Predictive Maintenance System (NASA CMAPSS)

📊 Overview

This project builds an end-to-end machine learning system to predict machine failures using time-series sensor data from the NASA CMAPSS dataset.

The system forecasts failures in advance, enabling proactive maintenance and reducing downtime.

---

🎯 Objective

• Predict machine failure before it occurs
• Use time-series sensor data (temperature, pressure, vibration)
• Provide interactive visualization and analysis

---

🧠 Key Features

• Time-series feature engineering (RUL calculation)
• Classification model using XGBoost
• Failure probability prediction
• Interactive Streamlit dashboard
• Per-engine analysis with trend visualization
• Dockerized for reproducible deployment

---

🏗️ Project Structure

predictive-maintenance-ml/
│
├── data/
│ ├── raw/ # Original NASA dataset
│ └── processed/ # Cleaned and feature-engineered data
│
├── src/
│ ├── data_preprocessing.py # Data cleaning + RUL creation
│ ├── train_model.py # Model training
│ ├── evaluate_model.py # Evaluation metrics
│ ├── predict.py # Inference logic
│
├── models/
│ └── model.pkl # Trained model
│
├── dashboard/
│ └── app.py # Streamlit dashboard
│
├── Dockerfile # Container setup
├── requirements.txt # Dependencies
├── README.md # Project documentation
└── .gitignore

---

⚙️ Tech Stack

• Python (Pandas, NumPy)
• Scikit-learn
• XGBoost
• Streamlit
• Docker

---

🚀 How to Run

Option 1: Run Locally

pip install -r requirements.txt
streamlit run dashboard/app.py

Option 2: Run with Docker

docker run -p 8501:8501 predictive-maintenance

Then open: http://localhost:8501

---

📈 Dashboard Features

• Upload dataset
• Predict failure probability
• View risky machines
• Engine-level analysis
• Failure trend visualization

---

🧠 Machine Learning Approach & Benchmark Evaluation

The system evaluates engine health using a dual framework:

1. Binary Failure Classification (Predicting whether an engine will fail within 30 cycles)
2. Remaining Useful Life (RUL) Regression (Predicting exact remaining operating cycles)

📊 Evaluated Metrics (Supported & Claimed):

• 🎯 Classification Metrics (Failure Risk):
  • Accuracy: Measure of total overall correct predictions across all healthy and failing engines.
  • Precision: Ratio of true positive failures out of all predicted failures (minimizes false alarms).
  • Recall & F1-Score: Evaluates true failure detection rate and overall harmonic balance.
• 📈 Regression Metrics (RUL Forecasting):
  • RMSE (Root Mean Squared Error): Measures standard deviation of prediction errors in operating cycles (penalizes large errors).
  • MAE (Mean Absolute Error): Average absolute cycle difference between predicted and actual RUL.
  • R² Score (Coefficient of Determination): Proportion of variance in RUL explained by engine sensor telemetry.

---

💡 Key Insight

• The model predicts increasing failure probability and decreasing continuous RUL as the machine approaches end-of-life cycles.

---

📦 Deployment

• Containerized using Docker
• Can be deployed on:
  • Streamlit Cloud
  • Render
  • AWS / GCP

---

📌 Future Improvements

• SHAP explainability
• Real-time IoT data streaming
• Multi-engine comparison dashboard

---

👨‍💻 Author

NIMALAN MANI M

---

⭐ If you like this project

Give it a star ⭐ on GitHub