XGBoost Regression Model for Remaining Useful Life Prediction - Case Study of Turbofan Engines

Abstract

This paper focuses on the prediction of Remaining Useful Life (RUL) for turbofan engines in the context of Predictive Maintenance (PdM) in Industry 4.0. The study is based on the NASA C-MAPSS dataset and focuses on the development of a predictive stage, including data preprocessing, feature engineering, normalisation, and sensor selection. An XGBoost regression model was evaluated on four benchmark subsets of increasing operational complexity. The obtained results show that prediction accuracy strongly depends on operating conditions, with RMSE values ranging from 20.97 to 29.64 cycles. The analysis indicates that operational variability has a greater impact on model performance than the number of fault modes alone. Additional investigation of sensor degradation trends, data distributions, and outliers provided insight into the factors affecting prediction quality. The results confirm that carefully designed machine learning pipelines can effectively support condition monitoring and maintenance planning in complex engineering systems.