Skip to main content
Machine Learning / Data Science·PROJECT_03

 

Predictive ML Jet Engine Analytics

Domain
Machine Learning / Data Science
Role
Lead ML Engineer
Stack
Python · XGBoost · Scikit-Learn · NASA C-MAPSS
Year
2026
Aero Guard project cover

Overview

Aero Guard is a predictive maintenance platform that forecasts Remaining Useful Life (RUL) of commercial jet engines from 21 multi-sensor telemetry channels, allowing proactive maintenance scheduling before structural degradation.

PythonXGBoostScikit-LearnPandasNumPyFeature Engineering
The Challenge

Unscheduled aircraft maintenance causes massive flight cancellations and financial losses. Traditional threshold alerts trigger only after degradation starts. The challenge was modeling non-linear thermal and pressure sensor degradation trajectories across hundreds of flight cycles.

System Architecture

Engineered rolling statistical aggregates (mean, std, EMA) across 21 engine sensors over multi-horizon windows. Trained and tuned an XGBoost gradient boosting regressor on NASA C-MAPSS simulation dataset with feature selection and hyperparameter optimization.

Validation Metrics

Validation Dataset
NASA C-MAPSS (FD001)
Model Architecture
Tuned XGBoost Regressor
Validation RMSE
11.42 Cycles (vs 14.5 baseline)
Telemetry Inputs
21 Sensor Rolling Aggregates
Key Highlights
Extracted non-linear degradation features from 21 multi-sensor telemetry channels.
Validation RMSE of 11.42 cycles — well below safety-critical threshold.
Early warning thresholds enable proactive scheduling before component failure.