Numerical investigation towards HiTAC conditions in laboratory-scale ethanol spray combustion

In the past 25 years high temperature air combustion (HiTAC) technology has been proved and utilized in industry as a promising way to increase thermal efficiency, create a relatively uniform temperature distribution, and reduce emissions of harmful pollutants such as NOX and CO. However, due to the complexity of fuel-oil combustion, to date HiTAC is mainly applied to gaseous fuel or coal, and little is known about spray combustion under HiTAC condition. In the present study, we numerically investigate the Delft Spray-in-Hot-Coflow (DSHC) using ethanol in high temperature diluted combustion air, and extend it to more co-flow conditions. We employ different temperatures and oxygen concentrations of the co-flow in order to dilute the oxidizer/fuel before it reacts with the fuel/ oxidizer. The pressure-swirl atomizer model with an Eulerian-Lagrangian approach was implemented for the spray modeling. Collision, coalescence, secondary breakup and evaporation of the drops were taken into account. The steady laminar flamelet model for the combustion of ethanol, the Discrete Ordinate model for radiation and the k-ε model for the turbulence with enhanced wall treatment were validated by the simulation of the NIST flame under conventional conditions and then used in the current study. The results indicate that the decreased peak temperature in many HiTAC applications with high temperature combustion air is mainly due to the reduced oxygen concentration by entraining flue gas. In the present study, a low oxygen concentration slows the evaporation process of droplets. It results in an enlarged combustion zone, a lowered peak temperature and minor NOX emission. However, decreasing the oxygen concentration may lead to problems of cracking, soot formation and flame extinction, especially for heavy oils. The optimization needs to be carried out based on the analysis of a specific fuel in order to create a HiTAC-like condition. Based on the results of the current study, the 1500 K and 6%vol oxygen concentration case is considered as a HiTAC condition.


