338 Proceedings of Combustion Institute Canadian Section Spring Technical Meeting Université Laval May 13-16, 2013 Combustion in anode baking furnaces: Comparison of two modeling approaches to predict variability François Grégoire +, Louis Gosselin * Département de génie mécanique, Université Laval, 1065, avenue de la Médecine, Québec City, Qc, G1V 0A6, Canada Houshang Alamdari Département génie des mines, métallurgie et matériaux, Université Laval, 1065, avenue de la Médecine, Québec City, Qc, G1V 0A6, Canada Abstract Carbon anode blocks, used in electrolysis cells of the Hall-Heroult process for the primary production of aluminum, are baked up to 1150 C in large-scale furnaces. These furnaces work similarly to counterflow heat exchangers where heated air flows in one direction and carbon anodes flow in the opposite direction as air supply and burner ramps are displaced on top of the furnace. Numerical modeling of the air flowing in the flue of an anode baking furnace (ABF) is challenging since it involves strongly coupled turbulence, combustion and radiation. Furthermore, the flue is rather large (~ 5 m 5 m 0.3 m) and its geometry is complex with multiple baffles. In this work, we compare the effect of modeling the effect of combustion in two different ways: 1) No combustion, instead the burners flames are mimicked by hot streams of air entering the flue channel at the burners inlets, and 2) The non-premixed mixture fraction approach is used. Specifically, we compare factors that are critical to the baking process homogeneity and to the furnace maintenance cost: the spatial distribution of the heat flux from the flue gas toward the baking anodes, the maximum temperature achieved in the flue and the presence of hot or cold spots. 1. Introduction Carbon anodes are used in the primary aluminum production to carry out the Hall-Héroult process. For each ton of aluminum produced, 0.4 ton of carbon is typically required [1]. Anodes are produced in dedicated plants inside or outside aluminum smelters. Their production line involves mixing of raw materials (i.e., coke, pitch, and recycled anode butts), compaction of the paste, baking of the green anodes, and rodding (i.e., an anode assembly is inserted into stubholes formed in the baked anode, and cast iron is poured in the gap between the carbon and the assembly). The present paper is focussed on the baking furnace. In a baking furnace, green anodes are stacked between refractory porous walls separated by a flue channel in which hot gases flow (Figure 1). The operation principle of a baking furnace is similar to a counter flow heat exchanger with the hot gases that would flow on one side, and the anodes on the other side, although in practice anodes do not move. In fact, it is the blowing ramps and burners that are moved every 24 hours. The baking cycle lasts 10-14 days. For the first 3-5 days, green anodes are preheated. During preheating, volatiles such as tar, CH 4 and H 2 are generated in the anodes and released in the flue where they burn. Then, for an additional 3 days, baking per se occurs. Burners are positioned at the top of the flue channel, and are controlled to maintain a prescribed temperature in the gases. Finally, the last step of the anode baking cycle is the anodes cool down (4-6 days). Natural gas burners Flue channel where hot gases flow Coke Anodes Figure 1. Schematic representation of the anode baking process * Professor, corresponding author, Louis.Gosselin@gmc.ulaval.ca + Doctoral Student
339 One of the main industrial challenges related to anodes is variability. Anode properties change significantly from an anode to another, and even within a single anode block. Inconsistent properties represent a serious problem in electrolytic cells where aluminum is produced, leading to instability, as well as carbon and energy overconsumption. Anode variability is caused in part by changes in raw materials, with a growing supply of non-traditional carbon sources due to the current market. Furthermore, baking induces variability since every anode in the furnace will experience a different baking curve. For example, it was reported that anodes located in the same pit can exhibit temperature differences as large as 100 C depending on their positions [2]. In order to help the industry to develop a better understanding of how variability is generated during anode baking, and eventually design better furnaces, anode baking furnace models have been developed in the past [1, 3]. These models vary greatly in complexity, from 1D analytical methods to advanced CFD models with turbulence, combustion, radiation, pyrolysis, etc. In this paper, we use a detailed 3D model and focus on the part of the furnace where burners are used to complete the baking process. Our main purpose is to study the impacts of combustion models on the prediction of variability in anode baking curves. 2. Description of anode baking furnace and uniformity A schematic representation of the domain of interest is shown in Fig. 2. It consists in a single pit containing 18 anodes with its corresponding flue channel. Due to symmetry, only half of the anodes and of the channel are considered. Air enters at a temperature T air,in, is deviated by baffles and tie bricks, and exits at the right-hand side outlet. Two burners are installed on the top of the flue channel. The heat generated by the combustion of natural gas in the flue channel is transmitted to the refractories wall, and then to the anodes. Natural gas inlets Air inlets Natural gas inlets Anodes Coke Refractories Flue channel y Gas outlets y z x z Anodes Coke Refractories Flue channel Figure 2. Schematic representation of the domain that is considered in the present study. Figure 3 shows the typical evolution of the air inlet temperature and of the desired baking curve for the anodes during the 36 hours when burners are on. The injection of natural gas at the burners is controlled so as to follow as close as possible the desired baking curve. During the plateau at 1200 C, gases exiting the domain must be at the same temperature as those coming in since the next flue section where the flow is going corresponds to the next 24 hours of the baking cycle. At the beginning of the firing, anodes have already been preheated by the combustion of volatiles in the flue. In order to simplify the simulations and because our interest is to assess the generation of nonuniformity during heating (i.e., when burners are on), it was assumed that the anodes were initially at a uniform temperature of 1000 C and that the injection of natural gas is at a constant flow rate (i.e., a control strategy has not yet been implemented in order to reproduce the 1100 C plateau of Fig. 3, and the anodes temperature keeps rising with time in our simulations).
