航空发动机

赵珍,许辰,袁伟.基于退化特征强化的民用涡轮发动机在翼剩余寿命预测[J].航空发动机,2026,52(2):13-23

基于退化特征强化的民用涡轮发动机在翼剩余寿命预测

Residual Life Prediction of Civil Aviation Turbine Engine On-Wing Based on Degradation FeatureEnhancement

DOI：10.12482/ISSN.1672-3147.20240709002

中文关键词: 剩余寿命预测特征强化改进白鲸算法双向长短时记忆网络民用航空涡轮发动机

英文关键词:RUL prediction Feature enhancement IBWO BiLSTM Civil Aviation Turbine Engine

基金项目:国家自然科学基金联合基金项目(U2333213)资助

作者	单位
赵珍	中国民航大学电子信息与自动化学院,天津 300300
许辰	中国民航大学电子信息与自动化学院,天津 300300
袁伟	天津赛象科技股份有限公司,天津 300384

摘要点击次数: 1731

全文下载次数: 292

中文摘要:

民用航空涡轮发动机结构复杂，退化信息难以提取，对其剩余寿命预测的精度不高。为提高涡轮发动机在翼剩余寿命预测精度，提出一种基于退化特征强化结合改进白鲸优化算法（IBWO）与双向长短期记忆网络（BiLSTM）的剩余寿命预测方法 IBWO-BiLSTM。为了给剩余寿命预测模型提供与设备退化过程关联紧密的建模数据，对过程数据进行2阶段退化特征强化。第 1阶段通过随机森林删除与退化关联较弱的变量，第2阶段通过主成分分析进一步强化对发动机退化的表征性。基于强化后的退化特征信息，搭建了IBWO-BiLSTM剩余寿命预测模型，通过IBWO对BiLSTM模型的超参数进行寻优，提高了模型预测准确度。使用C-MAPSS数据集对所提方法的有效性进行验证。结果表明：数据集中涡轮发动机的平均误差为13.87%，模型可以较为准确地预测民航涡轮发动机剩余寿命。

英文摘要:

The complex structure of civil aviation turbine engines poses challenges in extracting degradation information and achiev? ing high-accuracy remaining useful life (RUL) prediction. .To improve the prediction accuracy, an IBWO-BiLSTM method is proposed based on the degradation feature reinforcement, which combines Improved Beluga Whale Optimization (IBWO) and Bi-directional Long Short-Term Memory (BiLSTM) networks. Firstly, a two-stage feature enhancement method was applied to historical data to provide model? ing data closely related to the degradation process. In the first stage, variables weakly associated with degradation were removed using random forest. In the second stage, principal component analysis was employed to further enhance the characterisation of engine degrada? tion. Subsequently, the IBWO-BiLSTM residual life prediction model was constructed based on the enhanced degradation feature informa? tion. The hyperparameters of the IBiLSTM model were optimized by improved beluga whale optimization algorithm to improve prediction accuracy. The proposed method was validated using the C-MAPSS dataset. Simulation results demonstrate that The average prediction error of the turbine engines in the dataset was 13.87%. the model can accurately predict the remaining life of the civil aviation turbine engine.

查看全文 HTML 查看/发表评论下载PDF阅读器