Cascade modeling method and device for multi-task learning of aero-engine

Abstract

The invention relates to a cascade modeling method and device for aero-engine multi-task learning. The method comprises the following steps: acquiring detection data of each actual measurement point of each component structure of an aero-engine, a task association relationship among the component structures, gas path flow direction information and flight condition information of the aero-engine, and generating an engine task corresponding to each component structure; the method comprises the following steps: identifying task dependence information among component structures, constructing a low-fidelity network and a high-fidelity network, and constructing an engine performance model of the aero-engine based on the low-fidelity network, the high-fidelity network and gas path flow direction information among the component structures. By adopting the method, the simulation effect of constructing the engine performance model of the aero-engine can be improved.

Description

Technical Field

[0001] The present application relates to the technical fields of simulation modeling and multi-task learning, and particularly to a cascaded modeling method and device for multi-task learning of an aero-engine. Background Art

[0002] The modeling technology of aero-engines is the basis for carrying out tasks such as overall engine control, fault diagnosis, and fault-tolerant control, and has long received extensive attention from practitioners in the field of aero-engines. However, due to the multi-disciplinary coupling of the engine and the complexity of the system, it is extremely difficult to establish an engine performance model covering the entire process of the engine. Therefore, how to establish an engine performance model covering the entire process of the engine is the current research focus.

[0003] The traditional modeling method of the engine performance model is to use the data-driven method as the residual compensation model between the physical model and the actual operation data. However, such methods face the problem of insufficient generalization of the data-driven model and rely on large-scale real data, resulting in poor simulation effects in constructing the engine performance model of aero-engines. Summary of the Invention

[0004] Based on this, it is necessary to provide a cascaded modeling method, device, computer device, computer-readable storage medium, and computer program product for multi-task learning of an aero-engine in view of the above technical problems.

[0005] In a first aspect, the present application provides a cascaded modeling method for multi-task learning of an aero-engine, including:

[0006] Obtain the detection data of each actual measurement point of each component structure of the aero-engine, the task association relationship between each component structure, the gas path flow direction information between each component structure, and the flight condition information of the aero-engine, and generate an engine task corresponding to each component structure based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure;

[0007] Identify the task dependency information between each component structure based on the task association relationship between each component structure and the engine task corresponding to each component structure, and construct a low-fidelity network based on the flight condition information and the engine task corresponding to each component structure;

[0008] Construct a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between each component structure, and construct an engine performance model of the aero-engine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information between each component structure.

[0009] Optionally, generating an engine task corresponding to each component structure based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure includes:

[0010] Based on the detection data of each actual measurement point of each component structure, identify the detection type of each actual measurement point;

[0011] Based on the detection types of the actual measurement points, query the operation task type corresponding to each component structure in the component operation database;

[0012] Based on the operation task type corresponding to each component structure and the gas path flow direction information between each component structure, query the engine task of each component structure during the operation of the aeroengine in the engine operation database.

[0013] Optionally, identifying the task dependency information between each component structure based on the task association relationship between each component structure and the engine task corresponding to each component structure includes:

[0014] Based on the task association relationship between each component structure, identify the information transfer method between each component structure;

[0015] Based on the engine task corresponding to each component structure, identify the input requirement information corresponding to each component structure and the output information corresponding to each component structure, and based on the information transfer method between each component structure, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure, identify the data interaction information between each component structure;

[0016] Use the data interaction information between each component structure as the task dependency information between each component structure.

[0017] Optionally, constructing a low-fidelity network based on the flight condition information and the engine task corresponding to each component structure includes:

[0018] Obtain the baseline simulation data of the aeroengine;

[0019] Identify each flight condition data in the flight condition information, and based on each baseline simulation data and each flight condition data, train an initial low-fidelity network to obtain a low-fidelity network.

[0020] Optionally, constructing a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between each component structure includes:

[0021] Based on the data interaction information among the component structures, identify the sub-data interaction information among the detection data of the actual measurement points, and based on the flight condition data, obtain each model parameter through the low-fidelity network;

[0022] Based on the detection data of the actual measurement points, the flight condition data, and the model parameters, train the initial high-fidelity network to obtain the high-fidelity network.

