Free-Form Trap Design for Vibratory Feeders using a Genetic Algorithm and Dynamic Simulation

TLDR: A Self-Adaptive Genetic Algorithm based learning strategy that uses dynamic simulation to validate feeder designs that learns a free-form design needing little prior domain knowledge from the designer.

Abstract: The task of feeding parts into a manufacturing system is still extensively handled using classical vibratory bowl feeders. However, the task of designing these feeders is complex and largely handled by experience and trialand-error. This paper proposes a Self-Adaptive Genetic Algorithm based learning strategy that uses dynamic simulation to validate feeder designs. Compared to previous approaches of ensuring parts are oriented to a desired orientation by both deciding on a set of suitable mechanisms and then optimizing them to the specific part, this strategy learns a free-form design needing little prior domain knowledge from the designer. This novel approach to feeder design is validated on two different parts and it creates designs of hills and valley that reorients the parts to a single orientation. The found designs are validated both in simulation and with real-world experiments and achieve high success rates for reorienting the parts.