(145c) Soft Sensor Development for 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
The potential of “big data analytics”, characterised by volume, velocity, variety, veracity and value has, not unexpectedly, attracted the attention of the Dairy Processing industry. While classical big data techniques have been successfully applied to dairy herd management, genetics, with the aim to increase milk production from cows, the application of big data mining techniques to industrial process data is still relatively new.

One of the most important questions to ask is what you want to learn from taking a big data approach. Most big data applications focus on the ‘what’, trying to find trends or clusters in what is occurring without necessarily understanding the ‘why’. However, adding the ‘why’ into the equation changes the usage of big data landscape, and this is what we are interested in, in the milk powder processing industry. For example, why is powder occasionally off-specification on particular properties.

Industrial processing plants typically collect huge volumes of data, much of which is dumped to a historian, never to see the light of day. From our experience and the literature, the problem is that the big data value is extremely low. The big data does not contain enough or very little key process/quality variables strongly linking to the final product quality.

To address the above question, we developed a soft sensor which can predict a milk powder quality attribute, dispersibility, based on its morphology, so that the big data value is enriched. The functionality and performance of instant whole milk powder (IWMP) can be related to its physical and chemical properties such as particle size distribution (PSD), morphology and fat distribution etc. The dispersibility, a measure of rehydration, has become one of the most important quality characteristics of IWMP. Many milk powder properties may affect the dispersibility, based on the literature and our own experience, PSD and morphology may be the two most important factors. A soft sensor based on a quantitative relationship between the PSD/morphology and dispersibility was built using an artificial neural network. The results may help dairy industry to improve their milk powder online monitoring.