Diagnostic system combining real-time liquid rocket engine fault simulation model and artificial intelligence algorithm

Abstract

The invention particularly relates to a diagnosis system combining a real-time liquid rocket engine fault simulation model and an artificial intelligence algorithm, and belongs to the field of liquid rocket engine fault diagnosis. According to the system, models of parts such as a pump and a turbine are established according to physical characteristics of all parts of an engine, a dynamic system model of a three-layer architecture is constructed, normal data containing random noise are generated through a Broyden iteration method, and fault data are obtained through a fault factor injection method; and then an LSTM-Lightgbm fault diagnosis algorithm is established, and the data is divided into a training set and a test set to complete model training and testing. According to the method, the problems of limited feature extraction capability, insufficient generalization capability and the like of a traditional diagnosis technology are solved, the method has diagnosis accuracy, real-time performance and comprehensiveness, the accuracy of a training set and the accuracy of a test set reach 99.58% and 99.31% respectively, the functions of fault early warning, type positioning and the like can be achieved, and the engineering practicability is high.

Description

Technical Field

[0001] This invention relates to the field of liquid rocket engine fault diagnosis, and specifically to a diagnostic system that combines a real-time liquid rocket engine fault simulation model with artificial intelligence algorithms. Background Technology

[0002] The core design of traditional liquid rocket engine fault diagnosis systems relies on fixed thresholds or simplified physical models to achieve basic fault identification and alarms. However, as space missions increasingly demand reusability, thrust adjustment range, and reliability from engines, single fault identification functions are insufficient to meet multi-dimensional technical requirements. In complex scenarios such as long-duration test runs, variable thrust conditions, and repeated launches, it is necessary not only to accurately locate the fault type but also to provide early warnings, severity assessments, and propagation path tracing. In these situations, the real-time performance, accuracy, and comprehensiveness of fault diagnosis become paramount. To adapt to the stringent requirements of modern space missions, fault diagnosis systems integrating real-time status awareness and deep intelligent analysis have attracted widespread attention from academia and industry.

[0003] To address the core requirement of fault diagnosis in liquid rocket engines, several classic technical solutions exist, such as diagnostic methods based on physical analytical models, traditional statistical signal processing methods, and basic machine learning algorithms. However, these diagnostic techniques have significant limitations: traditional statistical methods have limited ability to extract features from early, minor faults and struggle to adapt to parameter fluctuations under varying operating conditions; basic machine learning algorithms are limited by the scarcity of fault samples and uneven data distribution, resulting in insufficient generalization ability and an inability to achieve real-time dynamic tracking of faults. Therefore, it is necessary to design a diagnostic system that combines a real-time liquid rocket engine fault simulation model with artificial intelligence algorithms. This system would recreate the changes in fault parameters under complex operating conditions through simulation models and achieve fault diagnosis through intelligent algorithms.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] Existing liquid rocket engine fault diagnosis technologies suffer from limitations in extracting early, minor fault features, difficulty adapting to fluctuations in operating parameters, insufficient generalization ability due to scarce fault samples and uneven data distribution, and inability to achieve real-time dynamic fault tracking. These limitations make it difficult to meet the real-time, accurate, and comprehensive fault diagnosis requirements of modern aerospace missions. This invention provides a diagnostic system that combines a real-time liquid rocket engine fault simulation model with artificial intelligence algorithms, thereby overcoming, to some extent, the shortcomings of existing technologies.

[0006] Other features and advantages of the invention will become apparent from the following detailed description, or may be learned in part by practice of the invention.

