MineD | 2025

🚀 Manufacturing Use Case Solution With AI

Developed By: Team BigBrains, Nirma University
For: JK Lakshmi Cements & Udaipur Cements

---

📌 Problem Statement

🚧 Challenges in Rail Wagon Inspection

• High Dependency on Manual Labor: Current inspections rely heavily on human effort, making them time-consuming and inconsistent.
• Risk of Human Error: Manual verification of wagon numbers and structural damages is prone to inaccuracies.
• Inefficient Volume Measurement: Estimating the material volume in loaded wagons lacks precision, leading to inefficiencies.

🔍 Our Solution

We have developed an automated AI-powered inspection system leveraging computer vision to enhance the safety, accuracy, and efficiency of rail wagon monitoring. Our system captures real-time images of wagons, analyzes them for:

• Structural damage detection
• Cargo verification
• Wagon counting
• Damage spot analysis
• Automated PDF report generation

🛠 Implementation

🔹 Phase 1: Data Preprocessing & Image Extraction

• Extracted images from raw rail wagon footage (incoming and outgoing wagons).
• Labeled training datasets using CVAT tool for supervised learning.

🔹 Phase 2: Deep Learning Model Training

• Leveraged Detectron2 framework to train a state-of-the-art image recognition model.
• Detected:
  • Damaged parts of wagons
  • Cement residue post-loading
  • Synthetic images generated by deep learning models

🔹 Phase 3: Wagon Counting Using OpenCV

• Implemented a highly accurate wagon counting algorithm using OpenCV & NumPy.
• Achieved a 95% accuracy rate with a low loss value of 0.1.

🔹 Phase 4: AI-Powered Report Generation

• Analyzed images and extracted insights using Google Gemini AI.
• Compiled structured, comprehensive PDF reports containing:
  • Damage analysis
  • Wagon count & tracking
  • Cargo residue identification

🎯 Why Choose Our Solution?

✅ High Accuracy & Reliability

• Built on Detectron2 and OpenCV frameworks.
• Delivers 95% accuracy in wagon counting & defect detection.

✅ Comprehensive Reporting

• Generates detailed PDF reports with images and AI-generated summaries.
• Enables quick decision-making and compliance verification.

✅ Scalability & Automation

• Fully *automated, reducing dependency on *manual labor.
• Suitable for large-scale manufacturing & transport logistics.

🚨 Limitations

🔹 Day-Night Contrast Issues: Slight degradation in accuracy due to lighting variations.
🔹 *Manual Labeling Overhead: Requires *manual dataset preparation using CVAT tool.
🔹 *Volume Detection Constraints: Further improvement needed in *precise volume estimation.

📂 Demo & Results

🚧 Project Demonstration

• 📹 *Video Demonstration: *(To be added)

• 🖼 Sample Images:

  

  

📜 Generated PDF Report:

https://github.com/say-het/MINeD-Hackathon-Team-BigBrains/blob/main/wagonReport_compressed.pdf

📌 Tech Stack & References

🔹 Deep Learning: Detectron2 (Facebook AI Research)
🔹 Computer Vision: OpenCV, NumPy
🔹 AI Text Processing: Google Gemini AI
🔹 Data Labeling: CVAT
🔹 Frameworks: PyTorch

Flowcharts:

Model Training:

Solution WorkFlow:

📑 References:

🏆 Team BigBrains

🔹 Het Modi
🔹 Raj Mistry
🔹 Krish Chothani
🔹 Mihir Khunt
🔹 Param Shankar

📌 Developed at Nirma University
📌 Part of JK Lakshmi Cements & Udaipur Cements AI Challenge

---

🚀 Contributions & Feedback
We welcome feedback and contributions! Feel free to open issues or contribute enhancements to the project.

💡 Let's revolutionize industrial AI together!