화학공학소재연구정보센터
Journal of Food Engineering, Vol.252, 1-9, 2019
Predictive modelling of instant whole milk powder functional performance across three industrial plants
The efficient production of instant whole milk powders with stringent functional requirements is the economic goal for large-scale milk powder plants. However, the physical predictors of functional properties are not sufficiently characterised to ensure quality at the time of manufacture, and given high production rates, the time delays for testing can lead to significant amounts of devalued product if functional tests are subsequently failed. This work combines big-data collection and alignment with multivariate methods on a re-sampled, balanced data set to provide a real-time proxy measurement_ A unique dataset was constructed to compare three large-scale plants of different designs, overcoming significant challenges in alignment of disparate data sources, across six years of production. Models were clearly able to differentiate plant and process changes, and by appropriate weighting of the unusual results, were able to predict functional test results, especially the production of off-spec product. These models can be used in real-time to significantly lower the likelihood of producing powder that will fail customer functional tests, and therefore substantially reduce cost of quality failure.