You are on page 1of 15

Energy 72 (2014) 521e535

Contents lists available at ScienceDirect

Energy
journal homepage: www.elsevier.com/locate/energy

Numerical simulation and process optimization of an aluminum


holding furnace based on response surface methodology and uniform
design
Ji-min Wang*, Shen Lan, Wen-ke Li
School of Energy and Environment, Anhui University of Technology, Maanshan 243002, Anhui, China

a r t i c l e i n f o a b s t r a c t

Article history: It is indispensable for strengthening smelting process and controlling ne to understand metallurgical
Received 11 June 2013 behaviors and mechanism of production process of aluminum holding furnaces. A CFD (computational
Received in revised form uid dynamics) compressive process model was developed and integrated with energy distribution
19 May 2014
regime using user subroutines based on FLUENT code. Integrated intelligence ideas with a combination of
Accepted 19 May 2014
response surface methodology and uniform design was employed to quantitatively achieve the inherent
Available online 13 June 2014
laws of thermal characteristics and mechanism behind various factors. The optimal scheme obtained
from the compromise of the two desirable responses was as follows: ue position of same side, oxidant
Keywords:
Aluminum holding furnace
type of pure oxygen, air preheated temperature of 719 K, natural gas velocity of 108.88 m/s, burner
Thermal characteristics height of 1150 mm, vertical burner angle of 5 , pool depth of 625 mm, horizontal spacing between
Numerical simulation burners of 2050 mm, height-radius ratio of 0.6, airefuel ratio of 1.04, and burner load ratio of 1:1. The
Process optimization maximal end aluminum temperature uniform coefcient of 12.85% and minimal energy consumption per
Response surface methodology ton of aluminum of 0.01092 kg ce/t-Al were obtained. Conrmed experiments demonstrated that such a
Uniform design combination of the CFD, RSM (response surface methodology) and UD (uniform design) is a powerful and
useful approach for process optimization of aluminum holding furnaces.
2014 Elsevier Ltd. All rights reserved.

1. Introduction temperature is kept constant. Aluminum holding furnaces with a


large box-shaped chamber are frequently used to store molten
Aluminum is the largest non-ferrous metal or metallic material aluminum for purication and alloying processing. Any type of
and the secondary aluminum becomes more and more ubiquitous aluminum holding furnaces is an extremely energy-intensive de-
under the exhaustion of raw resources all over the world. Due to vice. The furnace manufacturers are constantly striving to nd new
their high performance, light weight and low cost, aluminum alloys technology, retrot furnace design and explore optimum parame-
are widely used in many elds, e.g., aerospace, transportation, ters combination, which would provide enormous amounts of en-
package and architecture industries, and have been the research ergy effectively to heat the metals and greatly improve product
and development focus since the 19th century for scientists and quality, particularly with the recent escalation in natural gas prices.
engineers. Along with the rapid development of aluminum fabri- This requires detailed knowledge about the operation parame-
cation and applications in the world, the aluminum production in ters, combustion phenomenon and energy distribution. However, it
China also grows continuously in the last decade. The aluminum is signicant for the designer or operator to know the details in the
production and consumption in China have ranked No.1 in the furnace. Extensive studies have been carried out to analyze multi-
world since 2007. physics eld inside aluminum holding furnaces, optimize thermal
Excellent properties of secondary aluminum are typically parameters, and make holding systems. By running energy ef-
accomplished inside large reverb, natural-gas-red furnaces of ciency tests in aluminum combination melting and holding fur-
aluminum casting houses. After aluminum is molten inside a naces, Belt [1] addressed that the largest improvement is the
melting furnace, it is transferred to a holding furnace, where its reduction in re rate. Migchielsen et al. [2] discussed the main
consideration in designing a holding furnace with respect to me-
chanical, thermal-process and control aspects. Although the
* Corresponding author. Tel.: 86 555 2311571.
experimental techniques have been relied on almost exclusively for
E-mail address: endlessfree@163.com (J.-m. Wang). furnace research and development, they have been proven time-

http://dx.doi.org/10.1016/j.energy.2014.05.077
0360-5442/ 2014 Elsevier Ltd. All rights reserved.
522 J.-m. Wang et al. / Energy 72 (2014) 521e535

consuming and expensive. Fortunately, the advances in numerical factors. Uniform design is a collection of statistical and mathe-
techniques and computing power have given birth to CFD matical techniques that has been successfully used for developing,
(computational uid dynamics), which has reduced the difculties improving and optimizing process.
for full-scale prototype testing. For acquiring the details in Wang et al. [18] proposed that a novel approach with a combi-
aluminum holding furnaces with bottom porous brick purging nation of response surface methodology and uniform design was
system, Zhang et al. [3] adopted transient analysis scheme and employed to optimize the coagulation-occulation process. Wang
constant boundary temperature to nd out the potential optimal et al. [19] analyzed the effects of processing parameters of extrac-
operation schemes. Zhou [4] focused on melt ow driven by porous tion from Gynostemma pentaphyllum Makino on the yield of poly-
plug bubbles in aluminum casting furnace using EulereLagrange saccharides using uniform Design. By using the uniform design
approach, and discussed [5] the effects of gas ux on holding non-linear model method, Lam Leung et al. [20] evaluated quali-
furnace performance during heating and agitation process. Zhou tatively and quantitatively fabrication conditions of the ultraltra-
et al. [6] employed a uidesolid coupled method to predict three- tion membranes. Li et al. [21] systematically discussed a plasma
dimensional temperature distribution of molten aluminum and its spray process of Titanium nitride deposition using the uniform
inuencing factors inside an industrial aluminum holding furnace, design of experiments. Liu et al. [22] employed uniform design and
and elaborated [7] the effects of temperature of combustion air and partial least squares regression to quantitatively explore the effects
excess air ratio on temperature distribution of aluminum holding of plant oil and fatty acid as well as their addition amount and
furnaces. Zhou [8] built up a numerical model of aluminum holding addition time on the performance of Tuber melanosporum sub-
furnaces with bottom porous brick purging system based on merged fermentation.
FLUENT code, claried the inuences of operation parameters, and Particularly, production process of aluminum holding furnaces
accomplished decision-making in thermal scheme of the furnace. is a complicated system which is a nonlinear multivariable and
Ma [9] simulated numerically temperature eld of 25 tons coupling system, it is time-variable and uncertain. In the present
aluminum holding furnace. With an increasing emphasis on work, this paper concerns uid ow, fuel combustion, heat and
reducing greenhouse gas emissions due to more stringent regula- mass transfer, energy distribution regime, heat loss through
tory pressures, Golchert et al. [10] modeled the effect of changing a furnace walls inside aluminum holding furnaces. A CFD compre-
generic aluminum holding furnace from air-red to oxy-fuel red hensive process model was developed and integrated with energy
burners. Concerning with both the holding and pouring phases of distribution regime. Integrated intelligence ideas by combining the
aluminum casting, Pauty et al. [11] developed a numerical model merits of RSM and UD would be able to achieve the process opti-
for the ows and the transport of solid inclusions in an aluminum mization of a complex multivariate process with the fewest
holding furnace. multilevel experiments. The inherently thermal characterization
Earlier literature about melt quality is centered on the puri- and overall performance were achieved based on statistical prin-
cation of liquid aluminum alloy inside molten pool. However, there ciple and an integrative CFDeRSMeUD approach. End aluminum
is less study on aluminum holding equipments to explore the ef- temperature uniform coefcient and energy consumption per ton
fects on thermal characteristics. In recent years, numerical methods of aluminum were chosen as the dependent output variables. The
have been employed to assess thermal behaviors of aluminum compromise optimal conditions for these two responses were ob-
holding furnaces. It's important to note that these studies barely tained using the desirability function approach. It not only enriches
covered certain phase of aluminum holding furnaces, and do not comprehensive knowledge of thermal behaviors of furnaces and
consider that heat load varies and uctuates with melt tempera- kilns for nonferrous metallurgical engineering, but also provides a
ture, thus, the inherent laws of thermal characteristics of aluminum novel modeling and optimal control strategy of other complicated
holding furnaces can not thoroughly be revealed. What's more, the systems such as metallurgical engineering and other elds. The
optimum conditions are short of theoretical support as well. In corresponding results can also guide or improve industrial pro-
statistics, RSM (response surface methodology) [12] is used to duction practices.
quantitatively explore the relationships between several explana-
tory variables and one or more response variables. Response sur- 2. Mathematical formulation of aluminum holding furnace
face methodology has been widely used for the optimization of
various process engineering. The holding furnace is equipped with two burners. To avoid
Wang et al. [13] reported that CFD technique, in association with any proprietary issues, a typical burner involves a tube-in-tube
statistical experimental design was used to optimize melting pro- arrangement. Fuel (natural gas) would be injected in an inner
cess of aluminum melting furnaces. Natarajan et al. [14] focused on tube while the oxidant (air or pure oxygen) would be injected in
response surface methodology for multiple response optimization the outer tube. Nevertheless, it may still truly reect thermal
in micro-endmilling operation to achieve maximum metal removal phenomenon and main rules within aluminum holding fur-
rate and minimum surface roughness. Ko rbahti et al. [15] elabo- naces. The conguration description of simulation domains of
rated electrochemical treatment of a simulated water based paint the furnace is shown in Fig. 1. The combustion space dimensions
wastewater in the presence of NaCl on carbon electrodes using are 7  6.4  2 m3, which is a typical dimension of existing
response surface methodology. Abdullah et al. [16] intended to aluminum furnaces. The lower surface represents the melt
make use of response surface methodology process to maximize whose depth is 0.8 m. The inlets of oxidant and fuel as well as
the production of biodiesel from sunower oil using the conven- the outlet of the combustion products are located at the oppo-
tional transesterication method. site wall. The jets are directed against the refractory walls, and
Regrettably, in the traditional experimental design approaches the jet angle of 15 with respect to the horizontal axis was set.
used in RSM, such as central composite design, with an increase in The burners are located at 0.9 m away from the melt surface.
experimental factors, the number of experimental trials increases The height of the chimney is a little higher than that of the
exponentially. To overcome this shortcoming of RSM, UD (uniform burners.
design) concept [17] was proposed with number theory and sta- Aluminum holding furnaces belong to a typical metallurgical
tistics methods, which would reduce substantially fewer experi- reactor, and involve a drastically complex thermal process. Ac-
mental trials, since it determines the number of the experimental cording to the features of aluminum holding furnaces and process
trials only by the level of factors, rather than by the number of requirements, the following assumptions were made:
J.-m. Wang et al. / Energy 72 (2014) 521e535 523

