摘要
近年来,卷积神经网络(CNN)等深度学习方法的发展为发动机故障诊断和预测带来了新的思路;CNN具有局部连接、权值共享、池化操作以及多层结构等特点,能够有效提取局部特征,降低网络的训练难度,使CNN具有很强的学习能力和特征表达能力;开展了深度卷积神经网络故障预测方法研究,实现了面向发动机气路故障预测算法架构;利用基于发动机试验仿真数据对该方法进行了验证,并与其他几种常见的基于数据驱动的预测方法进行了比较,验证结果表明文章提出的基于卷积神经网络的预测方法具有较好的可行性和效果,可作为开展发动机PHM技术研究的参考。
In recent years,the development of deep learning methods have brought new ideas to engine prognosis and health management. CNN has the characteristics of sparse connectivity,shared weights,pooling operation and multi-layer structure.It can effectively extract local features,reduce the training difficulty of the network,and make CNN have strong learning ability and feature expression ability.The prognostic method based on convolutional neural network is studied,and the software platform of the algorithm for engine gas path fault prognosis is realized.Using the test data from engine simulation,verification study shows that prognostic method proposed has better feasibility and effectiveness for the prognostic technology of aero-engine compared with other data -driven prediction methods.
作者
元尼东珠
杨浩
房红征
Yuan Nidongzhu;Yang Hao;Fang Hongzheng(Computer College,Qinghai Nationalities University,Xining 810007,China;Beijing Aerospace Measure &Control Corp.Ltd,Beijing 100041,China;Beijing Key Laboratory of High-speed Transport Intelligent Diagnostic and Health Management,Beijing 100041,China;National and Local Joint Engineering Research Center of Equipment Life Cycle Condition Monitoring and Health Management Technology and Application,Beijing 100041,China)
出处
《计算机测量与控制》
2019年第10期74-78,共5页
Computer Measurement &Control
