Reformed Methanol Fuel Cell Systems - and their use in Electric Hybrid Systems

PEM fuel cells are widely regarded as a promising technology which has the potential to replace more polluting and less efficient internal combustion engines in many applications. They do, however, have the drawback that their hydrogen fuel is cumbersome and energy consuming to store and transport. Alternative system topologies that use a liquid fuel is therefore of great interest. One such topology is the Reformed Methanol Fuel Cell (RMFC) system where a mixture of liquid methanol and water is reformed via the steam reformation process to hydrogen and carbon monoxide. Most of the hydrogen is then used in a fuel cell and the rest is passed to a catalytic burner which supplies the process heat for the reformer. This makes RMFCs complex systems where the different parts of the system affect each other and it makes demands on the way they are integrated with the loads they supply. This PhD study has therefore been concerned with the module’s integration in a practical application and the optimization of the operating parameters of the system based on models of the system components.

The chosen application is a street sweeping machine which is a good case because they often operate in fleets with long periods of operation. Both of which are beneficial for the integration of RMFC systems. To analyze if the integration of an RMFC system is a good idea, a dynamic model of a street sweeping machine including approximate models of a battery and a RMFC system is produced. This model, along with a defined drive cycle, is then used in the context of the Outdoor Reliable Application using CLean Energy (ORACLE) project to predict the performance of a RMFC powered street sweeping machine before a prototype is made. After the prototype has been manufactured, the model is updated based on measurements and the performance of the vehicle is reanalyzed. It is concluded that the vehicle can operate for a full 8-hour working day without discharging the drive battery, if the vehicle is fitted with a 10 [kW] RMFC system. In this case 62.13 [L] of methanol is used if the standard hysteresis method is used to control the state of charge (SOC) of the battery. An analysis of the power through the drivetrain shows that most of the energy loss occurs in the RMFC system and that this loss could be minimized if a more constant lower power set point is used for the fuel cell. To achieve this, a SOC control is developed that minimizes fluctuations in the output power of the RMFC system. When this is done, the fuel consumption drops to 46.85 [L], which is a reduction of 24.6%.

It is further concluded that if the power consumption is minimized further it is realistic to reduce the RMFC power to 5 [kW] and the fuel consumption to 42.08 [L].

To be able to achieve the efficiency gain observed in the vehicle model, it is necessary to develop a controller that can control the output current of the RMFC system instead of the fuel cell current which is the standard procedure. To be able to do this, a model of the output current of an RMFC system is produced. This includes approximate models of the dynamics of the fuel cell and battery as well as the power consumption of the Balance Of Plant (BOP) consumers. The models are fitted on the basis of experiments and used to develop a PI controller with feedforward and anti-windup, which is tested experimentally with success. A model predictive controller is also developed based on the system models and it is tested in the model, but it was not possible to verify its functionality on the experimental setup. It is, however, believed that it could be an effective way to control output current of the module, especially if it is combined with an identification experiment during the startup of the module.

A series of models of the components of an RMFC system was also made to analyze how the operating points of the system affect the system efficiency. The first of these was an Adaptive Neuro-Fuzzy Inference System model of the cell voltage of an HTPEM fuel cell which is trained on an identification experiment spanning the expected operating range. The inputs of the model are the fuel cell temperature, the carbon monoxide content in the anode’s supply gas and the current density. Such a model has not been observed in literature before. The Mean Absolute Error (MEA) of the model is 0.94% and is it concluded that it is suitable for use in larger system models or for integration in a dynamic model of the fuel cell.

Subsequently ANFIS models of the carbon monoxide concentration and hydrogen flow in the output gas of the reformer are trained based on identification experiments. The carbon monoxide concentration model has an MAE of 0.323% and the hydrogen flow model has an MAE of 0.074%. These models are then combined with the ANFIS model of an HTPEM fuel cell and their combined efficiency is analyzed for different fuel cell currents and reformer temperatures. It is concluded that the system efficiency can be improved by an average of 1.47 percentage points across fuel cell currents and 4 percentage points at maximum fuel cell current at a fuel cell temperature of 170 [C]. If this efficiency gain is added to the gain achieved through the development of a controller for the battery SOC, the total efficiency gain achieved through modeling and control is increased to 28%.