Fig. 1. Conguration description of simulation domains of aluminum holding furnace.

(1) Owing to the low temperature differences between   


vrk m
aluminum loads, the melt velocity is low, thus, the liquid divrvk div m t grad k G  r (5)
vt Prk
aluminum can be assumed to be stagnant, i.e., natural con-
vection inside melt bath may not be considered. Chemical
  
reactions and mass transfer above the melt surface can be vr m
also negligible. Radiation and convection heat transfer at gas- divrv div m t grad c1 G  c2 r (6)
vt Pr k
melt interface are taken into account.
(2) According to the features of aluminum holding furnaces, The turbulent Prandtl and Schmidt numbers, Prt and Sct, and the
during heating phase and high re, the heat loss from the Prandtl number for the turbulence kinetic energy Prk and its
walls to the ambient may be neglected, therefore, the phase dissipation rate Pr, are empirical constants, as well as c1 and c2 in
is supposed to an adiabatic process. However, during the equation for the dissipation of the turbulence kinetic energy. In
standing stage and low re, convection heat exchange is the momentum equation, the modied total pressure P p  (2/3)
adopted as a boundary condition of external surface. Besides, [mef div v rk], takes into account the contributions due to the
total holding time is assumed to be 1 h. turbulent uctuations. The generation of the turbulence kinetic
(3) Based on the production experience of aluminum holding energy is given by G mt[grad v (grad v)T] grad v.
furnaces, uctuation factor of heat load ranges from Sh consists of sources of enthalpy due to chemical reaction and
around 5% to 5%. During holding phase, melt temperature is radiation. Models relating in particular to the determination of the
within the range of 1073 to 1078 K. In addition, the burning radiation contribution to the source term in the energy equation are
capacity of high re is ten times that of low re. categorized as radiation models. All radiation models employ the
RTE (radiative transfer equation) in one form or another. For ther-
mal process in aluminum holding furnaces, radiation heat transfer
2.1. Three-dimensional numerical model
plays the dominant role. The primary advantages of the DTRM
(Discrete Transfer Radiation Model) are threefold: it is a relatively
Since the ows in the furnace are turbulent and reactive, the
simple model, the accuracy can be improved by increasing the
Favre-averaged form of conservation equations is used in this study
number of rays, and it applies to a wide range of optical thickness.
to account for the effect of density change. For Lewis number
DTRM was chosen in most of the simulations in this study. This
Le Pr/Sc 1, the unsteady three-dimensional conservation
model is derived based on the assumptions that scattering is rela-
equations can be expressed as [23],
tively small in comparison to absorption and emission, and the
vr radiating gases are gray. The radiation transfer equation used by the
divrv 0 (1) DTRM is [23]
vt

vrv h  i dI asT 4
divrvv div mef gradv gradvT  grad p (2) aI (7)
vt ds p

   where, I is the radiation intensity, a is the absorption coefcient,


vrh m m
divrvh div t grad h Sh (3) and s is the StefaneBoltzmann constant. For increasing the accu-
vt Pr Prt racy, the DTRM may be utilized with an effective emissivity sub
   model, which takes into account the non-gray property of the
vrmi m m radiating gas. In this model, the absorption coefcient a is calcu-
divrvmi div t gradmi Ri (4)
vt Sc Sct lated by the Weighted Sum of Gray Gases Model as follow [23]:

In order to close the equation, the Reynolds stresses in the " #


momentum equation must be modeled. Models relating to theses Pn  ki ps

ln 1  i0 a;i T 1  e
stresses are categorized as turbulence models because they contain
physical information about the statistical behavior of turbulent a (8)
s
ows. The turbulent viscosity mt is dened in accordance with
widely used ke model as mt cmk2/, where cm is an empirical where, a,i is the emissivity weighting factors for the ith ctitious
constant in the model. The term Sh is the source term of momentum gray gas, ki is the absorption coefcient of the ith gray gas, p is the
equation. The conservation equations of the turbulent kinetic en- sum of the partial pressures of all absorbing gases, and s is the path
ergy k and its dissipation rate are [23] length.
524 J.-m. Wang et al. / Energy 72 (2014) 521e535

When modeling combustion process of natural gas and air, a Table 2


complex chemistry mechanism is involved. In the present study, Parameters for each reaction for the Arrhenius reaction rate expression.

