Automatic visual defect detection in industrial products using deep learning methods

This repository contains our team's (NTUA-INDUSTRY) project for the competition Industry 4.0 hosted by crowdpolicy

🏭 Project Description

Quantity and quality automation in manufacturing is a demand not a privilege. In this direction, our team proposed a system to resolve the material quality control problem. Human Inspection is an error prone & time consuming process and results in false positives
- Parts corectly detected as faulty provoke loss in upstream labor, consumables as well as revenue
- Parts wrongly detected as faulty provoke a market reaction and eventually a unrepairable damage to company's reputation
In that direction every automation system that can early detect false positives without human intervation, while being cost-effective for the customer, is preffered.

We automated defect detection using deep learning, and more specicially convolutional neural networks. The model is trained on steel images, since it is the most important building material of modern times.