基于静电信号变分模态分解和随机森林的气路故障识别方法

Gas-Path Fault Identification Method Based on Electrostatic Signal Variational Mode Decomposition and Random Forest

关键词:  航空发动机  气路  静电传感  随机森林  信号处理  故障识别

Key words:  Aeroengine  Gath path  Electrostatics monitoring  Random Forest  Signal processing  Fault identification

摘要:

气路静电监测是面向航空发动机健康诊断的全新技术，为使其后续可应用于气路故障识别，提出一种基于变分模态分解和随机森林的识别方法。首先分析了气路静电监测原理，针对静电信号噪声干扰问题，提出了一种基于变分模态分解和峭度-排列熵重构准则的静电信号增强方法，并提出故障特征集构造方法，帮助有效提取关键故障信息；进一步通过开展燃烧室积碳、叶片-机匣碰摩、部件掉块等模拟实验，获取故障静电信号和特征集，并构建了基于随机森林的故障类型识别模型。结果表明：所提方法测试集上识别精度达到90%以上，且所提出新特征的归一化重要度达到0.2以上，较传统特征更高，能为基于静电监测的气路部件故障识别提供有效途径。

Abstract:

Gas path electrostatic monitoring is a novel technology for aero-engine health diagnosis. To realize typical gas path fault identification by using this technology， a method based on variational mode decomposition and the random forest is proposed. Firstly， the principle and model of gas path electrostatic monitoring are analyzed. Aiming at the noise interference in electrostatic signal， a signal enhancement method based on variational mode decomposition and Kurtosis-Permutation Entropy reconstruction criterion are studied， and a features dataset construction method is proposed to extract the key fault information. The fault electrostatic signals and feature datasets are obtained through the simulated experiments， such as combustion chamber carbon deposition， blade-casing rubbing， and component drop， and the faults are identified based on the random forest model. The results show that the recognition accuracy of the proposed method is more than 90 % on the test dataset， and the normalized importance of the proposed new feature is more than 0.2， which is higher than the classical features， this method can provide an effective way for fault identification of gas path components based on electrostatic monitoring.