect problem in the decomposition process, thereby obtaining a better signal decomposition result. The energy principle method is used to eliminate the possible false IMF components in the decomposed IMF components, and the target IMF components after decomposition are obtained.

[0064] (2) Calculate the fuzzy entropy values of the first several order IMF components that contain most of the information in the target IMF component; based on the fuzzy entropy values of the first several order IMF components, form the feature vector of each fault, thereby completing the feature extraction of the fault signal.

[0065] The whole stage is divided into training stage and testing stage.

[0066] Training: Training a fault diagnosis model requires a large amount of historical data, samples, and multi-dimensional features. Some historical data can be labeled based on the current state (e.g., healthy / faulty) using mechanisms or the experience of experienced experts. However, industrial data often lacks labels; it is either entirely healthy data or completely unclear whether it is healthy or faulty data. These two scenarios correspond to supervised and unsupervised learning, respectively, and corresponding algorithms can be used to train a fault diagnosis model.

[0067] Testing phase: Test the current state identification - extract the characteristics of the equipment and put them into the trained fault diagnosis model to calculate the current state identification.

[0068] Furthermore, a training sample set is obtained from a sample database, and training sample feature data in the training sample set is extracted; wherein the training sample feature data at least includes spatial feature data and temporal feature data.

[0069] The network model is trained using the training sample feature data to obtain an initial fault monitoring model; and a test sample set is input into the initial fault monitoring model to obtain an initial fault perception result; the test sample set is a test sample set obtained from a sample database.

[0070] Finally, the initial fault perception result is compared with the actual result, and the initial fault monitoring model is adjusted based on the error function until the error function converges, thereby obtaining the target fault monitoring model.

[0071] Specifically, an algorithm based on a transfer learning algorithm as the core can be used to train the fault monitoring model to obtain the target fault monitoring model, so that the target fault monitoring model can predict spatial features and temporal features, and fuse the spatial feature prediction data and the temporal feature prediction data to obtain a better classification effect of the fault monitoring result; among them, the core of the transfer learning algorithm lies in the mechanism of knowledge transfer.

[0072] We adopt the transfer AdaBoost learning framework TrAdaBoost, which extends the transfer learning of AdaBoost. AdaBoost (Freund & Schapire, 1997) is a learning framework designed to improve the accuracy of weak learners by carefully adjusting the weights of training instances and accordingly learning classifiers. When using the transfer AdaBoost learning framework TrAdaBoost, in each iteration, if a training instance of a different distribution is mispredicted, then this instance is likely to conflict with the training data of the same distribution; then, we reduce its training weight by multiplying its weight by to reduce the effect of its conflict, where Therefore, in the next round, the influence of misclassified training instances of different distributions that are different from the training instances of the same distribution on the learning process will be less than that in the current round. After multiple iterations, the training weights of training instances of different distributions that fit well with the training instances of the same distribution are larger, while the training weights of training instances of different distributions that are not similar to the training instances of the same distribution are smaller; among them, the posture with a larger training weight will help the learning algorithm train a better classifier.

[0073] It should be noted that the above steps need to be completed before calling the fault monitoring model, that is, the establishment of the target fault monitoring model is completed before step 230 and stored in the rocket engine digital twin system.

[0074] Preferably, in step 230, the inputting of the multiple target feature data into the pre-trained target fault monitoring model to predict the fault monitoring result may include: inputting the spatial features into the pre-trained target fault monitoring model to obtain spatial feature prediction data; inputting the temporal features into the pre-trained target fault monitoring model to obtain temporal feature prediction data; fusing the spatial feature prediction data and the temporal feature prediction data to obtain the fault monitoring result; the data fusion includes at least data splicing and data weighting.

[0075] Specifically, the target fault monitoring model can be used to process the input spatial features and temporal features, and the Softmax layer can be used to weight and predict the extracted features to obtain spatial feature prediction data and temporal feature prediction data. Then, the spatial feature prediction data and the temporal feature prediction data are fused, such as data weighting or data splicing, to obtain the corresponding diagnostic results, achieving the maximum retention of the original features of the target test piece while fully combining spatial and temporal features to predict faults and obtain fault detection results, thereby broadening the scope of fault monitoring.