Introduction
High temperature air combustion (HTAC) is a promising technology for energy saving, flame stability enhancement and reduction of NO X emission. It has been applied in many experimental and industrial applications, and also developed and reported as "moderate or intense low oxygen dilution (mild) combustion", "flameless oxidation", or "colorless distributed combustion (CDC)" [1][2][3][4][5][6][7][8]. In such a combustion regime, oxygen/fuel stream is diluted by a substantial amount of hot inert flue gases before it reacts with the fuel/oxygen. This results in more uniform temperature distribution and lower NO X emission than in case of conventional combustion. To date most of the applications of HiTAC are for gaseous fuels [3][4][5] or solid fuels [9,10], but little is known about spray combustion under HiTAC condition [1,11]. H. Tsuji et al. [1] introduced the historical background of HiTAC technology, and described its development and practical application to different kinds of furnaces of importance in industry. Besides the gaseous and solid fuels, they investigated experimentally kerosene spray flames and reported qualitatively with photographs the states of spray flame combustion in the high temperature preheated diluted air (523 K to 1373 K), when the O 2 concentration is changed (13% to 3%), Although it was concluded that NO X emissions reduce in the same manner as gaseous fuel, cases when the O 2 concentration in highly preheated air is lower than 15% were not further discussed. Moreover, the experimental results from NKK Keihin using heavy oil A did not show a clear trend [1]. That can be explained by the complexity of spray combustion and many unclear fundamental aspects involved in spray combustion, in particular turbulent spray combustion. Modeling of turbulent spray combustion however, although challenging provides a better understanding of various phenomena in the processes involved. In a real turbulent spray flame, dispersion, continuous phase turbulence modification, dispersed phase inter-particle collisions, evaporation, mixing and combustion occur simultaneously. Dealing with all these complexities and their interactions poses a tremendous modeling task [11]. Jenny et al. [11] reviewed the modeling developments of turbulent spray combustion, together with the relevant experiments of spray flame configurations presented in a structured way, with the intention to provide a database for model validation and a guideline for future investigations. The ethanol spray flame modeling presented in the current study is one of the investigations using the Delft Spray-in-Hot-Coflow (DSHC) burner.
To address all the relevant phenomena, we employ a combination of models, such as models for turbulence, atomization, secondary break-up, collision and coalescence, evaporation, radiative heat transfer, combustion, etc. These models have been validated already by the simulation of the NIST flame using methanol under conventional condition [12] and then were applied for comparison with preliminary results from the ethanol spray-in-hot-co-flow [13]. The predicted mean velocity components of the gaseous flow and the droplets, the droplet number density, and the Sauter mean diameter (SMD) of the droplets at various heights showed good agreement with the experimental data. The method to set inlet boundary conditions for the spray has been discussed and compared with direct application of data regarding measured droplet size and velocity distribution. In the application for modeling the ethanol spray-inhot-co-flow case, relatively good agreement with experimental data of mean droplet velocity components and size distribution at various elevations has been observed, as well. It was recommended that under hot coflow conditions, the evaporation model and the auto-ignition and local extinction processes require more efforts in order to obtain more accurate prediction in the high strain region, which is in the vicinity of the injector exit. Ma et al. [14,15] further exploited more directly all available measured data and used a "conditional droplet injection model", in which large droplets are injected within a small range of angle around the main spary trajectory while small droplets are injected in a much wider range. In this way the prediction of SMD in low droplet number density region also showed good agreement with the experimental data, as well as in the vicinity of the injector exit.
However, with the DSHC setup, the co-flow temperature and oxygen dilution cannot be varied independently [16]. As a result, the temperature, velocity of gas and liquid, gas components in the co-flow and their distribution vary depending on the investigated test case. Furthermore, due to the performance of the atomizer under different coflow temperature conditions, the mass flow of ethanol can be different from the designed one. This can be verified by the reported injection pressure and mass flow rate of ethanol [17]. As a result, in order to investigate the influential parameters on combustion characteristics, we focus on a comparative numerical study of cases with different co-flow temperatures and O 2 concentrations using the validated models in a wider range than studied experimentally.
In the present research, we used different co-flow temperatures (300 K, 600 K, 900 K, 1200 K and 1500 K) and oxygen concentrations (21%, 18%, 15%, 12%, 9% and 6%) of the co-flow. The mass flow rate and pressure of the fuel-oil are kept constant. The simulation results from different cases, such as temperatures, flame profiles, droplet size distributions, etc., are studied with respect to HiTAC conditions.
As we discussed in [12], both the boundary condition and spray trajectory analysis are essential for validation of spray flames. The validation of the DSHC ethanol spray flame requires both a well-defined boundary condition of co-flow and a detailed analysis of spray trajectory. To clearly identify the influence of temperature and oxygen concentration in the coflow we here study two typical operation conditions of DSHC test rig, i.e. cold co-flow condition (300 K, 21% vol O 2 ) and hot co-flow condition (1500 K, 6% vol O 2 ), keeping the composition of the rest of the coflow identical. They represent spray flames under conventional condition and towards-HiTAC condition, respectively.

Numerical cases
The numerical cases are based on the experimental set up available at Delft University of Technology using the DSHC burner. Fig. 1 presents a schematic of the Delft spray burner [17,18]. An ethanol spray is produced by a pressure swirl atomizer, whereas the hot co-flow is assured by secondary burner whereby air and Dutch natural gas (DNG) mix and generate a matrix of 236 lean flames. The air/DNG ratios in combination with the effects of two perforated plates located along the pipe length dictate the temperature, oxygen and turbulence levels. Coflow composition and temperature are selected to emulate the conditions of flue gas in large scale furnaces. This design enables a wide range of co-flow characteristics with good axisymmetric properties without the need for elaborate safety systems and allows easy optical access to laser diagnostics.

Mathematical models
The mathematical formulation for turbulent spray combustion simulation consists of the consideration of the computational grid, turbulence model, spray model, radiation and combustion model, NO X model etc. Since these models and methods are expected to be validated and developed to generate the knowledge needed for the extension of the application of HiTAC technology to other fuel oils in various applications, for which detailed reaction mechanisms may still need to be investigated and various geometries of chambers/furnaces may be involved, the Reynolds averaged equations are employed, using ANSYS Fluent software. As a result, the models used in the study remain as general as possible in order to be implemented for various applications.

