CFD Study on the Heavy Oil Lance Positioning in the Blast Furnace Tuyere to Improve Combustion

Transcription

1 ISIJ International, Vol. 57 (017), ISIJ International, No. 11 Vol. 57 (017), No. 11, pp CFD Study on the Heavy Oil Lance Positioning in the Blast Furnace uyere to Improve Combustion Ari VUOKILA, 1) * Olli MAILA, ) Riitta Liisa KEISKI 1) and Esa MUURINEN 1) 1) University of Oulu, Pentti Kaiteran katu 1, Oulu, Finland. ) SSAB Europe, Rautaruukintie 155, P.O.Box 93, 9101 Raahe, Finland. (Received on April 18, 017; accepted on July 4, 017; J-SAGE Advance published date: October 3, 017) Auxiliary fuels are used to replace expensive coke in two ways, as a reducing agent for iron oxides and to provide energy for the blast furnace operation. Poor combustion of heavy oil produces soot and cenosphere, which plug the coke bed and lead to an increased pressure drop and lower the productivity of blast furnace. In the study a new combustion model for heavy oil in a blast furnace is explained. Droplet atomization is modelled with the wave breakup model and the combustion consists of droplet vaporization and gas phase combustion with a detailed combustion model. hree different lance positions were modelled to evaluate the effect of lance position on combustion and blast furnace operation. he study is limited to the tuyere-raceway area, where e.g. mixing between air blast and heavy oil, pressure drop and combustion have been investigated. he most promising position based on this study is to move the injection lance 10 cm downstream from the tuyere nose. he results show that the pressure drop increases when the lance is moved inside the tuyere, but the pressure drop growth is moderate until the lance is at least 10 cm downstream from the tuyere nose. According to the results the pressure drop versus lance position behavior is similar to an experimental study previously described in the literature. In addition the combustion behavior was found to be similar to the actual blast furnace. KEY WORDS: blast furnace; heavy oil; CFD; combustion; raceway. * Corresponding author: ari.vuokila@oulu.fi DOI: 1. Introduction Blast furnace is the most commonly used technology to produce pig iron. Auxiliary fuels (pulverized coal (PC), natural gas, heavy oil, tar, etc.) are injected into the blast furnace to reduce consumption of expensive coke, to decrease carbon dioxide emissions (higher H/C ratio), to stabilize the process and to increase productivity. Auxiliary fuels can replace coke in two ways, first as a reducing agent for iron oxides and second to provide energy for the blast furnace operation. 1) Extra heavy fuel oil combustion in high injection levels (over 10 kg/thm) tends to be incomplete and it produces small char particles (soot and cenosphere) in the raceway area, which plug the porous coke bed and reduce the gas flow through the blast furnace.,3) Effective combustion and increased oil injection rates require small droplets and good mixing as well as maximized residence time in the tuyere-raceway area. herefore, lance positioning is an important factor in the injection system design. Experimental measurements in the blast furnace tuyere-raceway area are extremely difficult due to the harsh environment (high temperature, increased pressure as well as solid and molten materials), which leads to the necessity of alternative research methods. 4) Computational Fluid Dynamics (CFD) modelling is used to find out the effect of lance positioning on the extra heavy oil combustion. he literature highlights some previous public studies considering heavy oil injection in blast furnace. 5 10) Droplet atomization and size distributions have been studied with lab scale experimental models. 7 9) Jordan et al. 6 8) used a small-scale test rig and a particle image velocimetry, and high speed imaging to investigate heavy oil atomization to improve their CFD modelling boundary conditions and to validate CFD results. Hakala & Paloposki 9) studied heavy oil atomization with a small-scale test rig, which was based on the SSAB blast furnace in Raahe. he small-scale system was developed using dimensional analysis to ensure similarity between experiments and the actual blast furnace conditions. he results were used to develop an equation for the droplet volume mean diameter based on the correlations between oil properties and operation parameters. Previously, the heavy oil combustion modelling has been made with equilibrium reaction modelling by Andahazy et al. 5) hey were comparing coke oven gas combustion with heavy oil. Liao s 10) thesis investigated raceway phenomena and oil injection in Raahe s blast furnace. In the study he found a good method to calculate the raceway size and shape, which matched well with the industrial results. He demonstrates problems encountered with high oil injection rates and some methods to improve combustion in the raceway. hesis also presents ISIJ

