基于降噪自编码器的航空发动机性能退化评估  

Assessment of performance degradation for aero-engine based on denoising autoencoder

作　　者：洪骥宇 王华伟[1] 倪晓梅 HONG Jiyu;WANG Huawei;NI Xiaomei(College of Civil Aviation,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)

机构地区：[1]南京航空航天大学民航学院,南京210016

出　　处：《航空动力学报》2018年第8期2041-2048,共8页Journal of Aerospace Power

基　　金：国家自然科学基金青年基金(71401073);国家自然科学基金与中国民航局联合资助项目(U1233115);江苏省研究生培养创新工程项目(SJZZ16_0060)

摘　　要：针对航空发动机性能退化的形式及规律,提出一种基于降噪自编码器的航空发动机性能退化评估方法。针对采集的航空发动机6个状态监测参数,采用降噪自编码器,利用贪婪逐层训练算法,挖掘各参数对发动机性能的深层影响,提取出更有利于评估的数据特征,进行性能退化评估。将提出的算法与BP(back propagation)神经网络以及支持向量机得到的结果进行测试比较,测试表明:提出的方法准确率有所提高,达到93.5%,具有较强的鲁棒性,在信噪比为10dB时准确率达到84.5%,并且能够防止航空发动机状态监测中小样本过拟合的问题。Targeting the form and law of the aero-engine＇s performance degradation,a degradation assessment method based on denoising autoencoder was proposed.On account of the collected six aero-engine condition monitoring parameters,denoising autoencoder and greedy layer-wise training algorithm were used to assess performance degradation in order to explore the deep influences of those parameters on engine＇s performances and extract the data characteristics more conducive to the assessment.The comparison between the proposed algorithm and back propagation（BP）neural networks as well as support vector machine showed that the proposed method had high accuracy and robustness.The accuracy of the proposed method was 93.5% and the accuracy just reduced to 84.5% when signal-to-noise ratio was 10 dB.The proposed method also can prevent the over-fitting of small samples in aero-engine condition monitoring.

关 键 词：状态监测 性能退化 降噪自编码器 深度学习 鲁棒性 

分 类 号：V231[航空宇航科学与技术—航空宇航推进理论与工程]