[0023] Optionally, constructing the engine performance model of the aero-engine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information among the component structures includes:

[0024] Obtain the physical information of each component structure of the aero-engine, and based on the physical information of each component structure, construct the physical model of the aero-engine;

[0025] Based on the low-fidelity network and the high-fidelity network, through a multi-task cascaded modeling strategy, construct the data-driven model of the aero-engine;

[0026] Take the physical model and the data-driven model as the engine performance model of the aero-engine.

[0027] In a second aspect, the present application also provides a cascaded modeling device for multi-task learning of an aero-engine, including:

[0028] An acquisition module, configured to acquire the detection data of each actual measurement point of each component structure of the aero-engine, the task association relationship among the component structures, the gas path flow direction information among the component structures, and the flight condition information of the aero-engine, and based on the detection data of each actual measurement point of each component structure and the gas path flow direction information among the component structures, generate the engine tasks corresponding to the component structures;

[0029] An identification module, configured to identify the task dependency information among the component structures based on the task association relationship among the component structures and the engine tasks corresponding to the component structures, and construct a low-fidelity network based on the flight condition information and the engine tasks corresponding to the component structures;

[0030] A construction module, configured to construct a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information among the component structures, and construct the engine performance model of the aero-engine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information among the component structures.

[0031] Optionally, the obtaining module is specifically configured to:

[0032] Based on the detection data of each actual measurement point of each component structure, identify the detection type of each actual measurement point;

[0033] Based on the detection types of the actual measurement points, query the operation task type corresponding to each component structure in the component operation database;

[0034] Based on the operation task type corresponding to each component structure and the gas path flow direction information between the component structures, query the engine tasks of each component structure during the operation of the aero-engine in the engine operation database.

[0035] Optionally, the identification module is specifically configured to:

[0036] Based on the task association relationships between the component structures, identify the information transmission methods between the component structures;

[0037] Based on the engine tasks corresponding to the component structures, identify the input requirement information corresponding to each component structure and the output information corresponding to each component structure, and based on the information transmission methods between the component structures, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure, identify the data interaction information between the component structures;

[0038] Use the data interaction information between the component structures as the task dependency information between the component structures.

[0039] Optionally, the identification module is specifically configured to:

[0040] Obtain the baseline simulation data of the aero-engine;

[0041] Identify the flight condition data in the flight condition information, and based on the baseline simulation data and the flight condition data, train the initial low-fidelity network to obtain the low-fidelity network.

[0042] Optionally, the construction module is specifically configured to:

[0043] Based on the data interaction information between the component structures, identify the sub-data interaction information between the detection data of the actual measurement points, and based on the flight condition data, obtain each model parameter through the low-fidelity network;

[0044] Based on the detection data of the actual measurement points, the flight condition data, and the model parameters, train the initial high-fidelity network to obtain the high-fidelity network.

[0045] Optionally, the building block is specifically configured to:

[0046] Obtain the physical information of each component structure of the aeroengine, and construct a physical model of the aeroengine based on the physical information of each component structure;

[0047] Based on the low-fidelity network and the high-fidelity network, construct a data-driven model of the aeroengine through a multi-task cascaded modeling strategy;

[0048] Use the physical model and the data-driven model as the engine performance model of the aeroengine.

[0049] In a third aspect, the present application provides a computer device. The computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the method according to any one of the first aspect are implemented.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium. A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of the first aspect are implemented.

[0051] In a fifth aspect, the present application provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of the first aspect are implemented.

[0052] The above cascade modeling method and device for multi-task learning of aero-engines obtain the detection data of each actual measurement point of each component structure of the aero-engine, the task association relationship between each of the component structures, the gas path flow direction information between each of the component structures, and the flight condition information of the aero-engine, and generate engine tasks corresponding to each of the component structures based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each of the component structures; identify the task dependency information between each of the component structures based on the task association relationship between each of the component structures and the engine tasks corresponding to each of the component structures, and construct a low-fidelity network based on the flight condition information and the engine tasks corresponding to each of the component structures; construct a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between each of the component structures, and construct an engine performance model of the aero-engine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information between each of the component structures. In this solution, by combining low-fidelity and high-fidelity data, the method can effectively improve the accuracy and robustness of the model without significantly increasing costs, and is applicable to resource-constrained operating environments. And this solution uses a multi-task cascade physical information neural network model. This method not only retains the interpretability of the physical model but also improves the fitting ability of the data-driven method, and is applicable to the modeling of complex engine systems. Finally, this modeling method overcomes the error problem between the traditional model and actual measurement by integrating the physical model and the data-driven model, and has higher prediction accuracy and correction efficiency. Thus, the simulation effect of constructing the engine performance model of the aero-engine is comprehensively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0054] Figure 1 It is a schematic flowchart of a cascade modeling method for multi-task learning of an aero-engine in an embodiment;

