Minerals Engineering 19 (2006) 299308 This article is also available online at: www.elsevier.

com/locate/mineng

Modelling of aluminium scrap melting in a rotary furnace

B. Zhou, Y. Yang *, M.A. Reuter, U.M.J. Boin
Department of Geotechnology, Delft University of Technology, Mijnbouwstraat 120, 2628 RX Delft, The Netherlands Received 7 April 2005; accepted 21 July 2005 Available online 19 September 2005

Abstract In a typical secondary aluminium process, the scrap feed is charged into a rotary furnace, melting and mixing under a salt layer in the furnace. The complexity in such a pyrometallurgical process is due not only to the high temperature eect and the complex chemical reactions, but also to the highly complex scrap feed with a distributed nature of aluminium types, compositions, sizes, shapes, paintings and other contaminations. In this study, user sub-models, which represent the distributed nature of the scrap feed, were developed and integrated into a computational uid dynamics (CFD) based process model of a rotary furnace. Aluminium scrap was classied into several groups depending on their properties, e.g., size, establishing a discretized population balance model (PBM). The melting behaviour of aluminium scrap was simulated with the exchange of information between the melting sub-model and the CFD calculations. In addition, the sub-model for scrap burn-o was also developed and integrated in the CFD framework providing distributed burn-o rates. Simulations of the melting process were made to model the ow and thermal phenomena in such a furnace, and the inuence of the scrap size, shape and quality, as well as burn-o rate were studied. 2005 Elsevier Ltd. All rights reserved.
Keywords: Modelling; Simulation; Computational uid dynamics; Population balance; Recycling

1. Introduction Secondary aluminium production has been used in a number of areas, such as transportation, building and packaging (EAA, 2004). Its production is increasing very rapidly in recent years and it will keep a steady growing in the future. Recycling is a critical component of the aluminium industry based on its favourable economic impact on production and its contribution to environment. Compared to the processing of primary aluminium, recycling of aluminium is highly benecial, saving approximately 95% of energy consumption required for primary aluminium production. Production of secondary aluminium also results in less gaseous emissions, water consumption and solid residues. In a typical secondary aluminium process, the rotary furnace functions simultaneously as a smelter and a
*

Corresponding author. E-mail address: Y.Yang@CiTG.TUDelft.NL (Y. Yang).

phase separator. It is capable to deal with heavily contaminated scrap (Aluminium Handbook, 2003). Fig. 1 illustrates the complex phenomena in such a furnace. The scrap feed is charged into a rotary furnace, passing through a salt layer, melting, mixing and being cleaned in the furnace. The rotary furnace is normally operated at a temperature around 800 C. Burning of natural gas with oxygen is used as the heat source. The produced liquid aluminium is tapped into a holding furnace, further rened and then directly transported to the industrial partners or cast into ingots. The salt slag with various contaminations should be further processed and reused. The complexity in the process is due not only to the high temperature eect and the complex chemical reactions, but also to the highly complex scrap feed with a distributed nature of aluminium types, sizes, shapes, compositions, paintings and other contaminations. Though the process for reclamation of aluminium scrap has been developed for many years, little public knowledge is available. In this study, a computational

0892-6875/$ - see front matter 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.mineng.2005.07.017

300

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

Fig. 1. Illustration of the rotary melting furnace and the complex phenomena inside the furnace.

uid dynamics (CFD) based process model of scrap melting in a rotary furnace was developed to predicting the melting rate and energy distribution in relation to the scrap types and properties, and to make improvements of the process. Turbulent uid ow, gas combustion, radiation, and conjugated heat transfer were simulated. To represent the distributed nature of the aluminium scrap feed with dierent types, sizes, shapes etc., a scrap melting sub-model with population balance modelling (PBM) was developed and integrated into the CFD based process model. The scrap melting sub-model was simplied from the previous developed numerical model for a single aluminium particle melting in molten melts (Zhou et al., 2003). Furthermore an aluminium burno sub-model was developed and integrated to take into account the heat generated due to the burn-o (oxidation) eect during the melting process. Finally a number of case studies were conducted to understand the inuence of particle size, shape and quality of the scrap.

