Implementation of sensor based on neural networks technique to predict the PEM fuel cell hydration state

Proton exchange hydrogen fuel cells have the potential to produce clean and environmentally friendly energy. However, this technique should be adapted to technical challenges, such as performance and durability prior to its marketing. These challenges are closely related to water management. In this research, a PEM fuel cell simulation model was designed for water management. This model consisted of a voltage evolution model based on electrochemical and dynamics gases. It also comprised a model of water activity to estimate the relative humidity. Meanwhile, in identifying the PEMFC hydration state, impedance was estimated by the humidity sensor model, which was based on neural network technology for diagnosis. This model predicted the changes of behaviour in the step response of load demand and the rate of water which flowed into the fuel cell. In the case of flooding or drying, the proposed neural network sensor model was executed through the estimation of internal resistance and biasing resistance values at high and low frequencies. These frequencies corresponded to the model of PEMFC electrical performance. As a result, it was found that the efficacy of this new neural network sensor model led to improved PEMFC hydration and a controlled humid airflow in the fuel cell. Overall, it was indicated that the proposed model can be used in the control system to improve water management by adjusting the relative humidity of supplied air.