natural gas was employed as fuel. Combustion process is supposed Reaction Pre-exponential factor Activation Concentration exponent
to comply with simple chemical reaction system, and fast reaction. Ai (m3 kmol1 s1) energy Ei
gCH4 gO2 gCO2 gCO gH2 O
Therefore, the kinetic reaction rate may be neglected, and com- 11
1 5.01  10 200 0.7 0.8 e 0 0
bustion process is mainly affected by the turbulent ow. A two-step
2 2.24  1012 170 e 0.25 0 1 e
mechanism and six species were selected, as shown in Table 1. The
stoichiometric coefcients for each reaction must be specied in
accordance to equations in Table 1. The reactions rate uk of the kth as a conducting solid. Eq. (13) is used to describe heat transfer
species is determined by the Arrhenius model, which is given by through the interface of uid and solid [23].
Ref. [23]

vT
   
Y l

h Tf  Tw s Tf4  Tw
4
(13)
uk hi Mi Ai T bi exp  Ei =Rc T Xi gi;k (9) vn w
i
where, l is the thermal conductivity of the solid. h is the local
where, the pre-exponential factor Ai, the temperature exponent bi, convective heat transfer coefcient. Tf is the uid temperature. Tw is
and the activation energy Ei are specied. Rc is the universal gas the temperature of the coupled wall. is the solid emissivity. s is
constant. Xi is the molar concentration of the ith species. gi,k is the the StefaneBoltzmann constant.
concentration exponent of the ith species. hi is the stoichiometric The major function of aluminum holding furnaces is to make
coefcient of the ith species. Mi is the molecular weight of the ith liquid aluminum be alloying within reasonable temperature range.
species. Alloying temperature of the melt usually is kept within the range of
To consider the turbulent effect on the reaction rate, the Mag- 1023e1123 K. Alloying process includes heating, holding and
nussen model was employed. The reaction rate is determined as the standing phase. During heating phase, liquid aluminum is heated to
smallest value between the two following expression [23] near alloying temperature. During holding phase, aluminum tem-
perature is kept constant by adjusting fuel mass ow. Burners start
Y high re and low re respectively when aluminum temperature
Ri:r hi;r Mi T br Ai exp  Ei =RT Xi gi;k (10)
drops below lower limit or exceeds upper limit. What's more,
i
holding time should meet alloying requirements. Turn off burners
2 P 3 and stop heating after holding process lasts for about 1 h.
mp Aluminum temperature gradually reduces to tapping temperature.
6 m j* p 7
Ri;r hi;r Mi K1 r min4 ;K P 5 (11) Heat load randomly varies during alloying process. Conceptual
k hj*;k Mj* 2 hp;k Mp schematic of energy distribution regime model is shown in Fig. 2.
p
The quality of temperature distribution is measured by aluminum
where, mp is the mass fraction of each species of the product, mr is temperature uniform coefcient, which can be expressed as follows:
the mass fraction of a particular reactant, R is the reactant which

gives the smallest value of Ri,r, K1 and K2 are empirical constants s Dtmax
equal to 4 and 0.5, respectively. The parameters in eqs. (9)e(11) are
q 1 (14)
x L
shown in Table 2, where bi was set as zero for all reactions.
Besides above conservation equations, the liquid aluminum in where, q is aluminum temperature uniform coefcient, s is stan-
the molten pool was modeled by solving the following conduction dard deviation, x is mean value, Dtmax is maximum temperature
equation according to assumption (1) [23]. difference of liquid aluminum, L is equivalent size. Thus, it is
dened as end aluminum temperature uniform coefcient at the
vrh end of production process. In the present study, L is dened as
divrvh divlgrad T (12)
vt

where, l is the thermal conductivity. V


L6 (15)
The turbulence models discussed in the previous section can be A
only applied with accuracy to core regions of the ow. In regions
where, V is volume of molten pool, A is surface area of molten pool.
close to the boundaries such as walls, the models tend to break-
down mainly because there are signicant changes in the level of
turbulence. For this reason, near-wall regions need special atten-
tion. One approach that is used to determine the values of variables
in the near-wall region is the wall function approach. In this
approach, the variables are obtained from semi-empirical formulas
that bridge the wall to the fully turbulent region.
By the principle known casting, the rise of melt temperature
mainly depends on heat transfer between gas and melt, which was
implemented by uid-solid coupled model in the study. According
to the assumption (1), in the current model, the melt was regarded

Table 1
Reactions.

Reaction

1 CH4 3/2O2/CO 2H2O


Fig. 2. Conceptual schematic of energy distribution regime of aluminum holding
2 CO 1/2O2/CO2
furnace.
J.-m. Wang et al. / Energy 72 (2014) 521e535 525

To use computational uid dynamics software FLUENT as a 2.2. Solution and independent test
platform, a CFD compressive process model was developed and
integrated with energy distribution regime model using the FLUENT solver parameters are listed in Table 3. According to the
hybrid programming method of FLUENT UDF and FLUENT working characteristics of the furnace, velocity, mass ow and
Scheme. The ow chart of energy distribution regime is shown in outow boundary conditions were assigned to natural gas inlet, air
Fig. 3. The user-dened model was used to randomly generate inlet and ue gas outlet, respectively. Non-slip and adiabatic
varying heat load according to melt temperature. The solution boundary conditions are specied for the walls. Interface boundary
procedure is shown in Fig. 4, and its explanations are given condition was set for coupled surface between combustion space
below: and molten pool. The initial temperature in the melt zone was set at
Melt temperature and its uniform coefcient, furnace temper- 1013 K. The composition of a typical natural gas can be also found in
ature, and heat load were computed by heat_load_al_temp_func Table 3. In this numerical study, a control volume approach is used
function. The change of the burning capacity was implemented and in which a non-staggered grid arrangement and a sequential pro-
the results were written to the le by fuel_proc procedure. heat_- cedure of the SIMPLE (semi-implicit method for pressure-linked
load_al_temp_func function and fuel_proc procedure exchange equations) algorithms are used to transform the governing differ-
holding information such as alloying status, heat load pulsing co- ential equations into computer solvable, algebraic equations.
efcient, and melt temperature. heat_load_al_temp_func function Interpolation is accomplished via a rst-order power law scheme.
sends the exit signal to fuel_proc procedure so that the program The system of algebraic equations related to the mass conservation
saves the results and exits FLUENT once melt temperature reaches and chemical species conservation were solved by Additive
to 1023 K. Correction Multi-grid. All other conservation equations were solved

Fig. 3. Flow chart of energy distribution regime.


