欢迎访问《兵工学报》官方网站,今天是

兵工学报 ›› 2024, Vol. 45 ›› Issue (1): 253-263.doi: 10.12382/bgxb.2022.0615

• • 上一篇    下一篇

一种基于TCN-LGBM的航空发动机气路故障诊断方法

吕卫民*(), 孙晨峰, 任立坤, 赵杰, 李永强   

  1. 海军航空大学, 山东 烟台 264000
  • 收稿日期:2022-07-07 上线日期:2024-01-30
  • 通讯作者:
  • 基金资助:
    山东省自然科学基金项目(ZR2021QE193)

A Gas Path Fault Diagnosis Method for Aero-engine Based on TCN-LGBM Model

LÜ Weimin*(), SUN Chenfeng, REN Likun, ZHAO Jie, LI Yongqiang   

  1. Naval Aviation University, Yantai 264000, Shandong, China
  • Received:2022-07-07 Online:2024-01-30

摘要:

长时间工作在高温高压、强振动等恶劣气路环境下的航空发动机经常面临部件疲劳、腐蚀和性能退化的问题,且其故障诊断时序逻辑性不强、故障参数耦合较深等特点十分明显,为此提出一种基于时间卷积神经网络(Temporal Convolutional Network,TCN)和轻量级梯度提升机(Light Gradient Boosting Machine,LGBM)的航空发动机气路故障诊断方法。故障诊断分为故障特征提取和分类诊断两个过程:引入TCN框架,在保证故障数据训练时序逻辑的基础上,实现对远层历史信息和当前层信息的特征融合构建,融合通道注意力机制增强了高质量特征的权重;基于LGBM模型实现对特征的快速分类,利用贝叶斯方法实现对模型超参数的快速优化。以基于PROOSIS软件建模的某军用小涵道比涡扇发动机故障仿真数据为例,对6种故障模式进行诊断识别。仿真结果表明了所提方法的有效性;通过与其他模型对比体现了该方法的优越性。

关键词: 航空发动机, 故障诊断, 时间卷积神经网络, 轻量级梯度提升机, 注意力机制

Abstract:

With the obvious characteristics of poor temporal logic in fault diagnosis and the strongly coupled feature parameters, the aero-engines working in the hostile gas path conditions of high temperature, pressure and strong vibration face with the degradation performance and structure defect problems such as fatigue and corrosion. And an aero-engine gas path fault diagnosis method based on temporal convolutional networks(TCN) and light gradient boosting machine(LGBM) is proposed to provide a feasible solution to the problems above. The diagnosis process can be divided into feature extraction and classification: TCN is introduced to guarantee the fault diagnosis training temporal logic and achieve the features fusion of distant layers and current layers, which is also strengthened by channel attention mechanism; the features are quickly classified based on LGBM model, and the Bayesian method is used to quickly optimize the model hyperparameters. Based on the aero-engine performance modelled by PROOSIS software, six types of fault mode are diagnosed and identified by taking a military low-bypass ratio turbofan engine as an example. The results indicate that the proposed model is effective for fault diagnosis and shows the superiority by comparing with other models.

Key words: aero-engine, fault diagnosis, temporal convolutional network, light gradient boosting machine, attention mechanism

中图分类号: