(39b) Disturbance Tracking in Industrial Milk Powder Process Systems

Boiarkina, I., University of Auckland
Young, B. R., University of Auckland
Prince-Pike, A., University of Auckland
Depree, N., University of Auckland
Yu, W., University of Auckland
Wilson, D., AUT
Milk powder quality has become increasingly important in the competitive commodity market, leading to a focus on technologies and systems that can enable the monitoring of quality in real time. A key aspect effecting milk powder quality are is process disturbances, however the effect on milk powder quality is not known. One known disturbance cause is the periodic switching in, and out, of equipment for cleaning during continuous milk powder manufacture, to maintain process hygiene. It was thought that these line swaps were the main contributor to non-steady state operation of the plant. However, although the timing of these line-swaps was known, their magnitude, duration and relative proportion in relation to total disturbance and manufacturing time, was not quantified. Therefore to date it has not been possible to infer whether they could affect quality without further a quantitative data analysis.

In this project, we used both conventional statistical monitoring tools such as statistical control charts, (SPC), and advanced modern regression techniques such as big data analytics including machine learning to analyse historical milk powder process plant data. Both methods performed equally well for tracking the historical process disturbances. The results showed that most disturbances in the total solids were correlated with line swaps, although not all. Furthermore the magnitude of the disturbances experienced during a line swap varied markedly, indicating that there is scope for minimising potential disturbances through specific operational intervention. However, there were a number of short, and long duration disturbances in the total solids that were not linked to line-swaps, indicating other significant mechanisms at play. This analysis helped us to reveal the line swap impacts on the quality variables, e.g. total solids in the concentrated milk; to identify different line swap effects on down-stream units; and to classify the different disturbances identified. This work has also indicated a more thorough analysis if the clean-in-place is warranted in order to hold the total solids content steady.