526 J.-m. Wang et al. / Energy 72 (2014) 521e535

Fig. 4. Numerical solution procedure of aluminum holding furnace.

by the TDMA (Tri-diagonal Matrix Algorithm) line by line algo- temperature depends on dominantly radiative heat transfer for
rithm, with block correction for speeding convergence. Since the low gas velocity. However, if the burners are switched from low
conservation equations are non-linear, relaxation factors were re to high re, convection heat transfer can be enhanced between
used. The global convergence criteria adopted was that residual melt and ue gas, which results in the decrease of the aluminum
tolerances all over the grid and time step were set to 109 for the temperature uniform coefcient. During standing period, the
energy and species equations, smaller than 104 for the other burners were turned off. The melt emits the heat to the ambient
equations. by radiation and convection, particularly, conduction heat transfer
According to geometric structure of the furnace, the multi- within melt only occurs. Therefore, the aluminum temperature
block unstructured grid was used in this simulation. Grid size uniform coefcient increases. The average aluminum temperature
selected is based on a compromise between computing time and uniform coefcients were compared under different girds and
accuracy, since a larger number of point results in an unreasonably time steps. Due to the small differences observed on the selected
large computing time. From Fig. 5, ve different grid sizes were quantities (lower than 1% between grid 4 and grid 5), grid 5 was
tested: grid 1 with 200 mm, grid 2 with 250 mm, grid 3 with chosen for the present investigation. Five time step independence
300 mm, grid 4 with 350 mm and grid 5 with 400 mm. Learning tests (10 s, 5 s, 2 s, 1 s and 0.5 s) have been also undertaken.
from steel billet stack cooling process [25], aluminum temperature Likewise, the time step has no inuence on the average aluminum
uniform coefcient was put forward to measure melt temperature temperature uniform coefcient. However, it is a little impact on
uniformity. During heating phase, because the burners always heat total holding time as shown in Fig. 5a. As a result, the optimum
liquid aluminum, the temperature differences between the upper mesh size and time step are 14,158 elements and 1.5 s for the
and the bottom of aluminum increase, which reduces the computation, respectively. The problem was solved in a 3.4 MHz
aluminum temperature uniform coefcient. During holding stage, Intel Core i3-2130 processor with 2 GB RAM. The time to obtain
the aluminum temperature uniform coefcient increases as the convergence solution for a typical case, with 3336 nodes, was
burners begin to re low. This is because the rising of aluminum approximately 10 h.

Table 3
Boundary and initial conditions.

Items Values

Liquid aluminum Initial temperature, K Holding temperature, K Final temperature, K


1013 1073e1078 1023
1
Air (79% N2 21% O2) Mass ow, kg s Temperature, K Hydraulic diameter, mm Turbulence intensity, %
High re 2.369 300 198 3.7
Low re 0.2369 3.1
Fuel (100% CH4) Velocity, m/s Temperature, K Hydraulic diameter, mm Turbulence intensity, %
High re 51.38 300 64 3.58
Low re 5.138 2.86
Wall Inner face Outer face Coupled face
Roof Floor Side
Emissivity 0.8 p.142 in Ref. [24] according to high re or low re 0.33
Convection heat transfer e e
coefcient
Flue Outow
Material physical properties (e.g., thermal conductivity, Appendices III and V in Ref. [24]
specic heat capacity, density)
J.-m. Wang et al. / Energy 72 (2014) 521e535 527

temperature of liquid aluminum and ue gas is 1089 K and 1343 K,


respectively. These values were very close to the average measure
values of 1078 K and 1329 K respectively. The temperature gradu-
ally increases rstly, and then periodically oscillates, nally, de-
creases again. This is attributed to process requirements of
aluminum holding furnaces. Due to heat conduction in melt merely,
the aluminum temperature reduces with the pool depth. The
simulated spatial distribution of furnace temperature is similar to
that of measurement points. However, the calculated results are
little larger than the measured results, which is mainly due to the
opening of the test poles and furnace door. As cold air is absorbed
into the combustion space under the negative pressure, the whole
temperature in the furnace decreases. Besides, the heat loss
through the furnace walls is neglected according to the assumption
(3), which causes the simulation results to increase. What's more,
the mathematical model simplies holding process of aluminum
holding furnaces such as fuel composition and reaction mechanism.
However, the computational model was proven to be reliable and
accurate. The results show that the basic validity of the numerical
model was conrmed. It is also indicated that the thermal phe-
nomenon and process optimization of aluminum holding furnaces
may be revealed thoroughly by the model.

3. Process optimization of aluminum holding furnace

Combined with specialty knowledge and production experience


of aluminum casting industry, as well as the brainstorm method,
cause and effect diagram was constructed to identify process pa-
rameters that may inuence the performance of aluminum holding
furnaces as shown in Fig. 7. Based on the diagram, ue position,
oxidant type, air preheated temperature, natural gas velocity,
burner height, vertical burner angle, pool depth, horizontal spacing
between burners, height-radius ratio, airefuel ratio, and burner
load ratio were selected to investigate their effects on the perfor-
Fig. 5. Independence test of grid number and time step. mance of aluminum holding furnaces.

3.1. Uniform design array and experiment results


2.3. Numerical model validation
The experimental design used in this study was similar to those
In order to prove the credibility of the developed mathematical used for screening studied qualitative and quantitative variables in
model, a comparison of the numerical predictions and experi- melting process of aluminum melting furnaces [13]. In this work,
mental data was conducted. The layout of measurement points was among the eleven independent variables, ue position and oxidant
presented in Fig. 1. By means of multi-point test, measurement data type were considered as categorical factors with two levels. The
were veried from different perspectives, as shown in Fig. 6 and others are indicator factors with ve levels. All of them were
Table 4 (high re). The measured positions of furnace temperature selected to investigate the inuence and interaction among the
with 1000 mm horizontal distances are 900 mm away from the factors. End aluminum temperature uniform coefcient and energy
melt surface. The locations of the test points of aluminum tem- consumption per ton of aluminum were selected as dependent
perature with 150 mm vertical spacing lie on 500 mm away from variables in order to represent the performance of aluminum
slag-off doors, and 100 mm away from the melt surface. The tem- holding furnaces. Even for moderate levels and factors, it is very
peratures were measured by means of thermocouple NiCreNi and expensive to search for an optimal uniform design. There are
the major gas-phase species concentrations were detected by ue several approaches to construct uniform design, that is, good lattice
gas analyzer from Testo in one point at the end of the furnace sets, cyclic Latin squares and orthogonal designs. KoksmaeHlawka
chamber where the waste gases enter the ue. Combustion product inequality is the basis of uniform design. Thus, obtaining experi-
samples were taken using a stainless steel probe. The furnace ment points that are most uniformly scattered in the domain is the
pressure is usually at heart level. Considering the features of pro- key step in uniform design. Discrepancy is a measure of uniformity
duction process of aluminum holding furnaces, average tempera- that is universally accepted. In this paper, a direct approach is to use
tures of various test points at three representative times of every a global optimization algorithm based on threshold accepting. The
stage were obtained using multiple measurements. It is seen from uniform table U20 (22  95) for process optimization of aluminum
Fig. 6c that frequency distribution of furnace temperature of test holding furnaces is obtained by using good lattice point method.
points in the chamber is similar to that of simulation results. The Hickernell method [26] was used to calculate centered discrepancy,
change trend of test temperature is in accordance with that of here, it is 0.6479. Based on the analysis of preliminary experimental
simulation temperature. It can be clearly seen that a good quali- tests, the optimal region of every factor has been determined.
tative agreement is obtained between the present numerical re- Therefore, the range and levels of each factor are listed in Table 5.
sults and the experimental data, and the relative error is within According to U20 (22  95) table in Pref. [17], twenty typical cases
10%. During holding stage, the simulation results show: the average were simulated and the results are listed in Table 6.
528 J.-m. Wang et al. / Energy 72 (2014) 521e535

Table 4
Comparison of simulation results and test values for aluminum holding furnace.