2 a combustion model for heavy oil injection. Most of the blast furnace studies are involving coal combustion and typically these models use only diffusion limited combustion models like the Eddy Dissipation Model (EDM) or the Eddy Break-Up (EBU) model ) hese kinds of models do not apply very well in blast furnaces because the temperature is even at around K and the amount of oxygen can be from 0.5 to 0.8 of the stoichiometric ratio, which means that gasification and dissociation reactions are important. 1) Ariyama et al. 15) have done experimental work on the pulverized coal injection (PCI) systems considering the single lance and double lance arrangements. hey found out that the lance arrangement influenced combustion efficiency since a double lance enhanced dispersion of coal particles and local O deficiency was relatively reduced. Mostly lance arrangement studies with CFD are only considered for PC injection systems or combined PC methane systems. 16 0) Chen 16) modelled three different lance arrangements for PCI without chemical reactions or turbulent dispersion. wo lance configurations and cooling gas types were studied by Yeh et al. 17) Devolatilization of coal is modelled with a model having two competing rates, gaseous combustion was modelled with a partial equilibrium combustion model and char combustion was modelled with a kinetic/diffusion limited rate model. Silaen et al. 18) used CFD to investigate three different lance designs for natural gas injection. Natural gas combustion was modelled with the Eddy Dissipation Concept (EDC) model. Majeski et al. 19) studied three cases with a varying lance configuration for combined PC and natural gas injection. he combustion model is a combined EDM and finite rate chemistry model, where the slower of those is used to limit reaction rate. Li et al. 0) reported a parametric study with CFD of the blast furnace raceway and tuyere areas. he gas phase combustion model in the study was the EBU-Arrhenius model. hey evaluated the burnout rate of PC and optimized 7 parameters by orthogonal experiments. he parameters included two different lance angles (0 and 11 ) with respect to the tuyere centerline. he aim of this study was to develop a combustion model for extra heavy oil that is able to capture limiting factors of the injection process and that can be used to optimize the lance position in the tuyere area. hree different lance locations are modelled. Atomization of the extra heavy oil is modelled with the wave breakup model, 1) where the breakup time and droplet size constants have been adjusted and validated using experimental results from a physical set-up. ) In high Weber number (We) flows (We > 100) Kelvin-Helmholz instabilities are considered to dominate atomization. he size of the child droplets is proportional to the wavelength of the fastest growing wave on the surface of the parent droplet. 1) Combustion of extra heavy oil can be divided into two stages, i.e. vaporization and gas phase combustion. Vaporization is modelled with a convection and diffusion controlled model. 3) he gas phase combustion model is based on GRI-MECH 1., which contains species and 104 reactions. 4,5) he extensive model is needed for high temperature gasification and dissociation reactions and to capture the important phenomena caused by large gradients in temperature and species in the tuyere-raceway area. he combustion model is compared with the probability density function based model and maximum adiabatic temperature is calculated.. Model Description In the blast furnace (Fig. 1), tuyeres are used to blow hot air into the lower part of the furnace to provide oxygen needed for combustion and gasification of coke and hydrocarbons. Air blast induces a cavity, a so called raceway, in the coke bed in front of the tuyere. he auxiliary fuel such as heavy oil is injected into the air blast through an oil lance in the tuyere. he blast furnace in SSAB Europe Oy steel factory in Raahe, Finland has 1 tuyeres distributed symmetrically to the lower part of a blast furnace of which one is chosen for analysis..1. Geometry and Operating Conditions he geometry of the CFD model consists of a tuyere, a raceway and a part of the coke bed (Fig. ). he inlet boundary for the air blast is set at the inlet of the tuyere and Fig. 1. Schematic picture of the bottom part of the blast furnace. Fig.. Conceptual drawing of computational geometry and coke bed porosities (a), lance positions in tuyere in different cases (b) and nozzle geometry (c). 017 ISIJ 191

3 the outlet is in the upper part of the coke bed. In Fig. (b) the lance positions for different cases are presented. Lance positions are y tuy =0 cm (Case 1), y tuy =10 cm (Case ) and y tuy =0 cm (Case 3) from the tuyere-raceway intersection towards the beginning of the tuyere. Oil injection is set as injection points in front of an oil lance, which represents holes in the nozzle. he optimum oil lance nozzle geometry was studied earlier by us and it was used in the calculations. ) he optimum nozzle has 8 holes with 30 angle from the oil lance centerline, which is presented in Fig. (c). he porosity in the deadman is ε=0.35 and in the rest of the coke bed it is ε=0.4. Other boundaries are considered as adiabatic no-slip walls. In Fig. (a) plane illustrates the boundary between the deadman and the rest of the coke bed. o simplify the model, porosity in the raceway area is set to ε=1, which is also porosity in the tuyere area. he porosities in the deadman and the rest of the coke bed are based on the core-drill measurements in the Raahe blast furnace. here are several different equations for calculating the raceway dimensions, but according to Liaos s (1998) studies the Nomura s (1986) correlation fits to the Raahe s blast furnace. 