Aircraft Engine RUL Prediction AI

Overview

This machine learning project estimates the Remaining Useful Life (RUL) of turbofan aircraft engines using NASA's C-MAPSS simulated degradation dataset. It demonstrates how multivariate sensor measurements and operational cycles can predict how many cycles remain before engine failure — enabling proactive maintenance planning instead of reactive repair.

Purpose — Why This Project Exists

Unscheduled engine failures are expensive, disruptive and dangerous, while fixed-interval servicing wastes good component life. This project shows how airlines can shift toward condition-based maintenance — using real sensor trends instead of a calendar to decide when an engine actually needs attention — by comparing several regression models on real NASA degradation data.

Animated Pipeline

1

Sensor Data

C-MAPSS temperature, pressure & speed readings.

2

Preprocessing

Cleaning and normalizing run-to-failure trajectories.

3

Feature Engineering

Degradation-relevant features extracted per cycle.

4

Model Training

Random Forest, XGBoost and LightGBM compared.

5

Evaluation

RMSE, R², learning curves and error analysis.

6

RUL Prediction

Cycles remaining until failure, per engine.

Key Features

Multivariate Sensor Analysis — temperature, pressure, rotational speed and other engine parameters.

Multiple ML Models — Random Forest, XGBoost and LightGBM regression compared side by side.

Comprehensive Evaluation — RMSE, R², feature importance, learning curves and error analysis.

Modular Codebase — separate data loading, preprocessing, training and evaluation modules.

NASA C-MAPSS Dataset — authentic turbofan run-to-failure simulation data.

Predictive Maintenance Focus — bridges sensor data with maintenance decision-making.

Benefits

Less Downtime

Predicts failures before they ground an aircraft unexpectedly.

Extends Component Life

Condition-based scheduling avoids servicing parts that still have life left.

Model Reliability

Three models compared side by side so the best performer wins.

Real Degradation Data

Trained on authentic NASA run-to-failure engine trajectories.

Evaluation Metrics

RMSE (Root Mean Squared Error) and R² Score are used to measure prediction accuracy and how well each model explains RUL variation.

Technology Stack

Python Pandas NumPy Scikit-learn XGBoost LightGBM Matplotlib Seaborn Kaggle

Explore the Repository