[0007] According to a first aspect of the present invention, a diagnostic system combining a real-time liquid rocket engine fault simulation model and an artificial intelligence algorithm is provided, the system being constructed using the following steps: Based on the physical characteristics of each engine component, models of each engine component are established, including pumps, turbines, pre-combustion chambers, pipelines, valves, and combustion chambers. The components are connected according to the engine flow diagram. In a simulation software environment, a three-level architecture engine dynamic system model is constructed by combining the models of various engine components. The three-level architecture includes models of various engine components, a solver, and an integrator. The solver consists of a Broyden solver and common working equations of various components. The integrator is used to calculate dynamic and steady-state data in the engine dynamic system model. An iterator was designed using the Broyden iteration method to generate normal data from an engine dynamic system model, and real engine data was simulated by adding uniformly distributed random noise. Fault-free data is generated using an engine fault simulation model, and fault data is obtained through simulation using the fault factor injection method. A fault diagnosis model based on gradient boosting decision tree Long Short-Term Memory Neural Network LSTM-Lightgbm is constructed. The LSTM-Lightgbm network is composed of a Long Short-Term Memory network and a Lightweight Gradient Boosting Decision Tree connected in series. Engine data under different faults are used as training and testing sets. After training the fault diagnosis model using the training set, the fault diagnosis model is tested using the testing set.

[0008] In some exemplary embodiments, the engine pump model involves parameters including the pump head. Pump efficiency and pump power The expressions are as follows:

[0009]

[0010]

[0011] in For pump outlet pressure, The pump inlet pressure, The average density of the propellant inside the pump. , , This is the pressure increase coefficient. For pump speed, This refers to the propellant mass flow rate within the pump. , , This is the pump efficiency coefficient.

[0012] In some exemplary embodiments, the engine turbine model involves metrics including the isentropic expansion work per unit working fluid in the turbine. Turbine efficiency Turbine tangential speed Theoretical injection speed of turbine gas and turbine power The expressions are as follows:

[0013]

[0014]

[0015]

[0016]

[0017] in For the specific heat capacity of the gas at constant pressure, The total temperature of the combustion gas at the turbine inlet. The total temperature of the exhaust gas at the turbine outlet. The adiabatic index of the gas. Let be the gas constant of the fuel gas. This represents the total pressure at the turbine inlet. The total pressure at the turbine outlet. , , The turbine efficiency coefficient. The diameter of the turbine blade. This refers to the turbine speed.

[0018] In some exemplary embodiments, the engine fluid piping model involves parameters including the pressure difference between the fluid piping inlet and outlet. Their expressions are respectively

[0019] in For the pipeline inlet pressure, For pipeline outlet pressure, The flow resistance coefficient of the pipeline. This refers to the density of the liquid propellant.

[0020] In some exemplary embodiments, the engine pre-combustion chamber model involves indicators including the air-fuel mixture ratio in the pre-combustion chamber. and the pressure of the pre-combustion chamber Their expressions are as follows:

[0021]

[0022] in This refers to the oxidant mass flow rate in the pre-combustion chamber. This refers to the fuel mass flow rate in the pre-combustion chamber. This is the theoretical characteristic velocity of the pre-combustion chamber. This represents the thrust coefficient of the pre-combustion chamber.

[0023] In some exemplary embodiments, the engine valve model involves indicators including valve flow rate. Relative opening of valve Their expressions are as follows:

[0024]

[0025] in The flow coefficient of the valve. This refers to the flow cross-sectional area of ​​the valve. The valve inlet pressure. For valve outlet pressure, It is the product of the valve's rated flow area and flow coefficient.

[0026] In some exemplary embodiments, the engine combustion chamber model involves metrics including the combustion chamber air-fuel ratio. and combustion chamber pressure Their expressions are as follows:

[0027]

[0028] in This refers to the oxidant mass flow rate in the combustion chamber. This refers to the fuel mass flow rate in the combustion chamber. This is the theoretical characteristic velocity of the combustion chamber. This represents the combustion chamber thrust coefficient.

[0029] In some exemplary embodiments, the components in the three-tiered engine dynamic system model satisfy a common working equation, the specific expression of which is:

[0030] in, The methane flow rate in the oxygen-enriched pre-combustion chamber, To regulate the outlet flow rate of the methane control valve, The oxygen flow rate in the oxygen-enriched combustion chamber. This refers to the outlet flow rate of the oxygen main valve. For the methane flow rate in the fuel-rich pre-combustion chamber, This represents the methane main valve outlet flow rate. For the oxygen flow rate in the rich combustion chamber, To regulate the outlet flow of the oxygen control valve, The speed of the methane turbine pump. For oxygen turbine pump speed, The moment of inertia of the methane turbopump. For methane turbine power, For methane pump power, The moment of inertia of the oxygen turbine pump. For oxygen turbine power, This refers to the oxygen pump power.

