基于XGBoost-DCNN的航空发动机剩余寿命预测  

Prediction of Residual Life of Aircraft Engines Based on XGBoost-DCNN

作　　者：刘国建 杜冬[1] 邢苗英 LIU Guojian;DU Dong;XING Miaoying(Civil Aviation Flight University of China,Guanghan 618300)

机构地区：[1]中国民用航空飞行学院,广汉618300

出　　处：《舰船电子工程》2023年第9期182-188,共7页Ship Electronic Engineering

基　　金：四川省大学生创新创业项目(编号:S202110624267)资助。

摘　　要：航空发动机作为飞机的核心部件,内部结构复杂,其剩余使用寿命(RUL)预测对生产维修和人身安全具有重要意义。采集的数据中,由于设备振动和环境的影响,噪声是不可避免的。为此,提出基于异常值剔除的深度卷积网络(DCNN)的航空发动机剩余寿命预测方法。首先,基于NASA提供的C-MAPSS航空发动机数据集,采用皮尔逊相关系数法筛选与航空发动机剩余寿命强相关的传感器数据,再利用XGBoost算法剔除异常数据,建立航空发动机剩余寿命模型数据集;最后,将数据Z-Score标准化构建时间序列数据作为DCNN模型输入构建航空发动机剩余寿命预测模型。为了验证方法的有效性,与改进卷积神经网络等方法对比,结果表明XGBoost-DCNN模型拥有更高的精度。As the core component of an aircraft,the aero-engine has a complex internal structure,and its residual service life(RUL)prediction is of great significance to production maintenance and personal safety.In the collected data,the noise is inevitable due to the influence of the equipment vibration and the environment.Therefore,the aviation engine residual life prediction method is proposed based on deep convolutional network(DCNN).Firstly,based on the C-MAPSS aviation engine data set provided by NASA,the pearson correlation coefficient method is used to screen the sensor data,and then it uses the XGBoost algorithm to establish the aviation engine residual life model data set.Finally,the data Z-Score standardized construction time series data is input as the DCNN model to construct the aviation engine residual life prediction model.To verify the effectiveness of the method,compared with the improved convolutional neural networks,the results show that the convolutional neural network with XGBoost outliers removed has higher accuracy.

关 键 词：航空发动机 剩余寿命 XGBoost异常检测 DCNN模型 分段线性退化 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]