Data-driven modeling and monitoring of fuel cell performance

Abstract

A mathematical framework that provides practical guidelines for user adoption is proposed for fuel cell performance evaluation. By leveraging the mathematical framework, two measures that describe the average and worst-case performance are presented. To facilitate the computation of the performance measures in a practical setting, we model the distribution of the voltages at different current points as a Gaussian process. Then the minimum number of samples needed to estimate the performance measures is obtained using information-theoretic notions. Furthermore, we introduce a sensing algorithm that finds the current points that are maximally informative about the voltage. Observing the voltages at the points identified by the proposed algorithm enables the user to estimate the voltages at the unobserved points. The proposed performance measures and the corresponding results are validated on a fuel cell dataset provided by an industrial user whose conclusion coincides with the judgement from the fuel cell manufacturer.