Item Measurement value Simulation result Relative error/%

Holding time/h 2.5 2.37 5.49


CO2/% 4.22 4.66 9.44
H2O/% 4.91 5.12 4.1
N2/% 74.66 77.01 3.05
O2/% 13.84 14.15 2.19
Pressure/Pa 14 16.51 15.2

3.2. Stepwise non-linear regression

RSM is a generic means of optimization. RSM was introduced in


this study to optimize the technique of aluminum holding furnaces.
In the optimization process the response was related to the selected
variables by a second-order polynomial model. The generalized
model proposed for the response is given in Eq. (16) below [12]:

X
j X
k i<j X
X
Y b0 bi X i bii Xi2 bij Xi Xj (16)
i1 i1 i j

where, Y is the response variable to be modeled, Xi and Xj are the


independent variables which inuence Y, b0, bi, bii and bij are the
offset terms, the ith linear coefcient, the quadratic coefcient and
the ijth interaction coefcient, respectively.
End aluminum temperature uniform coefcient is crucial to
improve molten aluminum quality. In statistics, stepwise regres-
sion includes regression models in which the choice of predictive
variables is carried out by an automatic procedure [27]. Usually, this
takes the form of a sequence of F-tests. A variable is considered to
be eligible for removal if its Fout is less than the F value for variable
removal, and vice versa. At each stage in the process, after a new
variable is added, a test is made to check if some variables can be
deleted without appreciably increasing the RSS (residual sum of
squares). The procedure terminates when the measure is (locally)
maximized. The following regression model is recommended by
stepwise regression method with Fin Fout 2,

C 3:46 0:015D  0:014O1U 0:0035O1D  0:016P1L


0:038O1L  0:0013AL
(17)
With Adj-R2 99.5%. R2 is a statistic that will give some infor-
mation about the goodness of t of a model. Adj-R2 is a modica-
tion of R2 that adjusts for the number of terms in a model. R2 always
increases when a new term is added to a model, but Adj-R2 in-
creases only if the new term improves the model more than would
be expected by chance. The closer the value Adj-R2 is to unity, the
better the empirical model ts the actual data. Adj-R2 was calcu-
lated as 99.5% for Eq. (17), indicating good agreement between the
experimental and the predicted values. The ANOVA (analysis of
variance) results show that the second-order polynomial model is
signicant and tted experimental results for the end aluminum
temperature uniform coefcient well. The items of P, O, T, U, H, A, B,
R, n, and L are not signicant in stepwise regression and was
eliminated. This indicates that the effects of them on the end
aluminum temperature uniform coefcient are of least signicance.
It is known from Table 7 that cross correlation is exist amongst
factors. Although the effect of P, O, U, A, and L on the end aluminum
temperature uniform coefcient is not signicant, cross-

Fig. 6. Comparison results of CFD simulation and eld tests of aluminum holding
furnace.
J.-m. Wang et al. / Energy 72 (2014) 521e535 529

Fig. 7. Cause and effect diagram of recycling non-ferrous metal reverb furnaces.

correlations between them are signicant. Therefore, the effect of temperature coefcient. In another article, Regression analysis and
each factor on experiment indices must be taken into account thermal characteristics of an aluminum holding furnace based on
during experiment. partial least squares method, we have covered what would happen
The following equation to describe energy consumption per ton when the size of the furnace is changed in more detail.
of aluminum was obtained using stepwise regression method with
the results of energy consumption from UD experiments: 3.3. Analysis of interaction effects

E 0:0180:00087O1A0:000000043T 2 0:038n2 (18) In order to visualize the relationship between the response and
experimental levels of each factor and to deduce the optimum
The mathematical model was expressed in terms of the values of conditions, the tted polynomial equation was expressed as sur-
all the independent variables and by neglecting the statistically face, contour and interaction plots. With the end uniform coef-
insignicant terms. The adjusted correlation coefcient of Adj- cient as the response, the contour and interaction plots of the
R2 84.32% indicated a close agreement between the experimental quadratic model with one variable kept at the optimum level and
and predicted values of the energy consumption per ton of the other two varying within the experimental ranges are shown in
aluminum, and the predicted model seemed to reasonably repre- Fig. 8. It is concluded from Table 7 that the mutual inuence be-
sent the observed values. Thus, the response was sufciently tween the vertical burner angle and burner load ratio is very sig-
explained by the model. The signicance of each coefcient was nicant. The interaction effect of the vertical burner angle and
determined using Student t-test and p-value in Table 8. The cor- burner load ratio (A*L) indicated there is the cooperation between
responding variables would be more signicant if the absolute t- this two factors. The vertical burner angle also demonstrated a
value becomes larger and p-value becomes smaller. It can be seen linear reduce when the burner load ratio is within the range of 1:1
that the variable with the largest effect is the interaction effects of to 1:12, which accords with the results in Table 7, and then small
oxidant type and vertical burner angle (O*A), followed by the vertical burner angle and low burner load ratio could bring low end
square terms of air preheated temperature (T), and airefuel ratio uniform coefcient. With increasing the burner load ratio, the end
(n). uniform coefcient decreases over a small range. This indicates that
It can be found from Eqs. (17) and (18) that the structure pa- the burner load ratio has a negative effect on the end uniform co-
rameters of the furnace such as burner height (H), vertical burner efcient. When the burner load ratio is higher, the heat load dif-
angle (A), horizontal spacing between burners (B), and ference between burners becomes greater. The strength level of
heighteradius ratio (R) have very little inuence on response var- heat transfer between melt and ue gas would be changed, which
iables, yet pool depth (D) shows signicant effect on end aluminum leads to be unequal heat mode. The direct result would be the slow-
down of the end uniform coefcient. As the vertical burner angle
elevates, mixing space of fuel and oxidant becomes small. Thus, the
Table 5
levels of variable simulation in the U20(22  95) uniform design. heat exchange capacity of high-temperature gases reduces. The end
uniform coefcient drops.
Variables Levels
The interaction effect of the natural gas velocity and oxidant
1 2 3 4 5 type on the end uniform coefcient is shown Fig. 9a. With the in-
Flue position (P) P1 (opposite P2 (same crease of the natural gas velocity, on the one hand, physical heat of
side) side) fuel enhances, which increases the furnace temperature. On the
Oxidant type (O) O1 (air) O2 (pure other hand, turbulent intensity in the furnace increases, and the
oxygen)
Air preheated temperature 300 400 500 600 700
temperature distribution tends to be even, which increases the RSD
(T), K (relative standard deviation) of furnace temperature. As a result,
Natural gas velocity (U), m/s 86.76 90.76 94.76 98.76 102.76 the end uniform coefcient is mainly affected by the furnace
Burner height (H), mm 650 900 1150 1450 1650 temperature and RSD of furnace temperature. When pure oxygen is
Vertical burner angle (A),  5 15 25 35 45
adopted as oxidant, oxygen-enriched atmospheres can cause nat-
Pool depth (D), mm 400 450 500 550 600
Horizontal spacing 1250 1750 2250 2750 3250 ural gas to burn more violently. This increases the furnace tem-
between burners (B), mm perature. The molten aluminum temperature rises quickly. The
Heighteradius ratio, R 0.5 0.57 0.61 0.64 0.68 heating time shortens substantially. The furnace temperature
Airefuel ratio, n 1.05 1.09 1.13 1.17 1.21 signicantly affects the end uniform coefcient, which cuts down
Burner load ratio, L 1:1 1:2.5 1:5 1:7.5 1:10
the end uniform coefcient accordingly. On the contrary, when air
530 J.-m. Wang et al. / Energy 72 (2014) 521e535

Table 6
UD and response results.

Runs Factors Responses

P O T U H A D B R n L End aluminum temperature Total energy consumption


uniform coefcient (C), % (E), kg ce/t-Al

N1 1 2 500 94.76 1400 35 400 1250 0.50 1.13 1:10 8.74 0.01675
N2 2 1 500 86.76 1650 45 500 1750 0.57 1.21 1:5 11.22 0.06694
N3 2 2 700 98.76 900 45 450 2250 0.68 1.17 1:7.5 9.63 0.01995
N4 1 2 300 102.76 1400 45 500 3250 0.64 1.09 1:5 10.37 0.02032
N5 1 2 600 90.76 1650 25 450 3250 0.57 1.17 1:1 10.05 0.01463
N6 2 1 300 90.76 1400 5 450 2250 0.50 1.05 1:5 10.35 0.03826
N7 2 1 400 94.76 1650 35 550 3250 0.61 1.05 1:10 12.09 0.04301
N8 2 2 600 102.76 1150 35 600 2250 0.50 1.09 1:1 12.19 0.01413
N9 1 2 400 98.76 650 5 400 2250 0.61 1.21 1:5 9.2 0.03949
N10 2 1 300 102.76 1150 25 450 1250 0.64 1.21 1:2.5 10.26 0.04575
N11 2 2 700 94.76 1400 5 550 2750 0.61 1.21 1:2.5 11.61 0.01244
N12 1 1 400 90.76 900 45 600 1750 0.61 1.17 1:1 13.01 0.07793
N13 1 2 400 86.76 1150 15 550 1750 0.68 1.05 1:2.5 11.56 0.01285
N14 1 1 700 98.76 650 25 500 1250 0.57 1.05 1:2.5 11.15 0.02398
N15 2 1 500 94.76 650 15 400 3250 0.68 1.13 1:1 9.39 0.03202
N16 1 1 500 102.76 900 15 550 2750 0.50 1.17 1:10 12.07 0.03319
N17 1 1 700 86.76 1150 35 400 2750 0.64 1.09 1:7.5 9.23 0.03072
N18 1 1 600 98.76 1650 5 600 1750 0.68 1.13 1:7.5 13.01 0.02525
N19 2 2 600 90.76 900 15 500 1250 0.64 1.09 1:10 10.57 0.01240
N20 2 2 300 86.76 650 35 600 2750 0.57 1.13 1:7.5 11.85 0.01642

is used as an oxidant, the RSD of furnace temperature dominantly It appears from Fig. 9c that increasing the burner load ratio
affects heat transfer between molten aluminum and high- causes a slight decrease in the end uniform coefcient. As
temperature gases. It is in favor of the improvement of the end mentioned analysis above, if air is to be replaced by pure oxygen,
uniform coefcient with the natural gas velocity. the end uniform coefcient would be increased.
The mutual inuence between the pool depth and oxidant type From Fig. 9d, the effects of the two variables, namely, the
on the end uniform coefcient is depicted in Fig. 9b. Whether the interaction effect of the ue position and burner load ratio on the
oxygen oxidant is air or pure oxygen, the end uniform coefcient end uniform coefcient shows linear interaction effect of P*L. It is
increases with increasing the pool depth. As increasing the pool indicated that great end uniform coefcient could be obtained
depth, molten aluminum amount increases. It need much more when the ue and burners locate at the same side or low burner
heat to be alloying treatment. Thus, the heating time would be load ratio. While the ue is on the same side of the burners, the
prolonged. In addition, the vertical distance between the burners residence time of high-temperature gases in the furnace is pro-
and melt surface reduces. This makes velocity distribution above longed. It enhances heat transfer between combustion gases. As a
melt surface more even. All these increase the end uniform consequence, the same side is appropriate for the increases of the
coefcient. end uniform coefcient.
When the energy consumption per ton of aluminum was
Table 7
selected as the response, two variables within experimental range
ANOVA for end aluminum temperature uniform coefcient. were depicted in one three-dimensional surface and interaction
plots while the two other variable kept constant at the middle level.
Source Sum of Degrees of Coefcient Standard Mean p-Value
squares freedom error square
Fig. 10a gives the mutual effect of the oxidant type and vertical

Constant 3.4648
D 9.7337 1 0.0147 0.2563 9.7337 0.0001
O*U 0.1654 1 0.0141 0.3237 0.1654 0.0006
O*D 0.2989 1 0.0035 0.3443 0.2989 0.0001
P*L 0.0449 1 0.0160 0.0685 0.0449 0.0365
O*L 0.0871 1 0.0380 0.1169 0.0871 0.0064
A*L 0.3399 1 0.0013 0.0917 0.3399 0.0001

Error 0.1073 13 0.0083


Total 31.4122 19

Table 8
ANOVA for energy consumption per ton of aluminum.

Source Sum of degrees Coefcient Standard Mean p-Value


squares of freedom error square

Constant 0.0180
O*A 0.0039 1 0.0009 0.0045 0.0039 0.0001
T2 0.0007 1 0.0000 0.0055 0.0007 0.0017
n2 0.0005 1 0.0378 0.0185 0.0005 0.0085

Error 0.0008 16 0.0001


Fig. 8. Contour plot of vertical burner angle and burner load ratio for end uniform
Total 0.0063 19
coefcient.
J.-m. Wang et al. / Energy 72 (2014) 521e535 531

Fig. 9. Interaction plots between important factors for end uniform coefcient.

burner angle on the energy consumption, and the interaction response surface, in which the optimal condition is exactly located
shows the complex interaction effect of O*A. When air is used as an inside the design boundary. With the increase of the airefuel ratio,
oxidant, mixing space of high-temperature gases enlarges with high-temperature gases increase, which strengthens heat transfer
increasing the vertical burner angle. Heat exchange between between combustion gases and aluminum loads, hence, the heating
combustion gases is improved. The energy consumption reduces time reduces. If the airefuel ratio is too large, excess cold air need
due to a short heating time. When pure oxygen is adopted as an absorb heat. What's more, these high velocities reduce the reten-
oxidant, the energy consumption gradually increases with the tion time of the hot gases in the furnace and thus reduce the
vertical burner angle. Radiation heat transfer is favorable for heat probability of these gases radiating their energy to the loads. As a
supply of molten aluminum. Radiant path length increases with the result, the heating time increases again. In short, the energy con-
increase of the vertical burner angle, which would weaken radia- sumption per ton of aluminum rstly decreases with the airefuel
tion heat transfer between ue gas and aluminum loads. Never- ratio, then increases. Similar results were obtained by Wang et al.
theless, it is concluded that oxy-fuel red burners reduces the [28].
energy consumption instead of air-red burners when the vertical
burner angle is below 40 , but increases above 40 . 3.4. Multi-objective optimization
Fig. 10b shows the variability of the energy consumption is
trivial with change of the air preheated temperature. This behavior End aluminum temperature uniform coefcient and energy
can be attributed to energy and physical properties of preheated air, consumption per ton of aluminum are two individual responses.
such as radiant ability. When the air preheated temperature is high, Thus, a compromise among the conditions for the two responses is
more energy would be added into the furnace. This would reduce desirable. The desirability function approach was used to achieve
energy consumption, yet maybe lead to insufcient emissivity. such a goal. The goals may apply to either factors or responses. The
From Fig. 11, the airefuel ratio has a great simulative effect on goals of factors, i.e., P, O, T, U, A, D, n and L are within the certain
the energy consumption. It is evidenced by the obvious peak in the range, and other factors are set to the exact value. The goal of end
532 J.-m. Wang et al. / Energy 72 (2014) 521e535

