A Recommender System for the Additive Manufacturing of Component Inventories Using Machine Learning

Abstract

Abstract To appropriately leverage the benefits of additive manufacturing (AM), it would be advantageous if a printing could be guaranteed before allocating the necessary resources. Furthermore, when considering AM for an inventory of existing components traditionally fabricated through traditional means, such a guarantee could result in significant technical and economic advantages. To realize such advantages, this paper presents a platform that allows for a successful and efficient transition of part-inventories to AM. This is accomplished using a novel design recommender system supported by machine learning, capable of making suggestions towards effective design modifications. This system uses an automatic AM feasibility analysis of existing parts and a clustering of the parts based on similarities in their AM-feasibilities to develop a set of recommendations for those part clusters whose current designs are deemed as infeasible and/or inefficient for AM. The design modifications leverage a redesign algorithm to address not only problematic geometric issues but also potential infeasibilities associated with resource consumption. The utility of the presented modification algorithm is demonstrated using a number of case studies.