中国民航大学学报 ›› 2025, Vol. 43 ›› Issue (1): 60-66.

• 航空维修 • 上一篇    下一篇

基于飞行数据的 MSCNN-LSTM 水平安定面系统状态监测方法

张 鹏 a ,胡芳语 b ,段照斌 a ,刘静静 b
  

  1. 中国民航大学 a.工程技术训练中心, b.电子信息与自动化学院,天津 300300
  • 收稿日期:2023-02-10 修回日期:2023-04-10 出版日期:2025-04-09 发布日期:2025-04-09
  • 作者简介:张鹏(1963— ),男,北京人,教授,硕士,研究方向为航空机载设备故障诊断等领域
  • 基金资助:
    中国航空工业集团公司西安飞行自动控制研究所项目(H04420180028)

MSCNN-LSTM method for monitoring the state of horizontal stabilizer system
based on flight data

ZHANG Penga , HU Fangyub , DUAN Zhaobina , LIU Jingjingb
  

  1. a. Engineering Techniques Training Center, b. College of Electronic Information Engineering, CAUC, Tianjin 300300, China 
  • Received:2023-02-10 Revised:2023-04-10 Online:2025-04-09 Published:2025-04-09

摘要:

针对真实飞行数据中故障样本匮乏、数据类间失衡且缺少标注问题,本文提出了一种基于多尺度卷积神经
网络(MSCNN,multi-scale convolutional neural network)与长短时记忆(LSTM,long short-term memory)网络
的水平安定面系统状态监测方法。 此方法不依赖于标注数据,利用无监督学习的方式对水平安定面系统进行
状态监测。 首先,利用 MSCNN-LSTM 对系统正常运行状态的快速存储记录器(QAR, quick access recorder)
数据从空间和时间两个维度进行特征提取,以实现舵面位置预测;其次,计算舵面位置预测值与舵面位置
实际值的残差,分析残差分布来确定系统健康状态的阈值;最后,利用某飞机的 QAR 数据进行验证。 实验
结果表明,本文所提方法能准确实现水平安定面系统飞行级的异常状态识别,并能在系统发生故障时,提
前 1 个飞行循环进行异常预警。

关键词:

Abstract:

To address the problems of insufficient fault samples, imbalanced data classes and lack of labeling in real flight data, a state monitoring method for a horizontal stabilizer system based on multi-scale convolutional neural network
(MSCNN) and long short-term memory (LSTM) network is proposed in this paper. This method does not rely on labeled data and uses unsupervised learning to monitor the state of the horizontal stabilizer system. Firstly, the quick
access recorder (QAR) data of the system in normal operation are extracted in both spatial and temporal dimensions
using MSCNN-LSTM to achieve rudder position prediction. Secondly, the residuals between the predicted and actual values of the rudder position are calculated and the distribution of the residuals is analyzed to determine the
threshold for the health state of the system. Finally, the QAR data of an aircraft is used for verification, and the experimental results show that the proposed method in this paper can accurately achieve the abnormal state identification of the horizontal stabilizer system at the flight level and can provide an abnormal alarm one flight cycle in
advance when system failure occurring.

Key words:


中图分类号: