2026
AeroPulse
Turbofan predictive maintenance
Remaining-useful-life forecasting for NASA C-MAPSS turbofan engines with XGBoost, calibrated uncertainty and explainable AI, end to end and fully local.
AeroPulse turns turbofan engine telemetry (the NASA C-MAPSS dataset) into a maintenance workspace. It estimates each engine's remaining useful life, ranks the fleet by urgency, shows the uncertainty of every estimate and explains which signals drove the prediction.
It is an end-to-end machine-learning product, not a notebook: reproducible data acquisition, leakage-safe temporal features, model comparison, calibrated uncertainty intervals, a typed FastAPI service and a React interface, all served locally.
Fleet by priority
All 100 test engines ranked by predicted remaining life and split into critical, watch and stable bands.
Windowed XGBoost
17.2-cycle RMSE on the official test set, compared against a Ridge baseline kept as a reference.
Uncertainty, unvarnished
A ±29.9-cycle interval from validation residuals. Measured test coverage was 82%, below the nominal 90%, and the app shows that gap.
Explainable predictions
Global feature importance plus per-engine local XGBoost contributions.