[0031] In some exemplary embodiments, the fault data obtained through simulation using the fault factor injection method is mathematically described as follows:

[0032] in, X= These are the state parameters of the component model. U = These are the input parameters for the component model and the input interface for the module. D For component model parameters; t For time; The output parameters of the component model, and the output interface of the module; , The functional relationships of the component model; It is a fault factor.

[0033] In some exemplary embodiments, engine data under different faults are used as training and testing sets. After training the fault diagnosis model using the training set, the fault diagnosis model is tested using the testing set. Specifically: Engine data under different faults are divided into training and testing sets. Each data sample in the dataset contains two aspects of information: measurement parameters and fault type label. In the fault diagnosis model, the LSTM network extracts the temporal features, trend changes and dynamic patterns of the data. Based on this, Lightgbm classifies the corresponding data samples into the corresponding fault categories according to the information extracted by LSTM, thereby realizing the fault diagnosis function. After training the fault diagnosis model using training set data samples, the measurement parameters of the test set data samples are then input into the fault diagnosis model to perform fault diagnosis, output the fault type, and compare it with the fault type of the test set data samples to obtain the diagnostic accuracy index.

[0034] The diagnostic system combining a real-time liquid rocket engine fault simulation model and an artificial intelligence algorithm provided in the embodiments of the present invention first establishes models of each component based on the physical characteristics of each engine component. Then, a three-tiered dynamic system model of the engine is constructed using these models. Static and dynamic data are generated using the Broyden iteration method, and fault data is obtained through a fault factor injection method. Finally, an LSTM-LightGBM fault diagnosis algorithm is built, and the generated normal and fault data are proportionally divided into training and testing sets for training and testing, thus realizing the simulation and diagnosis of liquid rocket engine faults. Compared with the prior art, it has the following beneficial effects: 1. Significantly Improved Diagnostic Accuracy: This invention innovatively adopts an LSTM-LightGBM cascaded architecture. The LSTM layer accurately captures the long-term dependencies of multi-dimensional time-series engine data, automatically extracting high-dimensional feature vectors containing dynamic patterns without manual feature design. The LightGBM algorithm then efficiently processes these high-dimensional sparse features, significantly improving fault classification accuracy through gradient iteration optimization and efficient node splitting strategies. Experimental results show that the model achieves a training set accuracy of 99.58% and a test set accuracy of 99.31%, significantly outperforming traditional SVM (test set accuracy 97.22%) and a single LSTM model (test set accuracy 95.14%). It can accurately identify normal states and various faults, effectively reducing misclassification and missed detections.

[0035] 2. Real-time Response Optimization: A three-tiered engine dynamic system model is constructed, employing the Broyden solver and the common working equations of each component to form the core solution unit. Compared to the traditional Newton-Raphson iterator, this significantly shortens the convergence time and improves the model's computational efficiency. Simultaneously, the Broyden iteration method rapidly generates normal data and simulates operational data under real-world noise conditions, ensuring the real-time nature of the fault diagnosis process. This system is adaptable to complex scenarios such as long-duration test runs, variable thrust conditions, and repeated launches, meeting the core requirements of early fault warning and real-time dynamic tracking.

[0036] 3. Enhanced data adaptability and generalization ability: By using the fault factor injection method, different degrees and types of fault states are accurately simulated in the engine fault simulation model, effectively solving the pain points of scarce fault samples and uneven data distribution in traditional diagnostic techniques, and generating sufficient and comprehensive training data; combined with the strong processing capability of the LSTM-Lightgbm algorithm for high-dimensional data, the system can still maintain stable performance under varying operating conditions and parameter fluctuations, and its generalization ability is significantly better than traditional statistical methods and basic machine learning algorithms.

