Christian Hähnel

Nonlinear Model Predictive Control of a PEM Fuel Cell System for Cathode Exhaust Gas Generation

Abstract : Polymer electrolyte membrane (PEM) fuel cells are highly efficient energy converters and provide electrical energy, cathode exhaust gas with low oxygen concentration and water. They are investigated as replacement for auxiliary power units (APU) that are currently used for electrical power generation on aircraft. For generation of oxygen depleted cathode exhaust air (ODA) oxygen concentration must be 10-11%. A challenging task is controlling the fuel cell system for this product and simultaneously keeping fuel cell stack degradation, voltage losses and stack damage as low as possible as well as keeping the system within operational limitations such as bounds and gradients on control parameters. This constrained control task for PEM fuel cell systems is attacked by a nonlinear model predictive control (NMPC) strategy. Simulation and experimental results are shown.

Power efficient operation of a PEM fuel cell system using cathode pressure and excess ratio by nonlinear model predictive control

Polymer electrolyte membrane (PEM) fuel cell systems convert chemical energy from hydrogen into electrical power via a reaction with oxygen. Fuel cells are highly efficient, nearly noiseless, and locally emission-free. However, safe and efficient operation requires special system temperatures, mass flow rates of the reaction gases, and a particular excess ratio and gas pressure on the anode and cathode. All values are current dependent and must fall within specific ranges for both stationary and dynamic loads. This paper deals with modeling fuel cell peripherals and with model predictive control (MPC) of gas pressures. This nonlinear MPC handles all kinds of dynamic load changes, while maximizing power efficiency. All experiments were verified on a test bench with a 4,4kW PEM fuel cell.

Offset-free nonlinear model predictive control of electrical power of a PEM fuel cell system using an Extended Kalman Filter

Polymer electrolyte membrane (PEM) fuel cell (FC) systems convert chemical energy from hydrogen into electrical energy via a reaction with oxygen. FCs are highly efficient, nearly noiseless, and locally emission-free. An efficient and safe operation demands several variables in certain ranges. However, the electrical characteristic of a FC systems shows a nonlinear behavior and an exact control of the electrical power using a model-based approach requires an accurate model. Due to assumptions and simplifications the model differs from the plant which leads to steady state errors. This paper deals with offset-free Nonlinear Model Predictive Control (NMPC) of electrical power of a PEMFC using an Extended Kalman Filter (EKF). The NMPC allows direct use of nonlinear models with respect to constraints whereas an EKF allows a model correction and thus an exact control without steady state error. All experiments were verified on a test bench with a 4.4kW PEMFC.

Optimal iterative learning control of a PEM fuel cell system during purge processes

This paper deals with an Optimal Iterative Learning Control approach for the anode pressure during the periodic purge processes of a fuel cell system. Due to accumulation of diffused nitrogen and water condensate inside the anode volume the chemical reaction is restrained. This adverse effect is avoided through the purge process, by which a short time opening of the exhaust valve forces the nitrogen and water out of the system. Unfortunately, the opening leads to a pressure drop along the anode volume that causes a force to the membrane. To avoid this mechanical stress the control aim is a constant anode pressure during the purge process by supplying additional hydrogen. An Iterative Learning Control is suited for the multiple times executed purge procedure. For this purpose, the nonlinear characteristics of the fuel cell system model are transferred into time-variant, linearized state-space models in discrete-time domain to create learning filters. In order to achieve a fast convergence by a strictly monotonously reduction of the control error, the control problem becomes a minimization problem and therefore an Optimal Iterative Learning Control is applied.

Online identification of an electric PEMFC model for power control by NMPC

Of course, fuel cells are also subjects to signs of wear. Over time of usage this is shown by lower power efficiency, the degradation. Degradation correlates with wear due to mechanical stress, high stack temperature, and too fast load changes. Generally it is influenced by operating the fuel cell with wrong conditions regarding to temperature, anode and cathode pressure. Despite this time variant behavior the aim in operation with fuel cell systems is to keep the quality of power control as high as possible. Modeling of long term time variations, especially such as degradation effects, can only hardly be realized. Therefore this paper deals with an online identification of the electrical model of a PEM fuel cell to support the model predictive control (MPC). Due to the amount of constraints to be respected in fuel cell operation, MPC is an appropriate control strategy. Identification by using only the current measured data leads to a very exact model for the actual range of operation, but great model errors outside of this range are occurring. Thus, an appropriate strategy for the storage of measured data in relation to the identification is shown. The aim of the data management is to achieve steady computation times and an accurate model over the entire operation range. Experimental results on a test bench with a 4.4kW PEM fuel cell are presented.