340 1250 1200 1150 1100 1050 1000 T_air,in T_anode 950 0 6 12 18 24 30 36 Soaking time (h) Figure 3. Air inflow temperature and desired anode baking curve evolution, for the 36 hours when burners are used. As mentioned in the introduction, the furnace design does not allow a completely uniform baking, and each anode experiences a somewhat different temperature evolution. The present model development ultimately aims at determining the extent of this non-uniformity in current furnaces. Few measurements are available in the anodes during their baking due to the harsh conditions and prohibitive costs. Therefore, models are useful in order to assess anode baking uniformity. 3. Models The numerical model has been developed with a commercial software [4]. All modeling details are not reported here due to space limitation and to the fact that they are available elsewhere [5]. Essentially, the anodes are represented by solids into which only the conduction equation is solved. In the flue channel, flow and energy equations are solved. The Realizable κ-ε turbulent model is used with enhanced wall treatment. The Discrete Ordinates radiation model is used, considering a gray medium and setting minimal values of θ and φ divisions to 2. The absorption coefficient of the flue gas varies with composition in the case of the mixture fraction model (Section 3.1), and is 0 in the case of the simplified hot air jet model (Section 3.2). Although air infiltration or exfiltration at the top of the furnace can occur, this phenomenon was not modeled since it is considered to be negligible in the firing sections. Density and specific heat of the gas vary with temperature, while thermal conductivity and viscosity are assumed to remain constant at values corresponding to air at 1200 C. Mesh is constituted of about 200,000 hexahedral cells. A constant time step of 5 minutes is used, for a total of 432 time steps to achieve 36 h of heating. The simulation is parallelized on 8 cores of 2 Intel Nehalem-EP processors. Two different approaches to include the energy generated by the natural gas combustion have been tested and are described below. 3.1 Mixture fraction model The mixture fraction model is a computationally cheap model where the combustion is simplified into a mixing problem. The model solves 2 variables: the mixture fraction and mixture fraction variance. The mixture fraction is a conserved scalar that represents the local mass fraction that originates from the natural gas inlets, and under the assumption of chemical equilibrium, its local value is related to the species mass fractions, density and temperature of the mixture. The variance of the mixture fraction is necessary in order to couple the model with turbulence and obtain statistically averaged values of the 3 scalars aforementioned. In our model, there are 3 air inlets having the same temperature and composition, and 2 natural gas inlets also having the same temperature and composition. In practice, the mass flow rate of natural gas is adjusted in order to maintain the desired thermal conditions in the flue channel. In the present paper, the mass flow rate was kept constant as a first approximation. Nevertheless, the simulations allow to assess the temperature gradient that is susceptible of developing in the firing sections of the furnace. 3.2 Simplified hot air jet model In order to reduce the computational burden of simulating anode baking furnaces, several authors have used simplified models to account for the thermal energy released by combustion. One of the simplest approaches is to eliminate the combustion model in the simulations, and simply replacing the burner by an inlet of hot air. Although this strategy does not provide any information in order to improve combustion efficiency and reduce natural gas
341 consumption, it is interesting to document the ability of this simple approach to provide realistic anode temperature evolution. To our knowledge, no comparison of this simplistic approach to a more advanced combustion model has been proposed in literature. In particular, the possibility to predict anode variability with the help of the hot air jet model has not yet been explored. In order to inject an amount of energy at the burners inlets equivalent to that injected with natural gas in the mixture fraction model, the following strategy is used: The air is injected at the same temperature as the maximum flame temperature (i.e., 2190 C) achieved with the mixture fraction model; the natural gas net calorific value is estimated to 50 MJ/kg [6]; knowing the enthalpy of air at 2190 C, the air mass flow rate is readily obtained by matching the energy injected with natural gas. 4. Results and discussions Since no control strategy was implemented on the natural gas injection in order to reproduce the 1100 C plateau of Figure 3, the simulations over predicted the temperature in the furnace. However, the results allow us to compare the two approaches used to reproduce the burners flames. 4.1 Gas temperature comparison One of the major differences that was noted between the two approaches is the shape of the flame. Although the methane and the hot air jets have the same velocity at the burners inlets, Figure 4 shows how the flames produced with the mixture fraction model extend further inside the flue channel than the hot zone created by the hot air jet model. This difference has a major impact on the heat distribution of the two models, as will be discussed in the following sections. The velocity of the air jets could eventually be raised in order to reproduce the flame shape of the combustion model, but if this approach was to be used, the amount of energy introduced by the air jets would need to be adjusted in order to maintain energy conservation. Mixture fraction model Hot air jet model Figure 4. Comparison of the flame shape produced by the mixture fraction model and the warmer zone achieved with the simplified model, after 36 hours of firing. 4.2 Hot and cold spots in refractories with the two models As a result of the larger flames obtained with the mixture fraction model, the hot spots at the top of the refractories wall are significantly larger and hotter than the hot spots produced by the hot air jet model (Figure 5). Maximum hot spot on the refractories is 1353 C and 1261 C for the mixture fraction model and hot air jet model, respectively. In both cases, the temperature is significantly higher at the top of the refractories wall, which is detrimental to the refractories in that portion of the furnace. The recommended maximal temperature for the refractory is 1250 C and it is likely that this limit is reached during the baking process, resulting in higher maintenance costs.
