Nonlinear non-isothermal distributed-parameter observer for PEM fuel cell systems

Abstract

One of the most promising technologies for climate-neutral energy production is the polymer electrolyte membrane fuel cell (PEMFC) when using green hydrogen as fuel. Despite advancing development, highly dynamic operations are still challenging due to the potential occurrence of adverse fuel cell states, e.g., membrane drying. To avoid such operating states, the fuel cell has to be monitored by measuring vital physical quantities of the cell. However, measuring internal fuel cell states with sensors is difficult. To still reconstruct internal states, state observers are used. State observers use measured inputs and outputs and a PEMFC model to estimate the actual internal states of the fuel cell. In this thesis, a state observer based on the extended Kalman filter (EKF) algorithm is developed that uses a non-isothermal quasi-2D PEMFC model. The focus lies on the estimation of the temperature distribution within the fuel cell. Additionally, the current density and membrane water content distribution as well as the species concentrations in the gas channels are estimated. The non-isothermal quasi-2D model of high order is used to predict the state distributions. For feasible and efficient estimation, the state update is computed using a reduced-order model, obtained by a suitable model order reduction method, that retains the high-order system’s dominant behavior. The EKF algorithm is explained and possible locations for sensors are discussed. An observability analysis of the system is carried out afterwards. The developed observer is validated against a simulated reality and compared to a simulation without an observer. The results show that the observer drives the states faster to the actual states than the simulation without an observer, even with wrong initialization, measurement noise, and multiple manipulable inputs. In the end, an estimation of selected model parameters is added to the EKF algorithm. The results show that the observer is also able to correct initially wrong parameters while estimating the PEMFC states accordingly.