part of the furnace. The rotation of the furnace, about 1.33 rpm, and the agitation of the paddles built in the furnace wall were not included in this model. An unstructured mesh was applied in this study, the total number of the meshes is 88,566 and ner meshes were used in some sensitive areas, e.g., burner and ame areas, while coarser meshes were applied in the end part of the furnace and in the furnace lining. The CPU time for a whole simulation on a Pentium IV, 2.66 GHz PC is about 35 h with the main time step set at 30 s. 2.2. Treatment of the scrap-salt zone The solidliquid region was regarded as a conducting solid, a mixture of scrap and salt solids. The phase change of scrap melting was handled by the userdeveloped melting sub-model. Thus the uid ow in this zone was not considered, while the eect on heat transfer in this zone was represented by a number of thermal parameters. It was assumed that scrap and salt are well mixed. The thermal properties of the mixed material were calculated based on the mass fraction and the phase state of the materials in the mixture. The eective density of this region was calculated based on densities of solid scrap and salt, and it was assumed not to change during the process, despite the eect of thermal expansion. Heat capacity was also dened in a similar way. Some augmentation coecients for the thermal conductivity of the mixture were applied here to take into account the inuence of the voidage in the scrap-salt zone, as well as the eect on heat transfer due to the uid ow in the scrap-salt zone, agitated by furnace rotation. 2.3. Initial and boundary conditions The initial temperature in the gas zone and the scrapsalt zone was set as 303 K. The initial temperature in the lining structure was imported from a previous steady simulation of heating the empty furnace. An initial size distribution of the scrap and salt particles was predened.

2. CFD framework of the process model 2.1. General information of the CFD based process model Metallurgical processes involve complex phenomena of momentum, heat and mass transport, which play important roles in reaction kinetics and reactor performance. CFD as a research tool was found useful in studying various metallurgical processes. In this study, a commercial CFD package, ANSYS-CFX 5.6 (2003), was used as a framework of the process model, coupled with user-developed sub-models. The industrial scale rotary furnace is 3.0 m in inner diameter, 3.65 m in outer diameter including the lining structure and 6.9 m in length. The model consists of a gas region with turbulent ow and combustion as well as radiative heat transfer in the upper part of the furnace, a solid region of the furnace lining, and a solid liquid region of salt and aluminium metal in the lower

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

301

The inlets of the burner for natural gas and oxygen are really small compared to the furnace body. In this case the inlet was simplied to reduce the computing time. The proles of velocity, temperature and mass ows of gases at the inlet were dened using CFX Expression Language (CEL), based on a previous simulation with a full burner structure and ner meshes. Pressure boundary condition was set for the outlet. Heat transfer coecient was applied for outside wall. In this case, it is set as 15 W/m2 K and the environmental temperature is set as 303 K. 2.4. Physical models For the simulation of the turbulent ow in the furnace gas zone, the widely used standard ke model was applied. Eddy dissipation model (ANSYS-CFX, 2003) was used for the combustion of the natural gas with oxygen. This model is based on the concept that chemical reaction is fast relative to the transport processes in the ow. It is applicable in many industrial combustion problems where reaction rates are fast compared to reactant mixing rates. Radiation has a large contribution of the heat transfer to the scrap-salt zone and the furnace wall. Discrete transfer model (DTM) (ANSYS-CFX, 2003) was applied. The emissivity of the inner furnace wall and the interface between the gas zone and scrap-salt zone was set as 0.8. The buoyancy ow of the gas was simulated with the full buoyancy model implemented in ANSYS-CFX (2003).