Error handling approach of a PEM fuel cell system by nonlinear model predictive control

For an efficient chemical reaction and a safe operation for both stationary and dynamic loads all values of anode and cathode gas pressures and stack temperature must fall within specific ranges. System error can result in values outside of these ranges, in turn causing serious damage to the fuel cell (FC). In addition the energy supply by the FC is at risk, because an emergency shutdown is possible. Therefore it is necessary to implement strategies for handling error cases in order to ensure safe operation. A common method for controlling the operation of PEM FC is a model predictive control, which allows for fault tolerance. In literature, error handling strategies are only considered for special parts of FC, but not for the whole system. This paper deals with real-time nonlinear model predictive control (NMPC) of electrical power with applied error handling strategies for a PEM FC system. Different types of errors are discussed in connection with various solution approaches. This fault tolerant NMPC handles all kinds of dynamic load changes and allows a safe operation of the FC in the event of errors, while maximizing power efficiency. Any emergency shutdown is not necessary and energy supply is still guaranteed. All experiments were verified on a test bench with a 4.4kW PEM FC.

State estimation of exhaust valve position by Kalman Filter in PEM fuel cell systems

Sensitive systems in control approaches, such as fuel cells, require exact operating conditions for high efficiency, safety and endurance of the system. A promising method in controlling fuel cell systems is model predictive control (MPC). For those, many states have to be measured very precisely within high sampling rates, because they are used as initial values for the following calculations. However, some states, such as pressures, are more significant for the efficiency of the system than others. In case of incorrect measurements or general errors in operation, false outputs are calculated and damages can occur to the fuel cell. Therefore an estimation of significant system states is necessary to cover failures for avoiding negative influence on the systems efficiency. It is established that a powerful state estimator is the Kalman filter. In connection to nonlinear and constrained applications, such as fuel cell systems, different types of the Kalman filter have been proposed, e.g. Extended and Unscented Kalman filter (EKF, UKF). This paper deals with the state estimation of the exhaust valve position by use of an Unscented Kalman filter (UKF) to support the pressure control of a 4.4kW PEM fuel cell system.

Iterative Learning Control of a PEM fuel cell system during purge processes

This paper deals with an Iterative Learning Control (ILC) approach of the anode pressure during purge processes of a fuel cell (FC) system. The purge processes are necessary because diffused nitrogen and water condensate cumulate in the anode volume and influence the chemical reaction. The temporary opening of the exhaust valve allows the purge process, which removes the water and nitrogen. ILC is suited for the multiple times executed purge procedure. The control aim is a constant anode pressure during the purge process by supply of additional hydrogen. For this purpose, the nonlinear characteristics of the FC system model are transferred into discrete-time, time-variant, linear state-space models to create learning filters.

Application of a Hardware-in-the-loop DC/DC converter and microgrid simulation to a PEMFC

Polymer electrolyte membrane fuel cells are efficient energy converters that convert chemical energy directly into electrical energy. These units are suitable for dynamic applications and can be used for stationary as well as mobile electrical power generation. The field of application, however, predetermines the type of electrical connection network, electrical storage devices and electrical loads that the fuel cell is attached to. This study puts the focus on the interaction of a physically available fuel cell with the electrical application such as storage and consumers. As an example, a microgrid connected through a DC/DC converter to the fuel cell is realized by a Hardware-in-the-loop simulation being linked to the fuel cell. The Hardware-in-the-loop simulation is compared to the physical system consisting of the fuel cell and a DC/DC converter. A model predictive control for electrical power is presented. Experimental results of the Hardware-in-the-loop simulation and the physical system are shown.

Iterativ Lernende Regelung des Anodendrucks während der Spülvorgänge eines PEM-Brennstoffzellensystems

Zusammenfassung Polymer-Elektrolyt-Brennstoffzellen sind Energiewandler, die aus reinem Wasserstoff und dem Sauerstoff der Umgebungsluft elektrische Energie bereitstellen. Zum Erhalt des effizienten Ablaufs der chemischen Reaktion ist es notwendig, das anodenseitige Gassystem regelmäßig von angesammeltem Stickstoff und Wasserkondensat zu befreien. Dieser Vorgang wird im vorliegenden System durch Spülen mittels Öffnen eines Ventils durchgeführt. Zur Druckregelung während dieses wiederkehrenden Vorgangs wird eine Iterativ Lernende Regelung eingesetzt. Dieser Beitrag zeigt die zu Grunde liegenden Modelle für die Erstellung der benötigten Lernfilter und experimentelle Ergebnisse an einem Prüfling mit bis zu 4.4 kW elektrischer Leistung. Für einen Vergleich werden Lernfilter mit und ohne Modellkenntnis gebildet und eine Optimierend Iterativ Lernende Regelung eingesetzt.