Cluster analysis for systematic database extension to improve machine learning performance in double-sided incremental sheet forming

Abstract

Incremental sheet forming is a process for the production of sheet metal parts in small batch sizes. Due to the relatively low geometrical accuracy and the lack of precise and fast finite element analysis simulations of the process, industrial use cases are rare. Recently, a vast amount of scientific approaches simulated the process by utilizing machine learning techniques. Their success is limited by the quantity and quality of the used process data. Research institutes are struggling to gather enough data without industrial cooperations. For maximizing the distribution of process data in an experimental series and therefore their applicability for machine learning, the authors present a novel cluster analysis approach to systematically extend an existing database. The whole established process database consisting of 70 forming experiments and their toolpaths and digitizations is published to be used as a foundation for similar research.