✈️ Remaining Useful Life (RUL) Prediction - CMAPSS Jet Engine Simulated Data

Source: NASA Prognostics Data Repository

📘 Project Overview

• This machine learning project predicts the Remaining Useful Life (RUL) of aircraft engines using NASA’s CMAPSS dataset.
• The goal is to estimate how many operating cycles each engine has left before failure, based on its sensor readings.
• The workflow follows a complete supervised regression pipeline, utilizing a Native Scikit-Learn Pipeline for seamless production deployment.

📊 Streamlit Web App – Remaining Useful Life (RUL) Prediction

This project includes an interactive Streamlit dashboard that visualizes the predicted Remaining Useful Life (RUL) for each engine unit in the NASA CMAPSS FD001 dataset.

🔗 Try the App Online: https://cmapss-rul-prediction.streamlit.app/

🛠️ Tech Stack & Environment

• Python 3.13+
• Libraries: Pandas 2.2.3 (Cloud Optimized), NumPy 1.26.4, Scikit-Learn 1.6.1, XGBoost 2.1.4, Plotly, and Streamlit.
• Serialization: joblib for native pipeline persistence and autonomous deployment.

🎯 This Project Focuses on FD001

The CMAPSS dataset contains four subsets (FD001–FD004) of increasing complexity.
This project uses FD001, providing a controlled environment to develop and validate the RUL prediction pipeline, allowing for:

• A clearer understanding of degradation patterns.
• Straightforward feature selection and model comparison.
• A solid performance baseline before extending to more complex scenarios.

⚙️ Workflow Summary

1. Data Preparation & Quality

   • Cleaned raw CMAPSS data using sep=r'\s+' for modern Pandas compatibility.
   • Handled column renaming and unit-specific life-cycle labeling.

2. Feature Engineering & Selection

   • Applied normalization and scaling using native ColumnTransformer.
   • Selected the Top 8 Sensors using Permutation Importance (PI) and correlation analysis.

3. Production-Ready Modeling (Stockout Strategy)

   • Abandoned custom classes in favor of a Native Scikit-Learn Pipeline.
   • Combined preprocessing (StandardScaler) and XGBoost into a single .pkl artifact.
   • Optimized for Streamlit Cloud by maintaining compatibility with Pandas < 3.0.

4. Retraining & Execution

   • Retraining: Automated script to update the model using native library components.
   • Execution: Production inference script that processes sensor data through the serialized pipeline without external dependencies.

🧩 Notebooks

#	Notebook	Description
1	01_setup.ipynb	Environment setup, library installation, and folder structure.
2	02_data_quality.ipynb	Data cleaning and initial integrity checks.
3	03_eda.ipynb	EDA: distributions, correlations, and degradation insights.
4	04_feature_engineering.ipynb	Scaling, normalization, and sensor transformations.
5	05_feature_preselection.ipynb	Feature selection using MI, RFE, and Permutation Importance.
6	06_modelling_regression.ipynb	Regression model training, tuning, and evaluation.
7	07_retraining_execution_scripts.ipynb	Production Core: Native pipeline assembly and retraining/execution logic.

📊 Key Metrics (FD001)

• MAE: 30.8 - Average prediction error ≈ 30 cycles.
• RMSE: 44.0 - Moderate error dispersion; the model generalizes well.
• R²: 0.62 - Explains ~62% of RUL variance.

📊 Dashboard Features

• Engine Selector: Choose any engine unit from the validation set.
• Interactive Plotly Chart: Dynamic theme adaptation for dark/light modes.
• Maintenance Thresholds:
  • 🟧 Warning: Below 80 cycles.
  • 🟥 Critical: Below 50 cycles.

📁 Files Used by the Dashboard

• app.py — Streamlit application code.
• /04_Models/nasa_model.pkl — The native production pipeline.
• /05_Results/predictions_validation_FD001.csv — Results for dashboard visualization.
• requirements.txt — Environment manifest.