One of the many challenges to the development of a fuel cell system for transportation applications is that of a fuel delivery system. While on-board hydrogen storage is desired, this option is frustrated by the insufficient energy density of existing storage technologies as well as the lack of a hydrogen distribution infrastructure. The alternative is to install an on-board hydrogen generation unit to convert high energy density hydrocarbon fuels. At the heart of most on-board fuel processors one finds a reforming reactor.

The current effort focuses on the Autothermal Reforming (ATR) reactor. Toward illustrating the control system objectives, consider the following scenario. If power demand at the fuel cell increases, then ATR throughput (H2 production) will also need to increase. While a quick response is desired, it must be done while maintaining tight specifications with respect to reformat composition (i.e., CO levels). To achieve the desired low CO concentrations, steam flow also need to be increase which will cause a significant drop in reactor temperature, potentially to the point of extinguish the reaction. While an increase in air flow (and the resulting increase in total combustion of the fuel) will offset the steam induced temperature drop, it should be done in such that the upper temperature limits of the reactor are not exceeded (i.e., the catalyst is not damaged).

In a previous effort [1], we developed a dynamic nonlinear CDF based model of the ATR reactor. Using this model we tested classic feedback and feed-forward control schemes [2], and concluded that such schemes (especially the linear feed-forward portions) were insufficient to achieve the fast responses required by the transportation application, and suggested the use of a nonlinear model when generating the desired feed-forward action. In anticipation of such a controller and its expected computational demands, we have developed a reduced order nonlinear dynamic model of the ATR [3], which for a simulation horizon of 20 seconds can be solved in less than 0.01 seconds (using Matlab and a standard desktop computer).

In this work, we will present a nonlinear model predictive controller (NMPC) for the ATR reactor. The proposed NMPC scheme is composed of three parts. The first is a steady state optimizer, which aims to maximize hydrogen yield for a given fuel flow rate while simultaneously observing upper and lower bounds on the manipulated variables as well as bounds on the solid temperature (the hotspot should be lower than 1200C and at 7mm the catalyst should be higher than 800C) and exit CO concentration. The second part of the controller is a nonlinear dynamic optimization scheme whose objective is to minimize the transition time to the new steady state. In addition to the enforcement of upper and lower bounds with respect to manipulated and output variables, rate bounds with respect to the manipulated variables are applied. Using this trajectory a time dependent linearization of the nonliear model is then determined. The third portion of the controller utilizes this linear model along with a closed-loop information structure to minimize a quadratic, trajectory tracking, objective function.

[1] Papadias, D., Lee, S. H. D., Chmielewski, D. J.(2006) “Autothermal reforming of gasoline for fuel cell applications: A transient reactor model,” Industrial and Engineering Chemistry Research, 45(17), p 5841-5858.

[2] Chmielewski, Hu and Papadias, “Start-up and Feedback Control of Autothermal Reforming Reactors,” paper 398a, Annual Meeting of the AIChE, Cincinnati, OH.

[3] Hu and Chmielewski, “Model Reduction and Optimal Control for the Autothermal Reforming (ATR) Reactor,” paper 561c, Annual Meeting of the AIChE, San Francisco, CA.