航空学报 > 2014, Vol. 35 Issue (6): 1612-1622   doi: 10.7527/S1000-6893.2013.0543

航空发动机气路部件故障融合诊断方法研究

李业波1, 李秋红1, 黄向华1, 赵永平2   

  1. 1. 南京航空航天大学 能源与动力学院 江苏省航空动力系统重点实验室, 江苏 南京 210016;
    2. 南京理工大学 机械工程学院, 江苏 南京 210094
  • 收稿日期:2013-07-09 修回日期:2014-02-18 出版日期:2014-06-25 发布日期:2014-06-20
  • 通讯作者: 李秋红,Tel.:025-84892200-2606 E-mail:lqh203@nuaa.edu.cn E-mail:lqh203@nuaa.edu.cn
  • 作者简介:李业波男,博士研究生。主要研究方向:航空发动机建模、控制及故障诊断。Tel:025-84892200-2606 E-mail:liyb1985@gmail.com;李秋红女,博士,副教授,硕士生导师。主要研究方向:航空发动机建模、控制及故障诊断。Tel:025-84892200-2606 E-mail:lqh203@nuaa.edu.cn
  • 基金资助:

    国家自然科学基金(51006052,61104067);航空科学基金(20110652003);中央高校基本科研业务费专项基金(NZ2012104);江苏省2012年度普通高校研究生科研创新计划(CXZZ12_0169)

Research on Gas Fault Fusion Diagnosis of Aero-engine Component

LI Yebo1, LI Qiuhong1, HUANG Xianghua1, ZHAO Yongping2   

  1. 1. Jiangsu Province Key Laboratory of Aerospace Power System, College of Energy & Power Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China;
    2. School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, China
  • Received:2013-07-09 Revised:2014-02-18 Online:2014-06-25 Published:2014-06-20
  • Supported by:

    National Natural Science Foundation of China(51006052,61104067); Aeronautical Science Foundation of China(20110652003); The Fundamental Research Funds for Central Universities(NZ2012104); Graduate Student Research and Innovation Program of Jiangsu Province(CXZZ12_0169)

摘要:

针对发动机气路部件故障,提出了一种基于模型和基于数据驱动的融合诊断方法。采用极端学习机(ELM)实现基于数据驱动的故障诊断。针对ELM随机选择输入层权值和隐含层偏置带来的缺点,采用改进微分进化(IDE)算法以训练样本的均方根误差(RMSE)和输出层权值的范数为评价标准对其进行优化,减少了ELM的隐含层节点数,提高了网络的泛化能力。同时,由于传感器数目的不足,采用基于奇异值分解(SVD)的Kalman(SVD-Kalman)滤波器实现基于模型的部件故障诊断。为了提高航空发动机部件故障诊断的精度,利用改进的迭代约简最小二乘支持向量回归机(IRR-LSSVR)算法对两种算法的估计结果在特征层进行定量融合。仿真结果表明,在发动机稳态状态下,与单独使用基于模型和数据驱动的诊断方法相比,采用特征层融合有效地提高了部件故障诊断的精度和准确率。

关键词: 航空发动机, 极端学习机, 部件故障, 微分进化, Kalman滤波器, 信息融合

Abstract:

A model-based and data-driven-based aero-engine component fault fusion diagnosis method is presented in this paper. The extreme learning machine (ELM) is used to the data-driven-based component fault diagnosis. However, the ELM input weights and hidden layer biases are generated randomly, which usually leads to more nodes in the hidden layer and poor generalization ability. To overcome these drawbacks of ELM, an improved differential evolution (IDE) is used to optimize the input weights and the hidden layer biases. The optimization is carried out according to the root mean squared error (RMSE) of the validation set and the norm of the output weights. Meantime, singular value decomposition (SVD)-based Reduced-dimensional Kalman (SVD-Kalman) filters are used to estimate the health parameters which can solve the problem of limited number of sensors in the model based diagnosis method. To improve the accuracy of fault diagnosis for an aero-engine component, the prediction results of health parameters based on both methods are fused by the improved recursive reduced least squares support vector regression (IRR-LSSVR). Simulation results show that compared with either the model-based or data-driven-based approach, the proposed fusion method improves the accuracy of fault diagnosis significantly.

Key words: aero-engine, extreme learning machine, component fault, differential evolution, Kalman filter, information fusion

中图分类号: