LES/flamelet/ANN of oxy-fuel combustion for a supercritical CO2 power cycle

Abstract

Large eddy simulation (LES) employing a flamelet model coupled with an artificial neural network (ANN), namely the LES/flamelet/ANN is developed for supercritical oxy-fuel combustion and its validity is investigated in detail. An ANN is introduced to reduce the size of the flamelet library for non-adiabatic three gas-stream conditions of a real supercritical carbon dioxide (CO2) combustor. The fuel and oxidizer are methane (CH4) and a mixture of oxygen (O2) and CO2, respectively. To consider the real gas effect, the Soave-Redlich-Kwong (SRK) equation of state is solved to generate the flamelet library. The LES/flamelet/ANN is applied to two combustion fields in the combustor, namely the upstream and full fields of the combustor with low and high mesh sizes, respectively. The LES/flamelet/ANN for the upstream field is conducted to examine its applicability to the non-adiabatic three gas streams condition by varying the resolution of the flamelet library and by comparing it with the results obtained by the conventional LES/flamelet. The LES/flamelet/ANN for the full field is also conducted to validate the approach by comparing its results with those obtained from the experiment. The results show that the LES/flamelet/ANN of the upstream field properly reproduces the supercritical combustion behavior generated by the conventional LES/flamelet while maintaining the same level of computational cost. Furthermore, the LES/flamelet/ANN for the full field reasonably predicts the supercritical combustion behavior in the experiment, and the predicted combustor outlet temperature agrees well with the experimental results.

Keywords

• Oxy-fuel supercritical CO2 cycle
• Large eddy simulation
• Flamelet/progress-variable model
• Artificial neural network