MaHi-HF28/turbofan-predictive-maintenance

🛩️ Turbofan Engine Predictive Maintenance System

   

A deep learning–powered predictive maintenance system trained on NASA's CMAPSS turbofan engine dataset. A dual-output LSTM model predicts Remaining Useful Life (RUL) in cycles and raises binary failure alerts when an engine enters a critical degradation window. The live dashboard visualizes fleet health with color-coded alert levels.

🚀 Live Demo

The dashboard is running above. Use the dropdowns to:

• —Select any engine from the fleet
• —Switch between sensor channels to inspect degradation curves
• —View the RUL gauge and alert status per engine
• —Sort and filter the full fleet alert table

🧠 Model Architecture

Input: sliding window of 30 sensor cycles  →  shape (30, n_features)
         │
   LSTM(128, return_sequences=True)  →  BatchNorm  →  Dropout(0.3)
         │
   LSTM(64,  return_sequences=False) →  BatchNorm  →  Dropout(0.2)
         │
   Dense(32, relu)   ←── shared representation
      ╱                                      ╲
Dense(1, linear)                      Dense(1, sigmoid)
  RUL output                            Alert output
(regression, MSE)               (binary, BCE, weight=0.5)

Two outputs trained jointly:

• —RUL regression — predicts exact cycles remaining (0–125)
• —Alert classifier — predicts probability of failure within 30 cycles

🚨 Alert Levels

📊 Results on FD001 Test Set

🗂️ Dataset — NASA CMAPSS

This Space uses FD001 — single fault mode, single operating condition, 100 train / 100 test engines.

Each row = one engine cycle: engine_id | cycle | 3 op-settings | 21 sensor readings

Download: NASA Prognostics Data Repository

📁 Space File Structure

├── app.py
├── Dockerfile
├── requirements.txt
├── saved_model/
│   ├── rul_model.keras
│   ├── scaler.pkl
│   ├── feature_cols.npy
│   ├── pred_rul.npy
│   └── true_rul.npy
└── CMAPSSData/
    └── test_FD001.txt

📄 License & Credits

License: MIT

Dataset: Saxena, A., Goebel, K., Simon, D., & Eklund, N. (2008). Damage Propagation Modeling for Aircraft Engine Run-to-Failure Simulation. NASA Ames Research Center, Moffett Field, CA.