基于双通道模型的航空发动机剩余寿命预测  

Remaining Life Prediction of Aero-Engine Based on Dual-Channel Model

作　　者：车鲁阳 高军伟[1,2] 付惠琛 CHE Luyang;GAO Junwei;FU Huichen(College of Automation,Qingdao University,Qingdao 266071,Shandong,China;Shandong Provincial Key Laboratory of Industrial Control Technology,Qingdao 266071,Shandong,China)

机构地区：[1]青岛大学自动化学院,山东青岛266071 [2]山东省工业控制技术重点实验室,山东青岛266071

出　　处：《空军工程大学学报》2023年第6期42-49,共8页Journal of Air Force Engineering University

基　　金：山东省自然科学基金(ZR2019MF063)。

摘　　要：针对现阶段航空发动机单一剩余使用寿命预测模型数据挖掘深度不足导致预测精度低的问题,提出一种双通道模型的预测方法。首先,构建双通道网络结构:通道一使用时间卷积网络,通过残差结构和空洞卷积使得网络具有更大的感受野和计算速度;通道二使用卷积长短时间记忆网络,提取多维时空特征,捕捉数据长期依赖关系。其次,利用多头注意力机制为双通道网络特征重新赋予权重。最后,将双通道网络进行特征融合输出,实现对航空发动机剩余寿命预测。使用涡扇发动机退化数据集进行实验验证,并与其它文献中提到的卷积双向长短时间记忆网络模型、多特征注意力模型、多头注意力模型、卷积门控单元循环神经网络模型进行对比。结果表明,所提模型在3种评价指标上均取得更好的表现,为航空发动机剩余寿命预测提供了一种新思路。Aiming at the problem of insufficient data mining depth of the Remaining Useful Life prediction model of aero-engine at this stage,a prediction method of dual-channel model is proposed.First,a dual channel network structure is constructed:channel one uses time convolutional networks,which enables the network to have a larger receptive field and computing speed through residual structure and hole convolution;channel 2 uses convolutional long short-term memory network to extract multidimensional spatiotemporal features and capture long-term dependencies of data.Then,the multi head attention mechanism is used to reassign weights to the features of the dual channel network.Finally,the dual channel network is used for feature fusion output to achieve prediction of the remaining life of aircraft engines.Experimental validation was conducted using the turbofan engine degradation dataset and compared with other CNNbiLSTM models,multi feature attention models,multi head attention models,and CNN-GRU models mentioned in literature.The results indicate that the proposed model performs better on all three evaluation indicators,providing a new approach for predicting the remaining life of aircraft engines.

关 键 词：航空发动机 寿命预测 时间卷积网络 卷积长短时间记忆网络 多头注意力机制 

分 类 号：V263.5[航空宇航科学与技术—航空宇航制造工程]