[0076] In a second aspect, the present invention provides a fault monitoring system for a rocket engine, which is applied to a digital twin platform of a rocket engine. The fault monitoring system uses the fault monitoring method described in the first aspect to monitor faults. Please refer to Figure 5 , Figure 5 , which is a schematic diagram of the main structure of a fault monitoring system for a rocket engine provided by the present invention. In Figure 5 , the system may include: a data acquisition module 510, a feature extraction module

[0076] 520, and a target fault monitoring module Figure 5 530.

[0077] The data acquisition module 510 is used to acquire a plurality of state data of the target test piece; the plurality of state data are all state data of the operation test of the target test piece during the physical test run of the rocket engine. The feature extraction module

[0076] 520 is used to extract a plurality of target features from the plurality of state data; the plurality of target features at least include spatial features and temporal features. The target fault monitoring module Figure 5 530 is used to input the plurality of target features into a pre-trained target fault monitoring model to predict and obtain a fault monitoring result.

[0078] Based on this, a fault monitoring system for a rocket engine provided by the present invention acquires a plurality of state data of the target test piece through the data acquisition module, and then uses the feature extraction module to extract a plurality of target features from the plurality of state data. The plurality of target features at least include spatial features and temporal features. Finally, the target fault monitoring module inputs the plurality of target features into a pre-trained target fault monitoring model to predict and obtain a fault monitoring result, realizing the use of the pre-trained target fault monitoring model to perform time-series advanced prediction on the physical test during the physical test run of the rocket engine, achieving the maximum retention of the original features of the target test piece while fully combining spatial and temporal features to predict faults and obtain fault detection results, thereby broadening the scope of fault monitoring.

[0079] Preferably, the target fault monitoring module Figure 5 530 may include a fault monitoring fusion unit Figure 5 533; the fault monitoring fusion unit Figure 5 533 may be used to fuse the high-dimensional spatial feature data and the high-dimensional temporal feature data to predict and obtain a fault monitoring result. <\

[0080] Specifically, the fault monitoring fusion unit 533 can perform data fusion and splicing based on the spatial features and temporal features sent by the feature extraction module to obtain a diagnosis result; that is, a fault monitoring result with a wider monitoring range than the prior art is obtained.

[0081] Preferably, the target fault monitoring module 530 may include: a data dimensionality reduction unit 531 and a model establishment unit 532. Among them, the data dimensionality reduction unit 531 can be used to perform data dimensionality reduction processing on the multiple state data according to a preset data-driven fault prediction method to obtain multiple target data; the multiple full-state data are the full-state data of the rocket engine's historical test runs; the results represented by the multiple target data are the fault data of the target test piece at different time sequences; the multiple target data include high-dimensional spatial features and high-dimensional temporal features. The model establishment unit 532 can be used to establish a target fault monitoring model based on the multiple target data.

[0082] It should be noted that the method of establishing the target fault monitoring model is the same as the method described in the first aspect and will not be elaborated here. Of course, in a specific application scenario, in order to improve the operation efficiency of the target fault monitoring module, the target fault monitoring module 530 may not need to include a data dimensionality reduction unit and a model establishment unit related to establishing the target fault monitoring model. Instead, a fault monitoring model established according to the method disclosed in the present invention outside the system can be embedded into the target fault monitoring module to realize fault prediction for the full-state data of the operation test of the target test piece during the physical test runs of multiple rocket engines. Of course, the target fault monitoring model can also be established by using the feature extraction unit and the model construction unit in the feature extraction module according to the method described in the first aspect; the present invention does not make specific limitations.

[0083] Furthermore, the feature extraction module 520 includes a spatial feature unit 521 and a first model construction unit 522; among them, the spatial feature unit 521 can be used to extract the high-dimensional spatial features in the multiple state data; the first model construction unit 522 can be used to construct a spatial feature model based on the high-dimensional spatial features of the multiple target data.

[0084] Specifically, according to the known full-state data and the characteristics of the convolutional neural network, the high-dimensional spatial features in the multiple state data can be extracted, and the high-dimensional spatial features are input into the convolutional neural network model for model training to construct a spatial feature model; and the spatial feature model is embedded into the spatial feature unit 521 to realize the extraction of the high-dimensional spatial features in the data collected at the limited measurement points during the physical test of the rocket engine by using the spatial feature unit 521; among them, after each convolutional layer, a normalization layer and an activation function layer are used for feature convergence and feature dimensionality reduction.