[0037] 4. Comprehensive Fault Diagnosis: Breaking through the limitations of traditional fault diagnosis which can only achieve basic identification and alarm, through refined component models (pumps, turbines, pre-combustion chambers, etc.) and dynamic system models, it can restore the changing patterns of fault parameters under complex operating conditions. It can not only accurately locate the fault type, but also assess the severity of the fault and trace the propagation path, providing a comprehensive and reliable basis for engine maintenance and repair, and helping to improve the reusability and operational reliability of the engine.

[0038] 5. Strong Engineering Practicality: The overall system design balances theoretical rigor with engineering feasibility. Each component model is built based on physical characteristics, with precise mathematical expressions that closely match real-world engineering scenarios. Algorithm hyperparameters are clearly configured (e.g., the number of LSTM layer units, the number of LightGBM leaf nodes), and training and testing processes are standardized (data is divided in a 10:4 ratio), facilitating engineering implementation and widespread application. Furthermore, the model possesses excellent scalability, allowing for flexible adjustment of component model parameters and algorithm hyperparameters based on the structural characteristics and operational requirements of different types of liquid rocket engines, adapting to diverse aerospace mission needs.

[0039] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Attached Figure Description

[0040] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0041] Figure 1 This is a flowchart of a diagnostic system that combines a real-time liquid rocket engine fault simulation model and an artificial intelligence algorithm, according to the present invention. Figure 2 This is the engine flow path diagram of the present invention; Figure 3 This is the engine dynamic system model of the present invention; Figure 4This is a comparison chart of the convergence times of the Broyden iterator and the Newton-Raphson iterator of this invention; Figure 5 This is a simulation diagram of normal data for the model part of the present invention; Figure 6 This is a diagram of fault simulation data for the model of the present invention; Figure 7 This is a basic structural diagram of the LSTM memory cell of the present invention; Figure 8 This is a schematic diagram of the LightGBM optimal priority tree growth principle of the present invention; Figure 9 This is a diagram showing the confusion matrix of the training set of the LSTM-LightGBM model of this invention. Figure 10 This is a diagram showing the confusion matrix results of the test set for the LSTM-LightGBM model of this invention. Detailed Implementation

[0042] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the invention will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0043] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0044] To address the shortcomings and deficiencies of existing technologies, this exemplary embodiment provides a diagnostic system that combines a real-time liquid rocket engine fault simulation model with artificial intelligence algorithms, referencing... Figure 1 As shown, the construction of the system may specifically include the following steps: Step S1: Based on the physical characteristics of each engine component, establish models of each engine component, including pumps, turbines, pre-combustion chambers, pipelines, valves, and combustion chambers, and connect each component according to the engine flow diagram; Step S2: In the simulation software environment, a three-level architecture engine dynamic system model is constructed by combining the models of each engine component. The three-level architecture includes the models of each engine component, a solver, and an integrator. The solver consists of the Broyden solver and the common working equations of each component. The integrator is used to calculate the dynamic data and steady-state data in the engine dynamic system model. Step S3: Use the Broyden iteration method to design an iterator, generate normal data through the engine dynamic system model, and simulate real engine data by adding uniformly distributed random noise. Step S4: Generate fault-free data using the engine fault simulation model, and obtain fault data through the fault factor injection method. Step S5: Build a fault diagnosis model of a long short-term memory neural network LSTM-Lightgbm based on gradient boosting decision tree. The LSTM-Lightgbm network is composed of a long short-term memory network and a lightweight gradient boosting decision tree connected in series. Step S6: Divide the engine data under different faults into training set and test set. After training the fault diagnosis model using the training set, test the fault diagnosis model using the test set.

[0045] The following will describe in more detail the various steps of system construction in this example embodiment with reference to the accompanying drawings and embodiments.

[0046] In step S1, models of each engine component are established based on their physical characteristics, including components such as pumps, turbines, pre-combustion chambers, pipelines, valves, and combustion chambers. The components are then connected according to the engine flow diagram, as shown in the engine flow diagram. Figure 2 As shown.

