A Machine Learning Approach for Device Design from Materials and Operation Data

Machine Learning allows for the modelling and analysis of complex systems for which little mechanistic knowledge is available and is therefore envisioned as a powerful tool for the development of new designs with applications in engineering problems. In this work, we propose a framework based on dimension reduction, clustering, and self-organizing maps for the modelling and analysis of devices from materials and operation data, from which useful information can be drawn to inform future designs and developments. We demonstrate the applicability of this approach by analysing a high-temperature polymer electrolyte membrane fuel cell (HT-PEMFC). It was found that out of the 12 input variables studied, temperature, oxygen stoichiometric ratio, and ionomer binder ion exchange capacity are the most influential forachieving high power HT-PEMFC. This framework could be extended as new data becomes available about the different device components.