10,6) Based on the discussions with SSAB Europe Oy the estimated depth of the raceway is d R =1.5 m. he estimated depth is used to calculate other dimensions of the raceway; therefore Eq. (1) is not needed and other dimensions can be calculated easily when the tuyere diameter is known. In the Nomura s correlation the raceway size and form are calculated with the following equations, d d R = ρ g0 V A g pa pg g 1 98 gdcρc (1) where d R is the raceway depth (m), d is the tuyere diameter (m), ρ g0 is the gas density under standard condition (kgm 3 ), V g is the volume flow of air blast (m 3 s 1 ), A is the area of the tuyere nose (m ), p a is the atmospheric pressure (Pa), p g is the air blast pressure (Pa), g is the air blast temperature (K), g is the gravity constant, d c is the coke particle diameter (m) and ρ c is the coke particle density (kgm 3 ). he raceway wih (w R ) is calculated using Eq. () wr dr =. 631 d d... () he Raceway height (h R ) is calculated from Eq. (3) 4h + d w R R R ( hd R ) d = 3. 1 d R (3) Dimensions based on the equations are as follows: depth of the raceway is d R =1.5 m, wih is w R =0.44 m and height is h R =0.85 m, when the diameter of the tuyere is d =0.14 m. he boundary conditions and oil parameters for the simulations are presented in able 1. Values are based on the actual blast furnace in Raahe. he oil injection level of kg/s is equivalent to 100 kg/thm. Ultimate analysis of the extra heavy oil that is used in the simulation is given in able. he heavy oil is simplified to contain only carbon and hydrogen, because other elements 1913 able 1. Boundary conditions and oil parameters. Variable Air blast flow rate Air blast O Air blast temperature Outlet pressure Oil mass flow rate Oil temperature Quantity.4 kg/s 8 vol-% 1 43 K 34 kpa kg/s 473 K Oil density 1 00 kg/m 3 Oil viscosity Oil surface tension 31.5 mpas 5 mn/m able. Ultimate analysis of extra heavy oil. Extra heavy oil [wt %] C 86.6 H 10.8 O 0.7 N 0.3 S 1.6 are present in small quantities and are not in the scope of the study... Gas Phase he CFD modelling has been done with Ansys Fluent It uses the Finite Volume Method (FVM) to discretize the flow domain. he governing conservation equations are integrated over each computational cell. 7) he continuity equation is written as ρ + ( ρv)= S m... (4) t S m is the source term for mass addition to the continuous phase from the dispersed phase. In this case it is the mass of fuel vapor from heavy oil droplets. he continuous phase is modelled as a compressible ideal gas, where the density is calculated from the following equation pop + p ρ = R M... (5) w where p is the calculated local static relative pressure to the operating pressure and p op is the operating pressure (Pa), R is the universal gas constant ( kg m s K 1 mol 1 ), M w is the molecular weight (g mol 1 ) and is temperature (K). he momentum conservation equation in an inertial frame of reference is ( ρv)+ ( ρvv)= ps + t µ ( v+ v ) vi + ρ... (6) g+ F 3 where p s is the static pressure (Pa), F is the external body 017 ISIJ

4 forces including the source terms from interaction with a dispersed phase and porous-media S i. μ is the dynamic viscosity (kg m 1 s 1 ) and I is the unit tensor. he coke bed is modelled as a homogenous porous reactive material. he porous media coke bed model has been previously used by Andahazy et al. 5) and Li et al. 0) he source term S i for the momentum equation for the porous media is Si = ( Rvµ vi 05. RIρvmagvi)... (7) where v i is the velocity in direction i and the v mag is the velocity magnitude (m s 1 ). Equations for the viscous resistance (R v ), which is the reciprocal of permeability and the inertial resistance (R I ) are derived from the Ergun equation 8) and are given below 5,0) and all the variables are presented in able 3 150* ( 1 ε ) Rv =... (8) 3 ψ dc * ε where ψ is the sphericity of the coke particles and d c is the coke particle diameter (m). 1 ε RI = (9) 3 ψ * dc * ε urbulence is simulated with the Realizable κ ε model. 9) It is a two equation model, which is based on the modelled equations for the turbulence kinetic energy (κ) and its dissipation rate (ε). Realizability ensures the positivity of turbulent normal stresses and the Schwartz inequality between any of the fluctuating quantities. 7,9) It prevents the turbulence model from producing unphysical results. In the Realizable κ ε model turbulent viscosity is calculated differently from the standard model, where C μ is a constant. More information about the model can be found from the article written by Shih et al. 9) he blast furnace model includes combustion, based on 7 species. his requires that the conservation equation for chemical species is added. he local mass fraction for each species Y i is solved with the convection-diffusion equation ( ρyi)+ ( ρν Yi)= Ji + Ri + Si... (10) t where R i is the net rate of production of species i by chemical reaction and S i is the rate of creation by addition from the dispersed phase (kg s 1 ). he mass diffusion in turbulent flows is solved by µ t Ji = Di m + Yi D i Sc t ρ,,... (11) where μ t is the turbulent viscosity (kg m 1 s 1 ), D,i is the able 3. Coke bed variables. Below the raceway Other areas Porosity (ε) Sphericity (ψ) 1 1 Particle diameter (d c) Viscous resistance (R v) Inertial resistance (R I) thermal diffusion coefficient for species i and Sc t is the turbulent Schmi number Sc = µ t, where D ρd t is the turbulent t diffusivity (m s 1 ). Laminar diffusion through concentration gradients is generally not necessary in