基于深度长短期记忆网络的发动机叶片剩余寿命预测

为了研究航空发动机转子叶片的剩余寿命预测问题，提出了一种基于多传感器信号融合的深度长短期记忆网络(DLSTM)预测模型。首先利用深度学习和长短期记忆的组合来构造DLSTM网络。然后，将多个传感器信号数据进行融合处理，从而通过深度学习发现各个传感器时序信号之间隐藏的长期依赖关系。进一步在给定网格搜索策略的情况下，通过自适应矩估计算法调整DLSTM的网络结构和参数，并且在DLSTM模型中引入了一种随机丢失策略，以缓解过度拟合问题并使预测模型规范化。最后利用CMAPSS涡扇发动机进行了实验验证，在一种故障模式和两种故障模式条件下，DLSTM网络预测模型相对于其他传统方法的评价指标Score分别下降了17.19%和14.37%，其他两个评价指标相对来说也较优，结果表明本文提出的方法具有更高的准确性以及稳定性。

Prediction of Residual Life of Engine Blades Based on Deep Short Term Memory Network

In order to study the residual life prediction of aeroengine rotor blades, a prediction model of deep long short term memory(DLSTM) based on multi-sensor signal fusion was proposed. First, the DLSTM network was constructed by the combination of DLSTM. Then, the multi-sensor signal data were fused to find the hidden long-term dependence between the sensor timing signals through deep learning. Furthermore, given the grid search strategy, the network structure and parameters of DLSTM were adjusted by the adaptive moment estimation algorithm, and a random loss strategy was introduced into the DLSTM model to alleviate the over fitting problem and standardize the prediction model. Finally, the CMAPSS turbofan engine was used to test and verify. Under one failure mode and two failure modes, the score of DLSTM network prediction model was 17.19% and 14.37% lower than that of other traditional methods, and the other two evaluation indexes were relatively better. The results show that the method proposed in this paper has higher accuracy and stability.