Low-cost tracking of assembly tasks in industrial environments

Abstract

The fourth industrial revolution brings a lot of new challenges to the production process in a so called smart factory. The flexible production process and many different product variants call for assisting systems to support workers during their assembly tasks. We conducted a Contextual Inquiry in a real-life production environment to find typical problems during assembling products. In our work we use the General Assembly Task Model (GATM) proposed by Funk et al. [13] to identify and assess potential assistance systems in how they can supports each phase of an assembly step. Our analysis revealed that tracking of assembly tasks is very helpful to automatically forward work instructions and to check if intended parts where taken from the Kanban bin and were used in the proposed order. We built in a further step two vision-based low-cost systems, one with Halcon machine vision and one with TensorFlow deep leaning, and one low-cost system based on ultrasound (i.e. Marvelmind) to track assembly tasks. This paper compares the three approaches with the aid of three prototypes, one for visual recognition of assembly parts, one for visual recognition of assembly parts and tools, and one for ultrasound-based tracking of picking assembly parts from a bin. Finally, we discuss selected findings which are relevant for an industrial application setting.