Computational grid, near-wall treatment and turbulence model
For the simulation of the Delft spray flame, a 2D axisymmetric simulation is employed. The grid independence was tested by introducing a series of different cell sizes. The same axial/radial aspect ratio of 3 was kept in order to avoid the interference of potential errors caused by the aspect ratio. A considerable larger aspect ratio however leads to large cells close to the outlet which reduces the accuracy of results, while a considerable lower aspect ratio enhances the amount of cells requiring unnecessary computational cost. The role of the near-wall treatment was analyzed, and a 2D mesh with about 0.6 million quadrilateral cells in combination with the second order upwind scheme was found suitable for this study. A standard k-ε turbulence model with the enhanced wall treatment is applied based on the comparative analysis. The use of the enhanced wall treatment can possess the accuracy of the standard two-layer (a viscosity affected region and a fully-turbulent region) approach for the fine near-wall mesh and at the same time, not reducing accuracy for the wall-function mesh.

Spray model
In spray combustion studies, often the focus is on the modelling of dilute spray combustion due to the complexity of the processes of injection and atomization [11,19]. The experimental data of droplets, i.e. velocity components and size distribution, are frequently used as boundary conditions. Those data are supposed to be measured at a certain axial location, where the atomization has already finished while the injected droplets have not been influenced by the airflow and only a minor part of the evaporation has occurred. However, on one hand, the measurement technique provides data on only part of the liquid flux close to the atomizer; on the other hand, at further downstream positions, the droplets have already interacted with the co-flow, especially in hot co-flow, and boundary conditions on droplets have to be complemented with boundary conditions on vapor concentration and gas velocity [12]. As a result, we here keep using the Linearized Instability Sheet Atomization (LISA) model to simulate the transition of ethanol from the internal injector flow to fully developed spray by a pressureswirl atomizer [12,20]. In this model Kelvin-Helmholtz waves grow on the sheet and eventually break the liquid into ligaments which are further break up into droplets due to varicose instability. The film formation follows the below relationship: where t is the thickness of the film, m ff · e is the effective mass flow rate, d inj is the injector exit diameter, and u is the mean axial component of velocity at the injector exit. The approach of Han et al. [21] is used to calculate the velocity magnitude of u: where k v is a dimensionless velocity coefficient and a function of the injector design and injection pressure [22]. If the pressure difference at the injector, P Δ is known, u can be calculated as where θ is the spray angle. The effects of the surrounding gas, liquid viscosity and surface tension on the breakup of the liquid sheet are also included in the LISA model for sheet breakup and atomization, which is based on the growth of sinusoidal waves on the liquid sheet.
The breakup from ligaments to droplets is assumed to occur according to Weber's analysis for capillary instability [23]. So the most probable diameter for droplet diameter distribution, d 0 , is determined from: where oh is the Ohnesorge number which is a combination of the Reynolds number and the Weber number. Once this most probable droplet size has been found, with a spread parameter and a dispersion angle, the droplet diameter distribution is determined by a Rosin-Rammler distribution. According to our previous investigation [12], a spread parameter of 3.5 and a dispersion angle of 10°are found suitable for such a spray modelling using one main trajectory of spray injection.
In many real applications, there may arise need for a few trajectories to be superimposed in order to resemble better the real droplet diameter distribution and droplet velocity. Moreover, even with the same injector, the droplet diameter distribution and droplet velocity vary due to thermal expansion of the injector under various operating conditions. This makes the validation challenging. On one hand, even if there are experimentally measured data of droplets, in the vicinity of the injector they are often inadequate for a complete set of droplet diameter distribution and droplet velocity, which means they have to be tuned and verified by data at downstream elevations. On the other hand, even if well-tuned droplet diameter distribution and droplet velocity for one case are obtained and have been verified by data at downstream positions, they have to be tuned and verified again if the operating condition varies, especially when the measured data are directly used [12].
In the present study, however, we assume that the droplet size distribution and droplet velocity from the injector keep the same in various cases. This allows us for the comparative investigation of the influence of only the co-flow conditions.
In the simulation, the fuel is assumed to be injected into the chamber as a fully atomized spray consisting of spherical droplets of various sizes. The motion of the droplets in the turbulent combustion flow field are calculated using a stochastic method. The equation of motion for one droplet is: In this equation, u p i , is the velocity of droplet (particle) i, U is a sampled gas velocity, μ is the molecular viscosity of the fluid, ρ is the fluid density, ρ p is the density of the particle, D p is the particle diameter, Re is the relative Reynolds number based on slip velocity and particle diameter, and the drag coefficient C D is a function of the particle Reynolds number. F i is a possible additional acceleration term. In practice a number of 'parcels', each representing a set of identical droplets, is tracked.
The Taylor Analogy Breakup (TAB) model is used for secondary breakup. It is based upon Taylor's analogy [24] between an oscillating and distorting droplet.
The algorithm of O'Rourke [25] is employed for droplet collision and coalescence. It uses the concept of a collision volume to calculate the probability of collision. In general, once two parcels collide, the outcome tends to be coalescence if they collide head-on, and bouncing if the collision is more oblique. The critical offset for the probability of coalescence is a function of the collisional Weber number and the relative radii of the collector and the smaller droplet.
The rate of vaporization is governed by gradient diffusion, with the flux of droplet vapour into the gas phase related to the difference in vapor concentration at the droplet surface and the bulk gas: where N i is the molar flux of vapour, C s , i is the vapour concentration at the droplet surface, and C , i is the vapour concentration in the bulk gas. K c is the mass transfer coefficient calculated from the Sherwood number correlation [26,27], defined as: where D m , i is the diffusion coefficient of vapour in the bulk, Sc is the Schmidt number. The concentration of vapour at the droplet surface is evaluated by assuming that the partial pressure of vapour at the interface is equal to the saturated vapour pressure, P at s , at the droplet temperature, T p : where R is the universal gas constant.

