基于多阶段注意力机制的多种导航传感器故障识别研究  

Multiple Navigation Sensor Fault Diagnose Research Based on Multi-stage Attention Mechanism

作　　者：王亚朝 赵伟[1] 徐海洋 刘建业[1] WANG Ya-Zhao;ZHAO Wei;XU Hai-Yang;LIU Jian-Ye(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106)

机构地区：[1]南京航空航天大学自动化学院,南京211106

出　　处：《自动化学报》2021年第12期2784-2790,共7页Acta Automatica Sinica

基　　金：国家自然科学基金(61533008,61374115,61603181);中央高校基本科研业务费专项基金(NS2018021);江苏高校优势学科建设工程项目资助。

摘　　要：导航传感器在使用过程中容易发生故障,针对传统方法对其间歇性和渐变性故障识别率低的问题提出了一种基于多阶段注意力机制的多传感器故障识别算法.该算法采用基于长短期记忆神经网络和注意力机制的编码器-解码器结构,根据多类导航传感器数据之间的空间相关性和时间相关性来进行多传感器的故障互判.经验证,该算法对多种类传感器的故障识别率高达97.5%,可以高效地实现故障的检测和分类.该方法可以准确识别出故障传感器和故障类型,具有很强的工程应用价值.Since navigation sensors may malfunction in use,a multi-sensor fault diagnosis algorithm based on multistage attention mechanism is proposed to solve the problem of low diagnosis rate of intermittent defect and gradual fault.An encoder-decoder structure based on the long short term memory(LSTM)neural network and attention mechanism is adopted in the algorithm,and fault mutual diagnosis between multiple navigation sensors is based on spatial and time correlation between the data of multiple navigation sensors.It is verified that the fault diagnosis rate of the algorithm for multi-type sensors is as high as 97.5%.Besides,sensor faults can be detected and classified effectively by this algorithm.This method which has a strong engineering application value can accurately identify the fault sensor and fault type.

关 键 词：多阶段注意力机制 长短期记忆神经网络 编码器-解码器 多类传感器 故障互判 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TP183[自动化与计算机技术—控制科学与工程]