[0055] Figure 2 It is a schematic diagram of the model structure of a data-driven model in an embodiment;

[0056] Figure 3 It is a schematic diagram of the first set of flight condition schematic curves in an embodiment;

[0057] Figure 4Schematic diagram of simulation results of PBM and ROE measurement points under the first set of flight conditions in an embodiment;

[0058] Figure 5 Schematic diagram of the second set of flight condition schematic curves in an embodiment;

[0059] Figure 6 Schematic diagram of the simulation results after correction based on MFPINN in an embodiment;

[0060] Figure 7 Schematic flow diagram of the cascade modeling example of multi-task learning for aero-engines in an embodiment;

[0061] Figure 8 Block diagram of the structure of the cascade modeling device for multi-task learning of aero-engines in an embodiment;

[0062] Figure 9 Internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0063] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0064] The cascade modeling method for multi-task learning of aero-engines provided by the embodiments of the present application can be applied to the application environment of cascade modeling for multi-task learning of aero-engines. Among them, the terminal can be but is not limited to various personal computers, laptop computers, mid-range computers, etc. Among them, by combining low-fidelity and high-fidelity data, the method can effectively improve the accuracy and robustness of the model without significantly increasing the cost, and is applicable to resource-constrained operating environments. And this solution uses a multi-task cascade physics-informed neural network model. This method not only retains the interpretability of the physical model but also improves the fitting ability of the data-driven method, and is applicable to modeling complex engine systems. Finally, this modeling method overcomes the error problem between the traditional model and actual measurement by integrating the physical model and the data-driven model, and has higher prediction accuracy and correction efficiency. Thus, the simulation effect of constructing the engine performance model of aero-engines is comprehensively improved.

[0065] In an exemplary embodiment, as Figure 1 shown, a cascade modeling method for multi-task learning of aero-engines is provided. Taking the application of this method to a terminal as an example, it includes the following steps S101 to S103. Among them:

[0066] Step S101: Obtain the detection data of each actual measurement point of each component structure of the aero-engine, the task association relationship between each component structure, the gas path flow direction information between each component structure, and the flight condition information of the aero-engine. Based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure, generate the engine tasks corresponding to each component structure.

[0067] In this embodiment, the terminal obtains the detection data of each actual measurement point of each component structure of the aero-engine. Among them, each actual measurement point is an actual engine task set on each component structure of the aero-engine and used to identify the component state of each component and the engine task corresponding to the component structure. Each component structure includes, but is not limited to, component structures such as a compressor, a combustion chamber, a turbine, and an exhaust device. The actual measurement points corresponding to each component structure, for example, the actual measurement points of the compressor include the outlet static pressure and the rotational speed, while the actual measurement points of the turbine are the outlet temperature. Through the flow direction of the air flow, the task of the compressor is defined as an upstream task, and the task of the turbine is a downstream task. The establishment of this task relationship reflects the causal dependence in the physical system. Then, the terminal obtains the task association relationship between each component structure, which includes the task execution order and the information transmission method between each component, etc. Then, the terminal obtains the gas path flow direction information between each component structure and the flight condition information of the aero-engine. Among them, the flight condition information includes flight condition data of multiple different flight condition types, such as flight power type, flight duration type, and flight power consumption type, etc. Finally, based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure, generate the engine tasks corresponding to each component structure. The specific generation process will be described in detail later.

[0068] Step S102: Based on the task association relationship between each component structure and the engine tasks corresponding to each component structure, identify the task dependence information between each component structure, and based on the flight condition information and the engine tasks corresponding to each component structure, construct a low-fidelity network.