where w = w(x, y, z, f1, f2, . . . , fj, t) is the distribution of a property, as function of x, y, z, f1, f2, . . . , fJ, t; vx, vy, vz are geometric velocities, vx = dx/dt, vy = dy/dt, vz = dz/ dt; fj({f1, f2, . . . , fJ}) are a specied set of properties of the particles, vj = dfj/dt is the time rate of change of the property fj; D is the death function of the property and B is the birth function of the property. For the application of the population balance model for aluminium scrap melting, the form of PBM can be greatly simplied. There is no input or output during the melting process. The birth and death of the scrap pieces due to break-up of the bigger ones, is dicult to obtain and has been neglected, and there is no coalescence of the solid particles. Then the factor left is the shrinking (melting) of the scrap particles. For example, considering the scrap property of size, for a certain scrap particle, it shrinks during the process, transfers from a bigger-size group to a smaller-size group, and nally disappears. Furthermore, in the current process model, the scrap-salt zone was regarded as a conducting solid. The eect of uid ow, furnace rotation and agitation was not taken into account directly, but represented by some model parameters, such as the Nusselt number for the scrap melting sub-model, the coecients of eective thermal conductivity in the process model. The eect of geometric velocities can be omitted, thus, Eq. (1) can be simplied as
J ow X o vj w 0 ot ofj j1

3. User developed sub-models 3.1. Population balance model for aluminium scrap melting Population balance model (PBM) (Sohn et al., 1979) is a very useful tool to represent the dynamic particle size variations as function of time, distributed physical properties and other process parameters. There are two forms of the population balance: a microscopic form and a macroscopic form. In this study, the simplied form of the microscopic population balance model was applied. The microscopic population balance model accounts for changes in a particle population in an innitesimal volume at any geometrical position, as function of time. The general form of the microscopic population balance is ow o o o vx w vy w vz w ot ox oy oz J X o vj w D B 0 ofj j1

where fj can be the properties of the scrap such as size, shape and thermal properties. In the case of aluminium scrap melting, the population balance modelling of aluminium scrap size distribution is in fact the calculation of the melting rate of the aluminium scrap. The melting rate of a certain scrap particle is dependent on the position, time, local temperature, local Nusselt number, scrap properties etc., and it can be calculated by the user-developed melting sub-model, which was introduced in the next section. 3.2. Scrap melting model for a single aluminium particle To calculate the melting rate of the scrap in such a two-melt (salt and aluminium metal) system in a rotary furnace, experimental study (Zhou et al., 2002) and numerical modelling (Zhou et al., 2003) was rstly conducted for a single aluminium particle melting in the molten metal and salt melts. The melting process of a scrap particle is dependent on the properties of the particle (such as the size, shape, composition, and initial temperature), the properties of the bulk melt (such as the temperature, liquid ow and agitation of the melt), and the situations when the particle was charged passing

302

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

through the salt layer (such as the thermal properties of the salt melt, residence time in the salt layer and speed passing through the salt layer). If the scrap particle is regarded as a sphere with an original radius of R0, energy conservation equations due to heat transfer in the system can be described as follows (Zhang and Oeters, 1998). For heating of the sphere, when 0 < r 6 R0:  2  oT o T 2 oT ap 3 ot or2 r or I.C. : T T 0 r < R0 ; T T m;s r R0 at t 0 B.C.1: oT =or 0 at r 0     oT oT kshell B.C.2 : kp or rR0 or rR
0

where Rsolid is the radius of the solid core, qp is the density of the particle, Tm,p is the melting point of the metallic sphere, and DHp is the latent heat of fusion of the sphere. Similarly, if the melting point of the sphere is lower than that of the shell, pre-melting of the sphere inside the salt shell may happen, then the boundary conditions of the governing equation, Eq. (3), can be expressed as follows, assuming that the temperature of the metal liquid inside the shell is unique and it equals the melting point of metal.
B.C.1: oT =or 0 at r 0     oT dRsolid R2 oT qp DH p B.C.2: kp 2 0 kshell dt or rRsolid or rR0 Rsolid

