A data-driven EIS approach for developing PEM fuel cell operation maps: A pathway towards applying neural networks

In this study, an empirical database is developed to represent the internal ionic resistances for the whole operation area of a proton exchange membrane fuel cell, using a data-driven modelling approach. Electrochemical impedance spectroscopy is employed to obtain experimental data for modelling and analysing the effects of key external operating parameters including air stoichiometry, cell temperature, and inlet air humidity, on fuel cell performance. The resulting model is used to create operational maps that visualise both the individual and combined impacts of these parameters, offering a novel representation of the fuel cell's impedance characteristics. These maps are further used to identify optimal operating regions. The study also outlines the mathematical formulation of a multivariate polynomial regression model and its associated components. The proposed method is scalable, allowing for the inclusion of additional operating parameters, and is adaptable to a wide range of experimental setups and fuel cell systems. A major advantage of integrating the empirical database with the operational maps lies in their potential application within fault diagnosis systems based on artificial intelligence. As such, the developed database can be leveraged to train neural networks to clarify fuel cell health states. • Design of experiments approach is applied to theoretically derive PEMFC operation data. • Empirical data base obtained based on experimental data and polynomial regression model. • PEMFC operation maps are designed based on the empirical data base. • Suitable PEMFC operation areas are defined based on the PEMFC operation data.