[0047] The engine pump model mentioned above involves indicators including pump head. Pump efficiency and pump power The pump head is defined as the energy gained by a unit mass flow rate of fluid passing through the impeller, expressed as pressure increase. The expressions are as follows: (1) (2) (3) in For pump outlet pressure, The pump inlet pressure, The average density of the propellant inside the pump. , , This is the pressure increase coefficient. For pump speed, This refers to the propellant mass flow rate within the pump. , , This refers to the pump efficiency coefficient; The aforementioned engine turbine model involves indicators including the isentropic expansion work per unit working fluid in the turbine. Turbine efficiency Turbine tangential speed Theoretical injection speed of turbine gas and turbine power The expressions are as follows: (4) (5) (6) (7) (8) in For the specific heat capacity of the gas at constant pressure, The total temperature of the combustion gas at the turbine inlet. The total temperature of the exhaust gas at the turbine outlet. The adiabatic index of the gas. Let be the gas constant of the fuel gas. This represents the total pressure at the turbine inlet. The total pressure at the turbine outlet. , , The turbine efficiency coefficient. The diameter of the turbine blade. Turbine speed; The engine fluid piping model mentioned above involves indicators including the pressure difference between the inlet and outlet of the fluid piping. Their expressions are respectively (9) in For the pipeline inlet pressure, For pipeline outlet pressure, The flow resistance coefficient of the pipeline. Density of the liquid propellant; The aforementioned engine pre-combustion chamber model involves indicators including the air-fuel mixture ratio in the pre-combustion chamber. and the pressure of the pre-combustion chamber Their expressions are as follows: (10) (11) in This refers to the oxidant mass flow rate in the pre-combustion chamber. This refers to the fuel mass flow rate in the pre-combustion chamber. This is the theoretical characteristic velocity of the pre-combustion chamber. This refers to the pre-combustion chamber thrust coefficient; The engine valve model mentioned above involves indicators including valve flow rate. Relative opening of valve Their expressions are as follows: (12) (13) in The flow coefficient of the valve. This refers to the flow cross-sectional area of ​​the valve. The valve inlet pressure. For valve outlet pressure, This is the product of the valve's rated flow area and its flow coefficient. The engine combustion chamber model mentioned above involves indicators including the combustion chamber air-fuel ratio. and combustion chamber pressure Their expressions are as follows: (14) (15) in This refers to the oxidant mass flow rate in the combustion chamber. This refers to the fuel mass flow rate in the combustion chamber. This is the theoretical characteristic velocity of the combustion chamber. This represents the combustion chamber thrust coefficient.

[0048] In step S2, a three-tiered engine dynamic system model is constructed in the simulation software environment, combining models of various engine components, including models of each engine component, a solver, and an integrator, such as... Figure 3 The image shows the established dynamic model of the engine.

[0049] The solver consists of the Broyden solver and the common working equations of each component. The integrator calculates the dynamic and steady-state data in the engine dynamic system model. The components described above satisfy a common working equation, the specific expression of which is: (16) in, The methane flow rate in the oxygen-enriched pre-combustion chamber, To regulate the outlet flow rate of the methane control valve, The oxygen flow rate in the oxygen-enriched combustion chamber. This refers to the outlet flow rate of the oxygen main valve. For the methane flow rate in the fuel-rich pre-combustion chamber, This represents the methane main valve outlet flow rate. For the oxygen flow rate in the rich combustion chamber, To regulate the outlet flow of the oxygen control valve, The speed of the methane turbine pump. For oxygen turbine pump speed, The moment of inertia of the methane turbopump. For methane turbine power, For methane pump power, The moment of inertia of the oxygen turbine pump. For oxygen turbine power, This refers to the oxygen pump power.

[0050] In step S3, as Figure 4 As shown, the computation time using the Broyden solver and the Newton-Raphson solver were compared, demonstrating that the Broyden solver can effectively reduce computation time. Subsequently, an iterator was designed using the Broyden iteration method to generate normal data from the engine dynamic system model. Real engine data was simulated by adding uniformly distributed random noise in the range [0, 0.005). Figure 5 The image shown is a simulation diagram of normal data for the model portion of this invention; The Broyden iteration method, specifically expressed as follows: (17) in This refers to the mass flow rate in the engine dynamic system model. This represents the residual vector of the engine dynamic system model. For residual vectors exist The approximation of the Jacobian matrix at the given time step; the Broyden solver completes the solution of the model equilibrium equation at the current time step, and inputs the flow rate calculated at the current time step into the engine model to further obtain the output of the engine model.