turbulent flows, because the turbulent diffusion is much faster. ransport of enthalpy due to diffusion is important in multicomponent mixing flows and is solved by the energy equation ( )+ ( ρe ν ( ρe + p )) = t... (1) keff + ( hj j j τ eff ν ) +S h j where k eff is the effective conductivity (W m K 1 ) k+k t (where k is thermal conductivity and k t is the turbulent thermal conductivity). S h includes chemical reactions and radiation heat sources. E is the energy, which is calculated by p ν E = h +... (13) ρ where sensible enthalpy h in the case of ideal gas is h = Yjhj... (14) Y j is the mass fraction of species j. Enthalpy for the individual species j is solved from Eq. (15) hj = cp, jd... (15) ref For the pressure-based solver, the reference temperature ref =98.15 K. j.3. Heat ransfer Due to the high air blast velocity of more than 00 m/s, forced convection is the dominating heat transfer phenomena in the raceway area. here are large differences (maximum 440 K) in temperatures in the raceway-tuyere area and the forced convection is weaker in the upper parts of the raceway and also in the coke bed, therefore the radiation heat transfer is included in the computation. Furthermore, according to Gonҫalves et al. 30) combustion modelling without radiative heat transfer produces highly inaccurate results, because combustion occurs as a balance in small volumes (finite-volume framework) and the radiative heat transfer involves the long distance interaction. he Discrete Ordinates method (DO) is chosen as the radiation heat transfer approach. he model includes the effect of the gray gases also and it can be used to model radiation in porous areas. Droplets, walls and the coke bed are assumed to be black bodies. he particle scattering factor is set to σ p =0.9, which is the recommended value for coal combustion modelling in the ANSYS User s guide. 31) In the ANSYS Fluent the droplet radiation interaction causes that scattering in the continuous phase is ignored. he radiation heat transfer by the DO method with the droplet interaction is modelled with equation ( Is)+ a+ a + σ I r, s an σ π 4 ( p p) = σ 4π 4π p + Ep + I( rs, ) Φ ( s s ) dω 0... (16) 017 ISIJ 1914

5 where I is the intensity (W m ), s is the direction, a is the gas absorption coefficient, a p is the equivalent absorption coefficient, σ p is the equivalent particle scattering factor, r is the position, n is the refractive index, σ is the Stefan-Boltzmann constant (5.67E-8 Wm K 4 ), E p is the equivalent emission, s is the scattering direction vector, and Φ and Ω are control angles. he weighted sum of the gray gases model (WGSSM) is used for absorption modelling of gray gases. he equation for the total emissivity over the along length s is I =, i 1 i= 0 κ ps a e i x... (17) where a є,i is the emissivity weighting factor for the ith gray gas, κ i is the absorption coefficient of the ith gray gas, p x is the sum of the partial pressures of all absorbing gases..4. Discrete Phase Heavy oil is modelled with the Lagrangian method, which is called DPM (Discrete Phase Method) in the FLUEN solver. his model is used to study the individual droplet behavior. In the Lagrangian method the droplet trajectories are computed for each droplet separately, therefore the droplet properties are known at each step. he particle force balance, which is required for the particle velocity ( u p), is calculated from equation dup g ( ρp ρ) = FD( u up)+... (18) ρ p where u is the fluid phase velocity, ρ p is the particle density and the drag force is calculated from equation CDRe FD = 3 µ... (19) 4ρ pdp where C D is the drag coefficient and Re is the droplets relative Reynolds number ρdp up up Re... (0) µ When the droplet trajectory is calculated, the instantaneous fluid velocity u = u + u is used to predict the turbulent dispersion of droplets. he droplet atomization model was chosen to be the wave breakup model. 1) In our previous study the droplet atomization model was validated and the droplet breakup size constant B 0 and droplet breakup time constant B 1 were defined as B 0 =0.54 and B 1 =1. ) he droplet vaporization is modelled with a convection and diffusion controlled model, which can be applied for high Re flows when convection is important. 3) dmp = ka c pρ ln( 1 + Bm )... (1) where m p is the droplet mass (kg), k c is the mass transfer coefficient (m/s), A p is the droplet surface area (m ), ρ is the density of bulk gas (kgm 3 ) and B m is the Spalding number Yis, Yi, Bm =... () 1 Yis, where Y i,s is the vapor mass fraction at the surface and Y i, is the vapor mass fraction in the bulk gas. When temperature reaches the boiling point of the droplet the vaporization equation changes to ( p ) = d d ρ h p fg k dp Red ( )+ εσ θ 4 4 p p R p... (3) Droplet temperature is calculated as a heat balance that relates the sensible heat change in the droplet to the convective and latent heat transfer between the droplet and the continuous phase p mc d p p dmp = hap( p) hfg + Apεσ p θr 4 p 4... (4) where c p is the droplet heat capacity (J kg 1 K 1 ), p is the droplet temperature (K), is the temperature of continuous phase (K), h fg is the latent heat (J kg 1 ), ε p is the particle emissivity and the convective heat transfer coefficient h is calculated from Eq. (5) hdp ln( 1+ B ) Nu = = + Red Pr k B ( )... (5) where d p is the particle diameter (m), k is the thermal conductivity of the continuous phase (W m 1 K 1 ), Re d is the droplet Reynolds number and Pr is the Pranl number of the continuous phase (c p μ/k ) and B =B m. In this study cenospheres are not included in the computation..5. Combustion and Gasification In the blast furnace raceway maximum temperatures are around K and the equivalence ratios may vary from 0.5 to 0.8. However, gasification and dissociation reactions are important and reactions occurs through radical reactions. 