A Method for Extending Abnormal Data of Flutter Signals of Aeroelastic System

Abstract

The invention relates to a method for expanding abnormal data of flutter signals of an aeroelastic system. Based on a time series signal generation algorithm of a generator in a generative adversarial network, only a small amount of flutter signals with coupled resonance appear in structural response signals of aeroelastic systems or aircraft. It is possible to expand the flutter data set through the pre-trained generator network, so as to balance the normal and abnormal data as much as possible, and reduce the impact of data samples on the method during the method verification process. Compared with the prior art, the present invention introduces the deep learning algorithm into the data set expansion of the vibration signal of the aeroelastic system, and provides a flutter signal data set expansion method based on the deep learning algorithm. Based on the pre-trained generative adversarial network, through a small amount of coupled resonance measured wind tunnel flutter test flutter signals to realize the transformation from random signals based on Gaussian distribution to flutter signals with flutter characteristics, so as to achieve the purpose of expanding the data set.

Description

technical field [0001] The invention belongs to a data processing method of an aeroelastic system, and relates to a method for expanding abnormal data of a flutter signal of an aeroelastic system. Background technique [0002] For aeroelastic systems, flutter is a structural instability phenomenon of aeroelastic systems caused by the mutual coupling of aerodynamic and elastic forces. The condition monitoring data for normal operation is relatively easy to obtain and the amount of data is huge. On the contrary, the data of condition monitoring under abnormal operation of the system is usually smaller than the normal data in terms of data volume. This data imbalance phenomenon makes a lot of flutter type determination and flutter boundary based on flutter signal and data-driven. Forecasting methods suffer from issues such as insufficient sample size. [0003] In the existing flutter signal processing, methods such as flutter determination and boundary prediction are usually r...

AI Technical Summary

Problems solved by technology

Since the simulation signal cannot completely simulate the structural response signal of the actual system, for some data-driven methods with high data sensitivity, the imbalance between the abnormal data and the normal data of the flutter signal of the aeroelastic system waiting to be solved

[0004] There are usually two main ways to obtain the flutter signal data of the aeroelastic system in the case of flutter. On the one hand, it is obtained through the simulation environment or the aeroelastic model of the wind tunnel test, and on the other hand, it is artificially obtained in the actual aircraft flight test However, under normal circumstances, the long-term abnormal operation of the aeroelastic system is destructive, which makes it difficult to obtain data. Therefore, the data obtained by the above method has the following two problems:

[0005] 1. There are deviations between the simulation data and the actual aircraft flight test data due to different systems, and the methods based on simulation data research usually have insufficient robust performance when applied in actual systems

[0006] 2. Since the measured abnormal data is relatively small compared to the normal data, generally speaking, there is an imbalance between normal data and abnormal data. This imbalance leads to flutter signal processing based on measured signals and data-driven Research methods usually have a better effect on the normal state, but the abnormal state is difficult to be completely effective due to the problem of sample size

[0007] For a well-designed aircraft, it is usually in a normal operating state, and it is rarely artificially reached to the critical speed of flutter for safety reasons. Generally speaking, the measured flutter signal at the flutter boundary is usually In the wind tunnel test, the aeroelastic model is forced to reach the coupling resonance by adjusting the critical wind speed to obtain the data of the flutter state, but relatively speaking, the amount of data obtained based on the wind tunnel test is small. For the data-driven method, Usually manifested as data imbalance, which makes it difficult to achieve the purpose of validating the method

Images