When to Collect What? Optimizing Data Load via Process-driven Data Collection

Abstract

Industry 4.0 and the Internet of Production lead to interconnected machines and an ever increasing amount of available data. Due to resource limitations, mainly in network bandwidth, data scientists need to reduce the data collected from machines. The amount of data can currently be reduced in breadth (number of values) or depth (frequency/precision of values), which both reduce the quality of subsequent analysis. In this paper, we propose an optimized data load via process-driven data collection. With our method, data providers can (i) split their production process into phases, (ii) for each phase precisely define what data to collect and how, and (iii) model transitions between phases via a data-driven method. This approach allows a complete focus on a certain part of the available machine data during one process phase, and a completely different focus in phases with different characteristics. Our preliminary results show a significant reduction of the data load comp ared to less flexible interval- or event-based methods by 39%.