1) In modelling the gasification and radical reactions the finite rate combustion model is needed and the Eddy Dissipation Concept (EDC) is chosen. 3) he volatile combustion model is based on the GRI-MECH 1. approach which contains species and 104 reactions. 4,5) he same reaction mechanism was used for gas phase combustion in our previous study, 33) but the studied fuel was PC. he combustion model was validated with experimental results and the mechanism worked well in that case. Based on the ultimate analysis of heavy oil (able ) the droplet vaporizing specie is C 37 H 55, which was added to the reactions. he volatile cracking method to smaller components is presented in the following reaction C H 18. 5CH + 9H, H = 5 17 kj / mol... (6) he activation energy for cracking is E a = J/kmol and the pre-exponential factor A = , which are taken from the FLUEN coal calculator kinetics for the two step coal combustion. Volatile cracking is expected to be similar to that of coal. Coke reactions are modelled in the porous coke bed. he reactions are modelled with a method that is presented in the article by Shen et al. 34) Coke reacts with O, CO and H O ISIJ

6 he reaction rate for coke with each species is calculated with the Field model, 35) which utilizes both diffusion and chemical reaction rate for the overall reaction rate dm k k X r p coke ( d + c ) g4π coke =... (7) pa where X g is the mole fraction of oxidizing species, r coke is the coke particle radius (m), p A is the atmospheric pressure (Pa) and diffusion of oxidizing species (k d ) on the coke surface is solved from equation k d D r = ref coke coke + ref p 075. p A... (8) where the D ref is the dynamic diffusivity ( kg m 1 s 1 ), coke is the coke temperature, ref is the reference temperature (93 K). he chemical rate coefficient (k c ) is calculated from equation c kc = A c exp... (9) coke where A c for O is kgm s 1 and c for O is K. A c and c for CO and H O are expected to be equal. he value for A c is kgm s 1 and for c is K. Other coke bed reactions are modelled as a heat sink in the same way as in Shen et al. 34) = ( [ ]) source = ha g coke g 0, 0 max 075. g, 1773 K Similarly to the Shen et al. 34) the h g is 18. W m K 1 and A coke is 153 m below the raceway and 141 m in the rest of the coke bed. In EDC, it is assumed that most of the reactions take place in the smallest scales of the turbulence, i.e. in the fine flow structures. 3) hese fine structures are treated as well-stirred reactors. Each computational cell is treated as a constant pressure reactor, with initial conditions taken as the current species and temperature. Equation for the length fraction of the fine structures is 14 / * νε ξ = C... (30) ξ κ where * indicates the fine structure quantities and C ξ is the volume fraction constant.1377 and ν is the kinematic viscosity. Reactions occur in the fine structures over a time scale 1 / * ν τ = Cτ ε... (31) he time scale constant is C τ = he source term in the conservation equation for chemical species i, is modelled as * ρ ξ * Ri = Yi Yi 3 * * τ 1 ( ξ )... (3) where Y i * is the fine structure species mass fraction after the reaction over the time τ *. he chemical equilibrium combustion model, which is a probability density function (PDF) model, is compared with the EDC combustion model to evaluate reliability of it. In the chemical equilibrium combustion model it is assumed that the chemistry is fast enough to reach chemical equilibrium and combustion is considered to depend only on the mixing rate. Reactions take place in thin layers where the molecular transport and the chemical source term balance each other. o describe combustion, PDF is needed for the mixture fraction Z at a point x and time t. he instantaneous averaged thermochemical state of the fluid depends on the local mixture fraction. Model contains species..6. Numerical Solution he computational meshes consist of about hexa- and tetrahedral computational cells. he mesh is the densest in the tuyere area, where the average cell size is 9 mm. In the raceway the average cell size is 13.5 mm. Other areas have larger cells (maximum 50 mm), but the mesh is designed so that the y + values near the walls are between 50 to 300, so that the wall functions are modelled correctly. he worst cell skewness is kept under 0.8 and the maximum aspect ratio is about 1. he continuous phase was solved using a steady-state solver and the discrete phase was time dependent. he time step size was s. Momentum and pressure are solved with a Semi-Implicit Method for Pressure Linked Equations (SIMPLE) solver. Gradients are solved with least squares cell-based gradient evaluation, which is a good method for irregular unstructured meshes. Initial solution was achieved using the first order upwind and standard pressure schemes. o get the final solution pressure was solved using the second order central differencing scheme and all the other quantities were solved with the second order upwind scheme. Under-relaxation factors were those recommended by Ansys Fluent. 