Predicting manipulated variables to control water content in continuous butter manufacture by an artificial neural network

Modeling and prediction techniques in continuous butter manufacture were developed using an artificial neural network (ANN) on the basis of 48 test data sets on an industrial scale. The manipulated variables to achieve the target water content were also investigated. The fat content of cream, the physical ripening time of cream, the cream feed temperature, the cream flow rate, and the shear rate on the beater were selected as input manipulated variables for modeling and it was confirmed that the ANN model developed had good predictive ability for industrial applications. The manipulated diagram, which shows the relationship between the cream flow rate and the shear rate to achieve the target water content, was made using the ANN model and the database which consists of the manipulated variables and the water content. The manipulated diagram showed a linear relationship in the middle range of water content. At the extremes of target water content, deviation from linearity occurred due to the existence of stationary regions of water content. Furthermore, the influence of cream fat content and cream feed temperature on the manipulated diagram was examined.