CN-122020519-A - Multi-source information fusion method and device for digital prototype of turboshaft engine

CN122020519ACN 122020519 ACN122020519 ACN 122020519ACN-122020519-A

Abstract

The invention relates to the technical field of data processing, in particular to a multisource information fusion method and device for a digital prototype of a turboshaft engine. The method comprises the steps of obtaining multi-source heterogeneous data and physical parameters, carrying out feature extraction according to data characteristics of various data in the multi-source heterogeneous data in different feature extraction modes, fusing the extracted features to obtain fusion features, adopting a preset data driving model to process the fusion features to obtain first predicted values, adopting a preset physical model to process the physical parameters to obtain second predicted values, and obtaining fusion prediction results based on the first predicted values and the second predicted values. According to the invention, the self-adaptive learning capacity of the data driving model and the mechanism reliability of the physical model are effectively combined by fusing the prediction results of the data driving model and the physical model, and the accuracy and the robustness of the digital prototype of the turboshaft engine in the aspects of performance prediction, health evaluation, fault early warning and the like are remarkably improved.

Inventors

PENG KAI
LI WEIXIA
LI RUIMIN
LIU YINFENG
HUANG XIANGLONG
LIU WU

Assignees

中国航发湖南动力机械研究所

Dates

Publication Date: 20260512
Application Date: 20260112

Claims (10)

1. The multi-source information fusion method for the digital prototype of the turboshaft engine is characterized by comprising the following steps of: acquiring multi-source heterogeneous data and physical parameters; performing feature extraction by adopting different feature extraction modes according to the data characteristics of various data in the multi-source heterogeneous data, and fusing the extracted features to obtain fused features; processing the fusion characteristics by adopting a preset data driving model to obtain a first preset quantity; Processing the physical parameters by adopting a preset physical model to obtain a second predicted quantity; and obtaining a fusion prediction result based on the first prediction amount and the second prediction amount.
2. The method of claim 1, wherein the multi-source heterogeneous data includes sensor data, historical operation and maintenance data, structural simulation data and environmental parameters, and the feature extraction is performed by adopting different feature extraction modes according to data characteristics of various data in the multi-source heterogeneous data, including: extracting characteristics of the sensor data by adopting a preset time sequence network; Performing feature extraction on the structural simulation data by adopting a preset convolutional neural network; and extracting the characteristics of the historical operation and maintenance data and the environmental parameters by adopting a preset unsupervised machine learning model.
3. The method of claim 1, wherein fusing the extracted features to obtain fused features comprises: and fusing the extracted features by adopting a multi-mode self-encoder comprising an attention mechanism to obtain fused features, wherein the multi-mode self-encoder comprises an encoder, a fusion layer and a decoder, the fusion layer comprises the attention mechanism, and the attention weight of the attention mechanism is dynamically adjusted based on the to-be-predicted quantity.
4. The method according to claim 1, wherein the method further comprises: Determining a deviation of the actual operating data from the second predetermined amount; determining parameter correction amounts corresponding to the deviation and the fusion characteristics by adopting a preset data driving model; and correcting the physical model parameters of the preset physical model by adopting the parameter correction quantity.
5. The method of claim 1, wherein the fusion prediction result comprises a performance prediction result, a health diagnosis result, and a fault pre-warning result, the method further comprising: a health score is determined based on the fusion prediction result and the actual operation data.
6. The method of claim 1, wherein deriving a fusion prediction result based on the first predicted amount and the second predicted amount comprises: and carrying out weighted fusion on the first predicted quantity and the second predicted quantity by adopting a preset weight to obtain a fusion predicted result, wherein the preset weight is dynamically adjusted according to time and state.
7. The method of claim 1, wherein before performing feature extraction by using different feature extraction modes according to data characteristics of each type of data in the multi-source heterogeneous data, the method further comprises: And (3) adopting a small sample experiment to evaluate the performances of different preset time sequence networks, different preset convolutional neural networks and different preset unsupervised machine learning models, and determining the network or model meeting preset conditions.
8. A multi-source information fusion device for a digital prototype of a turboshaft engine, the device comprising: the data acquisition module is used for acquiring multi-source heterogeneous data and physical parameters; The feature extraction fusion module is used for extracting features according to the data characteristics of various data in the multi-source heterogeneous data in different feature extraction modes, and fusing the extracted features to obtain fusion features; The data driving module is used for processing the fusion characteristics by adopting a preset data driving model to obtain a first preset quantity; The physical analysis module is used for processing the physical parameters by adopting a preset physical model to obtain a second preset quantity; And the result fusion module is used for obtaining a fusion prediction result based on the first prediction amount and the second prediction amount.
9. An electronic device, comprising: A memory and a processor, the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so as to execute the multisource information fusion method of the digital prototype facing the turboshaft engine according to any one of claims 1 to 7.
10. A computer-readable storage medium having stored thereon computer instructions for causing a computer to perform the multisource information fusion method for a digital prototype of a turboshaft engine according to any one of claims 1 to 7.

