Fuel cell vehicle simulations are built on detailed physics, but when a project needs many simulations of different drives, such as controls development, managing run time remains a consideration. Gamma Technologies explored this using GT-SUITE’s Machine Learning Assistant. The fuel cell flow circuit and cooling system, often the most demanding part of the vehicle model to calculate, were replaced with machine learning models called metamodels, while the rest of the vehicle stayed physics-based.
Metamodels were trained on realistic drive-cycle data, including cold and warm starts, to predict stack voltage, stack temperature, MWC and compressor power from current. Predictions were checked against unseen drive cycles. Battery state of charge and electric motor power remained essentially unchanged, and the approach aims to reduce run time, with lower run times reported across tested cases. This suggests machine learning can complement physics-based modelling.
Image generated by: ChatGPT