CN-121980225-A - Aeroengine surge diagnosis method, system and equipment based on space-time feature fusion

CN121980225ACN 121980225 ACN121980225 ACN 121980225ACN-121980225-A

Abstract

The invention discloses an aero-engine surge diagnosis method, system and equipment based on space-time feature fusion, and belongs to the technical field of aviation safety. The method comprises the steps of obtaining dynamic signal data of an aeroengine, carrying out first feature extraction processing on the dynamic signal data to generate frequency domain space features, carrying out second feature extraction processing on the dynamic signal data to generate time domain features, carrying out space-time feature fusion on the frequency domain space features and the time domain features by adopting a cross attention mechanism to generate fusion features, and diagnosing the working state of the aeroengine based on the fusion features to obtain a surge diagnosis result. According to the invention, by constructing parallel space-time feature extraction paths and innovatively adopting a cross attention mechanism to carry out depth fusion, key associated information among different feature modes can be adaptively focused, the identification capability of early weak surge fault precursors is effectively enhanced, and the accuracy and the robustness of a diagnosis model are remarkably improved under the condition of small sample data.

Inventors

SHEN YAFENG
HU CHUNYAN
ZENG QINGWEN
TANG ZHENCHAO
SUN JIAXIAN
Miao Keqiang

Assignees

中国科学院工程热物理研究所

Dates

Publication Date: 20260505
Application Date: 20260211

Claims (10)

1. An aeroengine surge diagnosis method based on space-time feature fusion, which is characterized by comprising the following steps: acquiring dynamic signal data of an aeroengine; performing first feature extraction processing on the dynamic signal data to generate frequency domain space features; performing second feature extraction processing on the dynamic signal data to generate time domain features; Performing space-time feature fusion on the frequency domain spatial features and the time domain features by adopting a cross attention mechanism to generate fusion features; And diagnosing the working state of the aero-engine based on the fusion characteristic to obtain a surge diagnosis result.
2. The method of claim 1, wherein the first feature extraction process comprises: Performing Fast Fourier Transform (FFT) on the dynamic signal data to obtain a frequency domain signal; And inputting the frequency domain signals into a preset convolutional neural network CNN model, and extracting the frequency domain spatial characteristics through convolution and pooling operation.
3. The method according to claim 1, wherein the second feature extraction process includes: performing variation modal decomposition VMD on the dynamic signal data to obtain at least one intrinsic modal function IMF component; inputting the at least one intrinsic mode function IMF component into a pre-established bidirectional long-short term memory network BiLSTM model, and extracting the time domain characteristics.
4. The method of claim 1, wherein the employing a cross-attention mechanism for spatio-temporal feature fusion comprises: respectively carrying out linear transformation on the frequency domain space features and the time domain features to obtain a query matrix, a key matrix and a value matrix; calculating an attention score between the query matrix and the key matrix; Calculating attention weight according to the attention score; and carrying out weighted summation on the value matrix by using the attention weight to obtain the fusion characteristic.
5. The method of claim 4, wherein the frequency domain spatial signature is taken as input to generate the query matrix and the time domain signature is taken as input to generate the key matrix and the value matrix.
6. The method of claim 1, wherein diagnosing the operational state of the aircraft engine based on the fusion feature comprises: Inputting the fusion characteristics to a full-connection layer for characteristic integration; And processing the output of the full connection layer through a Softmax layer to obtain the surge diagnosis result.
7. The method according to claim 1, characterized by further comprising, before the first feature extraction process and the second feature extraction process: and preprocessing the dynamic signal data by adopting a sliding window slicing method, and dividing the time sequence data into a plurality of overlapped data segments by setting the window size and the sliding step length so as to expand the data samples.
8. The method of claim 2, wherein inputting the frequency domain signal to a preset convolutional neural network CNN model comprises: and finishing the frequency domain signals obtained through the fast Fourier transform into a two-dimensional feature matrix, and taking the two-dimensional feature matrix as the input of the convolutional neural network CNN model.
9. An aero-engine surge diagnostic system based on temporal-spatial feature fusion, comprising: The data acquisition module is used for acquiring dynamic signal data of the aeroengine; the first feature extraction module is used for carrying out first feature extraction processing on the dynamic signal data to generate frequency domain space features; the second feature extraction module is used for carrying out second feature extraction processing on the dynamic signal data to generate time domain features; the space-time feature fusion module is used for carrying out space-time feature fusion on the frequency domain space features and the time domain features by adopting a cross attention mechanism to generate fusion features; and the diagnosis module is used for diagnosing the working state of the aeroengine based on the fusion characteristics to obtain a surge diagnosis result.
10. An electronic device, the electronic device comprising: And a memory communicatively coupled to the at least one processor; Wherein the memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Description