[0069] In this embodiment, the terminal identifies the task dependency information among the component structures based on the task association relationships among the component structures and the engine tasks corresponding to the component structures, and constructs a low-fidelity network based on the flight condition information and the engine tasks corresponding to the component structures. The method for constructing the low-fidelity network is to construct a neural network corresponding to the low-fidelity data through a multi-fidelity physics-informed neural network (MFPINN) for rapid engine model correction. The task dependency information among the component structures is the data interaction information among the detection data corresponding to the actual measurement points among the component structures, that is, the data interaction information among the component structures. The specific identification process and construction process will be described in detail later. The low-fidelity network is used to generate the model parameters for approximating the data-driven model under the baseline conditions.

[0070] Step S103: Based on the low-fidelity network, the flight condition information, and the task dependency information among the component structures, construct a high-fidelity network, and construct an engine performance model of the aeroengine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information among the component structures.

[0071] In this embodiment, the terminal constructs a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information among the component structures, and constructs an engine performance model of the aeroengine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information among the component structures. The method for constructing the high-fidelity network is to construct a neural network corresponding to the high-fidelity data through a multi-fidelity physics-informed neural network (MFPINN) for rapid engine model correction. The engine performance model includes a physical model and a data-driven model, and the formula of the data-driven model is as follows:

[0072]

[0073] where, NN L is the low-fidelity network, NN Hnl , and NN Hn are the high-fidelity networks. is the flight condition data for each type of flight condition. As Figure 2 shown, it is the schematic diagram of the model structure of the data-driven model.

[0074] Based on the above solution, by adopting the combination of low-fidelity and high-fidelity data, the method can effectively improve the accuracy and robustness of the model without significantly increasing the cost, and is applicable to resource-constrained operating environments. And this solution uses a multi-task cascaded physics-informed neural network model. This method not only retains the interpretability of the physical model but also improves the fitting ability of the data-driven method, and is applicable to the modeling of complex engine systems. Finally, this modeling method integrates the physical model and the data-driven model, overcomes the error problem between the traditional model and the actual measurement, and has higher prediction accuracy and correction efficiency. Thus, the simulation effect of the engine performance model for constructing an aeroengine is comprehensively improved.

[0075] Optionally, based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure, engine tasks corresponding to each component structure are generated, including: based on the detection data of each actual measurement point of each component structure, identifying the detection type of each actual measurement point; based on the detection types of each actual measurement point, querying the operation task type corresponding to each component structure in the component operation database; based on the operation task type corresponding to each component structure and the gas path flow direction information between each component structure, querying the engine task of each component structure during the operation of the aeroengine in the engine operation database.

[0076] In this embodiment, the terminal identifies the detection type of each actual measurement point based on the detection data of each actual measurement point of each component structure. Among them, the detection type includes but is not limited to the outlet static pressure detection type, the rotational speed detection type, the outlet temperature detection type, etc. Then, the terminal queries the operation task type corresponding to each component structure in the component operation database based on the detection types of each actual measurement point. Among them, in the component operation database, the detection type range corresponding to each operation task type is included. The operation task type is used to characterize the operation task of the component structure.

[0077] Then, based on the operation task type corresponding to each component structure and the air path flow information between component structures, the terminal queries the engine tasks of each component structure during the operation of the aero-engine in the engine operation database. Among them, the engine operation database includes the corresponding relationships between different engine tasks, air path flow information, and operation task types. Then, based on this corresponding relationship, the terminal identifies the engine tasks of each component structure during the operation of the aero-engine. For example, the key measurement points of the compressor include the outlet static pressure and rotational speed, while the key measurement point of the turbine is the outlet temperature. Through the flow direction of the air flow, the compressor is located upstream of the air flow, and the outlet static pressure and rotational speed type of the compressor correspond to the gas compression task. The terminal identifies the engine task of this compressor as the upstream task by querying the engine operation database. And through the flow direction of the air flow, it can be known that the turbine is located downstream of the air flow, and the outlet temperature of the turbine corresponds to the gas transmission type. The terminal identifies the engine task of this turbine as the downstream task by querying the engine operation database. The establishment of this task relationship reflects the causal dependence in the physical system.

[0078] Based on the above solution, by combining the operation task type corresponding to each component structure and the air path flow information between component structures, the engine tasks corresponding to each component structure are identified, improving the accuracy of identification.