31) he calculations have been made with the Environmental and Chemical Engineering research unit s computing cluster located at the University of Oulu. he cluster has 5 cores. Each case is parallelized and calculated using four cores. he execution time is about 10 days. 3. Results and Discussion 3.1. Grid Independency he grid independency was studied with three meshes with the Case configuration. he mesh in the area of interest i.e. the tuyere-raceway area was changed. In the coarse mesh the cell size is 10 mm in the tuyere area and 15 mm in the raceway area. In the medium mesh the cell size is 9 mm in the tuyere area and 13.5 mm in the raceway area. In the fine mesh the cell size is 8 mm in the tuyere area and 1 mm in the raceway area. he coarse mesh has cells, the medium mesh has cells and the fine mesh has cells. he grid independency was studied by comparing velocity, temperature, droplet penetration and O mole fractions with the difference tool in the CFD-Post. here were differences between the droplet penetration, temperature and O mole fractions between the coarse and medium meshes, but the differences became very small in the fine mesh case. 017 ISIJ 1916

7 3.. Comparison between Combustion Models he EDC (case a) and the PDF (case b) combustion models have been compared to evaluate reliability of the heavy oil combustion model. he coke bed reactions are not used in the combustion model comparison cases to clarify the difference between two combustion models. emperature contours of the case a and the case b are shown in Fig. 3. It can be seen that the combustion begins already in the tuyere in the case b and also the combustion occurs inside the heavy oil spray. A photo of oil injection in the blast furnace tuyere is shown in Fig. 4. It can be seen that the combustion begins after the tuyere nose and combustion does not occur in the heavy oil spray, which is opposite to the results from the PDF model. In the EDC case the combustion begins after the tuyere and combustion takes place on the spray surface, which appears to be more realistic than the PDF model. he highest temperature in the case a is 973 K and in the case b 910 K. he EDC model produces about % higher maximum temperature than the PDF model. Adiabatic flame temperature ( ad ) is calculated with Cantera..1 to evaluate, if the maximum temperature levels are realistic with the CFD combustion model. Initial temperature in the adiabatic flame temperature calculation is 1 43 K, which is the air blast inlet temperature. Results for ad as a function of equivalence ratio with initial hydrocarbon, oxygen volume fraction and pressure are given in Fig. 5. A stoichiometric mixture is presented with the cut line where the equilibrium ratio is 1. he well-known phenomena i.e. the maximum value of ad on the rich side of the fuel equivalence ratio in the mixtures of hydrocarbon and air can be seen in Fig ) It can be seen also that the pressure hinders dissociation, which moves the peak temperature closer to the stoichiometric ratio. he maximum ad in the blast furnace initial conditions (1 43 K, 3.6 bar, 8% O ) is 981 K, which is 8 K higher than the results from the CFD model, which seems reasonable, when considering that the oxidizer stream heats further in the blast furnace tuyere-raceway area Effect of Lance Position Figure 6 presents the contours of velocity with different lance positions. It can be seen that the air blast velocity increases, when the injection lance is drawn inside the tuyere. he lowest maximum velocity 1 m/s is in the case 1 and the highest velocity 5 m/s is in the case 3. he difference in the maximum velocity of air blast inside the raceway is about 16%, between the cases 1 and 3. In the case the maximum velocity is 30 m/s. High vaporization rate of the heavy oil and gas phase combustion increase the pressure drop in the tuyere in the case 3, which results in an increased velocity in the tuyere-raceway area, because the air blast mass flow rate is kept constant throughout the calculation. he air blast velocity affects both the droplet residence time in the raceway and the mixing phenomena. After 10 cm flight in each case, the heavy oil is concentrated in an area, the diameter of which is 3.5 cm, 4 cm, 6 cm in the cases 1, and 3, respectively. his is consistent because the air velocity is also lower in the beginning of the tuyere, so the heavy oil is spread over a larger area. he heavy oil tracking is presented in Fig. 7. he maximum droplet penetration range is 1.60 m in the case 1, Fig. 3. Contours of temperature in the cases of a) EDC and b) PDF combustion model. (Online version in color.) Fig. 5. Adiabatic flame temperature of different hydrocarbons as a function of equivalence ratio. Fig. 4. Photo of oil injection through the tuyere viewing window. Fig. 6. Velocity contours in the tuyere-raceway area in the cases 1, and 3. (Online version in color.) ISIJ

