Maximising 3D printed supercapacitor capacitance through convolutional neural network guided Bayesian optimisation

Abstract

A convolutional neural network (CNN) guided Bayesian optimisation framework is introduced to strategically maximise the surface to volume ratio of 3D printed lattice supercapacitors. We applied Bayesian optimisation on printing parameters to exploit regions where uniform and narrow lines are printed. A line shape classifying CNN model guided the optimiser’s search space to straight-line printed regions, minimising optimisation time and cost. An automatic scoring method allowed each iteration to be conducted within two minutes with accurate and precise measurements. The optimisation process has been demonstrated with graphene oxide (GO) and poly(3,4-ethylenedioxythiophene):polystyrene sulphonate (PEDOT:PSS) inks. The results were compared to the parameters that follow the conventional methodologies of direct ink writing (DIW) 3D printing. For each printed line of GO and PEDOT:PSS inks, irregularities decreased by 61.8% and 18.9% and average widths decreased by 39.0% and 28.6%. PEDOT:PSS lattice supercapacitor printed using optimised result showed a 151.0% increase in specific capacitance.

Keywords

• 3d printing
• bayesian optimisation
• convolutional neural network
• supercapacitor