[0079] Optionally, based on the task association relationship between component structures and the engine tasks corresponding to each component structure, the task dependence information between component structures is identified, including: identifying the information transfer method between component structures based on the task association relationship between component structures; identifying the input requirement information corresponding to each component structure and the output information corresponding to each component structure based on the engine tasks corresponding to each component structure, and identifying the data interaction information between component structures based on the information transfer method between component structures, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure; using the data interaction information between component structures as the task dependence information between component structures.

[0080] In this embodiment, the terminal identifies the information transfer method between component structures based on the task association relationship between component structures. Among them, this information transfer method is the data transfer method of the actual measurement points data between component structures.

[0081] Then, based on the engine tasks corresponding to each component structure, the terminal identifies the input requirement information corresponding to each component structure and the output information corresponding to each component structure, and based on the information transfer method between component structures, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure, identifies the data interaction information between component structures. Among them, the data interaction information is the data interaction logic between two adjacent component structures obtained by establishing a data interaction strategy between the input information and the output information between two adjacent component structures in the order of engine tasks. Finally, the terminal uses the data interaction information between component structures as the task dependency information between component structures.

[0082] Based on the above solution, through the construction of reasonable task dependencies, the information transfer between different tasks is optimized, and the complexity of directly using end-to-end neural network learning is significantly reduced. By modeling the dependencies between tasks, the model can better capture the correlations between tasks, thereby improving the accuracy and reliability of engine performance prediction.

[0083] Optionally, based on the flight condition information and the engine tasks corresponding to each component structure, a low-fidelity network is constructed, including: obtaining the baseline simulation data of the aeroengine; identifying each flight condition data in the flight condition information, and training the initial low-fidelity network based on each baseline simulation data and each flight condition data to obtain the low-fidelity network.

[0084] In this embodiment, the terminal obtains the baseline simulation data of the aeroengine. Then, the terminal identifies each flight condition data in the flight condition information, and trains the initial low-fidelity network based on each baseline simulation data and each flight condition data to obtain the low-fidelity network. Specifically, the low-fidelity network NN L is trained based on the engine baseline simulation data and is used to approximate the data model under baseline conditions. Its input is the flight conditions P 0L , T 0L , W fL , and the output is the predicted model parameter Y L .

[0085] Based on the above solution, by training the low-fidelity network, the accuracy of setting the model parameters of the data-driven model is improved.

[0086] Optionally, a high-fidelity network is constructed based on the low-fidelity network, flight condition information, and task dependency information among component structures, including: identifying sub-data interaction information among the detection data of each actual measurement point based on the data interaction information among component structures, and obtaining each model parameter through the low-fidelity network based on each flight condition data; training an initial high-fidelity network based on the detection data of each actual measurement point, each flight condition data, and each model parameter to obtain the high-fidelity network.

[0087] In this embodiment, the terminal identifies the sub-data interaction information among the detection data of each actual measurement point based on the data interaction information among component structures, and obtains each model parameter through the low-fidelity network based on each flight condition data. Among them, each model parameter is the model parameter of this data-driven model. Then, the terminal trains the initial high-fidelity network based on the detection data of each actual measurement point, each flight condition data, and each model parameter to obtain the high-fidelity network. Specifically, the high-fidelity network consists of NN Hnl , NN Hn two sub-networks, which are respectively used to approximate the linear and non-linear correlations between flight conditions and engine measurement points in the real data. Its input is flight condition P 0H , T 0H , W fH and the model parameter Y L predicted by the low-fidelity network, and the output is the engine measurement point Y H predicted by the high-fidelity network.

[0088] Based on the above solution, by combining low-fidelity and high-fidelity data sources, the proposed multi-task cascaded MFPINN method can significantly improve the prediction accuracy of the model while ensuring computational efficiency. This model structure not only enhances the interpretability of the model, but also reduces the dependence on high-cost operation data, thus ensuring the efficiency and economy of the model. The multi-fidelity architecture of MFPINN can effectively utilize data sources of different qualities, combine prior knowledge of the engine physical system, and further optimize the overall performance of the model. This method passes the prediction results of the low-fidelity network to the high-fidelity network through a multi-task cascaded manner, ensuring information flow and collaborative optimization between different network levels.

[0089] Optionally, based on the low-fidelity network, the high-fidelity network, and the air flow direction information between component structures, an engine performance model of an aeroengine is constructed, including: obtaining the physical information of each component structure of the aeroengine, and constructing a physical model of the aeroengine based on the physical information of each component structure; constructing a data-driven model of the aeroengine based on the low-fidelity network and the high-fidelity network through a multi-task cascaded modeling strategy; using the physical model and the data-driven model as the engine performance model of the aeroengine.