8 1.6 m in the case and 0.45 m in the case 3. he droplet penetration increases drastically when the injection lance is closer to the tuyere nose. he reason for this behavior is most likely that the heavy oil mixes well with air blast in the case 3, which leads to a higher surface area and faster vaporization rate. In the cases 1 and the aerodynamic forces are higher than in the case 3, which reduces the mixing ability between the blast and heavy oil and leads to a slower vaporization rate. emperature contours are shown in Fig. 8. It can be seen that the temperature contours vary for the investigated cases and the maximum temperatures are located at different areas. In the case 1 temperature is highest under the heavy oil spray. Mixing is poorest in this case and most of the combustion occurs on the spray surface and only moderately in the upper part of the raceway. In the case the highest temperature is located on the surface of the spray, but high temperature exists also in the upper part of the raceway. In the case 3 the temperature field is the most uniform on the spray surface and the mixing is the best of these three cases. In the case 3, combustion begins already in the tuyere, which might cause wear and ablation in the tuyere walls. Combustion does not occur in the heavy oil stream, because vaporization and heating of the droplets use the excess heat. he domain outlet pressure level is fixed to 34 kpa, but because of the pressure drop in the coke bed the inlet pressure adjusts to the level of 360 kpa with the mass flow of.56 kg/s and no injection. he air blast pressure cannot be increased above 400 kpa with the existing system, so it Fig. 7. Heavy oil tracking in the tuyere-raceway area in the cases 1, and 3. is important to study the pressure losses in different cases. he inlet pressure varies within each case, because heavy oil vaporization and combustion increase the pressure drop. In addition the lance position has some effect on pressure, since the cross-sectional area varies in different parts of the tuyere. Figure 9 shows the change in the inlet pressure. he zero point is 366 kpa, which is the inlet pressure in the case 1. he pressure increases slightly, when the lance is moved 10 cm back from the tuyere tip, but increases drastically when the lance is moved 0 cm from the tuyere tip. he lowest pressure drop was observed in the case 1, where vaporization takes place completely in the raceway area. According to experimental results of the former Koverhar blast furnace by Helle and Saxen, 37) the blast furnace pressure loss increases, when the heavy oil lance is pulled back from the tuyere tip. Our calculations show a similar behavior. he increase in the pressure drop, which was observed in the case 3, might cause problems in the actual blast furnace, where 1 tuyeres are used instead of just one. he mole fractions of O, H O, H, CO and CO are presented in Fig. 10. It can be seen that gasification gases move towards the tuyere nose, when the lance is moved 0 cm away from the tuyere tip. Mixing is the best in the case 3 when the lance is moved back, indicating that more oxygen is reacting and gasification starts earlier. Gasification takes place in the areas with low oxygen contents and high temperatures. Gasification reactions are endothermic and it can be seen that the areas with high CO and H levels also have lower temperatures and no oxygen. According to these results the most effective lance position is 10 cm inside the tuyere. Even though the combustion is the best in the case 3, the problem is that the combustion begins inside the tuyere, which might lead to the wearing and ablation of the tuyere walls.furthermore blast velocities over 30 m/s (case 3) cause high degradation of coke, 38) producing fine coke dust and lowers the permeability in the dead man. he lance at the tuyere raceway connection is the best solution from the point of view of pressure drop and there will be no combustion inside the tuyere. he mixing in that case is the poorest and the coke bed might be reacting more than in the other two cases. his might result in an incomplete combustion and accumulation of fine particles in the coke bed. he 10 cm case is a compromise between two cases, but optimized position of the lance would need more simulations around the suggested lance position. Based on the discussions with SSAB Raahe and also the tuyere videos (Fig. 4), the model produces similar results Fig. 8. Contours of temperature in the cases 1, and 3. (Online version in color.) Fig. 9. Absolute pressure at the inlet with different lance positions. 017 ISIJ 1918

9 Fig. 10. Mole fractions of O, CO, CO, H and H O in the cases 1,, and 3. (Online version in color.) seen in the SSAB Raahe process. here is no visible flame when the lance is 10 cm inside the tuyere. It should be noted also that increasing the heavy oil injection rate above 100 kg/thm, with the current setting leads to unburned fuel, this causes clogging of the coke bed, leading to an unstable process. he same phenomena occur both in the actual blast furnace and the CFD model with the same injection level, which is important, when considering the reliability of the model. Even though the auxiliary fuel is changing from heavy oil to PC, the interest is still high for liquid fuels in SSAB Raahe steel mill. In the future, if it becomes cost-effective, bio-oil is possible substitute for PC, because it is considered carbon neutral renewable fuel. Heavy oil model can be used to model bio-oil combustion, because of their similar qualities ISIJ

