空时模型融合自注意力机制的航空管路故障诊断新方法  

New approach to diagnose faults in aero-pipelines based on spatial-temporal model fused with self-attention mechanism

作　　者：杨同光 袁晟友 周献文 韩清凯[1] 于晓光[3] YANG Tongguang;YUAN Shengyou;ZHOU Xianwen;HAN Qingkai;YU Xiaoguang(School of Mechanical Engineering and Automation,Northeastern University,Shenyang 110819,China;School of Mechatronic Engineering and Automation,Foshan University,Foshan 528225,China;School of Mechanical Engineering and Automation,University of Science and Technology Liaoning,Anshan 114051,China)

机构地区：[1]东北大学机械工程与自动化学院,沈阳110819 [2]佛山科学技术学院机电工程与自动化学院,广东佛山528225 [3]辽宁科技大学机械工程与自动化学院,辽宁鞍山114051

出　　处：《振动与冲击》2024年第10期299-310,共12页Journal of Vibration and Shock

基　　金：国家自然科学基金(51775257)。

摘　　要：航空发动机外部液压管路结构复杂且管路故障信号中伴随着强大的噪声干扰,导致航空管路故障特征提取十分困难,同时目前诊断模型参数和计算量比较庞大,不适合高效地移植到移动和嵌入式设备。面对上述挑战,提出一种轻量化空时模型融合注意力机制的航空发动机液压管路故障诊断新方法,并命名为S-Bi-ATM-Net模型,首先,设计轻量化管路空间特征提取模块,从而融合管路故障数据的空间特征,抽取数据的细粒度特征;然后,设计轻量化管路时间特征提取模块,从细粒度特征中继续融合管路粗粒度特征,实现粗细粒度特征融合;此外在空时模型中融合自注意力机制进行优化,使得最后的决策更加的聚焦,进一步提高所提模型的诊断精度。基于同一数据集,将所提方法与目前主流方法对比分析发现,所提方法能更准确地识别航空管路不同的故障状态,证明了所提方法的优越性和稳定性。The structure of the external hydraulic pipeline of an aero-engine is complicated and the fault signal of the pipeline is accompanied by strong noise interference,which makes it difficult to extract the fault characteristics of the aviation pipeline.Meanwhile,the numerous great number of parameters and large calculation amount of the current diagnostic model are not suitable for efficient transplantation to mobile and embedded equipments.In the face of these challenges,a new method for the fault diagnosis of aero-engine hydraulic pipelines was proposed based on a lightweight spatial-temporal model united with self-attention mechanism,by the name of S-Bi-ATM-Net model.Then,the lightweight pipeline time feature extraction module was designed,and the coarse-grained features of the pipeline were continuously fused with the fine-grained features to achieve the fusion of coarse-fine-grained features.In addition,the self-attention mechanism was integrated into the spatial-temporal model for optimization,which makes the final decision more focused and further improves the diagnostic accuracy of the proposed model.Based on the same data set,comparing the proposed method with the current mainstream methods,it is found that the proposed method can more accurately identify different fault states of air pipelines,which proves the superiority and stability of the method.

关 键 词：故障诊断 空时融合 轻量化 注意力机制 航空管路 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3