航空发动机状态监控和预测性维护应用研究

Applied Research of Status Monitoring and Predictive Maintenance for Aeroengine

中文关键词:  特征提取  深度学习  故障预测  健康管理  剩余寿命预测

英文关键词:feature extraction  deep learning  fault prediction  health management  remaining life prediction

中文摘要:

      为了深化飞参数据的应用价值，通过研究发动机转动件故障预测、剩余寿命预测以及气路健康等，为发动机保障决策和预测性维护提供参考。采用经验模态分解(EMD)结合相对向量机(RVM)、灰度模型(GM)用于发动机转动件、气路监测的状态监控和故障预测，选取波音某型飞机故障数据验证了模型的准确性，平均绝对百分比误差(MAPE)能达到8.46%；采用卡尔曼滤波(KF)结合梯度提升决策树(GBDT)的方法对数据进行降噪并预测剩余寿命，通过美国国家航空航天局（NASA）的航空发动机仿真数据集验证了模型能达到91.3%的准确率；采用核主成分分析(KPCA)结合深度置信网络(DBN)的方法建立发动机气路健康监控模型，经过大量QAR数据验证和测试，预测相对误差为0.43%。针对基于数据挖掘的航空发动机故障诊断算法开展研究，设计了相应的算法，开展了实验验证，通过有效的数据预处理和模型参数调节，使得故障诊断性能达到较高水准，为航空发动机的预测性维护提供了重要参考。

英文摘要:

      In order to deepen the application value of flight parameter data,reference is provided for engine support decision-making and predictive maintenance through research on fault prediction of engine rotating parts,residual life prediction,and gas path health.Empirical mode decomposition (EMD) combined with relative vector machine (RVM) and gray model (GM) are used for condition monitoring and fault prediction of engine rotating parts and gas path monitoring.The fault data of a Boeing aircraft are selected to verify the accuracy of the model,and the average mean absolute percentage error(MAPE) reaches 8.46%.Kalman filter (KF) combined with gradient boosting decision tree(GBDT) is used to denoise the data and predict the remaining life.The accuracy of the model is verified to be 91.3% by NASAs aeroengine simulation dataset.The method of kernel principal component analysis (KPCA) combined with deep confidence network (DBN) is used to establish the engine gas path health monitoring model.After a large number of QAR data validation and testing,the prediction relative error is 0.43%.Research is conducted on aeroengine fault diagnosis algorithms based on data mining,corresponding algorithms are designed,and experimental verification is conducted.Through effective data preprocessing and model parameter adjustment,fault diagnosis performance reaches a high level,providing important reference value for predictive maintenance of aeroengines.