where r is the particle radius, T is the particle temperature, t is the time, Tm,s is the melting point of the shell, kp and kshell are the thermal conductivities of the sphere and the shell, respectively, and ap is the thermal diusivity of the metallic sphere. Heat balance at r = R0, the interface between the sphere and the shell, is expressed as B.C.2 in Eq. (3). A shell is formed when a cold metal particle charged into a hot melt. For the shell development and remelting, when R0 < r 6 Rshell:  2  oT o T 2 oT ashell 4 ot or2 r or I.C.: Rshell R0 ; T T m;s at t 0 B.C.1: T T m;s at r Rshell   oT dRshell hT f T m;s B.C.2: kshell qshell DH shell or rRshell dt where Rshell is the radius of the shell, qshell is the density of the shell, Tf is the temperature of the bulk salt melt, DHshell is the latent heat of the phase change, h is the heat transfer coecient from the bulk melt to the solidifying shell, and ashell is the thermal diusivity of the shell. Heat balance at r = Rshell, the interface between the shell and the melt, is expressed as B.C.2 in Eq. (4). If the bulk melt is changed during the process, e.g., an aluminium particle passing through the salt layer and entering the metal bath, there may be two shell layers formed. If the melting point of the sphere is higher than that of the shell, there will be no pre-melting of the sphere inside the shell. After the shell is re-melted, the metal solid sphere is further heated up and begins to melt. In this situation, the governing equation is the same as Eq. (3) but with dierent boundary conditions: B.C.1: oT =or 0 at r 0   oT dRsolid B.C.2 : kp hT f T m;p qp DH p or rRsolid dt

By solving these equations with a nite dierence method (Zhou et al., 2003), the melting process of a scrap particle under certain conditions can be simulated and the size change of the particle against heating time can be obtained. The melting behaviour of a metal particle as well as the salt shell formation and re-melting was studied. The results were compared with the previous experimental study (Zhou et al., 2002), and reasonable agreement was obtained. As an example shown in Fig. 2, an aluminium particle, its equivalent radius is about 1.0 cm, was screwed on a steel rod, pre-heated to 280 C, dipped into the salt melt (NaClKCl 70%30%wt base salt system plus Cryolite 5%wt, stagnant, 800 C) and then taken out after a certain immersion time. The weight and thickness of the salt shell formed on the aluminium solid was measured. The triangles in Fig. 2 are the measured thickness of the salt shell, compared with the model prediction of the rst 80 s. In this case, pre-melting of the aluminium core inside the salt shell happens, as shown in the gure. It should be noted that in this case, there is no metal melt in the system, because it is dicult to conduct and control the experiment when two melts exist. More details of the experimental study (Zhou et al., 2002) and the numerical modelling (Zhou et al., 2003) for aluminium melting as well as salt shell formation and re-melting can be found in the previous publications.

Fig. 2. Experimental and numerical results of salt shell formation and re-melting on Al particles in molten salt bath.

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

303

3.3. Implementation of the scrap melting sub-mode For a multi-size particle system, the scrap is classied into certain groups depending on the scrap size. It can also be classied by other criteria, such as shape, composition, and scrap thermal properties, as well as the combinations of those criteria. For each cell in the scrap-salt zone, it is assumed that it has the same initial size distribution. A scrap melting sub-model was developed and integrated in the CFD based process model, it handles each size group the same as a single solid particle, calculates the melting rate based on the conditions in each cell in the scrap-salt zone at any time, based on the information exchange with the CFD framework. The melting sub-model provides the CFD framework with the information due to the melting of solid scrap, e.g., the heat sink due to melting, the amount of liquid metal and solid scrap, and the size distribution of scrap. At the same time, the CFD framework provides the information needed for the phase change calculations, e.g., the local temperatures. The scrap melting sub-model was simplied from the previous work of modelling a single particle melting in molten melts (Zhou et al., 2003), in order to reduce the computing time. The uid ow in the scrap-salt zone and the agitation due to furnace rotation were not fully included in the CFD based process model, and the resolidication process was not simulated in the simplied melting sub-model and its eect on heat transfer were taken into account by some model parameters, such as the Nusselt number for the scrap melting model, the coecients of eective thermal conductivity of the scrap-salt zone in the process model. Since the thermal conductivity of the aluminium metal is very high, the temperature dierence within the particle can be ignored: oT/or = 0. When the temperature of the solid particle reaches its melting point, the heat transferred from the environment to the particle is totally used for melting of the solid metal. The heat balance at the interface between the bulk melt and the solid, (r = Rsolid) can be expressed as follows: qp DH p dR hT f T m;p dt 5