[0051] In step S4, fault-free data is generated using the engine fault simulation model, and fault data is obtained through simulation using the fault factor injection method, such as... Figure 6 The figure shown is a fault simulation data diagram of the model of the present invention, which includes one fault of six different fault levels.

[0052] When the mathematical model of a component remains unchanged, but the model parameters change, the model parameters can be multiplied by a certain coefficient (called a fault factor) to characterize the impact of faults in the engine component. Furthermore, by modifying the magnitude of the fault factor, the degree of engine component failure can be simulated. Here, we illustrate this with an example of a fault form incorporating a fault factor. When an engine component fails, its mathematical description is: (18) in, X= State parameters of the component model U = These are the input parameters for the component model and the input interface for the module. D For component model parameters; t For time; The output parameters of the component model, and the output interface of the module; , The functional relationships of the component model; It is a fault factor.

[0053] In step 5, a fault diagnosis model based on gradient boosting decision tree Long Short-Term Memory Neural Network (LSTM-Lightgbm) is built.

[0054] The LSTM-LightGBM network consists of a Long Short-Term Memory network cascaded with a Lightweight Gradient Boosting Decision Tree. The LSTM layers capture the long-term dependencies of the engine's multi-dimensional data. The model input is a standardized feature sequence processed by a sliding window. (M is the window length, (where is the feature dimension), and the output is the fault classification for the current window.

[0055] like Figure 7 The diagram shows the basic structure of an LSTM memory unit. LSTM selectively retains or forgets temporal information through a "gating mechanism." Its core components include a forget gate, input gate, cell state, and output gate, enabling it to capture long-range dependencies in a sequence. The LSTM gating mechanism update formula is as follows: (19) in It represents the Hadamah accumulation. This is the weight matrix. This is a bias term.

[0056] like Figure 8The diagram illustrates the tree growth principle of the Light Gradient Boosting Machine (LightGBM). LightGBM is an improved algorithm based on Gradient Boosting Decision Tree (GBDT), which excels at handling high-dimensional sparse features and improves training speed and classification accuracy through an efficient node splitting strategy. Based on the gradients of the previous model, multiple decision trees are iteratively trained, and the results of all trees are ultimately converted into class probabilities.

[0057] LSTM extracts features and inputs them into the LightGBM model. LightGBM calculates the probability of each feature parameter in each class and iterates continuously until the maximum number of iterations is reached. The probability calculation formula is: (20) In the formula: This represents the number of iterations. This represents the overall category in a classification problem.

[0058] Then, the negative gradient of the feature parameters is calculated. In each iteration, the negative gradient is the target that the new tree needs to fit, and its calculation formula is: (twenty one) In the formula: This represents the true probability of the polarization feature parameters. After obtaining the negative gradient value, it is necessary to calculate the leaf node values ​​after model splitting. The node values ​​determine how samples are divided into different leaf nodes, directly affecting the performance of the LSTM-LightGBM fault diagnosis model. The calculation formula is: (twenty two) In the formula: This represents the number of leaf nodes; This is the set of samples at the leaf nodes.

[0059] After obtaining the values ​​of the split leaf nodes, the LSTM-Lightgbm fault diagnosis model is updated and optimized according to equation (22): (twenty three) In the formula: The learning rate set for the model; The index of a sample set for a given leaf node; This represents the number of leaf nodes in the decision tree.

[0060] The final LSTM-LightGBM fault diagnosis model is in the following form: (twenty four) In the formula: The total number of decision trees constructed.

[0061] In the LSTM-LightGBM network, the network hyperparameters are shown in Table 1. LSTM is responsible for temporal feature extraction, and LightGBM is responsible for nonlinear classification. LSTM captures the dynamic dependencies in temporal data, encoding high-dimensional temporal signals into feature vectors containing "dynamic patterns," eliminating the need for manual design of temporal features.

