Pattern Recognition for Knowledge Transfer in Robotic Assembly Sequence Planning

Abstract

The autonomous assembly of customized products is highly demanded in future manufacturing scenarios. This requires robotic systems being able to adapt to individual products without increasing overall production time. However, increasingly complex assemblies lead to a growing number of potential assembly sequences that have to be considered. To cope with this, we present an algorithm that is able to transfer previously identified assembly constraints to novel product variants. This reduces the search space, and thus planning times. The approach consist of three main steps. 1) Deduct semantic assembly constraints, from an analysis of feasible and unfeasible solutions. 2) Match key features of assemblies on a semantic level, by performing graph matching in the representation of the assemblies. 3) Use pattern recognition and classification based on machine learning techniques to transfer the knowledge of constraints for sub-assemblies into the complete assembly. We demonstrate our contributions on a two-armed robotic setup that assembles product variants out of aluminum profiles.