摘要
目的 解决传统航空发动机剩余寿命预测方法精度较低的问题。方法 将变分模态分解(Variational Mode Decomposition,VMD)与长短期记忆网络(Long Short-Term Memory,LSTM)进行深度融合,构建新型混合预测框架。通过信号分解与深度学习相结合的技术路径,提升飞参数据特征提取能力和航空发动机寿命预测的准确性。首先,对航空发动机数据集中的各参数进行VMD;然后将分解完毕的数据输入LSTM网络中,结合对应的剩余寿命标签进行训练;最后,通过参数对比和优化实现参数寻优,提高模型的性能表现。结果 相较于CNN、DCNN、RNN和GRU,VMD-LSTM模型在涡扇发动机RUL预测任务中取得了较高的预测精度,在提取时频域特征及捕捉长期依赖关系方面展现出强大能力。结论 VMD-LSTM模型提升了对复杂动态变化的捕捉能力,实现了对航空发动机RUL的高精度预测。
The work aims to solve the problem of low accuracy of existing aero engine residual life prediction methods.A new hybrid prediction framework was constructed by deep fusion of variational mode decomposition(VMD) and long short-term memory network(LSTM).Through the technical path combining signal decomposition and deep learning,the scheme effectively improved the extraction capability of engine degradation features and the accuracy of life prediction.Firstly,Variational Mode Decomposition(VMD) was performed for each parameter in aero engine data set.The decomposed data was input into the Long Short-Term Memory(LSTM) network and trained with the corresponding remaining life label.Finally,parameter optimization was realized through parameter comparison and optimization to improve the performance of the model.The experimental results of C-MAPSS showed that the VMD-LSTM model proposed in this paper achieved higher prediction accuracy in RUL prediction tasks of turbofan engines compared with CNN,DCNN,RNN and GRU,demonstrating the powerful capability of the VMD-LSTM model in extracting time-frequency domain features and capturing long-term dependency relationships.The VMD-LSTM model improves the ability to capture complex dynamic changes,and realizes high-precision prediction of aero engine RUL.
作者
齐阳
祝华远
吴士博
QI Yang;ZHU Huayuan;WU Shibo(Naval Aviation University Qingdao Campus,Shandong Qingdao 266041,China)
出处
《装备环境工程》
2025年第5期92-102,共11页
Equipment Environmental Engineering
关键词
变分模态分解
航空发动机
剩余寿命
预测
长短期记忆网络
性能退化
variational mode decomposition
aero engine
residual life
prediction
long short-term memory network
performance degradation