Fig. 11. Response surface of airefuel ratio and burner load ratio for energy
consumption.

most important, you could change it to 5 pluses (). In the


present work, the weight and importance of all goals are set to be
default. For several responses and factors, all goals get combined
into an overall desirability function.
In desirability based approach, different best solutions are
obtained. The solution with high desirability is preferred.
Maximum desirability of 1.0 was as follows: ue position of same
side, oxidant type of pure oxygen, air preheated temperature of
719 K, natural gas velocity of 108.88 m/s, burner height of
1150 mm, vertical burner angle of 5 , pool depth of 625 mm,
horizontal spacing between burners of 2050 mm, heighteradius
ratio of 0.6, airefuel ratio of 1.04, and burner load ratio of 1,
respectively. The corresponding end aluminum temperature
uniform coefcient and energy consumption per ton of
aluminum were 12.76% and 0.01017 kg ce/t-Al, respectively. A
conrmation experiment under the compromised conditions
was carried out. The end aluminum temperature uniform coef-
cient, and energy consumption per ton of aluminum were ob-
tained as 12.85% and 0.01092 kg ce/t-Al, respectively. These
results were in good agreement with the model predictions.
Compared with the initial condition as shown in Table 9, the end
aluminum temperature uniform coefcient and energy con-
sumption per ton of aluminum were improved 24.15%, and
72.81% respectively. For the initial case, during heating phase and
holding phase, the amount of heat supplied is 2.64e7 kJ and
2.24e7 kJ, respectively. In the case of the optimal case, it is
Fig. 10. Interaction plots between important factors for energy consumption.
9.43e6 kJ and 2.24e7 kJ respectively. Therefore, it is critical that
energy saving measures were adopted during heating process.
aluminum temperature uniform coefcient and energy consump- Judging from Fig. 12, it is quite clear that the heating time under
tion per ton of aluminum is maximized and minimized, respec- optimum condition is drastically shortened due to oxy-fuel com-
tively. A minimum and a maximum level must be provided for each bustion. Thus, the energy consumption per ton of aluminum also
parameter included in the optimization. The numerical optimiza- reduces obviously. During heating phase, the aluminum tempera-
tion nds a point that maximizes the desirability function. The ture uniform coefcient rstly drops, then increases and decreases
characteristics of a goal may be altered by adjusting the weight or alternately during holding period, nally, linearly rises during
importance. A weight can be assigned to each goal to adjust the standing stage. However, during heating phase, the statistic com-
shape of its particular desirability function. The default value of one parison before and after optimization shows that the aluminum
creates a linear ramp function between the low value and the goal temperature uniform coefcient is evidently increased, which
or the high value and the goal. Increased weight (up to 10) moves would increase melt quality. Under optimum parameters, the
the result towards the goal. Reduced weight (down to 0.1) creates aluminum temperature uniform coefcient rises quickly during
the opposite effect. The importance of each goal can be changed in holding and standing process. Therefore, the end aluminum tem-
relation to the other goals. The default is for all goals to be equally perature uniform coefcient of optimum condition is far greater
important at a setting of 3 pluses (). If you want one goal to be than that of initial condition.
J.-m. Wang et al. / Energy 72 (2014) 521e535 533

Table 9
Comparison of initial results and optimum results.

Initial condition

P O T U H A D B R n L
P1 O2 300 102.76 900 15 450 2250 0.5 1.05 1:1

End uniform coefcient 10.32%


Energy consumption per ton of aluminum 0.04016

Optimum condition

P O T U H A D B R n L
P2 O2 719 108.88 1150 5 625 2050 0.6 1.04 1:1

End uniform coefcient Simulation 12.85% Predicted 12.76% Relative 0.7%


Energy consumption per ton of aluminum Values 0.01092 Values 0.01017 Errors 6.87%
Improved performance End uniform coefcient Energy consumption per ton of aluminum
24.15%[ 72.81%Y

4. Conclusions characterization of the CFDeRSMeUD integrated intelligence ideas


in process optimization and thermal behaviors of aluminum
Based on CFD process models, a three-dimensional numerical holding furnaces. The main conclusions of this study are summa-
model of aluminum holding furnaces was established and imple- rized as follows:
mented using multi-language programming technology. The reli-
ability of the mathematical model was experimentally veried (1) The results of the conrmation experiment agreed closely
from multi-point test and different perspectives. This paper has with predictions. This demonstrated that such a combination
presented an application of the parameter design and inherent of the CFD, RSM and UD is a powerful and useful approach for
process optimization of aluminum holding furnaces with the
least number of experiments and a short period of time.
(2) Pool depth has important inuence on aluminum tempera-
ture uniform coefcient. Cross-correlations between ue
position, oxidant type, fuel velocity and burner load ratio are
signicant. Oxidant type and vertical burner angle are the
largest interaction effect on energy consumption per ton of
aluminum, followed by the square terms of air preheated
temperature, and airefuel ratio.
(3) With increasing burner load ratio or vertical burner angle,
end uniform coefcient decreases. Air works as an oxidant or
burners and ue locate on the same side, leading to the in-
crease of end uniform coefcient. As pool depth increases,
end uniform coefcient increases. End uniform coefcient
depends on the interaction effect between furnace temper-
ature and its relative standard deviation with the increase of
fuel velocity. Air preheated temperature has a trivial effect on
energy consumption per ton of aluminum. Energy con-
sumption per ton of aluminum increases and decreases with
vertical burner angle respectively when oxidizer is adopted
as pure oxygen and air. Energy consumption per ton of
aluminum rstly decreases with airefuel ratio, then
increases.
(4) Both aluminum temperature and furnace temperature
gradually increase rstly, and then periodically oscillate,
nally, decrease again. During heating phase, aluminum
temperature uniform coefcient reduces. During holding
stage, aluminum temperature uniform coefcient increases
and decreases alternately. During standing period, aluminum
temperature uniform coefcient increases.
(5) By desirability function approach, an optimal condition of
same side, pure oxygen, air preheated temperature 719 K,
natural gas velocity 108.88 m/s, burner height 1150 mm,
vertical burner angle 5 , pool depth 625 mm, horizontal
spacing between burners 2050 mm, heighteradius ratio 0.6,
airefuel ratio 1.04, and burner load ratio 1:1 was obtained
form the compromise of the two desirable responses. In this
condition, the simulation results of end aluminum temper-
Fig. 12. Aluminum temperature uniform coefcient and energy consumption per ton
of aluminum for before optimization and after optimization. ature uniform coefcient and energy consumption per ton of
534 J.-m. Wang et al. / Energy 72 (2014) 521e535