to the CFD framework as an energy source term, which inuences the temperature distribution in the scrap-salt zone and heat transfer from the combustion gas or furnace wall to the scrap-salt zone. In this way, the melting rate of the scrap at any position and time can be obtained, and therefore the population balance of aluminium scrap melting can be established, based on the coupling of the scrap melting modelling and the CFD simulation of the furnace. 3.4. Development of the scrap burn-o sub-model Aluminium is a very reactive metal, thus oxidation is always occurring during its life. In secondary aluminium process, the oxidized aluminium can never be reclaimed in secondary aluminium process and contributes to the losses. During the melting process, aluminium scrap is sometimes exposed in a high temperature and oxidation atmosphere, despite the presence of a protecting salt layer. The burn-o rate is dependent on the operation, salt amount, scrap quality and many other factors. The generated heat amounts to between 1/3 and 1/2 of the total energy input generated by burning of the natural gas (Boin et al., 2004). This indicates that to build a valid process model, scrap burn-o must be taken into consideration. The scrap burn-o is composed of several dierent sources and reactions (Boin et al., 2004). These reactions can be summarised as follows: 2Al + 3/2O2 = Al2 O3 2Al + 3H2 O = Al2 O3 + 3H2 2Al + 3CO2 = Al2 O3 + 3CO 4Al + 3CH2 = Al4 C3 + 3H2 2CH2 + 3O2 = 2CO2 + 2H2 O 7 8 9 10 11

where the heat transfer coecient, h, can be calculated based on the particle size, uid ow condition and bulk melt properties, and the bulk melt temperature, Tf, can be obtained from the CFD simulations. Thus the size change of the particle can be calculated as follows: dR hT f T m;p dt qp DH p 6

New particle size for each group in each cell and the total amount of melted particle in each cell at the current time step can be obtained. The heat sink for each cell due to phase change can be calculated and returned

It includes the direct oxidation reaction of aluminium metal, as expressed in Eq. (7), and the reaction between the aluminium metal and the moisture, as expressed in Eq. (8), or carbon dioxide, as expressed in Eq. (9). The aluminium metal also reacts with the contamination materials, e.g., plastic materials, and these reactions can also be regarded as the scrap burn-o, which generates a large quantity of heat and increase the metal loss, as expressed in Eq. (10), if the structure formula of the contamination materials can be approximately expressed as CH2. Burning of the contamination materials and organic components attached on the scrap also has a similar eect on the melting process, as expressed in Eq. (11). Related to the distributed nature of the scrap, the scrap size distribution, surface to volume ratio, and content of contamination have a big inuence of the scrap

