Prediction of engine failure time using principal component analysis, categorical regression tree, and back propagation network

In aviation industry, prediction of engine failure time is very important. The engine is the main power device for aircraft flight and the consequences if they fail could be extremely serious. Predicting engine failure is a key task in airline repair control. This study predicts engine failure using principal component analysis (PCA). Variable replacement is a well-known technique to improve PCA prediction performance, hence we propose a hybrid method incorporating PCA, categorical regression tree and a back propagation network (PCA–CART–BPN) to engine failure time prediction. The proposed PCA–CART–BPN method effectively improved prediction accuracy and achieved satisfactory results. The proposed method first applies PCA analysis and classification and regression tree (CART) clustering, and then inputs these outcomes into a back propagation network (BPN) for training and testing. Experimental and analytical comparisons show that the proposed PCA–CART–BPN method provided significantly improved prediction accuracy to the PCA–CART method. The proposed PCA–CART–BPN method improved the performance by 2.9% (MAPE) and 39% (RMSE) over PCA–CART alone. © 2018, Springer-Verlag GmbH Germany, part of Springer Nature.