aluminum were 12.85% and 0.01092 kg ce/t-al, respectively, mef effective viscosity (Pa s)
which was reasonable consist with 12.76% and 0.01017 kg ce/ mt turbulent viscosity (Pa s)
t-al of the predicted values. r density (kg m3)
s StefaneBoltzmann constant (W m2 K4), standard
Acknowledgments deviation (K)
t time (s)
The project is nancially supported by National Nature Foun- uk reaction rate of the kth species (kmol m3 s1)
dation Committee of P. R. China (51306001 and 51206001) and
Natural Science Foundation of Anhui Province (1408085QG138). Abbreviations
We also wish to thank School Nature Foundation of Anhui Uni- ANOVA analysis of variance
versity of Technology (QZ201303 and QS201304) and Student ACM Additive Correction Multi-grid
Research Training Program of Anhui University of Technology ce coal equivalent
(AH201310360120) for the partial support of this study. CFD computational uid dynamics
DTRM discrete transfer radiation model
Nomenclature RSD relative standard deviation
RSM response surface methodology
A surface area of molten pool (m2) RSS residual sum of squares
Ai pre-exponential factor (m3 kmol1 s1) RTE radiative transfer equation
b0 offset term SIMPLE semi-implicit method for pressure-linked equations
bi ith linear coefcient TDMA Tri-diagonal Matrix Algorithm
bii quadratic coefcient UD uniform design
bij ijth interaction coefcient UDF user-dened function
Ei activation energy (MJ kmol1) WSGGM weighted sum of gray gases model
h total enthalpy (J kg1)
h convective heat transfer coefcient (W m2 K1) References
I radiant intensity (W m2 K4)
K1, K2 empirical constants [1] Belt C. Energy efciency tests in aluminum combination melting and holding
furnaces. TMS annual meeting. Charlotte: USA; 2004. pp. 613e7.
L equivalent size (m)
[2] Migchielsen J, Grab HW, Schmidt T. Design considerations for holding and
Mi molecular weight of ith species casting furnaces. TMS annual meeting. New Orleans: USA; 2008. pp. 593e6.
mi mass fraction of the ith species [3] Zhang JQ, Zhou NJ, Zhou SH. Combustion and energy balance of aluminum
mp mass fraction of each species of the product holding furnace with bottom porous brick purging system. J Cent South Univ
Technol (English Edition) 2012;19:200e5.
mr mass fraction of a particular reactant [4] Zhou SH. Numerical simulation of ow eld and temperature eld in the
P modied total pressure process of gas stirring in aluminum holding furnace. Light Metal 2011;12:
p pressure (N m2) 256e60.
[5] Zhang JQ, Zhou NJ, Zhou SH. Numerical simulation on melt ow with bubble
Pr Prandtl number stirring temperature eld in aluminum holding furnace. J Cent South Univ
Prt turbulent Prandtl number Technol (English Edition) 2011;18:1726e32.
Prk Prandtl number of turbulence kinetic energy [6] Zhou NJ, Zhou SH, Zhang JQ, Pan QL. Numerical simulation of aluminum
holding furnace with uid-solid coupled heat transfer. J Cent South Univ
Pr Prandtl number of turbulent dissipation rate Technol (English Edition) 2010;17:1389e94.
Rc universal gas constant (kJ mol1 K1) [7] Zhou NJ, Zhou SH. Numerical simulation of combustion and uid-solid
Ri,r reaction rate of the ith species (kmol m3 s1) coupled heat transfer on the aluminum reverb holding furnace. Ind Heat
Sh source term (w m3) 2010;39:9e12.
[8] Zhou SH. Numerical simulation and optimization on coupled multi-eld in a
Sc Schmidt number tilting-type aluminum holding furnace. Changsha, China: Central South Uni-
Sct turbulent Schmidt number versity; 2010.
[9] Ma HP. Numerical simulation and operating parameters optimum research on
s radiation path length (m)
aluminum alloy melting furnace and holding furnace. Changsha, China: Cen-
T temperature (K) tral South University; 2011.
Tf uid temperature (K) [10] Golchert BM, Metwally H, Kumar A. Computational analysis of the conversion
v velocity vector (m/s) of a generic aluminum holding furnace from air-red to oxy-red burners.
TMS annual meeting. Warrendale: USA; 2006. pp. 733e6.
Xi, Xj molar concentration (kmol m3), independent variables [11] Pauty E, Laboudigue B, Etay J. Numerical simulation of the ow and the solid
x mean value (various unit) transport when tilting a holding furnace. Metal Mater Trans B Process Metal
Y response variable (various unit) Mater Process Sci 2000;31:207e14.
[12] Myers RH, Montgomery DC, Anderson-Cook CM. Response surface method-
V volume of molten pool (m3) ology: process and product optimization using designed experiments.
Hoboken: John Wiley & Sons, Inc.; 2009.
Greek symbols [13] Zhou JM, Wang JM, Yan HJ, Li SX, Gui GC. Multiple-response optimization for
melting process of aluminum melting furnace based on response surface
a absorption coefcient methodology with desirability function. J Cent South Univ Technol (English
a,i emissivity weighting factors of the ith ctitious gray gas Edition) 2012;19:2875e85.
bi temperature exponent [14] Natarajan U, Periyanan PR, Yang SH. Multiple-response optimization for
micro-end milling process using response surface methodology. Int J Adv
gi,k concentration exponent of the ith species Manuf Technol 2011;56:177e85.
Dtmax maximum temperature difference of liquid aluminum (K) [15] Korbahti BK, Aktas N, Tanyola A. Optimization of electrochemical treatment
turbulent dissipation rate (m2 s3), solid emissivity of industrial paint wastewater with response surface methodology. J Hazard
hi stoichiometric coefcient of ith species Mater 2007;148:83e90.
[16] Abdullah A, John F, David B. The effects of using biodiesel on CI (compression
q aluminum temperature uniform coefcient ignition) engine and optimization of its production by using response surface
k turbulent kinetic energy (m2 s2) methodology. Energy 2013;59:56e62.
ki absorption coefcient of the ith gray gas [17] Pham H. Handbook of engineering statistics. Berlin: Springer-Verlag London
Limited; 2006.
l thermal conductivity (W m1 K1) [18] Wang JP, Chen YZ, Wang Y, Yuan SJ, Yu HQ. Optimization of the coagulation-
m absolute viscosity (Pa s) occulation process for pulp mill wastewater treatment using a combination
J.-m. Wang et al. / Energy 72 (2014) 521e535 535

of uniform design and response surface methodology. Water Res 2011;45: [23] Fluent Inc. Fluent 6.3 user's guide. Pennsylvania: ANSYS Inc.; 2006.
5633e40. [24] Mei C. Handbook of non-ferrous metallurgical furnaces. Beijing: Metallurgical
[19] Wang ZJ, Luo DH, Ena C. Optimization of polysaccharides extraction from Industry Press; 2000.
Gynostemma pentaphyllum Makino using uniform design. Carbohydr Polym [25] Sun XJ. Numerical simulation and optimization of steel billet stack cooling
2007;69:311e7. process. Changsha, China: Central South University; 2010.
[20] Leung LSY, Chan WH, Luk CH. Optimization of fabrication conditions of high- [26] Hickernell FJ. A generalized discrepancy and quadrature error bound. Math
efciency ultraltration membranes using methods of uniform design and Comput 1998;67:299e322.
regression analysis. Chemom Intell Lab Syst 2000;53:21e35. [27] Draper N, Smith H. Applied regression analysis. New York: John Wiley & Sons,
[21] Li JF, Liao H, Normand B, Cordier C, Maurin G, Foct J, Coddet C. Uniform design Inc.; 1981.
method for optimization of process parameters of plasma sprayed TiN coat- [28] Wang JM, Yan HJ, Zhou JM, Li SX, Gui GC. Optimization of parameters for an
ings. Surf Coatings Technol 2003;176:1e13. aluminum melting furnace using the Taguchi approach. Appl Therm Eng
[22] Liu RS, Tang YJ. Quantitative analysis for the effect of plant oil and fatty acid 2012;33e34:33e43.
on Tuber melanosporum culture by uniform design combined with partial least
squares regression. Appl Microbiol Biotechnol 2010;87:1689e97.

You might also like