Manufacturing Defect Detection

Manufacturing Defect Detection is an advanced computer vision system designed to automatically identify and classify defects in manufactured products, specifically focusing on cups and television sets. Our team "Deep Dreamers" developed this project during an intensive 3-day hackathon at Thakur College of Engineering and Technology (TCET), where we won 2nd prize in the Mind Bender Club Hackathon.

The project utilizes the YOLO v5 (You Only Look Once) object detection architecture, which offers real-time processing capabilities critical for production line integration. What makes this project unique is the custom dataset we collected and annotated, consisting of hundreds of images of cups and TVs with various types of manufacturing defects. This manual data acquisition process ensured that the model could be trained on highly relevant examples that accurately represent real-world quality control scenarios.

The system achieves high accuracy in detecting subtle defects such as scratches, dents, color inconsistencies, and structural abnormalities. It efficiently classifies items as either "good" or "broken/defective" with clear visual indicators. We trained and optimized the model on local hardware, demonstrating how effective AI solutions can be developed within tight timeframes and without relying on expensive cloud resources.

Our team of four second-year students (Ishan Naik, Kunal Pawar, Akshay Vennikkal, and Anusha Yadav) collaborated intensively during the hackathon, dividing responsibilities across data collection, model training, optimization, and presentation. This collaborative approach allowed us to deliver a complete solution within the challenging 3-day timeframe.