sddec21-17 • Automatic Detection of Paint Defects using Deep Learning

This theme uses "root" links to dependencies and assets (CSS, JS, images, etc.). Place all theme files and directories at the root of your web server for this page to render as intended.

Senior Design Team sddec21-17 • Automatic Detection of Paint Defects using Deep Learning

Problem Statement

Danfoss Central Paint and Packaging manufactures painted pumps. Over an 18-month period, there have been a large number of defective pumps produced where the paint job has either been incomplete or poor. Approximately one-eighth of these defects were classified as high-impact defects which indicates that they were most likely scrapped. The combined cost of scrap and rework for both high-impact and low-impact defects for this period is very high, and Danfoss would like to reduce, or even completely eliminate, this cost.

Pump Classifications

The different classifications of pumps (non-defective or type of defect) are shown below.

Solution Approach

Our solution approach is to implement a system which captures images of pumps and classifies the pumps as defective or non-defective. When a defective pump is detected, the system notifies staff members on the floor so that they can shut down the line and repair the painting robot.