342 Mixture fraction model Hot air jet model Figure 5. Temperature at the surface of the refractories wall after 36 h of firing. 4.3 Maximum and minimum temperature in anodes As could be expected from the temperature mapping at the surface of the refractories (Section 4.2), the warmest zones in the anodes is located at the top of the pit. Figure 6 shows the influence of the models on the temperature at the center plane of the anodes. Although the heat diffusion in the solids has smoothed to some extent the hot and cold spots present in the refractories, the temperature difference between the top and bottom anodes is still significant. Figure 7 shows the temperature evolution of the coldest and hottest points in the stack of 18 anodes. After 36 hours, the minimum and maximum temperatures in the anode stack are 1141 and 1222 C with the mixture fraction model, and 1121 and 1170 C with the hot air jet model. The evolution of the temperature difference within the anodes (shown in the right-hand side frame of Figure 7) as obtained by the combustion model after 36 h of firing is thus 65 % higher than the one obtained with the hot air jet model. Furthermore, this temperature difference is rising faster with the combustion model. Mixture fraction model Hot air jet model Figure 6. Temperature at center plane of the anodes after 36 h of firing. 1240 T_min,mix.fraction T_min,air jet T_max,mix.fraction T_max,air jet 90 mixture fraction air jet 1200 80 1160 1120 1080 Tmax-Tmin ( C) 70 60 50 40 1040 6 12 18 24 30 36 Soaking time (h) 30 6 12 18 24 30 36 Soaking time (h) Figure 7. Evolution of the minimum and maximum temperatures in the stack of 18 anodes.
343 4.4 Computational time The purpose of using a simplified hot air jet model to reproduce the natural gas flames is to reduce the overall computational effort. The computational time for the baking simulations is 2 hours for the hot air jet model and 3 hours for the mixture fraction model. That is, both models solved in similar and acceptable times. The hot jet model only allows a 1/3 reduction of the computational time of the mixture fraction model. On the other hand, considering that the full baking furnace model will simulate at least 144 h of baking (6 days), including 3 days where the burners are on, and might include other chemical or physical aspects, this time saving could be useful in some context. 5. Conclusions Although the results presented in this work overpredict the temperature in the furnace, they allow a first comparison of two approaches to model the natural gas burners in an anode baking furnace. The mixture fraction combustion model produced a more realistic flame shape than the simplified hot air jet model. Because the flames mimicked by the hot streams of air are much smaller than the flames of the combustion model, the results between both models differ significantly. It is possible that the hot air jet model could be calibrated in order to reproduce the flame shape of the combustion model, but it is unlikely that the amount of energy injected at the burners could be related to actual natural gas consumption this way. For future work, many details remain to be adjusted in the simulations: control strategy for natural gas flow rate, air infiltration/exfiltration at the top of the furnace, heat source from the combustion of volatiles in the preheating sections, heat losses by surfaces, etc. Acknowledgements The authors are grateful to the Natural Sciences and Engineering Research Council of Canada (NSERC) and to ALCOA for their financial support via a Cooperative Research and Development (CRD) grant entitled Amélioration de l'efficacité énergétique de l électrolyse d aluminium par l optimisation du procédé de fabrication d'anodes. References [1] A. Charette and Y. Kocaefe, Le carbone dans l industrie de l aluminium, Les Presses de l aluminium (2012). [2] V. Piffer, M. Miotto, C. Kato, M.A. Silva, M. Meier, R. Perruchoud, and P. Sulger, Process optimization in bake furnace, Light Metals (2007) 959 964. [3] F. Grégoire, L. Gosselin, and H. Alamdari, Sensitivity of Carbon Anode Baking Model Outputs to Kinetic Parameters Describing Pitch Pyrolysis, Industrial & Engineering Chemistry Research (2013), 52(12) 4465 4474. [4] ANSYS Fluent, Release 14.5, ANSYS, Inc. (2012). [5] ANSYS Fluent Theory Guide, Release 14.5, ANSYS, Inc (2012). [6] F. Keller and P.O. Sulger. Anode Baking - Baking of Anodes for the Aluminum Industry, R&D Carbon Ltd. (2008).