Abstract Efficient production of milk powder with good quality that can pass strict functional tests is of primary concern for most milk processing plants, making the online prediction of production… Click to show full abstract
Abstract Efficient production of milk powder with good quality that can pass strict functional tests is of primary concern for most milk processing plants, making the online prediction of production quality a popular research topic in recent years. Rehydration is an important quality indicator for instant whole milk powder and is mainly affected by process variables and morphology of the powder. This work investigated the feasibility of using both morphology metrics and process variables to develop on-line, or at-line, sensors for the prediction of rehydration of instant whole milk powder. Two key properties, namely dispersibility and slowly dissolving particles are the focus of this study. Light microscopy with image processing were used to obtain quantitative information on different shape factors of the milk powders, before using resampling to solve the class imbalances in the original dataset. Various partial least squares models constructed from (i) process variables only, (ii) shape factor variables only, and (iii) process variables combined with shape factor variables were used to compare which variables are important in developing soft sensors for predicting the dispersibility and slowly dissolving particles of instant whole milk powder. It was found that the dispersibility of instant whole milk powder mainly depends primarily on the shape factor variables while the process variables and shape factor variables are both important to predict slowly dissolving particles of instant whole milk powder. The good performance of the models (the Q2 is 0.77 and 0.94, respectively, while the R2 is 0.93 and 0.97, respectively) developed by process variables and shape factor variables also indicated that this approach could be used in real-time to measure the rehydration properties of milk powder and could be used for developing online model-based process monitoring.
               
Click one of the above tabs to view related content.