[0090] In this embodiment, the terminal obtains the physical information of each component structure of the aeroengine, and constructs a physical model of the aeroengine based on the physical information of each component structure. Then, the terminal constructs a data-driven model of the aeroengine based on the low-fidelity network and the high-fidelity network through a multi-task cascaded modeling strategy. Finally, the terminal uses the physical model and the data-driven model as the engine performance model of the aeroengine. Among them, the multi-task cascaded modeling strategy is a construction strategy for constructing a data-driven model through a multi-task cascaded model.

[0091] In practical applications, in the non-linear dynamic physical model, by adjusting a series of component parameters, such as the mass flow rate and efficiency of typical components, to simulate the mismatch between the actual engine and the model. The typical adjustable component parameters of the physical model include the efficiency and flow capacity of each component, the pressure drop of the pipeline, the efficiency of combustion, etc. In the implementation process, it is considered that the efficiency and flow rate of the baseline model are always 1, and the coefficient of each factor is changed to simulate the real engine.

[0092] The simulated flight plan is as Figure 3 shown, and the simulation results of the physical model (PBM) and the actually operating engine (ROE) are obtained through flight condition simulation. Figure 4 Shows some measurable parameters N1 and N2 in the flight plan, expressed as normalized results. The root mean square error (RMSE) of each parameter is greater than 2% and less than 6%, indicating that for engines of the same model, the mismatch is acceptable.

[0093] Based on the real data of this flight plan and the PBM simulation data of multiple flight plans, the MFPINN network is trained. At the same time, the trained network is tested according to the second group of flight plans, and the second group of flight plans is as Figure 5 shown. Figure 6 Shows the comparison between the final prediction result and the actual result. As can be seen from the figure, the proposed method can significantly reduce the error of the physical model.

[0094] Based on the above solution, by integrating the physical model and the data-driven model, the error problem between the traditional model and the actual measurement is overcome, and higher prediction accuracy and correction efficiency are achieved.

[0095] This application also provides a cascaded modeling example for multi-task learning of aeroengines, as Figure 7 shown, and the specific processing process includes the following steps:

[0096] Step S701: Obtain the detection data of each actual measurement point of each component structure of the aeroengine, the task association relationship between each component structure, the gas path flow direction information between each component structure, and the flight condition information of the aeroengine.

[0097] Step S702: Based on the detection data of each actual measurement point of each component structure, identify the detection type of each actual measurement point.

[0098] Step S703: Based on the detection types of each actual measurement point, query the corresponding operation task type of each component structure in the component operation database.

[0099] Step S704: Based on the corresponding operation task type of each component structure and the gas path flow direction information between each component structure, query the engine task of each component structure during the operation of the aeroengine in the engine operation database.

[0100] Step S705: Based on the task association relationship between each component structure, identify the information transfer method between each component structure.

[0101] Step S706: Based on the engine tasks corresponding to each component structure, identify the input requirement information corresponding to each component structure and the output information corresponding to each component structure, and based on the information transfer method between each component structure, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure, identify the data interaction information between each component structure.

[0102] Step S707: Take the data interaction information between each component structure as the task dependency information between each component structure.

[0103] Step S708: Obtain the baseline simulation data of the aeroengine.

[0104] Step S709: Identify each flight condition data in the flight condition information, and based on each baseline simulation data and each flight condition data, train the initial low-fidelity network to obtain the low-fidelity network.

[0105] Step S710: Based on the data interaction information between each component structure, identify the sub-data interaction information between the detection data of each actual measurement point, and based on each flight condition data, obtain each model parameter through the low-fidelity network.

[0106] Step S711: Train an initial high-fidelity network based on the detection data of each actual measurement point, each flight condition data, and each model parameter to obtain a high-fidelity network.

[0107] Step S712: Obtain the physics information of each component structure of the aero-engine, and construct a physical model of the aero-engine based on the physics information of each component structure.

[0108] Step S713: Based on the low-fidelity network and the high-fidelity network, construct a data-driven model of the aero-engine through a multi-task cascaded modeling strategy.

