AI Quality Assurance - Real-Time Defect Detection System

This project is built for Cyfuture Hackathon 1.0 – Challenge 3: Quality Assurance with AI, focused on real-time defect detection in manufacturing using computer vision.

---

🚀 Overview

An end-to-end AI-powered quality assurance system that:

• Detects visual defects in manufacturing (bottle caps/necks) using autoencoders
• Works in real-time using webcam or uploaded images
• Generates reports and QR codes for traceability
• Visualizes reconstructions and anomaly maps for insights

---

🧠 Core Features

• 🔍 Real-time Anomaly Detection (autoencoder + MSE)
• 📷 Camera & Image Upload Support
• 📈 Live Confidence Scoring
• 🧾 Automated Report Generation
• 📦 Unique QR Code for Traceability
• 🧊 Freeze-Frame Analysis + Redo Button
• 📊 Defect Visualization: Original vs Reconstructed

---

📷 Application Preview

Dashboard

Dashboard

Quality Passed

Quality Pass

Defective

Defective Image Screen with Visuals

Real-Time Checking

Real-Time Checking with visuals

---

📸 Demo

📹 Demo Video Link

---

📂 Folder Structure

├── app.py                         # Flask backend server
├── script.js                     # Frontend logic
├── style.css                     # UI styling
├── index.html                   # Main dashboard UI
├── train_autoencoder_bottle.py   # Model training script
├── detect_anamolies_bottle.py    # Model evaluation logic
├── bottle_autoencoder.h5         # Pre-trained model
├── product_traceability_bottle_log.csv # Product history log
├── qr_codes_bottle/              # Generated QR codes
└── bottle/                       # MVTec dataset (test images)

---

⚙️ How It Works

1. Flask backend receives image via upload or webcam frame.
2. The image is preprocessed (resized, normalized).
3. Autoencoder reconstructs it; MSE is calculated.
4. Threshold is applied to determine defect.
5. UI updates results, visualizations, and allows QR/report generation.

---

🧪 Tech Stack

• Frontend: HTML5, CSS3, JavaScript (Vanilla)
• Backend: Python Flask
• AI Model: TensorFlow Autoencoder
• Visualization: Matplotlib + Base64 for web
• Data Storage: CSV

---

🏗️ Setup Instructions

# 1. Clone the repo
$ git clone https://github.com/yourusername/ai-quality-inspection
$ cd ai-quality-inspection

# 2. Install dependencies
$ pip install -r requirements.txt

# 3. Train (Optional)
$ python train_autoencoder_bottle.py

# 4. Run app
$ python app.py

# 5. Open browser
Visit http://127.0.0.1:5000/

---

📌 Limitations & Next Steps

• Current model works well with MVTec dataset
• Real-world webcam accuracy may vary (needs fine-tuning or retraining)
• Real-time prediction is enhanced with freeze-frame + manual re-detect
• Future improvement: YOLO-based cap detection, cloud traceability

---

🧠 Authors & Acknowledgements

• Developed by - CodeClan
• Special thanks to the Cyfuture Hackathon 1.0 team for this opportunity
• Dataset from MVTec Anomaly Detection Dataset

---

📄 License

This project is licensed under the MIT License.
You are free to use, modify, and distribute this software for personal or commercial purposes, as long as proper credit is given to the original author.

Please do not remove the original license or claim authorship.
For attribution, retain the following: © 2025 preritasaini1

---

Made with ❤️ for the future of smart manufacturing.