304

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

burn-o. For the scrap of good quality, which has a relatively small surface to volume ratio, less contaminations and/or higher metal content, the burn-o is normally less. And the total amount of scrap burn-o during the melting process is also inuenced by a number of other variables in the system, e.g., temperature in the furnace, free-oxygen in the furnace, moisture content of scrap, melting status (liquid and solid ratio of metal). It makes very dicult to obtain the total amount of burn-o directly from the burn-o reactions and reaction kinetics. In this study, it was estimated by a mass and energy balance model with data reconciliation (Boin et al., 2004). Data reconciliation is a technique by which the mass and energy balances can be closed by adjusting the measured data, while the measurements should be adjusted as little as possible. The adjusted data should give a more consistent representation of the actual process, which then forms the basis for any subsequent modelling that covers energy balances, statistics, kinetic modelling, neural nets and CFD modelling. For twenty-six furnace cycles, the burn-o rate for each furnace cycle was obtained through data reconciliation by closing the mass and energy balance with minimum of standard deviation of errors. These cycles were then split into three groups based on the properties of the scrap, and the calculated results are listed in Table 1 (Boin et al., 2004). It indicates that for a certain type of the scrap feed, e.g., the scrap with a metal recovery rate of 80%, the scrap burn-o is 2.69%, and the heat generated due to scrap burn-o is about 657 MJ per ton of the scrap feed. The scrap burn-o sub-model translates this part of heat, which contributes in both of the gas zone and the scrap-salt zone, into heat sources. It is too complex to consider all the inuencing factors by far and here only a simplied denition of the burn-o is present: Total amount of heat generated by scrap burn-o is pre-dened based on the mass and energy balance calculations (Boin et al., 2004), in relation to the scrap type and quality. Kinetics involved in burn-o was not taken into account yet, however this can be assumed to be rapid. The overall heat generation rate was dened based on the general industrial observations and the measured
Table 1 Dierent burn-o rates for dierent scrap groups Groupa 1 2 3
a

Fig. 3. Pre-dened burn-o heat source (solid line), and the industrially recorded o-gas temperature.

o-gas temperature, which can be used to indicate the extra heat generation in the furnace. Therefore a scrap burn-o function was dened with the function parameters roughly estimated based on a statistic calculation of several typical cycles. Fig. 3 shows the pre-dened curve of the burn-o heat source together with the measured o-gas temperature. The scrap burn-o reactions and their following reactions generate heat both in the gas zone and the scrap-salt zone, while the ratio of the two parts is difcult to determine. It was thus studied as a model parameter. It is assumed that the burn-o reactions are position independent, and it is evenly distributed in the gas zone or in the scrap-salt zone. The eect of burn-o on the mass balance was ignored, which is very small.

4. Results 4.1. General results Turbulent uid ow, gas combustion, radiation, and conjugated heat transfer in the rotary furnace were simulated in the CFD framework. The detailed information of the uid ow in the gas zone, the temperature distribution in the furnace and energy ows of the process can be obtained. As an example, Fig. 4 shows the gas combustion and ow in the gas zone and Fig. 5 shows the temperature eld in the furnace, at the 9600th second of the process when about 80% of the scrap has been melted. One of the main purposes of the simulation is to obtain the total melting time under a certain condition. Fig. 6 shows the changing history of the total solids remaining, including the scrap and salt, against the heating time. It should be noted that the plotted remaining ratio are calculated by the total weight of the feed, which includes both scrap and salt. The melting curve can be regarded as the main criterion for the melting process,

Average expected metal yield (%) 65 80 90

Burn-o rate (%) 4.42 2.69 2.14

Group 1: dross; Group 2: granules, shredder residue, turnings; Group 3: packages (bottle caps, cooling elements, turnings), shredder residue.

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

305

Fig. 4. Gas combustion and ow in the gas zone at t = 9600 s.

Fig. 5. Temperature eld in the furnace at t = 9600 s.

which indicates the time the scrap starts to melt and the time it has been melted completely. For melting of about 13 tons of scrap and 4 tons of salt ux, the total melting time is about 4.0 h in this case, assuming that the burno rate is 2.7% and 50% of the burn-o heat (about 8540 MJ in total) contributes in the gas zone. 4.2. Inuence of scrap size and shape On the one hand, the experimental and numerical results (Zhou et al., 2003) suggest that an aluminium particle with a larger size needs longer time to be melted. On the other hand, if the total weight of the feed is xed, a combination with smaller sizes of scrap has a larger total

Fig. 6. Melting curve of the solid (scrap and salt) in the scrap-salt zone (ratio by total weight of feed).

