A novel multi-physics and multi-dimensional model for solid oxide fuel cell stacks based on alternative mapping of BP neural networks

Abstract

Solid oxide fuel cell (SOFC) stacks have complex multi-physics and multi-dimensional properties that are very difficult to observe in experiments. To solve this problem, a new modeling concept called alternative mapping is introduced in this paper. Using an artificial neural network (ANN), a multi-physics and multi-dimensional stack model can be decomposed into two bonded layers. Both can be solved much more rapidly and robustly than conventional models without losses in accuracy. A SOFC stack model is developed based on this concept. It is calibrated and validated by data from a single cell test and 30-layer stack experiments. The model is robust and rapid. The stack model results show that different physics and dimensional properties are closely related. Changes in each may cause chain reactions in the others, influencing the electrical efficiency or operation ranges. In this report, a fully 3D multi-physics and multi-dimensional dynamic simulation of a SOFC stack is implemented for the first time. Its dynamic behavior is a composite of electric-chemical reactions, gas transport, and heat transfer. These factors vary considerably in different positions and operational conditions.