🚀 Predictive Maintenance ML Platform

A production-style, containerized machine learning system for real-time predictive maintenance simulation.

This project demonstrates how to design, deploy, monitor, and simulate an ML-powered system using modern MLOps and backend engineering practices.

---

📌 Overview

This system simulates an industrial predictive maintenance platform where multiple machines continuously stream sensor data to a containerized ML API that:

• Performs real-time inference
• Applies hybrid risk logic (ML + domain rules)
• Tracks machine lifecycle states
• Logs incidents
• Exposes monitoring endpoints
• Persists critical failure events

The system is built with production-oriented architecture principles rather than a notebook-only ML demo.

---

📸 System Preview

Live Fleet Monitoring Dashboard

REST API Documentation

Event Retrieval Endpoint

🏗 System Architecture

    +-----------------------+
    |  Multi-Machine        |
    |  Streaming Simulator  |
    +-----------+-----------+
                |
                v
        +----------------+
        |  FastAPI ML    |
        |  Inference API |
        +--------+-------+
                 |
                 v
       +-------------------+
       | Hybrid Risk Engine|
       | (ML + Rules)      |
       +--------+----------+
                |
                v
    +---------------------------+
    | Fleet Monitoring Dashboard|
    +---------------------------+
                |
                v
       +--------------------+
       | Event Persistence  |
       | events_log.json    |
       +--------------------+

---

⚙️ Tech Stack

• Python
• XGBoost
• FastAPI
• MLflow
• Docker
• Rich (Live Console Dashboard)
• Threading (Concurrent simulation)
• Structured Logging

---

🔥 Key Features

✅ ML model training with MLflow tracking
✅ Containerized inference service (Docker)
✅ Real-time REST API (/predict)
✅ Observability:

• Request logging
• Latency tracking
• /health
• /metrics ✅ Multi-machine concurrent simulation
  ✅ Hybrid risk scoring (ML + rule-based logic)
  ✅ State transitions: NORMAL → WARNING → CRITICAL
  ✅ Incident persistence (events_log.json)
  ✅ Event retrieval API (/events)
  ✅ Live fleet monitoring dashboard

---

🧠 Hybrid Risk Logic

The system does not rely purely on ML probability.

It combines:

• Model prediction probability
• Tool wear thresholds
• Temperature spikes
• Torque stress

This simulates realistic industrial monitoring behavior where AI is integrated with domain knowledge.

---

📊 API Endpoints

Endpoint	Description
/predict	Perform ML inference
/health	Service health check
/metrics	Request statistics
/events	Retrieve persisted critical events

---

🐳 Running With Docker

Build Docker Image

docker build -t predictive-maintenance-api .


## 🌍 Live Deployment

This service is deployed on AWS EC2.

Public endpoints:

- /health
- /predict
- /events
- /docs

Deployment includes:

- Docker container
- SQLite persistent volume
- Public security group configurationg