Radiation and combustion model
In this study, the Discrete Ordinates (DO) radiation model with a variable absorption coefficient, weighted-sum-of-gray-gases model (WSGGM) is employed, which has been discussed in [1] and validated in our previous investigation [12].
According to the relative fast chemistry of ethanol, the laminar flamelet method provides a feasible way to include detailed chemical reactions in turbulent combustion simulations without a considerable increase in computational time. It assumes that in the gaseous phase combustion, the diffusion coefficients for all species are equal, and then the species mass fraction and temperature are mapped from physical space to mixture fraction space and can be uniquely described by two parameters: the mixture fraction ξ and the scalar dissipation rate χ . The Favre-averaged values of quantities in the turbulent flame are then obtained through the use of Favre-averaged probability density function, The detailed reaction mechanism for ethanol employed in the present study was developed by Marinov [28] consisting of 57 species and 383 reactions.
In the model, the heat gain/loss to the system is assumed to have a negligible effect on the species mass fractions, and adiabatic mass fractions are used [29,30]. The flamelet profiles are then convoluted with the assumed PDF as in Eq. (9), and then tabulated for look-up. Assuming that fluctuations of scalar dissipation rate can be neglected, only the mean value has to be considered in Eq. (9) and the relevant PDF is the PDF of mixture fraction which is assumed to be a β-function. The equations for the mean mixture fraction, mixture fraction variance, and mean enthalpy are solved. The mean scalar dissipation field is calculated from the turbulence fields ∼ k ,ε and the mixture fraction variance ∼ ξ Prime2 as follows: The mean values of temperature, density, and species mass fraction are obtained from the PDF look-up table. It should be pointed out that the β-function assumption in case of spray flames is a less appropriate than in gas phase combustion [31,32] and extensions proposed by Ge and Gutheil [31] could be better. Because only the standard β-function is available in ANSYS-Fluent and because for the purpose of comparative study of impact of different co-flow conditions on HiTAC conditions the difference is not expected to be large we have not explored those extensions.

