⚡ Luminous Predictive Maintenance (PdM) Command Center

APOGEE Innovation Challenge | BITS Pilani

📌 Project Overview

Relays and contactors in inverters do not have a fixed expiry date; their lifespan is dictated by switching cycles, electrical load, and thermal stress. This project delivers an AI-driven, Physics-of-Failure based solution to estimate the Remaining Useful Life (RUL) of these components.

By correlating power quality, thermal dynamics, and load currents, the system provides real-time predictive diagnostics and early warning alerts at both the Edge (Inverter level) and the Cloud (Command Center dashboard) to prevent unplanned inverter downtime.

👥 Developers

Deepak R * A. Thanigai Malai

🏗️ System Architecture

This solution is built as a real-time simulation model utilizing a dual-layer IoT architecture:

Machine Learning Model (model.py): A RandomForestRegressor trained on historical physics-of-failure patterns to correlate Temperature, Inverter Current, Mains Voltage, and Switching Cycles into a highly accurate RUL percentage and Time-to-Failure (TTF) cycle count.
Edge Intelligence (edge_system.py): Acts as the Inverter’s local processing unit. It simulates the full 17-register Luminous MODBUS map, runs the ML model locally for fast inference, and publishes the telemetry via MQTT.
Cloud Command Center (app.py): A Streamlit dashboard that acts as the cloud/server UI. It fetches MQTT data to visualize degradation trends, monitor power quality, and trigger critical UI alerts for Service Engineers.

✨ Key Features

Full Modbus Support: Maps and transmits all 17 specified Luminous inverter registers.
Edge AI Inference: Performs the heavy ML predictions locally to save cloud bandwidth.
Early Warning System: Triggers predictive alerts (e.g., “Failure predicted within 500 cycles”) before hardware breaks.
Dynamic Visualization: Live tracking of Grid Stability (Voltage/Hz), Load vs. Temp dynamics, and RUL trajectories.

🛠️ Tech Stack

Language: Python 3.x
Machine Learning: Scikit-Learn (RandomForestRegressor), NumPy
IoT Protocol: Paho-MQTT (MODBUS RTU mapped over MQTT)
Frontend: Streamlit
Data Processing: Pandas, JSON

🚀 Getting Started

1. Prerequisites

Ensure Python is installed, then install the required dependencies:

pip install paho-mqtt streamlit pandas scikit-learn numpy joblib

2. Execution

This project requires running the system in split terminal windows to simulate the Edge and the Cloud simultaneously:

Terminal 1 (Train the AI - Run Once):

python model.py

(This generates the relay_model.pkl file used by the Edge device).

Terminal 2 (Start the Edge Brain):

python edge_system.py

Terminal 3 (Launch the Cloud Dashboard):

streamlit run app.py

📋 Modbus Register Telemetry Map

Developed for APOGEE 26