Discussion: Process data streams aggregation versus product samples aggregation

The article by Zwetsloot and Woodall (2019) opportunely addresses an updated and relevant topic that has been escaping the attention of the data-centric/evidenced based communities. As data collectors become widely available across industrial processes and along the entire value chain, the issue of aggregation becomes not only a need for handling high-throughput data streams but, as I show here, an opportunity to improve the performance of classical data analytics by exploring an extra tuning dimension: data granularity (or level of aggregation). In this discussion, I begin by clarifying what I consider to be two distinct perspectives of data aggregation to be taken into account, here designated as process data streams aggregation (or the process-oriented perspective) and product samples aggregation (or the quality-oriented perspective). Next, the topic of handling variables with different levels of aggregation is addressed, as well as the potential advantages of selectively introducing aggregation, even when raw data are available at a single aggregation level. Finally, the topic of adaptability in data aggregation is commented upon within the scope of multiscale methods for statistical process monitoring. Across my exposition, I also try to convey the perspective of the process industries (PI), such as oil and chemicals, biofuels, pharmaceutical, pulp and paper, food and drink, semiconductors, and microelectronics, among others we have been involved with, and where the topic discussed in Zwetsloot and Woodall (2019) is very relevant.