[0062] Table 1 Network Hyperparameter Settings

[0063] In step 6, after generating normal and fault data using the constructed dynamic model, the engine data under different faults are divided into training and test sets in a 10:4 ratio. Each data sample in the dataset contains two aspects: measurement parameters and fault type label. In the diagnostic model, the LSTM network extracts the temporal features, trend changes, and dynamic patterns of the data. Based on this, LightGBM classifies the corresponding data samples into the corresponding fault categories according to the information extracted by the LSTM, thus realizing the fault diagnosis function. After training the diagnostic model with the training set data samples, the measurement parameters of the test set data samples are input into the model for fault diagnosis, and the model outputs the fault type. The model is then compared with the fault type of the test set data samples to obtain the diagnostic accuracy index.

[0064] Table 2 Comparison of Intelligent Fault Diagnosis Algorithms

[0065] The real-time model proposed in this invention was used to generate engine data under normal and fault conditions, and fault diagnosis was performed using Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and LSTM-Lightgbm, respectively. The comparison results of the three algorithms are shown in Table 2. LSTM-Lightgbm has the highest accuracy on both the training and test sets under the same test data compared to the other algorithms.

[0066] The verification results of the LSTM-LightGBM network are as follows: Figure 9 , Figure 10 As shown, where Figure 9 The simulation results show the confusion matrix of the training set. A small portion of normal data in the training set was incorrectly classified as oxygen main valve faults, while all other fault types were correctly diagnosed, achieving a final accuracy of 99.58%. The network was then tested using the test set, and the results are as follows: Figure 10 As shown, only one sample with an oxygen main valve malfunction was incorrectly classified into the normal data category, resulting in an accuracy rate of 99.31%.

[0067] It should be noted that the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of the present invention, and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Furthermore, it is readily understood that these processes may, for example, be executed synchronously or asynchronously in multiple modules.

[0068] It should be noted that although several modules or units of the device for performing actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0069] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention herein. This application is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0070] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is defined only by the appended claims.

Claims

1. A diagnostic system combining a real-time liquid rocket engine fault simulation model and artificial intelligence algorithms, characterized in that, The system is constructed using the following steps: Based on the physical characteristics of each engine component, models of each engine component are established, including pumps, turbines, pre-combustion chambers, pipelines, valves, and combustion chambers. The components are connected according to the engine flow diagram. In a simulation software environment, a three-level architecture engine dynamic system model is constructed by combining the models of various engine components. The three-level architecture includes models of various engine components, a solver, and an integrator. The solver consists of a Broyden solver and common working equations of various components. The integrator is used to calculate dynamic and steady-state data in the engine dynamic system model. An iterator was designed using the Broyden iteration method to generate normal data from an engine dynamic system model, and real engine data was simulated by adding uniformly distributed random noise. Fault-free data is generated using an engine fault simulation model, and fault data is obtained through simulation using the fault factor injection method. A fault diagnosis model based on gradient boosting decision tree Long Short-Term Memory Neural Network LSTM-Lightgbm is constructed. The LSTM-Lightgbm network is composed of a Long Short-Term Memory network and a Lightweight Gradient Boosting Decision Tree connected in series. Engine data under different faults are used as training and testing sets. After training the fault diagnosis model using the training set, the fault diagnosis model is tested using the testing set.

2. The diagnostic system according to claim 1, characterized in that, The engine pump model mentioned above involves indicators including pump head. Pump efficiency and pump power The expressions are as follows: in For pump outlet pressure, The pump inlet pressure, The average density of the propellant inside the pump. , , This is the pressure increase coefficient. For pump speed, This refers to the propellant mass flow rate within the pump. , , This is the pump efficiency coefficient.

