Onboard diagnostics concept for fuel cell vehicles using adaptive modelling

Abstract

Fuel cell vehicles and fuel cell research is one of the newer areas in automotive technology. This paper describes an approach that utilizes artificial neural networks to alleviate the task of onboard diagnostics for fuel cell vehicles. The basic idea is an online learning scenario that trains a power train model with every-day driving data; this model can then be used to estimate a characteristic curve by feeding it with predefined input variables corresponding to the constant conditions of a stationary workshop test. In this way, a major obstacle for on-line diagnosis, namely the multitude of varying nuisance variables, can be compensated for. For a diagnosis algorithm, it is considerably easier to compare the resulting predicted characteristic curve with an ideal reference curve, rather than to directly deal with all the influence factors.