Deep learning-based optimization of a microfluidic membraneless fuel cell for maximum power density via data-driven three-dimensional multiphysics simulation.

A deep learning-based method for optimizing a membraneless microfluidic fuel cell (MMFC)performance by combining the artificial neural network (ANN) and genetic algorithm (GA) was for the first time introduced. A three-dimensional multiphysics model that had an accuracy equivalent to experimental results (R2 = 0.976) was employed to generate the ANN's training data. The constructed ANN is equivalent to the simulation (R2 = 0.999) but with far better computation resource efficiency as the ANN's execution time is only 0.041 s. The ANN model is then used by the GA to determine the inputs (microchannel length = 10.040 mm, width = 0.501 mm, height = 0.635 mm; temperature = 288.210 K, cell voltage = 0.309 V) that lead to the maximum power density of 0.263 mWcm-2 (current density of 0.852 mAcm-2) of the MMFC. The ANN-GA and numerically calculated maximum power densities differed only by 0.766%.