PEM FUEL HYDROGEN LEAK DIAGNOSTIC USING ARTIFICIAL INTELLIGENCE

摘要

When a proton exchange membrane (PEM) fuel cell runs short of hydrogen, it suffers from a reverse potential fault that, when driven by neighboring cells, can lead to holes in the membrane due to local heat generation. As a result, hydrogen leaks through the electrically-shorted membrane-electrode assembly (MEA) without being electrochemically reacted and a reduction in fuel cell voltage is noticed. The voltage reduction caused by the leaked hydrogen recombining with air to give water and reduce the oxygen concentration can be detected by using electrochemical impedance spectroscopy (EIS). In this paper, the EIS measurement is used in training a neuro-fuzzy (NF) system with different hydrogen leak rates in a commercial short stack. Hydrogen leakage through the stack is controlled by adjusting the pressure drop across anode/cathode. Under different operating conditions, the proposed NF diagnostic system was able to recognize the hydrogen leak and identify its severity with high accuracy.