Data-based approach for identifying repeatability, reproducibility, and stability of industrial processes in series production

Abstract

Industrial series production is a complex environment with many factors, which can affect production machinery and the quality of products. Planned maintenance or even the start of a new material batch can have unpredictable and delayed effects on the plant. Typical monitoring systems use control limits to prevent damages in case of critical and sporadic changes, but leak any metrics on repeatability or stability of the process. Having these characteristics helps to identify unintended changes in the process and machinery, and to avoid system failures and quality issues. This work presents a data-based approach, which allows the evaluation of the repeatability of manufacturing processes. Further, it enables identification of process changes along the time or component process geometry and process stability evaluation. Therefore, the high-dimensional process raw data are divided into segments, standardized within these sections, and processed accordingly using the Local Outlier Factor anomaly score. An interactive heat map is used to display the changes along the process. The approach makes use of the fact that the majority of the manufactured products satisfy the production standard and the required quality. For evaluation, data sets from welding series production are used.