European Symposium on Computer Aided Process Engineering – 13 A. Kraslawski and I. Turunen (Editors) © 2003 Elsevier Science B.V.

All rights reserved.


An MINLP Model for the Conceptual Design of a Carbothermic Aluminium Reactor
Dimitrios I. Gerogiorgis and B. Erik Ydstie Dept. of Chem. Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA Abstract Systematic design and optimization set the standard in today’s petrochemical industries, but their impact is limited when considering most processes in metallurgical industries. The present study presents an MINLP model developed for design and optimization of the electrode heating system of a new multistage high-temperature carbothermic reactor. The focus here is the optimization of electrode placement and imposed voltage profile; the model and conclusions can effectively assist ongoing multiscale process modeling. 1. Introduction Metallurgical industries are historically characterized by a high degree of complexity (high-temperature reactions, multiphase flows, complex physicochemical phenomena) and a predominantly experimental, iterative nature of novel process design efforts, both resulting in lack of formal accurate models that could facilitate optimization endeavors. Thus, advances in this field can definitely improve profitability in the globalization era, exactly as petrochemical industries have benefited from MINLP problem formulations and routine use of optimization computer tools that yield significant investment savings. The present paper presents a MINLP model for a novel complex metallurgical process, illustrating its application for the effective design of the reactor electric heating system. Mixed Integer Nonlinear Programming (MINLP) theory and applications have pervaded the field of computer-aided design and optimization of traditional chemical processes. Exemplary processes studied include numerous petrochemical unit operations, such as heat-integrated distillation (Novak et al., 1996; Caballero and Grossmann, 1999; Yeomans and Grossmann, 1999), azeotropic distillation (Bauer and Stichlmair, 1998), reactive (Ciric and Gu, 1994) and cryogenic distillation (Boisset-Baticle et al., 1994), extraction and dehydration (Lelkes et al., 2000; Diaz et al., 2000; Alonso et al., 2001), supercritical fluid extraction (Kalampoukas and Dervakos, 1996), membrane separation systems (Qi and Henson, 2000), wastewater treatment (Galan and Grossmann, 1998), heat exchanger networks (Yee and Grossmann, 1990; Aggarwal and Floudas, 1992; Daichendt and Grossmann, 1994; Galli and Cerda, 1998; Bjork and Westerlund, 2002), mass exchanger networks (Papalexandri and Pistikopoulos, 1994) and reactor networks (Esparta et al., 1998; Stein et al., 1999; Rooney and Biegler, 2000; Pahor et al., 2001). Applications include power plants (Manninen and Zhu, 1998), reactive power planning (Chattopadhyay and Chakrabarti, 2002), pump systems (Westerlund et al., 1994), copolymer production (Mantzaris et al., 1999), paper industry (Harjunkoski et al., 1999) and leather industry (Graells et al., 1992), but not the design of metallurgical processes.

Al4C3 injection from the third stage assists in shifting the chemical equilibrium towards Al and in avoiding active carbon depletion. Carbon monoxide and aluminium vapors (Al/Al2O) generated are fed to the third stage.132 2. The Carbothermic Aluminium Reactor Carbothermic reduction of aluminium oxide for the production of pure aluminium is an energy-efficient and environmentally benign reduction process with industrial potential. dense electrode placement and high voltage result in excessive superheating (a catastrophic effect causing aluminium evaporation. Liquid Al and CO gas are rapidly generated. where Al and Al2O vapors react with C to form Al4C3 (unless gas Al species are recovered countercurrent to the incoming solid feed. Johansen and Aune. both in the first pre-reduction smelting stage as well as in the second reduction stage. major yield reduction and losses) while sparse electrode placement and low voltage fail to achieve adequate slag heating (an equally undesirable situation resulting in limited conversion and low productivity). metal loss has a catastrophic impact on the process economics). 4. Nevertheless.. Nevertheless. reaction complexity poses notable technical obstacles to implementation. Carbon and aluminium oxide pellets are continuously fed to the electric arc smelter. The third stage consists of a vapor recovery reactor (VRR). The multistage. where dissolved C and entrained Al4C3 particles can be removed by proprietary technology to recover pure Al.. Obviously. its structural and operational complexity entails many design challenges. flow and other concurrent multiphase phenomena (Al liquid and CO gas generation) introduce further modeling complications which are not considered in the present study. The fourth (final) stage of the process is the purification zone: the liquid Al alloy produced floats and flows through an overflow weir to a tank. Submerged horizontal electrodes also act as obstacles to horizontal molten slag flow. melt and form a viscous molten slag contained under inert atmosphere and oil cooling. 2002) is a new idea addressing known issues so as to achieve the reactor scaleups that would permit the desired capital cost savings. The first stage of the process is the pre-reduction zone where slag formation occurs. multielectrode high-temperature carbothermic reactor recently proposed (Johansen et al. The second stage is the high-temperature slag reduction zone: the first-stage molten slag flows slowly into the multielectrode submerged arc main reactor. avoiding local surface superheating caused in open arc reactors. as the electric heating necessary for the endothermic reaction is achieved using independent electrode pairs. The design challenge is to optimize electrode positions and the imposed voltage profile so as to achieve the reaction advance without unnecessary reactor space or energy use. Figure 1 depicts the structure of the four-stage carbothermic aluminium reactor studied: 1. The reaction of Al2O3 with an excess of C to form the Al4C3-rich slag is thus written as: 2Al2O3(s) + 9C(s) → (Al4C3+Al2O3)(slag) + 6CO(g) (T > 1900 °C) (1) 2. The decomposition of the Al4C3-rich slag to form the Al-rich metal phase is written as: (Al4C3 + Al2O3)(slag) → (6Al + Al4C3)(metal) + 3CO(g) (T > 2000 °C) (2) 3. This method has been proposed as a feasible alternative to the prevalent Hall-Héroult electrolytic process. 2000. The most interesting one is the distributed nature of the process. 1989). where it is heated to a higher temperature. several reactor designs have been tested (Motzfeldt et al. .

