Cream Cheese Fermentation pH Prediction

Abstract

Cream cheese is a popular dairy product; it is made from milk and cream with the help of lactic acid producing bacteria. The product quality is highly correlated to the pH at the end of fermentation. However, the pH change during the fermentation is dependent on many factors including: the natural variation of the milk composition and the starting culture, the activity change of the lactic acid producing bacteria during the fermentation, and the buffering effect of the milk. In order to accurately predict the pH change and thus the finishing time for the cream cheese fermentation, this research used a kinetic-model aided artificial neural network (ANN) with minimum input data required, and with the inputs of the change in bacteria biomass, lactose and lactic acid concentrations during the fermentation to predict the change of pH. Training of the ANN was accomplished using kinetic models verified by the experimental data. The kinetic model chosen to build the ANN achieved an average R2 value of around 0.9. The biomass experimental result deviated from the kinetic model by the greatest degree because of the inaccuracy inherently generated from the pouring plate method of biomass measurement. In terms of the pH prediction, the LSTM network was more stable and accurate compared to the NARX network, where the predicted fermentation finishing time differences were under three minutes and with an overall R2 of over 0.99. The outcome of the research provided a trained LSTM network that was able to accurately predict the pH change and the fermentation finishing time, which enables better product quality control and the possibility of improving the plant efficiency by accurately scheduling the upstream and downstream processes. It also proposed a pH prediction method for other similar fermentation systems.