CFD-based optimization of a transient heating process in a natural gas fired furnace using neural networks and genetic algorithms

In the present study a transient heating process in a natural gas fired test furnace, used for fire resistance tests of construction and building materials, was investigated by computational fluid dynamics (CFD). To ensure the reproducibility of a fire resistance test, the thermal exposure of the tested fire safety material has to be homogeneous and, thus, the temperature distribution is of high importance. For that purpose, a CFD-based optimization of the transient heating process was carried out using different optimization algorithms. Based on the furnace setup, parameters with a potential to improve the temperature distribution were identified and used for the optimization procedure. CFD results were used to create system response surfaces, which represent the temperature distribution in the furnace as a function of the chosen design parameters. The system response was approximated by neural networks and genetic algorithms, and represents the basis for the optimization. Since the duration of the transient process was 35 min, the calculation time of the gas phase combustion and heat transfer is high. Therefore, a novel CFD-based approach was used to investigate and improve the process by converting the transient heating problem to steady-state cases. A comparison of the initial and the optimized furnace configuration showed an improved temperature distribution, where the maximum temperature difference in the furnace at the measurement position was decreased from approx. 200 K to 162 K. This approach showed that the transient simulation can be optimized, and further used for other applications where a transient simulation is computationally too demanding.