. voltage and intensity profiles are considered piecewise constant along the reactor axis.. as the presence of metallic solvent (Cu/Sn) in their study is not affecting reaction kinetics. The coarse finite volume decomposition and drastic assumptions are necessary because of the lack of experimental data. 2002). A number of simplifying approximations are used for the development of this model. and to ensure a computationally manageable model. the relative concentrations within each volume only depend on the temperature therein. The assumption of instant thermodynamic equilibrium (extremely high temperatures) can be alternatively used to elucidate concentrations within each of the finite volumes. (1989) has been used. the latter have been previously reported in a relevant study (Gerogiorgis et al. the generated liquid and gas streams are considered immiscible with the molten slag. The advance of the reversible reduction towards Al (liquid) and CO (gas) is considered governed by the overall reaction proposed in a recent kinetic study (Frank et al. horizontal backmixing and vertical recirculation effects are neglected. The Joule heating effect of each electrode is thus confined within the respective volume. 1989): Al2O3(s) + 3C(s) → 2Al(l) + 3CO(g) (3) The pseudo-first order kinetic model as reported by Frank et al. considering only the production of Al (liquid) and CO (gas) without any phase changes. The latter product streams are taken to completely escape the slag at each finite volume. 2002). the use of a multiscale model is necessitated in that case (Gerogiorgis and Ydstie. Carbothermic Reactor Modeling The present study employs a simple finite volume model of the carbothermic reactor: a relatively coarse one-dimensional structured discretization of the domain is considered and a steady state CSTR reactor model is used in each resulting rectangular volume to probe mass. The potential presence of suspended C/Al4C3 particles in the incoming slag is ignored. 2002).133 3. The presence of electrode pairs (inert conductors) is modeled using binary variables. species diffusion. Figure 1: The conceptual carbothermic aluminium reactor (Johansen and Aune.. heat and molar balances using temperature-dependent physical properties.