[0085] Further, the feature extraction module 520 may further include a temporal feature unit 523 and a second construction unit 524. Among them, the temporal feature unit 523 may be used to extract high-dimensional temporal features from multiple pieces of the state data; the second construction unit 524 may be used to construct a temporal feature model based on the high-dimensional temporal features of multiple pieces of the target data.

[0086] Specifically, the high-dimensional temporal features in multiple pieces of state data can be extracted by using the temporal feature unit; the temporal feature model can be obtained in the following two ways:

[0087] Way 1: Input the temporal features of multiple pieces of the target data into a long short-term memory neural network model for model training to construct a temporal feature model.

[0088] Way 2: Use a residual network model, stack multiple different residual blocks to form a deep residual network model; input the temporal features of multiple pieces of the target data into a long short-term memory neural network model and a deep residual network model for model training to construct a temporal feature model. In this way, by combining the deep residual network composed of stacking multiple different residual blocks with the long short-term memory neural network model for model training, the problem of network degradation generated during the process of increasing the depth of the neural network can be suppressed.

[0089] Embed the temporal feature model into the temporal feature unit 523, so that the high-dimensional temporal features in the data collected at finite measurement points during the physical test of the rocket engine can be extracted by using the temporal feature unit 523.

[0090] It should be noted that the long short-term memory network (LSTM, Long Short-Term Memory) is a type of recurrent neural network in time, which is specifically designed to solve the long-term dependence problem existing in general RNNs (recurrent neural networks). All RNNs have a chained form of repeating neural network modules.

[0091] As a specific example, please refer to Figure 6 , Figure 6 which is a schematic structural diagram for fault monitoring of vibration data in a fault monitoring system of a rocket engine provided by the present invention; that is Figure 6 is a schematic structural diagram for taking the vibration data of the rocket engine as specific state data for fault prediction to obtain a fault detection result.

[0092] In Figure 6Among them, first, vibration data detected at limited measuring points on the rocket engine (thrust chamber) is obtained through an acquisition module. The stacked convolutional layers in the spatial feature unit are used to extract spatial features from the mechanical vibration fault signals of the thrust chamber, obtaining high-dimensional spatial features. Among them, after each convolutional layer, a normalization layer and an activation function layer are used for feature convergence and feature dimensionality reduction. At the same time, according to the temporal characteristics of the vibration data, a temporal feature model established based on a residual network and a long short-term memory network in the temporal feature unit is used to extract temporal features from the mechanical vibration fault signals of the thrust chamber, obtaining high-dimensional temporal features. Finally, a fault monitoring fusion unit performs feature fusion on the high-dimensional spatial features and high-dimensional temporal features, and the extracted features are weighted through a Softmax layer to obtain the corresponding diagnostic results, thereby realizing the diagnosis of faults by fully combining spatial and temporal features while retaining the original features of the mechanical vibration of the test components to the greatest extent, and broadening the fault monitoring range of the test components.

[0093] Although the present invention has been described in connection with various embodiments, however, in the process of implementing the claimed invention, those skilled in the art can understand and achieve other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the appended claims. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "one" does not exclude a plurality. A single processor or other unit can implement several functions recited in the claims. Certain measures are recited in mutually different dependent claims, but this does not mean that these measures cannot be combined to produce good results.

[0094] Although the present invention has been described in connection with specific features and their embodiments, it is obvious that various modifications and combinations can be made without departing from the spirit and scope of the present invention. Accordingly, this specification and the drawings are merely exemplary illustrations of the invention defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the present invention. Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

Claims

1. A fault monitoring method for a rocket engine, applied to a digital twin system of a rocket engine, characterized in that Including: Obtaining multiple state data of a target test piece; The multiple state data are the full state data of the operation test of the target test piece during the physical test run of the rocket engine; Extracting multiple target features from the multiple state data; The multiple target features at least include spatial features and temporal features; Inputting the multiple target features into a pre-trained target fault monitoring model to obtain a fault monitoring result.

