State estimation in bioprocesses -: Extended Kalman Filter vs. Neural Network

In biotechnology the demand for process control strategies has increased during the last decades. As fermentation processes become more and more complex, increasing requirements are posed to the control tools. A high-level process control depends on a real-time knowledge of process states, which cannot easily be provided by hardware measurement sensors. Especially the amount of biomass, often the control variable in the process, is difficult to determine online. Hence other strategies have to be developed in order to identify this important process state. Very promising approaches are the application of observers or software sensors in order to estimate the amount of biomass based on online-measurement of other process values [1, 2]. This work compares two possible observer designs for the estimation of biomass in the fermentation process of the bacteria Streptococcus thermophilus, which is an important agent in milk and dairy product industry [3, 4]. First a model-based approach using an Extended Kalman Filter is applied to the process. This observer design is then compared to the estimation of biomass using Neural Networks. Using the conductivity as real-time online-measurement, biomass as well as substrate and product concentrations can be estimated.