3. The diagnostic system according to claim 1, characterized in that, The engine turbine model mentioned above involves indicators including the isentropic expansion work per unit working fluid in the turbine. Turbine efficiency Turbine tangential speed Theoretical injection speed of turbine gas and turbine power The expressions are as follows: in For the specific heat capacity of the gas at constant pressure, The total temperature of the combustion gas at the turbine inlet. The total temperature of the exhaust gas at the turbine outlet. The adiabatic index of the gas. Let be the gas constant of the fuel gas. This represents the total pressure at the turbine inlet. The total pressure at the turbine outlet. , , The turbine efficiency coefficient. The diameter of the turbine blade. This refers to the turbine speed.

4. The diagnostic system according to claim 1, characterized in that, The engine fluid piping model mentioned above involves indicators including the pressure difference between the inlet and outlet of the fluid piping. Their expressions are respectively in For the pipeline inlet pressure, For pipeline outlet pressure, The flow resistance coefficient of the pipeline. This refers to the density of the liquid propellant.

5. The diagnostic system according to claim 1, characterized in that, The aforementioned engine pre-combustion chamber model involves indicators including the air-fuel mixture ratio in the pre-combustion chamber. and the pressure of the pre-combustion chamber Their expressions are as follows: in This refers to the oxidant mass flow rate in the pre-combustion chamber. This refers to the fuel mass flow rate in the pre-combustion chamber. This is the theoretical characteristic velocity of the pre-combustion chamber. This represents the thrust coefficient of the pre-combustion chamber.

6. The diagnostic system according to claim 1, characterized in that, The engine valve model mentioned above involves indicators including valve flow rate. Relative opening of valve Their expressions are as follows: in The flow coefficient of the valve. This refers to the flow cross-sectional area of ​​the valve. The valve inlet pressure. For valve outlet pressure, It is the product of the valve's rated flow area and flow coefficient.

7. The diagnostic system according to claim 1, characterized in that, The engine combustion chamber model mentioned above involves indicators including the combustion chamber air-fuel ratio. and combustion chamber pressure Their expressions are as follows: in This refers to the oxidant mass flow rate in the combustion chamber. This refers to the fuel mass flow rate in the combustion chamber. This is the theoretical characteristic velocity of the combustion chamber. This represents the combustion chamber thrust coefficient.

8. The diagnostic system according to claim 1, characterized in that, In the three-tiered engine dynamic system model, each component satisfies a common working equation, the specific expression of which is: in, The methane flow rate in the oxygen-enriched pre-combustion chamber, To regulate the outlet flow rate of the methane control valve, The oxygen flow rate in the oxygen-enriched combustion chamber. This refers to the outlet flow rate of the oxygen main valve. For the methane flow rate in the fuel-rich pre-combustion chamber, This represents the methane main valve outlet flow rate. For the oxygen flow rate in the rich combustion chamber, To regulate the outlet flow of the oxygen control valve, The speed of the methane turbine pump. For oxygen turbine pump speed, The moment of inertia of the methane turbopump. For methane turbine power, For methane pump power, The moment of inertia of the oxygen turbine pump. For oxygen turbine power, This refers to the oxygen pump power.

9. The diagnostic system according to claim 1, characterized in that, The fault data obtained through simulation using the fault factor injection method can be mathematically described as follows: in, X= These are the state parameters of the component model. U = These are the input parameters for the component model and the input interface for the module. D For component model parameters; t For time; The output parameters of the component model, and the output interface of the module; , The functional relationships of the component model; It is a fault factor.

10. The diagnostic system according to claim 1, characterized in that, Engine data under different fault conditions are used as training and testing sets. After training the fault diagnosis model using the training set, the fault diagnosis model is tested using the testing set. Specifically: Engine data under different faults are divided into training and testing sets. Each data sample in the dataset contains two aspects of information: measurement parameters and fault type label. In the fault diagnosis model, the LSTM network extracts the temporal features, trend changes and dynamic patterns of the data. Based on this, Lightgbm classifies the corresponding data samples into the corresponding fault categories according to the information extracted by LSTM, thereby realizing the fault diagnosis function. After training the fault diagnosis model using training set data samples, the measurement parameters of the test set data samples are then input into the fault diagnosis model to perform fault diagnosis, output the fault type, and compare it with the fault type of the test set data samples to obtain the diagnostic accuracy index.

Images