Reduced Rework Through Data Analytics and Machine Learning – A Three Level Development Approach

Abstract

In many production companies, rework of products that do not meet the defined quality level is a common practice. Usually organized at the end-of-line in form of test and rework cycles the goal is that no defect \textquotedblescapes\textquotedbl to the customer. Only in rare cases, a systematic problem solving is triggered through a rework process to identify and remove the root cause. In the EIT Manufacturing project IVE a data-driven problem-solving tool is being developed with the focus on rework. Through a three-level development approach, the goal is to significantly reduce rework effort as well as the time to detect and analyze a problem. In the first level, an error score is being developed to give the shop floor team an overview about error clusters and to help them decide which errors to focus on. In the second level, machine learning is used to give action recommendations to the rework teams and to predict the needed work force depending on the production program. In contrast to the first two levels, which aim to reduce the amount of rework, the third level aims to completely avoid rework through in-line action recommendations based on data from preceding processes. To validate and demonstrate the usefulness of a data-driven approach as well as the implementation of machine learning algorithms to companies, the developed tool will be implemented in the ``process learning factory CiP''. For this purpose, a rework station will be created and designed to meet the specific needs. If the implementation is successful, the concept can also be transferred to other learning factories.