[0109] Step S714: Use the physical model and the data-driven model as the engine performance model of the aero-engine.

[0110] It should be understood that although each step in the flowcharts involved in the above-described embodiments is shown in sequence according to the indication of the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same moment, but can be executed at different moments. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0111] Based on the same inventive concept, an embodiment of the present application further provides a cascaded modeling device for aero-engine multi-task learning for implementing the above-mentioned cascaded modeling method for aero-engine multi-task learning. The solution provided by this device for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the cascaded modeling device for aero-engine multi-task learning provided below can refer to the limitations on the cascaded modeling method for aero-engine multi-task learning in the above text, and will not be repeated here.

[0112] In an exemplary embodiment, as Figure 8 shown, a cascaded modeling device for aero-engine multi-task learning is provided, including: an acquisition module 810, an identification module 820, and a construction module 830, where:

[0113] An acquisition module 810, configured to acquire the detection data of each actual measurement point of each component structure of an aeroengine, the task association relationship between each of the component structures, the air path flow direction information between each of the component structures, and the flight condition information of the aeroengine, and generate an engine task corresponding to each of the component structures based on the detection data of each actual measurement point of each component structure and the air path flow direction information between each of the component structures;

[0114] An identification module 820, configured to identify the task dependency information between each of the component structures based on the task association relationship between each of the component structures and the engine task corresponding to each of the component structures, and construct a low-fidelity network based on the flight condition information and the engine task corresponding to each of the component structures;

[0115] A construction module 830, configured to construct a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between each of the component structures, and construct an engine performance model of the aeroengine based on the low-fidelity network, the high-fidelity network, and the air path flow direction information between each of the component structures.

[0116] Optionally, the acquisition module 810 is specifically configured to:

[0117] Identify the detection type of each actual measurement point based on the detection data of each actual measurement point of each component structure;

[0118] Query the operation task type corresponding to each component structure in the component operation database based on the detection types of the actual measurement points;

[0119] Query the engine task of each component structure during the operation of the aeroengine in the engine operation database based on the operation task type corresponding to each component structure and the air path flow direction information between each of the component structures.

[0120] Optionally, the identification module 820 is specifically configured to:

[0121] Identify the information transfer method between each of the component structures based on the task association relationship between each of the component structures;

[0122] Identify the input requirement information corresponding to each component structure and the output information corresponding to each component structure based on the engine task corresponding to each component structure, and identify the data interaction information between each of the component structures based on the information transfer method between each of the component structures, the input requirement information corresponding to each component structure, and the output information corresponding to each component structure.

[0123] Use the data interaction information between the component structures as the task dependency information between the component structures.

[0124] Optionally, the recognition module 820 is specifically configured to:

[0125] Obtain the baseline simulation data of the aeroengine;

[0126] Identify each flight condition data in the flight condition information, and train an initial low-fidelity network based on each of the baseline simulation data and each of the flight condition data to obtain a low-fidelity network.

[0127] Optionally, the construction module 830 is specifically configured to:

[0128] Based on the data interaction information between the component structures, identify the sub-data interaction information between the detection data of each actual measurement point, and based on each of the flight condition data, obtain each model parameter through the low-fidelity network;

[0129] Train an initial high-fidelity network based on the detection data of each actual measurement point, each of the flight condition data, and each of the model parameters to obtain a high-fidelity network.

[0130] Optionally, the construction module is specifically configured to:

[0131] Obtain the physical information of each component structure of the aeroengine, and construct a physical model of the aeroengine based on the physical information of each component structure;

[0132] Based on the low-fidelity network and the high-fidelity network, construct a data-driven model of the aeroengine through a multi-task cascaded modeling strategy;

[0133] Use the physical model and the data-driven model as the engine performance model of the aeroengine.

[0134] Each module in the above cascaded modeling device for multi-task learning of aeroengines can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in or independent of the processor in a computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0135] In an exemplary embodiment, a computer device is provided. The computer device can be a terminal, and its internal structure diagram can be as Figure 9As shown in the figure. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface, the display unit, and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a mobile cellular network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a cascaded modeling method for multi-task learning of an aero-engine. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covered on the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0136] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0137] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, it implements the steps corresponding to the cascaded modeling method for multi-task learning of an aero-engine.

