Applying Transfer Testing to Identify Annotation Discrepancies in Facial Emotion Data Sets.

Production data in industrial environments is frequently presented and stored in figures and diagrams. The original numerical raw data needs to be regenerated to use this data in advanced data analysis. Current software solutions still struggle to convert specific line curves automatically into accurate numerical data. Following the design science research paradigm, we present a novel approach for the automated regeneration of graphical data into its numerical representation by combining robotic process automation (RPA) with document image analysis (DIA). We evaluated the developed solution using a real-world dataset of quality inspection charts from a small and medium-sized manufacturing enterprise (SME). The results demonstrate that the data extraction, compared to other software-based methods, significantly reduces the time required compared to manual methods. Our approach provides a generally applicable, time-efficient, and easy-to-implement solution to increase data availability for technologies that require efficient data extraction processes.