dc.description.abstract | Membrane biological reactor (MBR) is an emerging technology adopted for domestic and industrial wastewater treatment. The high selectivity of the semi permeable membrane to water results in membrane fouling. Fouling is a highly nonlinear phenomenon that affects the MBR stability, productivity, and performance. Many techniques are proposed to minimize and control fouling such as backwashing and aeration. However, such techniques may lead to high operation costs and high energy consumption rates. Therefore, optimization of MBR operating conditions is essential in order to achieve effective, stable, and economical MBR operation. In this study, a rigorous mathematical model is developed for the MBR. The model is derived by performing a mass balance on three major parameters, the substrate concentration, the biomass concentration, and the Oxygen concentration. Then, kinetic models representing the major reactions in the MBR are stated. Four kinetic parameters namely, the maximum specific biomass growth rate, net biomass yield, Monod constant, and endogenous decay coefficient are estimated using experimental data. Further, an empirical flux model is proposed representing the flux exponential decline behavior. The flux model also requires the estimation of two constants representing the cake growth. Hence both the kinetic and flux parameters are estimated using POLYMATH nonlinear regression. Combining the mass balance equations along with the corresponding kinetic models and the estimated kinetic parameters yields to a system of first order nonlinear coupled ODEs. Hence solving such system online is inefficient due to the large computational time and effort. Thus artificial neural networks (ANNs) are suggested for MBR modeling. The input and output variables are first selected for the ANN model. The selected input variables are the backwash pressure, vacuum pressure, and ratio of vacuum-to-backwash time, while the flux is selected as the output variable. Different ANN models are developed by training different input variables and the response of flux is observed and compared with experimental data. A better ANN predicted flux is attained when increasing the number of inputs to the ANN model. Accordingly, advanced control strategy is used to control and stabilize the MBR performance. A Model Predictive Control (MPC) is implemented with the optimum ANN model using Neuro-MPC toolbox of MABLAB/SIMULINK. The NN-MPC demonstrates the effectiveness of the proposed methodology in stabilizing the MBR and optimizing its performance. The NN-MPC performance depends on the prediction horizon and control horizon. The control prediction is performed through minimizing an objective or cost function to track a predefined set-point trajectory. Short prediction and control horizons are used for a more aggressive controller. Aggressive controller settings reduce the computational time and effort but may lead to process instabilities. Therefore, two weighting parameters are set at a large value to avoid oscillations. The NN-MPC demonstrated a good servo performance (set-point tracking) within the constrained inputs range. On the other hand, the conventional linear PID controller demonstrated unrealistic manipulated variable moves due to unconstrained manipulated variables, and the difficulty of tuning the control parameters due to lack of deterministic models. | en_US |