j C P j.t.S and Fi. Fi. Thus. The free energy of formation (∆G°) for the reaction is published by Frank et al.S ⋅ C P i −1.P are the slag and product molar flows exiting finite volume i (CSTR). 4).jT 2 + a3. Vi   s. Vi is the imposed voltage on a pair. heat and molar species balance constraints.(13) (14) (15) . 2. (1989). P ⋅ x j.i = σ i ⋅ (E i ) 2 . i .S = 1 . xj. Ti are the temperatures. N is the total number of isothermal CSTR reactors considered (a sensitivity variable). S + Fi . E i = yi ⋅ Vi / L . (3). CP i.134 4. i . QH.i . i −1. 0 < x < 1 Here. respectively. QC. each finite volume is assumed a CSTR with perfect separation of reactants and products. Q R .i + Q R . concentrations Cj. QR.(23) 2173 < T < 2573 (K). 2). (1989).j + a1.i (6) Fi −1.jT-2) (20) .i. Cj. ∑ x j..S = 0 .(19) ∑ x j. (4) Fi −1.(17) σi = σ i (Τi ) = a σ + β σ ⋅ ln(Τi ) Q H . P ⋅ (Ti − TR ) − Q H .(11) ∑ x j. A and Tc are cooling rate and parameters. Temperatures Ti . Indices i and j refer to position and species (1:Al2O3. P + c j ⋅ R r ⋅ Vo (7) (8) . U.S + Fi .S ⋅ C P i . Q C.jT 3 + a-2.S ⋅ x j j. P = 1 . i are the molar species fractions. i are the concentrations.i is the reaction heat consumption using the kinetic model proposed by Frank et al.jT + a2.i.S = Fi.i are the state variables considered. i x ( j= 3.S / P ⋅ C P j. L is the lateral tip distance. S = Fi . Ei is the field intensity.P are the specific heat capacities. 3:Al. V° is the finite volume of the CSTR. P  i=  y i . and CP i.i and fractions xj. 0 < V < 100 (Volts). i = R r ⋅ V o ⋅ ∆H o  − ∆G o  o  R r = k 0 exp  R ⋅ T  ⋅ C j=1.i.S ⋅ (Ti − TR ) + Fi . P ⋅ Fi. yi are binary variables indicating absence (yi = 0) or presence (yi = 1) of electrode pairs and parameter Y denotes the maximum number of electrode pairs allowed in the reactor. The MINLP Model for Electrode Heating System Design This section presents the formulation of the MINLP finite volume model: the goal here is to perform electrode placement as well as imposed voltage profile optimization for maximization of Al production under mass. .S/P = (18) . 2:Al4C3.i + Q C . The mathematical formulation is based on a CSTR series steady state process model. i. N CSTRs) ∑ yi = Y ≤ N (12) .S/P = (a0.S . P ⋅ C P i . the Al maximization objective function and the balance constraints are written as:  N  max  ∑1 x j = 3. ∆G = a G + b G ⋅ T i   C P i.(21) (22) .. i. i. i . σi(T) is the slag electrical conductivity. P (5) Fi −1.S ⋅ x j j. cj are the stoichiometric coefficients in Eq. 4:CO).i is the Joule heat production. subscripts S and P denote slag (Al2O3 + C) and product (Al + CO) streams.i = U ⋅ A ⋅ (Ti − Tc ) . respectively.S / P . P = 0 (i = 1. x ( j=1.S ⋅ (Ti −1 − TR ) = Fi .

02 0.00 1 2 3 4 FINITE VOLUME # (i) 5 6 0. Al(l).30 0. CO(g)) in all fluid streams. 8. More complicated objectives (simultaneous minimization of Al(g) losses and energy use) can also be addressed by augmenting the model with VL equilibrium equations that will account for all 6 species (Al2O3(l). IMPOSED ELECTRODE VOLTAGE 2600 100 80 ABSOLUTE TEMPERATURE 2400 2200 2000 V (V) 60 40 20 0 1 2 3 4 FINITE VOLUME # (i) 5 6 T (K) 1800 1600 1400 1200 1000 1 2 3 4 5 FINITE VOLUME # (i) 6 HORIZONTAL (SLAG) FLOWS 6 5 4 3 2 1 0 1 2 3 4 FINITE VOLUME # (i) 5 6 6 5 VERTICAL (PRODUCT) FLOWS F P ( kmol. giving insight for preliminary heating system design.min ) F S ( kmol.16 0.135 5.10 0. 2002). .20 0. Production increases with power input (6. Al4C3(l).70 0. temperature increases as slag flows and limiting reactant (Al2O3) deplete. Al(g). for two different electrode number constraints (Y = 3.12 MOLAR FRACTION OF Al2 O 3 MOLAR FRACTION OF Al 0. The finite volume process model and the conclusions of this study can also be used for multiscale modeling of this complex chemical process (Gerogiorgis and Ydstie.80 0.06 0.06 kmol.14 0. MINLP Model Results The most important results obtained using the MINLP model are presented in Figure 2.min-1 / Y = 6).20 0.04 0.50 X Al 2O 3 X Al 0. Y = 6). Al2O(g). but heating is most important right at the reactor inlet.60 0.00 1 2 3 4 FINITE VOLUME # (i) 5 6 Figure 2: MINLP model results for Y = 6 electrodes (black) and Y = 3 electrodes (gray).min-1 / Y = 3. Electrode positions. Electrode number is equal to the maximum.10 0.min ) -1 -1 4 3 2 1 0 1 2 3 4 FINITE VOLUME # (i) 5 6 0. 6. Conclusions and Future Work The MINLP model outlined for electrode voltage optimization in a carbothermic reactor captures macroscopic phenomena.18 0.80 kmol.08 0. voltages and temperatures for maximization of liquid Al production are illustrated in Figure 2.40 0.

