| The on-board fuel processor, which converts gasoline to hydrogen through autothermal reforming (ATR), is one of the most promising technologies of providing hydrogen feedstock for fuel cell vehicles nowadays. However, due to the strong nonlinearity of the on-board reactor, the control of such fuel processors still remains one of the most challenging hurdles to commercialize. Specifically, the start-up with water injection will drastically decrease the reactor temperature and thus extinguish the fuel processor. Inversely, the start-up with extra air may boost the catalyst temperature over the upper bound and thus damage the catalyst. As such, the goal of this dissertation is to develop a robust control system to maintain the reactor temperature within bounds.;The classic feedback control is investigated in this work. The feedback controller employs air feed rate as the manipulated variable and a measurement of catalyst temperature as the control variable. However, the feedback only method is insufficient for the unique challenges associated with on-board operation. While the feed-forward configuration improves performance, significantly, a fair amount of sensitivity with respect to model-mismatch is found.;A computationally efficient advanced control strategy named nonlinear multivariable predictive controller (NMPC) with state estimation is presented. The proposed NMPC scheme is based on a fast reduced order nonlinear model which is derived from the original PFR model. There are three parts for the proposed strategy. The first is a steady state optimizer, which aims to minimize fuel flow for a given hydrogen demand while simultaneously observing the lower bound on the reactor temperature to maintain the ignition of the reactor. The second part of the controller is to find desired input/output trajectories via solving a nonlinear dynamic optimization problem subject to process constraints. The third portion of the controller determines a successive linear model around the references and minimizes an unconstrained quadratic, trajectory tracking problem of which analytical solution is obtained. A linear Kalman Filter (KF) is employed to estimate process states with the presence of unmeasured disturbance, model mismatch and system/measurement noises. The proposed NMPC is compared with the classic feed-forward controllers via simulation results and the improved performance is demonstrated. |
