Development of a Machine Vision System to Localise a Zinc Die Cast Product

The thesis presents a system to automate a manual/repeatable process in an Auckland, New Zealand manufacturing facility using predominantly machine vision techniques. An overview of the research \cite{BUT2019} has been accepted into The 2nd International Conference on Control and Computer Vision (ICCCV 2019). A manual/repeatable process has previously been required in the production of zinc die cast products. Where a worker stands at the end of a conveyor and picks up incoming die cast outputs for processing. To automate the process, a machine vision proof of concept was developed including four elements. The proposed system decides whether the incoming die cast objects are face up or down on the conveyor, determines the robot pick location and object orientation and conducts a quality control measure to check whether the correct cast is in production. The proposed system was successfully capable of checking the cast face, determining the robot pick location and orientation along with checking for error for a set of four die cast samples provided by the company. In some cases, small levels of error were corrected for using post vision manufacturing processes including mechanical nests and custom built robot gripping tools