10 4. Conclusion he work in this study focused on the development of a numerical model for the gas phase combustion for extra heavy oil. he effect of lance positioning on the heavy oil combustion in the blast furnace tuyere-raceway area was studied. hree different cases were modelled, where the lance position was varied between 0 and 0 cm from the tuyere tip. hese cases have been used to determine the best injection position for the extra heavy oil in the tuyere. Combustion is modelled with the EDC model, which includes detailed chemical reaction mechanisms. Based on the results the combustion model produces realistic maximum temperature. It also produces results similar to those obtained in industrial experiments. he most effective lance position was found to be 10 cm inside the tuyere. Anyway, mixing of the air blast with the fuel was poor in all cases, but the best results were found for the case 3. Most of the fuel is concentrated in the center of the air blast in a small area (maximum diameter 6 cm in the case 3). Pressure drop increases drastically when the lance is moved 0 cm upstream, which might cause problems in the actual blast furnace. From the results it can be concluded that the PDF model do not work well to simulate the tuyere-raceway area combustion, because the combustion begins too early and it also takes place inside the fuel stream. his is important, because the PDF model overestimates the combustion of heavy oil, which means that the coke combustion is underestimated. his would lead to over optimistic injection levels. he developed CFD and combustion model provides a tool for the future studies to improve understanding of the auxiliary fuel injection in a blast furnace operation. In the future cenospheres are added into the combustion model. Acknowledgements his work was written as a part of Finnish Metals and Engineering Competence Cluster (FIMECC) s System Integrated Metals Processing (SIMP) program. he authors wish to acknowledge CSC-I Center for Science, Finland, for computational resources. REFERENCES 1) K. Ishii: Advanced Pulverized Coal Injection echnology and Blast Furnace Operation, Elsevier/Pergamon, Oxford, (000), 34. ) P. Moszkowicz, L. Witzel and G. Claus: Chem. Eng. Sci., 51 (1996), ) W. Bartok and A. F. Sarofim: Fossil Fuel Combustion: A Source Book, John Wiley, New York, (1991), ) R. Nightingale, R. Dippenaar and W. Lu: Metall. Mater. rans. B, 31 (000), ) D. Andahazy, S. Slaby, G. Löffler, F. Winter, C. Feilmayr and. Bürgler: ISIJ Int., 46 (006), ) C. Jordan, M. Harasek, A. El-Gohari, C. Feilmayr and S. Schuster: 5th European Conf. on Computational Fluid Dynamics (CFD 010) - Book of Abstracts, Vol. 1, EccoMAS, Lisbon, (010), ) C. Jordan, M. Harasek, F. Zauner, F. Kirchbacher, C. Feilmayr and S. Schuster: Chem. Eng. rans., 1 (010), ) C. Jordan, C. Maier, Z. Harsfalvi, B. Kiss and M. Harasek: Chem. Eng. rans., 39 (014), ) J. Hakala and. Paloposki: Small Scale Model Experiments on the Injection of Heavy Fuel Oil into the Blast Furnace, Helsinki University of echnology, Espoo, (1996), ) D. Liao: Licentiate thesis, University of Oulu, (1998), ) Y. S. Shen,. Shiozawa, P. Austin and A. Yu: Miner. Eng., 63 (014), 91. 1) B. O. d. A. Santos, B.. Maia, F. S. Garajau, M. d. S. L. Guerra, P. S. Assis and J. E. M. Barros: J. Mater. Res. echnol., 3 (014), ) M. Gu, G. Chen, M. Zhang, D. Huang, P. Chaubal and C. Q. Zhou: Appl. Math. Model., 11 (010), ) B. Guo, P. Zulli, H. Rogers, J. G. Mathieson and A. Yu: ISIJ Int., 45 (005), ). Ariyama, M. Sato, Y. Yamakawa, Y. Yamada and M. Suzuki: ISIJ Int., 34 (1994), ) C. Chen: Appl. Math. Model., 9 (005), ) C. Yeh, S. Du, C. sai and R. Yang: Energy, 4 (01), ) A. K. Silaen,. Okosun, Y. Chen, B. Wu, J. Zhao, Y. Zhao, J. D Alessio, J. C. Capo and C. Q. Zhou: Proc. Iron Steel echnology Conf. and Exposition (AISech 015), AIS, Warrendale, PA, (015), ) A. Majeski, A. Runsteler, J. D alessio and N. Macfadyen: ISIJ Int., 55 (015), ) Y. Li, X. Zhang, J. Zhang, J. Zhou and H. Yan: Appl. herm. Eng., 67 (014), ) R. D. Reitz: At. Spray echnol., 3 (1987), 309. ) A. Vuokila, M. Riihimäki and E. Muurinen: Steel Res. Int., 85 (014), ) S. S. Sazhin: Prog. Energy Combust. Sci., 3 (006), 16. 4) M. Frenklach, H. Wang, M. Goldenberg, G. P. Smith, D. M. Golden, C.. Bowman, R. K. Hanson, W. C. Gardinera and V. Lissianski: GRI-Mech An Optimized Detailed Chemical Reaction Mechanism for Methane Combustion, Gas Research Institute, Des Plaines, IL, (1995), ) M. Frenklach, H. Wang, C. L. Yu, M. Goldenberg, C.. Bowman, R. K. Hanson, D. F. Davidson, E. J. Chang, G. P. Smith, D. M. Golden, W. C. Gardiner and V. Lissianski: GRI Mech 1., (accessed ). 6) S. Nomura: ISIJ Int., 6 (1986), ) ANSYS Inc: Ansys Fluent 16.0 heory Guide, ANSYS Inc, Canonsburg, PA, (014), ) S. Ergun: Chem. Eng. Prog., 48 (195), 89. 9).-H. Shih, W. W. Liou, A. Shabbir, Z. Yang and J. Zhu: Comput. Fluids, 4 (1994), 7. 30) R. Gonçalves dos Santos, M. Lecanu, S. Ducruix, O. Gicquel, E. Iacona and D. Veynante: Combust. Flame, 15 (008), ) Ansys Fluent 16.0 User s Guide, ANSYS Inc, Canonsburg, PA, (014), 69. 3) B. F. Magnussen: 19th American Instifute of Aeronautics and Astronautics Aerospace Sciences Meeting, Reston, VA, (1981), 4. 33) A. Vuokila, O. Mattila, R. L. Keiski and E. Muurinen: Ind. Combust., (016), Artide No , 1. 34) Y. S. Shen, B. Y. Guo, A. B. Yu, P. R. Austin and P. Zulli: Fuel, 90 (011), ) M. Field, D. Gill, B. Morgan and P. Hawksley: he Combustion of Pulverized Coal, British Coal Utilization Research Association, London, (1967), ) C. K. Law, A. Makino and. F. Lu: Combust. Flame, 145 (006), ) M. Helle and H. Saxen: ISIJ Int., 4 (00), ) Y. Iwanaga and K. akatani: rans. Iron Steel Inst. Jpn., 8 (1988), ISIJ 190