Databased learning of influencing factors in order specific transition times

Abstract

Abstract Manufacturing companies are faced with increasing customized product individualization in the last decade. To flexibly adjust manufacturing systems to customized orders, producing factories are challenged to achieve high adherence to delivery dates (ADD) and short throughput times (TPT). TPT mainly consists of non-value-adding transition times (TT), which make up to 95% of TPT in production. In factories, the calculation of TT is mostly made without considering all situational shopfloor conditions and influencing factors. Results are increasing deviations between predicted and real TT (respectively TPT) and thus decreasing ADD. This paper presents a methodology for databased identifying influencing factors in order specific TT. Scientifically, the methodology roots on the Cross-Industry Standard Process for Data Mining (CRISP-DM) and includes filter and wrapper as feature selection methods for data mining (DM). The methodology considers factory specific characteristics, identifies relevant and non-redundant features as key factors and continuously integrates factory employees in the learning process.