Plastic injection parameter optimization strategies for mold set-up reductions via soft computing techniques in a multi product system

For changeover time reduction, the famous methodology, named single minute exchange of dies (SMED), introduced by Shigeo Shingo, has been applied in production plants for years in lean manufacturing scope. The main philosophy of this technique can be summarized as "make it simple and keep it simple". However, this simplification is not possible for all the steps of a changeover. In the last part of a changeover, which is also named as trial runs and adjustment, injection parameters are manipulated to resolve quality issues that occur during trial productions after the mold is changed. This session requires a deep knowledge and experience, hence, only dedicated personnel called setup experts can handle this session. In this dissertation, two popular soft computing techniques, fuzzy inference system (FIS) and multilayer neural networks (MLNN), are used to capture this domain expertise. The primary objective is to distribute the domain experience to non-expert plastic injection personnel to eliminate expert scarcity and to increase flexibility. A systematic elimination of defect cases is presented to define the core scope that handles all the possible quality issues and their solutions with fewer rules. Proposed soft computing solutions are implemented in a well-known international wiring device manufacturing plant together with a project team of setup experts and production staff. The results show that both FIS and MLNN could generate correct defect resolution actions on injection parameters.