SteelEye

AI-Powered Ladle Tracking System for Modern Steel Plants

Steel Ladle
Ladle with Number
UI
UI

SteelEye Overview

An end-to-end computer vision solution that automates ladle tracking in steel manufacturing, delivering real-time analytics and operational intelligence through advanced computer vision and an intuitive web interface.

SteelEye

Intelligent ladle tracking system designed to revolutionize steel plant operations

YOLOv8 FastAPI React.js RAG Chatbot

Revolutionizing Steel Plant Operations

SteelEye is an intelligent ladle tracking system designed to revolutionize steel plant operations by automating the detection and monitoring of ladles throughout their lifecycle. By leveraging computer vision and deep learning, SteelEye eliminates manual tracking inefficiencies and provides real-time operational insights.

The system detects handwritten ladle numbers using a fine-tuned YOLOv8 model, maps them to critical parameters like temperature and circulation time, and visualizes comprehensive analytics through an intuitive web dashboard. An integrated RAG-based SQL chatbot enables operators to query ladle data conversationally, making tracking accessible and actionable.

  • Automated Ladle Number Detection with YOLOv8
  • Real-Time Analytics and monitoring
  • History Tracking and lifecycle management
  • Intelligent SQL Chatbot with RAG technology
  • Database Integration with FastAPI and MySQL

Current Status: Prototype successfully tested with single-camera deployment; multi-camera setup will be tested on site soon

The Problem

Optimizing ladle logistics and temperature management in steel plants

Why Ladle Tracking Matters

In steel manufacturing, ladles transport molten steel from converters to casting mills. Maintaining optimal temperature (superheat) throughout this journey is critical for casting quality. However, traditional tracking methods face several challenges:

  • Manual Monitoring: Inefficient, error-prone, unreliable and labor-intensive
  • Temperature Loss: Extended transit times reduce superheat, affecting product quality
  • Ladle Life: Excessive heat exposure erodes linings, shortening ladle life
  • Operational Inefficiency: Lack of real-time location and history data hampers planning

Impact on Operations

Without automated tracking, operators cannot:

  • Assess ladle health before converter processes
  • Minimize temperature loss in a cycle through optimized scheduling
  • Monitor ladle circulation and turnaround times
  • Access historical maintenance and repair data instantly
  • Find time-consuming bottlenecks in the operation and investigate other issues to reduce temperature loss etc.
Steel ladle with handwritten number
Steel ladles with handwritten identification numbers

Our Solution

Real-time detection, tracking, and analytics powered by Computer Vision

Key Innovation

SteelEye introduces a vision-based ladle tracking system that uses multiple camera feeds to detect ladle numbers (handwritten or printed) and maps them to teh camera location in the plant, hence tracking the ladle from unit to unit.

Other solutions like GPS and RFID come with problems of device cost, installation cost, maintenance cost since the steel ladle is at high temperature, for any device to be put on it, it will need a cooling system, and frequent charging.

Hence, our solution is to put up a camera unit at every unit/sub-unit in the plant where ladle would arrive for further processing and detect the ladle number hence track each ladle's location.

Key Parameters Tracked

  1. 1. Ladle Circulation Time – Total time from tapping (loading metal into the ladle from the BOF Converter) to return
  2. 2. Ladle Life – Cumulative usage and remaining lifespan
  3. 3. Ladle Return Time – Duration from casting completion to availability
  4. 4. Ladle Turnaround Time – Time between consecutive uses
  5. 5. Ladle Temperature & Superheat – Real-time thermal monitoring

System Capabilities

  • Automated Detection: Fine-tuned YOLOv8 model detects ladle numbers with high accuracy
  • Real-Time Analytics: Live dashboard displays ladle locations, temperature, and workflow status
  • Historical Tracking: Complete ladle history, maintenance records, and lifecycle analytics
  • Intelligent Chatbot: RAG-powered conversational interface for querying ladle data
  • Database Integration: FastAPI backend with MySQL database for robust data management
Real-time ladle detection
Our Ladle Number and Ladle Detection Testing on-site
Ladle detection demo
Ladle Detection and Tracking Demo

System Architecture

Technology stack and deployment architecture

Technology Stack

  • Computer Vision: YOLOv8 object detection model fine-tuned on 200+ images of steel plant ladles
  • Backend: FastAPI (Python) with MySQL database integration
  • Frontend: React.js for dynamic, responsive user interface
  • AI Assistant: RAG chatbot with SQL query generation for conversational data access
  • Deployment: Multi-camera support with centralized processing

Model Performance

Our YOLOv8 model was fine-tuned on a custom dataset that we captured and labeled at Rashtriya Ispat Nigam Limited (Vizag Steel). Every ladle and handwritten number was annotated in-house before training, and we incorporated diverse augmentation strategies to handle challenging conditions such as:

  • Varying lighting and angles
  • Motion blur and occlusion
  • Handwritten number variations
  • Industrial environmental noise
System architecture
Our System Integration in General Steel Plant Layout
Detection pipeline
Overall System Architecture
Data flow diagram
Real-Time Data Ingestion, Detection, Database and Web App Integration
Deployment architecture
SQL Chatbot

Project Impact

Benefits SteelEye delivers across the steel plant workflow

Benefits to Steel Plant Operations

  • Reduced Manual Labor: Eliminates hand-written logs and duplicate data entry
  • Energy and Time Efficiency Since temperature loss will be minimal and reheating would not be needed before casting, time and energt are saved
  • Temperature Management: Minimizes superheat loss by poitning out inefficiencies
  • Extended Ladle Life: Protects refractory linings by tracking cumulative exposure
  • Enhanced Transparency: Real-time visibility for shop-floor operators and plant leadership
  • Data-Driven Decisions: Historical analytics unlock predictive maintenance insights
  • Improved Productivity: Faster turnaround times by enabling proactive planning

Web Application

Comprehensive dashboard for monitoring all ladle operations in real-time

Comprehensive Dashboard

The SteelEye web application provides a centralized control center for monitoring all ladle operations in real-time. Key features include:

  • Live Tracking: Real-time ladle location visualization across plant zones
  • Temperature Monitoring: Current and historical temperature data with alerts
  • Analytics Dashboard: Circulation time, turnaround metrics, and efficiency insights
  • Maintenance History: Complete lifecycle records and repair schedules
  • RAG Chatbot: Natural language queries for instant data retrieval
Dashboard overview
Real-Time Dashboard With Live Ladle Tracking
Analytics view
Ladle Movement Between Units
Temperature monitoring
Ladle Details Page
Historical data
Ladle History and Maintenance Records
Chatbot interface
RAG SQL Chatbot for Conversational Data Queries
Location tracking
SMS Units of Steel Plants in UI
Alert system
Manage Ladles Page
Reporting module
Manage Cameras in Sub-Units Page
Admin panel
Ladle Movement Tracking Display

Project Team

The people and mentors behind SteelEye

Team Lead

Prudhvi Gudla

Project Lead & Computer Vision Engineer steering the project

Team Members

  • Ujjawal Kumar Web Developer and GenAI Engineer
  • Shreyansh Vansh Verma Principal Computer Vision Engineer
  • Subhrojyoti Mukherjee Data Scientist and Industry Outreach Coordinator

Supervision & Acknowledgments

Prof. Sudip Misra
Department of Computer Science and Engineering
Indian Institute of Technology Kharagpur

Mr. Biswajit Ghosh
Project Mentor

SWAN Lab (Smart Wireless Applications and Networking)
Indian Institute of Technology Kharagpur

We gratefully acknowledge the Visakhapatnam Steel Plant (RINL) for access to facilities and support during data collection.