•  
  •  
 

Proposal

Predictive maintenance in aviation and aerospace applications is among the most explored problems in machine learning (ML) and artificial intelligence (AI), and datasets such as NASA’s C-MAPPS turbofan engine degradation simulation data have proven invaluable, helping researchers explore numerous questions on engine performance, maintenance, and failure. The purpose of this study was to extend the current research on predicting the remaining useful life (RUL) of engines and their risk classification. Starting with simple yet under-investigated nonlinear survival and random forest models, the analysis implemented eXtreme Gradient Boosting (XGBoost) and long short-term memory (LSTM) from TensorFlow’s Keras library. For both regression and classification, valuable insights were gained from principal component analysis (PCA), lifelines survival analysis, k-nearest neighbors (KNN) algorithms, and random forests. The regression results obtained from XGBoost and LSTM architectures were comparable to those reported in related studies. The XGBoost regressor outperformed LSTM, achieving a root mean squared error (RMSE) as low as 21.29 and an R-squared score of 0.74. The XGBoost classifier yielded accuracy scores as high as 90%, matching the performance of LSTM and random forest classifiers.

Download

Share

COinS