Impact of Radiation Models in Coupled Simulations of Steam Cracking Furnaces and Reactors Guihua Hu, † Carl M. Schietekat, ‡ Yu Zhang, †,‡ Feng Qian,* ,† Geraldine Heynderickx, ‡ Kevin M. Van Geem,* ,‡ and Guy B. Marin ‡ † Key Laboratory of Advanced Control and Optimization for Chemical Processes of Ministry of Education, East China University of Science and Technology, Shanghai 200237, China ‡ Laboratory for Chemical Technology, Ghent University, Technologiepark 914, B-9052 Ghent, Belgium ABSTRACT: As large floor-fired furnaces have many applications in refinery and (petro-) chemical units and about 80% of heat transfer in these furnaces is by radiation, the accurate description of radiative heat transfer is of the most importance for accurate design and optimization. However, the impact of using different radiation models in coupled furnace/reactor simulations has never been evaluated before. Therefore, coupled furnace/reactor simulations of an industrial naphtha cracking furnace with a 130 kt/a capacity have been conducted. Computational fluid dynamics simulations were performed for the furnace side, while the one-dimensional reactor model COILSIM1D was used for the reactor simulations. The Adiabatic, P-1, discrete ordinates model (DOM), and discrete transfer radiation model (DTRM) were evaluated for modeling the radiative heat transfer. The results with DOM and DTRM are very similar both on the furnace and the reactor sides. The flue gas temperature using DOM is higher than when using the P-1 radiation model, resulting in higher incident radiation. Comparing the simulated results of all radiation models to the industrial product yields and run lengths shows that DOM and DTRM outperform the others. As DOM has a broader application range than DTRM, and because the current implementation of DTRM in FLUENT/14.0 cannot be run in parallel yet, DOM is the recommended radiation model for run length simulations of steam cracking furnaces. 1. INTRODUCTION Large floor-fired furnaces have many applications in refinery and (petro-) chemical units. One of the most important applications is in the hot section of steam cracking units for the production of olefins and aromatics. In this case, several tubular reactors are suspended in the furnace and the heat released by the burners in the furnace is transferred to the reactor tubes by convection, radiation, and conduction. This heat is on the one hand required for the production of steam and evaporation of the liquid feed. More importantly, the heat is needed to drive the endothermic thermal cracking reactions and to overcome the insulating effect of the forming cokes layer on the Fe-Cr- Ni heat resistant steel reactors. 1,2 The total heat absorbed by the reactor coils in the radiation section of a steam cracking furnace is about 42-47% for 100% floor firing. 3 This value, which is also known as furnace thermal efficiency, consists of both radiative heat transfer and convective heat transfer. Because of the high temperature in the furnace, i.e., above 1300 K, radiation dominates the heat transfer process. 4,5 Hence, an accurate prediction of radiative heat transfer is a prerequisite for a correct simulation of these furnaces. In the past decades, many researchers have used the Lobo- Evans method, 6 the Belokon’s method, 7 and zone methods 4,8,9 for simulating industrial steam cracking furnaces. The fuel combustion was not simulated rigorously, instead a predefined heat release rate was imposed to estimate the composition and temperature of the flue gas. Furthermore, convective heat transfer to the reactor tubes was often ignored. These simplifications obviously cause a certain error, which may result in inaccurate design optimization. By virtue of the development of accurate models and continuously growing computational power, computational fluid dynamics (CFD) has steadily grown to become an important and indispensable simulation tool for the chemical industry. More particularly, for the simulation of steam cracking furnaces, different CFD models have been evaluated by many researchers over the past 2 decades. Wang and Zhang 10 used the P-1 radiation model to calculate radiative heat transfer. Zhou and Jia 11 adopted the discrete ordinates model (DOM), but they introduced empirical formulas for the calculation of the flue gas radiative properties, which could introduce large errors in the results. Coelho 12 and Stefanidis et al. 13 assessed the influence of adopting nongray radiative properties of the flue gas mixture. More recently, Hu et al., 14,15 Yang et al., 16 and Hassan et al. 17 performed coupled simulations of the furnace and the reactor tubes, in which DOM was applied in the furnace simulation. However, in all these studies only a single radiation model was applied and the results can therefore not be used for the comparison of the performance of various radiation models. Keramida et al. 18 compared the discrete ordinates and six-flux radiation model for a natural gas diffusion flame and concluded that the two models performed similarly, both showing good agreement with the experimental data. Li et al. 19 compared different radiation models for heat transfer in a vertical pipe. Mendes et al. 20 adopted DOM and Rosseland model for the Received: November 3, 2014 Revised: February 10, 2015 Accepted: February 13, 2015 Published: February 13, 2015 Article pubs.acs.org/IECR © 2015 American Chemical Society 2453 DOI: 10.1021/ie5042337 Ind. Eng. Chem. Res. 2015, 54, 2453-2465