🔧 Predictive Maintenance using Machine Learning

📌 Project Overview

How do we know a bearing is about to fail… before it actually does?
This project explores that question using Machine Learning applied to real vibration data from experimental setups.

The goal: Develop a predictive maintenance system that detects early fault detection in bearings, helping industries reduce downtime, lower costs, and improve safety.

---

⚙️ Dataset

Dataset	Duration	Files	Channels	Faults Observed
Set 1	Oct 22, 2003 – Nov 25, 2003	2156	8 (2 per bearing: x- and y-axis)	Bearing 3: Inner race defect; Bearing 4: Roller element defect
Set 2	Feb 12, 2004 – Feb 19, 2004	984	4 (1 per bearing)	Bearing 1: Outer race defect
Set 3	Mar 4, 2004 – Apr 4, 2004	6324	4 (1 per bearing)	Bearing 3: Outer race defect

---

🧠 Methodology

1. Data Preprocessing → Cleaning, balancing, cutoff handling (e.g., special case for roller element defect in Set 1).
2. Feature Engineering → Time-domain features from vibration signals.
3. Model Training → Compared multiple ML classifiers.
4. Evaluation → Accuracy, Precision, Recall, F1-score.

---

🎯 Results

• Final model achieved ~87% accuracy on unseen test data.
• Key finding: Roller element defect in Set 1, Bearing 4 showed early critical signs at 81.3% of lifetime, unlike other bearings.

---

💡 Why It Matters

• 💰 Cost Reduction → Planned maintenance vs. costly emergency repairs
• 🛡️ Safety Enhancement → Avoid catastrophic failures & hazards
• ⚙️ Equipment Longevity → Extend machinery lifespan
• 📈 Production Efficiency → Minimize downtime, optimize operations

---

🛠️ Tech Stack

• Python 🐍
• Scikit-learn 🤖
• Pandas & NumPy 📊
• Matplotlib 📉

---

📚 References & Dependencies

• Dataset:

  • NASA Bearing Dataset: Kaggle
  • IMS Bearing Dataset: Raw vibration signal data from the NSF I/UC Center for Intelligent Maintenance Systems (IMS), University of Cincinnati

• Libraries Used:

  • Data Processing:
    • numpy: Numerical computing
    • pandas: Data manipulation
  • Visualization:
    • matplotlib: Basic plotting
    • seaborn: Statistical visualization
  • Machine Learning:
    • scikit-learn: Random Forest implementation and metrics
    • pathlib: File path handling

---

📢 Connect

• LinkedIn