306

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

surface area, which means more salt ux is needed in operation. Here, the melting sub-model has been developed and coupled with the CFD based process model. Population balance modelling of scrap with dierent size distributions can be established and the inuence of the size distribution on scrap melting can be studied. The initial size distributions as well as their change during the melting process are shown in Figs. 711. The distribution is plotted as the weight percentage to the total weight of scrap, against particle size, and the number of scrap groups is 25. In reality, the size distribution may be more complicated, dependent on the scrap type and the feed recipe. Here the initial size distributions studied are dened as follows:

Fig. 10. Simulated changing history of size distribution D.

Fig. 11. Simulated changing history of size distribution E. Fig. 7. Simulated changing history of size distribution A.

Size distribution A: 0.00.5 m in diameter, with a larger portion of smaller size of scrap. Size distribution B: 0.00.5 m in diameter, with a larger portion of bigger size of scrap. Size distribution C: uniform size, 0.1 m in diameter. Size distribution D: uniform size, 0.4 m in diameter. Size distribution E: 0.00.5 m in diameter, randomly distributed. Fig. 12 shows the melting curve of the aluminium scrap with dierent size distributions. For the scrap feed with larger portion of smaller particles (size distributions C and A, which can be regarded as a bad quality

Fig. 8. Simulated changing history of size distribution B.

Fig. 9. Simulated changing history of size distribution C.

Fig. 12. Melting curve of the solids in scrap-salt zone, with dierent size distributions.

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

307

scrap), the melting is faster than that with larger portion of bigger particles (size distributions D and B, which can be regarded as a good quality scrap) in the earlier stage. While the total melting time for all these cases is almost the same, about 14,400 s (4.0 h). It indicates that the melting process is mainly dependent on the heat transfer but scrap size. Here it should be noted that, in reality, scrap with smaller sizes normally has a lower metal recovery, causes a higher burn-o rate and requires more salt ux. These consequent eects have not been taken into account in the cases presented here. The aluminium scrap also has a large variety of shapes. The shape factor was applied and dened as the ratio of the surface area of the particle to the surface area of a sphere with the same volume, e.g., it is 1.0 for the spheres and it is 1.24 for the cubes. The results indicated that there is little inuence of the shape, the shape factor ranges from 1.0 to 20.0, if other parameters are not changed. While in reality, larger surface area to volume ratio of scrap would result in a larger requirement of salt ux, which may inuence the scrap melting process. 4.3. Inuence of scrap quality Metal burn-o during the melting process is one of the main reasons of metal loss in secondary aluminium, while it also generates a large amount of heat. According to the observations and data measurements in the plant, as well as the mass and energy balance calculations, the scrap burn-o rate has an obvious relationship to the scrap type or scrap quality (Boin et al., 2004). Normally, for the good quality of scrap, which has a relatively small surface to volume ratio, less contamination and/ or higher metal content, the metal recovery rate is higher and the metal loss is normally less. The burn-o model is only implemented preliminarily, chemical reactions and their mechanisms are not taken into account, and here the study is focused on the energy aspect. For a total input of 13 tons of scrap feed, the scrap melting behaviour was simulated with dierent scrap burn-o rates, 0.0% (A, no burn-o), 1.5% (B), 2.0% (C), 2.7% (D) and 3.5% (E), and the heat generated in total is 0 MJ, 4750 MJ, 6330 MJ, 8540 MJ and 11,070 MJ, respectively. It assumes that the heat generated in the gas zone and that in the scrap zone are the same, 50% to 50%. Fig. 13 shows the melting curves for these cases. The total melting time is about 6.25 h, 5.25 h, 5.0 h, 4.35 h, and 3.5 h, respectively, which indicates that the total amount of burn-o has a big inuence on scrap melting. A higher burn-o rate results in a shorter melting time, but it is at the expense of metal loss. The generated heat due to scrap burn-o contributes both in the gas zone and the scrap-salt zone, as discussed in the previous paragraphs, while the ratio of these two parts is dicult to decide. It was studied as a parameter in the model. Fig. 14 shows the melting curves with dif-

Fig. 13. Inuence of the scrap burn-o rate, the total amount.