Aeroengine surge diagnosis method, system and equipment based on space-time feature fusion Technical Field The invention belongs to the technical field of aviation safety, and particularly discloses an aero-engine surge diagnosis method, system and equipment based on space-time feature fusion. Background Aeroengines are the heart of an aircraft, whose performance, reliability and safety are directly related to success or failure of flight tasks and life safety of flight personnel. During operation of an aircraft engine, surge (Surge) is an extremely dangerous aerodynamic instability phenomenon. The device is characterized in that the device is in low-frequency and high-amplitude intense oscillation generated by the air flow in the air compressor along the axial direction, the thrust of the engine can be reduced sharply, the structural parts bear huge dynamic load and thermal load, and even in a very short time, the catastrophic consequences such as blade fracture, casing damage and the like are caused, so that the flight safety is seriously threatened. As modern aeroengines are moving toward higher thrust-weight ratios, wider operating envelopes, and better fuel economy, their internal operating environments are becoming increasingly severe, and the design margins for the core components of the compressor, etc., are continually compressed, so that the risk of surge when the engine is subjected to complex flight tasks (e.g., severe maneuvers, acceleration and deceleration) or encounters external disturbances (e.g., intake distortion) increases. Therefore, the development of a high-precision and high-reliability surge diagnosis and early warning technology has important significance for guaranteeing the safe operation of an engine, implementing Condition-based maintenance (CBM) and prolonging the service life of the engine. The existing surge detection methods can be mainly divided into two types, namely model-based and data-based driving. Model-based methods typically require the establishment of an accurate physical or mathematical model of the engine, and the determination of whether surge has occurred can be made by comparing the deviation of the model output from the actual sensor measurements. However, aeroengines are extremely complex nonlinear, time-varying systems, creating high-fidelity models that accurately describe their dynamics is extremely difficult, and the models are poorly adapted to changes in operating conditions and individual differences. In contrast, data-driven based methods do not rely on accurate physical models, but rather learn and mine failure modes directly from massive amounts of historical operating data, which has received widespread attention in recent years. Conventional data-driven based approaches typically rely on signal processing techniques (e.g., fourier transforms, wavelet transforms) and machine learning algorithms (e.g., support vector machines, decision trees). These methods typically require complex manual feature extraction and selection by a field expert, i.e., converting the raw sensor signal (e.g., pressure, temperature, vibration) into a set of feature vectors that effectively distinguish between normal and surge conditions. This process is not only time consuming and labor intensive, but the quality of the extracted features directly determines the final performance of the diagnostic model. In addition, early precursor features of surge tend to be very weak, easily submerged in strong background noise and operating condition fluctuations, and manually designed features may not be effective in capturing these subtle changes, resulting in insufficient diagnostic sensitivity and accuracy. In recent years, artificial intelligence technology represented by deep learning has made breakthrough progress in the fields of image recognition, natural language processing and the like, and the powerful automatic feature learning and nonlinear mapping capability of the artificial intelligence technology also brings new solutions to aeroengine fault diagnosis. Researchers have begun to attempt to directly process raw sensor data using deep learning models such as Convolutional Neural Networks (CNNs), recurrent Neural Networks (RNNs), and their variant long short term memory networks (LSTM), to achieve "end-to-end" fault diagnosis. For example, CNNs are able to efficiently extract spatial or local structural features in signals, while LSTM is good at capturing long-term dependencies in time series data. However, existing research still has the following challenges and limitations: limitations of single-mode features most studies focus on only a single domain feature of the signal, e.g. processing a frequency domain representation of the signal using only CNN (e.g. spectrograms), or processing a time domain sequence using only LSTM. Surging is a complex fluid-solid coupling phenomenon, and precursor information of the surge can be simultaneously contained in the time domain