基于集成学习方法的机电作动器故障诊断框架  

A framework for fault diagnosis of electromechanical actuator based on ensemble learning method

作　　者：张潇[1] 刘沐阳 ZHANG Xiao;LIU Muyang(China Academy of Space Technology,Beijing 100094,China;School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China)

机构地区：[1]中国空间技术研究院,北京100094 [2]西北工业大学航空学院,西安710072

出　　处：《航天器环境工程》2023年第5期559-566,共8页Spacecraft Environment Engineering

摘　　要：针对飞行器综合性能不断提高的发展需求,对机载机电作动器(EMA)进行健康管理尤为关键。文章以EMA作为研究对象,重点研究基于集成学习方法的故障诊断框架来解决飞行器可能存在的健康管理问题:对比不同集成学习策略间的优劣,提出一种以Boosting集成学习方法为核心的故障诊断框架。该方法的建立以XGBoost、LightGBM和CatBoost模型为基础,相较于时下流行的深度学习框架,其占用的计算资源更少,模型的可解释性更强。试验结果表明,该框架相较于传统机器学习方法准确率提高10%,相较于深度学习方法训练时间减少75%,且内存占用率更低,具有较强的工程应用价值。In order to meet the increasing demand for comprehensive performance of aircraft,a health management of airborne electromechanical actuator(EMA)is essential.Taking EMA as the research object and with a focus on studying a fault diagnosis framework based on ensemble learning method,this article aims at solving the issue that aircraft health management may have.By comparing the advantages and disadvantages of different ensemble learning strategies,a fault diagnosis framework based on Boosting ensemble learning was proposed.This method was constructed based on XGBoost,LightGBM and CatBoost models.Compared with the popular deep learning frameworks,it consumes less computing resources and has stronger interpretability.The experimental results show that the framework has a 10%improvement in accuracy compared to traditional machine learning methods,a 75%reduction in training time compared to deep learning methods,and a lower memory usage,indicating a high engineering application value.

关 键 词：机电作动器 永磁同步电机 健康管理 故障诊断 集成学习 

分 类 号：TP275[自动化与计算机技术—检测技术与自动化装置]