An agile layer-resolved SOFC stack model using physics-informed neural network

Abstract

Solid Oxide Fuel Cell (SOFC) stacks are one of the most critical modules in industrial SOFC energy conversion systems. Although the detailed multiphysics distribution has been elaborately studied with accurate 3-dimensional (3D) models, development and validation of agile stack model is yet inadequate. Due to slow and tedious meshing and simulation, fast prototyping of stacks remains a challenge. Therefore, a 30-cell stack was tested at varied temperatures and gas flowrates and a real-time transient layer-resolved stack model is established and calibrated using the measured data, which gives a Root-Mean-Square (RMS) prediction error of 2.21% for measured voltages and 1.53 °C for measured temperatures. With layer resolution, the stack model shows the voltage, average temperature, as well as fuel and air flowrates of each cell. Moreover, the stack model reveals the relation of voltage distribution at varied fuel flowrates with temperature distribution. Furthermore, the stack model is potentially applicable to stack scaling effects, or designing Balance of Plant (BoP).