[0138] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the processor, it implements the steps corresponding to the cascaded modeling method for multi-task learning of an aero-engine.

[0139] In an embodiment, a computer program product is provided, including a computer program. When the computer program is executed by the processor, it implements the steps corresponding to the cascaded modeling method for multi-task learning of an aero-engine.

[0140] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0141] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0142] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered to be within the scope described in this specification.

[0143] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A cascade modeling method for multi-task learning of aircraft engines, characterized in that: The method comprises: Acquire detection data of each actual measurement point of each component structure of the aircraft engine, task association relationships between the component structures, gas path flow direction information between the component structures, and flight condition information of the aircraft engine, and generate engine tasks corresponding to the component structures based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between the component structures; Based on the task association relationship between the component structures and the engine tasks corresponding to the component structures, identifying the task dependency information between the component structures, and constructing a low-fidelity network based on the flight condition information and the engine tasks corresponding to the component structures; A high-fidelity network is constructed based on the low-fidelity network, the flight condition information, and the task dependency information between the component structures, and an engine performance model of the aircraft engine is constructed based on the low-fidelity network, the high-fidelity network, and the gas flow direction information between the component structures.

2. The method according to claim 1, characterized in that The generating of the engine tasks corresponding to each component structure based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between each component structure includes: Based on the inspection data of each actual measurement point of each component structure, identifying the inspection type of each actual measurement point; Based on the detection type of each actual measurement point, query the operation task type corresponding to each component structure in the component operation database; Based on the operation task type corresponding to each component structure and the gas path flow direction information between the component structures, the engine task of each component structure during the operation of the aircraft engine is queried in the engine operation database.

3. The method according to claim 2, characterized in that The identifying task dependency information between the component structures based on the task association relationship between the component structures and the engine tasks corresponding to the component structures includes: Based on the task association relationship between the component structures, identifying the information transmission mode between the component structures; Based on the engine tasks corresponding to the component structures, input requirement information corresponding to each component structure and output information corresponding to each component structure are identified, and based on the information transmission mode between the component structures, the input requirement information corresponding to each component structure and the output information corresponding to each component structure, data interaction information between the component structures is identified; The data interaction information between the component structures is used as the task dependency information between the component structures.

4. The method according to claim 1, characterized in that The constructing of a low-fidelity network based on the flight condition information and the engine tasks corresponding to each of the component structures includes: Acquiring baseline simulation data of the aircraft engine; Each flight condition data in the flight condition information is identified, and an initial low-fidelity network is trained based on each baseline simulation data and each flight condition data to obtain a low-fidelity network.

5. The method according to claim 4, characterized in that The constructing of a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between the component structures includes: Based on the data interaction information between the component structures, identifying the sub-data interaction information between the detection data of the actual measurement points, and based on the flight condition data, acquiring the model parameters through the low-fidelity network; Based on the detection data of each actual measurement point, each flight condition data, and each model parameter, an initial high-fidelity network is trained to obtain a high-fidelity network.

6. The method according to claim 1, characterized in that The engine performance model of the aircraft engine is constructed based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information between the component structures, including: Acquiring physical information of the structure of each component of the aircraft engine, and constructing a physical model of the aircraft engine based on the physical information of the structure of each component; Based on the low-fidelity network and the high-fidelity network, a data-driven model of the aircraft engine is constructed through a multi-task cascade modeling strategy; The physical model and the data-driven model are used as the engine performance model of the aircraft engine.

7. A cascade modeling device for multi-task learning of aircraft engines, characterized in that: The device comprises: an acquisition module, used to acquire detection data of each actual measurement point of each component structure of the aircraft engine, task association relationships between the component structures, gas path flow direction information between the component structures, and flight condition information of the aircraft engine, and generate engine tasks corresponding to the component structures based on the detection data of each actual measurement point of each component structure and the gas path flow direction information between the component structures; an identification module, configured to identify task dependency information between the component structures based on the task association relationship between the component structures and the engine tasks corresponding to the component structures, and to construct a low-fidelity network based on the flight condition information and the engine tasks corresponding to the component structures; A construction module is used to construct a high-fidelity network based on the low-fidelity network, the flight condition information, and the task dependency information between the component structures, and to construct an engine performance model of the aircraft engine based on the low-fidelity network, the high-fidelity network, and the gas path flow direction information between the component structures.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images