16 (S). 24 (12). I.. Comput. 1994. 1994. 2002. 24 (9-10). Grossmann. Comput. F. Dervakos. K. 2001. Comput.. Pistikopoulos. Met. S217. Ind. Comput..D. 18 (11-12). Comput. G. Chem. 1135. Lassahn. Chem.I. Chem. Sweden (#192). Coral Springs. Pettersson. Daoutidis.. K.. 2001. 1999. Eng. Biegler. L.Y. Carbothermal Production of Aluminium – Chemistry and Technology. Pettersson. 25 (2-3). Westerlund. L. Johansen. J.. Z. I. C. Kvande. 14 (10). A. G. U. Chem. 2003. M.H. 22 (S). Srienc. Comput. T. 24 (2-7). Ciric. Grossmann. 679. 2069. Chem.R. Eng. Dusseldorf.G. E. Grjotheim.C. Zhu. Eng. Yeomans. Aune. Eng. B. Eng. 23 (10). Al Verlag.F. Diaz. Finn. 1999. Mat. Chem. Chakrabarti. 1271. Eng. 765. Johansen. Gerogiorgis. Westerlund. 161. Floudas. 1997.E. B. Manninen. D. A. 1999.R..A. T. Comput. I. Eng. S99. Comput. Yee. 37. S221. Eng.S. Germany. Obertopp. Comput.J. G.N.. Chem. B.E.. Kalampoukas. Kienle.. M. Chem. 22 (9). Esparta et al. Eng. 1992. N. Kelley.136 7. A. 2055. 1581. A. 23 (9).V. 4-6. Espuna. K. Comput. Comput. Comput. 521. Seetharaman. Chem. 1997. Chem. Frank. Eng. Chem. J. Chem. 2000. Ydstie and S. 47 (3). Ydstie. 2002. J. Comput. 21 (5). Chem. Eng.. Comput. Eng. Chem. Chem. Harjunkoski. Res. Elec.S.X. Eng.A.. I..A. C. Gilles. Proceedings of the Computer Modeling of Minerals.S. Kravanja. E. Stockholm. Chem.. L. Aune. Eng. . 2002. M. 1331. Proceedings of the Foundations Of ComputerAided Process Operations Meeting (FOCAPO 2003). 18 (8). Chem.E. 2001. Comput. Daichendt. Stein. I. 16 (2). F..E. Int.I. I. Eng. E. Rooney. M. Seattle.. Proceedings of the Sixth International Conference on Molten Slags.. M.I. Eng. R. Brignole. Z.R. Grossmann. 581. Schei. 2000. Comput. 1999. 1996. Fluxes and Salts.E. Grossmann.. J.E.. Gros.P. Mantzaris. P. Bruno. C. 18 (S). D. Chem.. Stichlmair. L. D.. Comput. 20 (12). M. 267.. Gruhn. DOE). Gu. 2719. T. Chattopadhyay. Graells. Lelkes. Comput. Z.M. 1994. H. B.) gratefully acknowledges an Institute of International Education Fulbright fellowship as well as an Alexander S. K. Eng..B.E. Schei.193 (to Alcoa Inc. Eng. X. WA. Szitkai.. A. Eng. Z.W.I. 1479. Comput. 26 (11). 727. Kravanja. for the present study (part of the carbothermic aluminium production project co-funded by the U.. A. Porn. S537. Chem. 1990. Westerlund. Bedenik. 2000. Chem. R. A. Stichlmair.F.I. Eng. F. 1989.T.H. Pibouleau. Z. 2000. Galan. D. Chem. 1992. Power 24 (3). 40 (9). 20B (4). E. Esparta. 1425. B.. 18 (9). Papalexandri. Trans.H. 24 (9-10). R. 845.. Rev et al. S1383. K.. Gerogiorgis. Bauer. B. Novak. 185. References Aggarwal.M. Eng.. 22 (S). S903. Eng. Pahor.. J. Latge.. Qi. Chem. Puigjaner. A. Grossmann. Comput. Comput. The first author (D. 273. Metals & Materials Processing Meeting (TMS2002). T. Bjork. and Elkem ASA). Chem. 2000. AIChE J. Comput. Alonso. Elliott. T. Motzfeldt. 1994. Onassis Foundation doctoral fellowship. S. 1545. 2002. 20 (B). Acknowledgments The authors acknowledge the financial support of ALCOA Inc.A. S671. Henson. Eng. Frey. 21 (S). 1165. Westerlund. 89. J. Grossmann. Eng. A.J. T.440. 1125. T. 23 (S). E. H. W. Chem. 1998. 4036. Chem. Eng. N.A. K... AIChE J. Comput. H. FL. I.. 1998. Bauer.E.. 1998. A. A. Comput. Patent 6. Boisset-Baticle.

Sign up to vote on this title
UsefulNot useful