故障检测作为航空发动机健康管理的主要内容之一,是保证航空发动机的安全性、可靠性和经济性的重要手段.针对发动机发生概率最大的气路部件故障,提出了一种长短期记忆神经网络(Long Short Term Memory,LSTM)和自编码器(Auto-Encoder,AE)结合的故障检测方法,利用T-MATS(The Toolbox for the Modeling and Analy-sis of Thermodynamic Systems)热力学仿真软件包中的航空发动机模型,进行仿真实验,仿真了单一和混合故障以及不同故障程度的数据样本,用健康数据样本训练构建故障检测模型,故障数据样本进行模型测试,根据核密度估计和箱型图结合的方法确定统计量和控制限,从而进行故障检测.实验结果表明了所提方法的可行性和有效性,具有一定的工程应用价值,与传统自编码器方法检测结果进行对比,LSTM自编码器的故障检测率提高最高达20%.
Aero Engine Fault Detection Based on LSTM Autoencoder
As one of the main contents of aero engine health management,fault detection is an important means to ensure the safety,reliability and economy of aero engine.A fault detection method combining Long Short Term Memory(LSTM)neural network and Auto-Encoder(AE)was proposed to solve the problem of gas-path component failure with the highest probability of occurrence.Aero-engine model in T-MATS(The Toolbox for the Modeling and Analysis of Thermodynamic Systems)was used for simula-tion experiments.The data samples of single and mixed faults and different fault degrees are simulated,and the fault detection model is constructed by training the health data samples.The fault data samples are tested,and the statistics and control limits are determined according to the method of kernel density estimation and box diagram,so as to carry out fault detection.The experimental results show that the pro-posed method is feasible and effective,and has certain engineering application value.Compared with the detection results of the traditional autoencoder method,the fault detection rate of the LSTM autoencoder is increased by up to 20%.
fault detecLSTMautoencoderaeroengineengine health management
万方数据
维普
