Prediction and estimation of the optimal time for harvesting Streptococcus thermophilus in a fermentation process

In this paper a new method for the estimation and prediction of relevant state variables of the cultivation process of the bacterium Streptococcus thermophilus as starter culture is presented. In order to improve the process and to maximize the quality of the product, the biomass should be harvested when the acidification activity has reached its maximum. The activity can be interpreted as the capability of the biomass to acidify and thicken milk and is an important quality attribute for the use of Streptococcus thermophilus as starter culture in dairy industry. In the production process of the bacteria, the pH-value of the medium is controlled by means of base added. The amount of base (control input) is assumed to directly reflect the activity of the biomass. The maximum activity is reached when the acidifying rate reaches its maximum, i.e. the second time derivative of the control input crosses zero. Thus the optimal harvesting time coincides with the inflection point of the control input. The optimal harvesting time based on estimates of the second derivative of the control input is estimated by an artificial neural network. The pattern recognition method is used to predict this point already in early process stages. Therefore the decline of the measured pH-value is evaluated to give important information about the velocity of the process. The performance of the presented approaches is shown by comparing prediction results with the detected zero-crossing of the estimated second derivative in an online fermentation process.