A Data Collection System for Future Automation Applications

Thumbnail Image
Hall, Grant
Journal Title
Journal ISSN
Volume Title
University of Guelph

A local automotive manufacturer in Guelph, Ontario is responsible for the production of automotive powertrain components and is aiming to automate its inspection process for one of their pinion gears. Their current inspection process is both time and labour intensive and is subject to human error such as improper training or worker fatigue. To combat this issue, the manufacture wishes to automate the inspection process and integrate it with their current workflow. The first step in implementing any machine learning algorithm is to build a database of images from which the system can learn. In order to acquire the required images, a data collection test cell was fabricated to collect images of defective parts and was installed on site at the manufacturer’s facility. A total of 19,619 images across 448 parts were collected and labeled, and are to be used in a defect detection algorithm.

Data Collection, Machine Vision, Machine Learning, Defect Detection