Online estimation of stirred-tank microalgal photobioreactor cultures based on dissolved oxygen measurement

Photobioreactor sensing presents unique challenges not met in conventional fermentors. First and foremost, photobioreactor processes are governed by photosynthesis, and hence many parameters important to photobioreactors are not considered in conventional fermentors. Furthermore, photobioreactor processes are typically associated with stringent cost constraints, and thus the use of complex sensing hardware is precluded. This calls for innovative approaches to address sensing problems in photobioreactors. Here, we report the development of an effective model-based estimator that is capable of tracking key culture states in stirred-tank microalgal photobioreactor systems. A marine micro-alga Dunaliella salina was used as a model organism in this study. Extended Kalman filter (EKF) was applied here to provide optimal estimates of photobioreactor states, based on a dynamic process model in conjunction with online dissolved oxygen measurement. The process model consists of a growth model and a light transport model. The former associates growth to average light intensity in the reactor, while taking into account both photoinhibition and oxygen inhibition. The latter is based on a modified radial model to estimate the average light intensity. The estimator is capable of estimating biomass density, specific growth rate, dissolved oxygen concentration, photosynthetic efficiency, and average light intensity in the photobioreactor illuminated either with constant incident light at different intensity levels or with time-varying incident lights. For the latter, an auxiliary internal model EKF was used to accurately track the variation rate of the incident lights. This paper also presents a detailed analysis on the tuning of EKF for optimal estimation. This state estimation system offers a cost-effective means for monitoring the process dynamics of microalgal photobioreactor cultures online, through which the productivity of such a process could be optimized.