Modeling, Simulation and Control of Biotrickling Filter for Removal of Air Pollutants

Abstract

Stringent environmental regulations for control of pollutants have led to the use of effective air pollution control strategies. Biotrickling filter, one of the biological reactors, offers a great advantage of being a cost effective and environmental friendly technology. This emerging technology has not yet received widespread application. Moreover, there is still a need to develop an appropriate biotrickling filter model for general acceptance and equally important to design an optimum control strategy before utilizing this technology on a large scale. Hence, this thesis aims to develop a representative dynamic model for the biotrickling filter based on the review of existing models, provide accurate analytical and numerical solution of the model under different conditions, and also select an optimum control strategy amongst the different control systems designed in this study. A theoretical model was selected, validated and modified to account for continuous, larger biotrickling filter. The modified model was solved using the pseudo-steady state assumption to reduce computational effort and time. Based on sensitivity analysis of the modified model, it was found that gas velocity and inlet concentration had strong effect on the outlet concentration of biotrickling filter. To implement the control strategies, simple data driven models were obtained using the data from simulation of the modified model. These data driven models were needed since the modified model simulation would require considerable computational effort and time. In particular, transfer function and neural network models were successfully obtained with R2 values above 0.97. Five control strategies were designed, implemented and analyzed through set-point and disturbance changes. Three of the five controllers were based on transfer function biotrickling filter model while the rest used steady state neural networks as the biotrickling filter plant model. Overall, it was found that the proportional-integral, proportional-integral with feedforward and the transfer function based model predictive controllers provided satisfactory system performance. In case of the neural network based model predictive controller, excellent set-point tracking had been observed but an offset error had been observed in case of disturbance change. While the addition of an integral controller to the neural network based model predictive controller eliminated the offset errors, large overshoots had been observed in response to both set-point and disturbance changes. Search Terms: biotrickling filter, mathematical modeling, step response model, neural network model, biotrickling filter control strategies, conventional control strategies, advanced control