Operating-Regime-Aware Feature Engineering and Tree-Based Learning for Remaining Useful Life Prediction on NASA C-MAPSS FD002

Research Paper Description This research investigates Remaining Useful Life (RUL) prediction of aircraft turbofan engines using the NASA C-MAPSS FD002 dataset. It explores machine learning approaches for estimating engine degradation and predicting the number of operational cycles remaining before failure. The study focuses on operating-regime-aware feature engineering, incorporating sensor measurements, temporal differences, rolling statistical features, K-means-based operating-regime identification, and regime-conditioned sensor normalization. Multiple regression and gradient-boosting algorithms, including Random Forest, XGBoost, CatBoost, and LightGBM, are implemented and evaluated. Experimental results demonstrate that LightGBM achieved an RMSE of 16.1187 cycles, MAE of 11.1703 cycles, and R² of 0.8512 on the internal validation dataset using a capped RUL target. The research also examines methodological limitations and opportunities for improving predictive reliability through independent test-set evaluation and feature ablation studies. Research Focus: Machine Learning, Predictive Maintenance, Remaining Useful Life Prediction, Feature Engineering, Operating-Regime Analysis, and Industrial AI.