基于多头注意力机制的飞机发动机寿命预测研究  

Remaining Useful Life Prediction of Aeroengine Based on Multi-Head Attention

作　　者：聂磊[1] 徐诗奕 张吕凡 尹业寒 董正琼 周向东 NIE Lei;XU Shi-yi;ZHANG Lyu-fan;YIN Ye-han;DONG Zheng-qiong;ZHOU Xiang-dong(Hubei Modern Manufacturing Quality Laboratory,School of Mechanical Engineering,Hubei University of Technology,Wuhan 430068,China)

机构地区：[1]湖北工业大学机械工程学院湖北省现代制造质量实验室,湖北武汉430068

出　　处：《推进技术》2023年第8期192-200,共9页Journal of Propulsion Technology

基　　金：国家自然科学基金(51975191);襄阳湖北工业大学产业研究院项目(XYYJ2022B01)。

摘　　要：针对飞机发动机监测参数多和预测模型不能充分提取监测数据的有效信息等问题,基于一维卷积神经网络(1DCNN)、时序卷积神经网络(TCN)和多头注意力机制,提出一种新的网络结构以实现飞机发动机剩余寿命的准确预测。对多维特征参数分别建立一个1DCNN-TCN模型,利用两层1DCNN对飞机发动机的多元传感器信号进行特征提取,利用TCN对特征量的时序信息进行记忆,通过多头注意力机制对多个1DCNN-TCN的输出分别进行加权处理,并拼接最终结果。分析结果表明,采用本文方法得到的RMSE和Score值比目前文献中最优值分别降低了6.84%,63.41%。该方法显著提升了飞机发动机剩余寿命预测的准确性。Aiming at the problems of multiple parameters of aeroengine and insufficient extraction of effec-tive information from monitoring data by prediction model,this paper proposes a new network structure based on one-dimensional convolutional network(1DCNN),temporal convolutional network(TCN)and multi-head atten-tion mechanism to accurately predict the remaining life of aircraft engines.Firstly,1DCNN-TCN models are es-tablished for multi-dimensional characteristic parameters,two-layer 1DCNN was used to extract features from multi-source sensor signals of aircraft engines,and TCN was used to memorize timing information of features.Fi-nally,multiple 1DCNN-TCN outputs are weighted by multi-head attention,and the final results are spliced.The experimental results show that the proposed method can reduce the RMSE and Score by 6.84%and 63.41%re-spectively on the basis of the existing model,which significantly improves the accuracy of aircraft engine remain-ing life prediction.

关 键 词：飞机发动机 卷积神经网络 时序卷积神经网络 多头注意力机制 剩余寿命 

分 类 号：V233.7[航空宇航科学与技术—航空宇航推进理论与工程]