Description

Multi-source information fusion method and device for digital prototype of turboshaft engine Technical Field The invention relates to the technical field of data processing, in particular to a multisource information fusion method and device for a digital prototype of a turboshaft engine. Background With the continuous improvement of the intellectualization and digitization level of the aviation power plant, the digital prototype (Digital Prototype) technology has become an important tool for the full life cycle management of the turboshaft engine. The digital prototype not only can support design optimization and virtual verification of products, but also can provide virtual mapping and data support at various stages of engine manufacturing, assembly, testing, operation, maintenance and the like. Particularly, for a turboshaft engine with a complex structure and a harsh running environment, dynamic performance prediction, health state evaluation and intelligent fault diagnosis are realized by utilizing a digital prototype, so that the turboshaft engine becomes a key means for improving the reliability and safety of aviation equipment. The existing digital prototype of the turboshaft engine has the problem that the digital prototype is difficult to integrate and manage uniformly in a high-efficiency way when facing multi-source heterogeneous data. Data from sensors, simulation platforms, maintenance records and other different sources, different formats and different frequencies can be generated in the whole life cycle process of the engine. The data types are various, the formats are not uniform, the time sequences are inconsistent, so that the traditional digital prototype platform is difficult to integrate effectively, and the comprehensive perception and analysis of the running state of the engine are affected. Disclosure of Invention The invention provides a multisource information fusion method and device for a digital prototype of a turboshaft engine, which are used for solving the problem that the digital prototype of the turboshaft engine in the prior art is difficult to integrate and manage uniformly in a high-efficiency way when facing multisource heterogeneous data. The invention provides a multisource information fusion method for a digital prototype of a turboshaft engine, which comprises the steps of obtaining multisource heterogeneous data and physical parameters, extracting features according to data characteristics of various data in the multisource heterogeneous data in different feature extraction modes, fusing the extracted features to obtain fusion features, processing the fusion features by a preset data driving model to obtain a first predicted value, processing the physical parameters by a preset physical model to obtain a second predicted value, and obtaining fusion prediction results based on the first predicted value and the second predicted value. According to the invention, the self-adaptive learning capacity of the data driving model and the mechanism reliability of the physical model are effectively combined by fusing the prediction results of the data driving model and the physical model, and the accuracy and the robustness of the digital prototype of the turboshaft engine in the aspects of performance prediction, health evaluation, fault early warning and the like are remarkably improved. The method not only utilizes multi-source heterogeneous data to excavate complex nonlinear rules, but also guarantees the rationality of the prediction result by depending on a physical principle, thereby providing more reliable and intelligent decision support for key tasks such as state monitoring, fault early warning, and optionally maintenance of the engine. In an optional implementation manner, the multi-source heterogeneous data comprises sensor data, historical operation and maintenance data, structural simulation data and environmental parameters, and the characteristic extraction is performed in different characteristic extraction modes according to the data characteristics of various data in the multi-source heterogeneous data, wherein the method comprises the steps of performing characteristic extraction on the sensor data by adopting a preset time sequence network, performing characteristic extraction on the structural simulation data by adopting a preset convolutional neural network, and performing characteristic extraction on the historical operation and maintenance data and the environmental parameters by adopting a preset unsupervised machine learning model. In the invention, aiming at multi-source heterogeneous data with different characteristics such as sensor data, structure simulation data, historical operation and maintenance data, environmental parameters and the like, an optimal time sequence network, a convolutional neural network and an unsupervised learning model are respectively adopted for processing, so that deep features and key information contained in variou