AI-Driven Predictive Maintenance: Remaining Useful Life Estimation for Turbofan Engines via Ensemble Learning, LSTM, and Uncertainty Quantification

Unplanned industrial equipment failures cause significant financial losses and
safety hazards across aviation, energy, and manufacturing sectors. Predictive Maintenance
(PdM) addresses this by estimating the Remaining Useful Life (RUL) — the number
of operational cycles before a machine requires intervention — enabling proactive, costoptimal scheduling. This paper presents a complete end-to-end machine learning pipeline
for RUL prediction on the NASA C-MAPSS turbofan engine degradation dataset [17]. We
compare four models — Linear Regression, Random Forest, Gradient Boosting, and a twolayer Long Short-Term Memory (LSTM) network implemented in pure NumPy  under
rigorous Group K-Fold cross-validation to eliminate engine-level data leakage. Our feature
engineering pipeline transforms 21 raw sensor channels into 37 temporal features including
rolling statistics, a composite health index, and degradation rate. Random Forest
achieves the best performance: RMSE = 9.95 cycles, R2 = 0.9398, MAE = 5.77,
and NASA PHM asymmetric score = 1.9. We further contribute bootstrap prediction
intervals achieving 90.0% empirical coverage at the 95% nominal level, and a maintenance
scheduling module projecting $430,000 in cost savings on a 20-engine fleet.