Estimation of fungal biomass using multiphase artificial neural network based dynamic soft sensor

Interest in low cost cellulase production has become a major challenge in recent years for biorefineries. Fed-batch fermentation of Trichoderma strains for the production of low cost cellulase is carried out on complex media that has various soluble and insoluble substrates. The lack of direct estimation of biomass in the presence of insoluble substrates is one of the major concerns for controlling bioprocesses in industries. In this paper, a Multiphase Artificial Neural Network (MANN) based dynamic soft sensor is developed to predict the biomass concentration of Trichoderma during fed batch fermentation in the presence of insoluble substrates. The soft sensor has three Nonlinear Auto Regressive with eXogenous input (NARX) models to capture the complete dynamics of lag, log and stationary phases of the microbe. At different phases, a particular neural network model is triggered based on the period of operation. Each NARX model estimates biomass concentration using online measurements such as pH, substrate concentration and agitation speed. The predicted output of the proposed model and single ANN model are compared against real-time biomass sensor data. The results demonstrated indicate that the proposed MANN based soft sensor shows good performance with focus on the dynamic behavior of the bioreactor. Also, the developed model recursively predicts the biomass concentration with acceptable deviation with respect to realistic measurement. The results summarized could offer a new methodology in estimating fungal biomass accurately, thereby increasing the productivity of cellulase in industries.