Fig. 14. Inuence of the scrap burn-o rate, the ratio in the gas zone and the scrap-salt zone.

ferent ratios of the heat contributed in the gas zone and the scrap-salt zone, 0%, 20%, 50%, 80% and 100% in the gas zone, respectively. For the same scrap burn-o rate, 2.7% here, larger ratio in scrap-salt zone results in a faster melting of scrap. For these cases, the total melting time ranges from 4.0 h to 4.5 h.

5. Concluding remarks A CFD based process model was developed, in which uid ow, heat transfer, natural gas combustion, and radiative heat transfer were simulated for the scrap melting process. Data measured in industry were applied in the model as initial and boundary conditions, as well as for model validation. User sub-models for scrap melting and scrap burn-o were developed and integrated into the CFD framework. A simplied population balance model for aluminium scrap melting was established by classifying the scrap feed into a number of scrap groups. The melting rate for each group in each cell at each time step was calculated with the exchange of information between the melting sub-model and the CFD based framework. Thus the distributed nature of the complex scrap feed can be taken into account. The distributed properties of scrap also result in distributed burn-o rates. The scrap burn-o sub-model was

308

B. Zhou et al. / Minerals Engineering 19 (2006) 299308

developed to take into account the inuence of the burno heat on the scrap melting process. Various case studies were carried out focussing on the relationship between the melting process and the scrap properties, e.g., scrap size, shape, burn-o and quality. It shows that the scrap size and shape only have a small inuence on melting, if ruling out the other consequent eects due to size and shape dierence. The distributed burn-o rate, which represents the scrap quality here, is one of the important factors for the melting process. For the scrap with a higher burn-o rate, which normally is the scrap with a poor quality, the total melting time and the gas consumption can be reduced, while it is at the expense of more metal loss. The melting sub-model can be further improved, e.g., taking into account the re-solidication, and rening the model parameters. The scrap burn-o sub-model can also be improved by rening the burn-o function, taking into account the chemical composition, distributed kinetics, etc. Moreover, the sub-models are not completely coupled with each other yet, e.g., the scrap burn-o rate has not been directly related to the distributed scrap properties but pre-dened, which should be improved in the future. Acknowledgements This work is part of the E.E.T. (Economy, Ecology and Technology) project (http://www.batchcentre.

tudelft.nl) supported by the Dutch government. The nancial support from the E.E.T. program is gratefully acknowledged. Special thanks to Karl Konzelmann Metallschmelzwerke GmbH, Hanover, for the sharing of the information, knowledge and data, and for their support, cooperation and hospitality.

References
Aluminium Handbook, 2003. Forming, Casting, Surface Treatment, Recycling and Ecology, vol. 2. Aluminium-Verlag, Dusseldorf. ANSYS-CFX manual and documentation, version 5.6, 2003. ANSYS, Inc. Boin, U., Reuter, M.A., Probst, Th., 2004. Measuringmodelling: understanding the aluminium scrap melting processes inside a rotary furnace. Erzmetall 57 (5), 266271. European Aluminium Association (EAA), 2004. Available from: <http://www.aluminium.org/>. Sohn, H.Y. et al., 1979. Rate Processes of Extractive Metallurgy. Plenum Press, New York. Zhang, L., Oeters, F., 1998. Melting and Mixing of Alloying Agents in Steel Melts, Methods of Mathematical Modelling. Verlag Stahleisen GmbH, Dusseldorf. Zhou, B., Yang, Y., Reuter, M.A., 2002. Study of melting behaviour of aluminium scraps in molten melts. In: Proceedings TMS Fall 2002 Extraction and Processing Division Meeting, Lulea, Sweden, June 1620, pp. 527537. Zhou, B., Yang, Y., Reuter, M.A., 2003. Modelling of melting behaviour of aluminium metal in molten salt and metal bath. In: Proceedings Yazawa International Symposium on Metallurgical and Materials Processing, San Diego, California, USA, March 36, pp. 12491258.