Systematic development and mesh sensitivity analysis of a mathematical model for an anode baking furnace

More Info


The anode baking process is developed and improved since the 1980s due to its importance in Aluminium industry. The process is characterized by multiple physical phenomena including turbulent flow, combustion process, conjugate heat transfer, and radiation. In order to obtain an efficient process with regards to quality of anodes, soot-free combustion, reduction of NOx and minimization of energy, a mathematical model can be developed. A mathematical model describes the physical phenomena and provides a deeper understanding of the process. Turbulent flow is one of the important physical phenomena in an anode baking process. In the present work, isothermal turbulent flow is studied in detail with respect to two turbulence models in COMSOL Multiphysics software. The difference between wall boundary conditions for these models and their sensitivity towards the boundary layer mesh is investigated. A dimensionless distance in viscous scale units is used as a parameter for comparison of models with and without a boundary layer mesh. The investigation suggests that the boundary layer mesh for both turbulence models increase the accuracy of flow field near walls. Moreover, it is observed that along with the accuracy, the numerical convergence of Spalart-Allmaras turbulence model in COMSOL Multiphysics is highly sensitive to the boundary layer mesh. Therefore, development of converged Spalart-Allmaras model for the complete geometry is difficult due to the necessity of refined mesh. Whereas, the numerical convergence of k-ε model in COMSOL Multiphysics is less sensitive to the dimensionless viscous scale unit distance. A converged solution of the complete geometry k-ε model is feasible to obtain even with less refined mesh at the boundary. However, a comparison of a developed solution of k-ε model with another simulation environment indicates differences which enhance the requirement of having converged Spalart-Allmaras model for complete geometry.