2. The method according to claim 1, wherein Before the step of inputting the multiple target feature data into the pre-trained target fault monitoring model, it includes: Obtaining a training sample set from a sample database; Extracting training sample feature data from the training sample set; the training sample feature data at least includes spatial feature data and temporal feature data; Using the training sample feature data to perform model training on a network model to obtain an initial fault monitoring model; Inputting a test sample set into the initial fault monitoring model to obtain an initial fault perception result; the test sample set is a test sample set obtained from the sample database; Comparing the initial fault perception result with the actual result, and adjusting the initial fault monitoring model based on an error function until the error function converges, to obtain the target fault monitoring model.

3. The method according to claim 2, characterized in that, Before the step of obtaining the training sample set from the sample database, it includes: Establishing a sample database; The establishment of the sample database includes: Obtaining the historical full state data of the target test piece; Performing data preprocessing on the historical full state data based on a preset data-driven fault prediction method to obtain multiple target data; the results represented by the multiple target data are the fault data of the target test piece at different time sequences; Based on the multiple target data, establishing the sample database; the multiple target data at least includes spatial features and temporal features.

4. The method according to claim 3, wherein The step of performing data preprocessing on the historical full state data based on a preset data-driven fault prediction method to obtain multiple target data includes: Performing signal dimensionality reduction processing on the historical full state data to obtain multiple target feature vectors; the multiple target feature vectors are low-dimensional feature vectors including state identifiers obtained by performing dimensionality reduction processing on the high-dimensional feature vectors in the historical full state data; the dimensionality reduction processing at least includes signal processing, feature extraction, feature dimensionality reduction, and pattern recognition; among them, the method of signal processing and feature extraction is a method combining the EMD method and the fuzzy entropy theory; Performing state classification and recognition on the multiple target feature vectors to obtain the multiple target data.

5. The method according to claim 1, wherein The step of inputting the multiple target feature data into the pre-trained target fault monitoring model to obtain a fault monitoring result includes: Inputting the spatial features into the pre-trained target fault monitoring model to obtain spatial feature prediction data; Inputting the temporal features into the pre-trained target fault monitoring model to obtain temporal feature prediction data; Performing data fusion on the spatial feature prediction data and the temporal feature prediction data to obtain the fault monitoring result; the data fusion at least includes data splicing and data weighting.

6. A fault monitoring system for a rocket engine, applied to a digital twin platform of a rocket engine, characterized in that The fault monitoring system adopts the fault monitoring method described in any one of claims 1 to 5 for fault monitoring; the system includes: a data acquisition module, a feature extraction module, and a target fault monitoring module; The data acquisition module is used to acquire a plurality of state data of the target test piece; the plurality of state data are all state data of the operation test of the target test piece during the physical test run of the rocket engine; The feature extraction module is used to extract a plurality of target features from the plurality of state data; the plurality of target features at least include spatial features and temporal features; The target fault monitoring module is used to input the plurality of target features into a pre-trained target fault monitoring model to predict a fault monitoring result.

7. The system according to claim 6, wherein The target fault monitoring module includes: a data dimensionality reduction unit, which is used to perform data dimensionality reduction processing on the plurality of all state data according to a preset prediction method for data-driven faults to obtain a plurality of target data; the plurality of all state data are all state data of the historical test run of the rocket engine; the results represented by the plurality of target data are the fault data of the target test piece at different time series; the plurality of target data include high-dimensional spatial features and high-dimensional temporal features; a building unit, which is used to build a target fault monitoring model based on the plurality of target data.

8. The system according to claim 6, wherein The feature extraction module includes a spatial feature unit and a first construction unit; The spatial feature unit is used to extract high-dimensional spatial features from the plurality of state data; The first construction unit is used to build a spatial feature model based on the high-dimensional spatial features of the plurality of target data.

9. The system according to claim 6, wherein The feature extraction module includes a temporal feature unit and a second construction unit; The temporal feature unit is used to extract high-dimensional temporal features from the plurality of state data; The second construction unit is used to build a temporal feature model based on the high-dimensional temporal features of the plurality of target data.

10. The system according to claim 6, wherein The target fault monitoring module includes a fault monitoring fusion unit; The fault monitoring fusion unit is used to perform fusion processing on the high-dimensional spatial feature data and the high-dimensional temporal feature data to predict a fault monitoring result.

Images