NO X model
Since the fuel, i.e. ethanol, used in the present study contains no nitrogen, NO X formation mainly consists of thermal NO X and N 2 O intermediate mechanism [33][34][35]. Although prompt NO X is also taken into account in the simulation, our numerical results have shown that it has little contribution to the total NO X formation. This was also observed by other researchers [36].
Thermal NO X is calculated by the extended Zeldovich mechanism [37,38], containing the following reactions governing the formation of thermal NO X from molecular nitrogen: These reactions are considered to be reversible and the rate constants with units of m mol s ( / · ) 3 are as follows: where "f" and "r" in superscript denotes forward and reverse rates, respectively. T means temperature in Kelvin.
In order to compute the NO concentration, concentrations of nitrogen radical [N], oxygen radical [O] and hydroxyl radical [OH] must be known.
Concentrations of [N] can be assumed in a quasi-steady state according to its nearly immediate consumption after creation. This Nradical formation is the rate limiting factor for thermal NO production, due to the extremely high activation energy of nitrogen molecule dissociation, which is caused by a triple bond between two nitrogen atoms as shown in Eq. (11). Hence the NO formation rate m mol s ( / · ) 3 becomes: The formation of N 2 O intermediate mechanism takes into account two reversible elementary reactions: where M is a general third body. N 2 O can be assumed to be at quasi-steady-state, and the rate of [N 2 O] and the rate of NO X formation via the N 2 O-intermediate mechanism are: where . The transport equation for mean mass fraction of NO is solved with the mean source term obtained using the PDF of temperature fluctuations. As alternative to the determination of OH and O from partial equilibrium assumptions, they could be retrieved from the flamelets. However due the assumption made above in the nonadiabatic flamelet computation that the species concentration in the flamelets is independent of the heat loss, with only temperature changing when enthalpy changes, the procedure using the values from partial equilibrium probably provides the more accurate NO source term.

Boundary conditions for modelling
The computational domain is illustrated in Fig. 2, which gives an image of a typical experimental flame. Further illustration of the computational domain, including spray angle (SA) and dispersion angle (DA) of the fuel spray are given in Fig. 3. As discussed in our previous investigation [12], SA needs to be analyzed for the estimation of spray trajectory based on measured droplet concentration at various heights. Since in the current comparative study the focus is on the influences of co-flow temperature and O 2 concentration in the co-flow on combustion characteristics, we use the design angle of 60°according to manufacturing data for all cases. DA is a presumed angle according to previous modelling [12], in which 10°is used and it is kept the same for all cases.
As shown in Fig. 3, ambient air is taken into account in order to investigate its influence on the spray combustion. The ambient air is entrained by the co-flow coming out from the pre-combustor, and may penetrate the co-flow and take part in the combustion process. The condition in which the spray combustion is most likely influenced by the ambient air is the one whose co-flow characteristics have the largest gradients with the ambient air (300 K and with 21% vol of O 2 ), i.e. the largest differences of O 2 concentration, temperature and velocity.
In the present study, we keep the velocity of the co-flow at 3.5 m/s and vary the O 2 concentration (6%, 9%, 12%, 15%, 18% and 21% vol) and temperature (300 K, 600 K, 900 K, 1200 K and 1500 K) of the coflow. As a result, the spray flame, which has 6% O 2 in the co-flow with a temperature of 1500 K, is most likely to be influenced by the ambient air.
The ethanol spray characteristics are kept the same in various cases, as shown in Table 1. There is always sufficient oxygen amount for complete combustion. Besides, the various amount of H 2 O and CO 2 in co-flow in actual conditions are replaced by N 2 for the current comparative study.
In order to investigate the influence of the ambient air on the spray combustion, we applied different hydraulic diameters of the ambient air inlet, ranging from 6 to 14 times diameter of the co-flow inlet. This was done for the simulation of the case with 1500 K and 6% O 2 in the coflow. With the employed vertical height of 2.5 m for the computational domain, we found that when the outer diameter of ambient air inlet is smaller than 8 times the outer diameter of the co-flow inlet, the back flow through the defined lateral outlet cannot be avoided and the computational stability is reduced. When diameter is larger than 12 times the diameter of the co-flow inlet, the recirculation zone caused by the co-flow and the back flow through the defined top outlet also increase the instability of the calculation. Therefore, a hydraulic diameter of 1600 mm (10 times the diameter of the co-flow inlet) is used in the present study and it results in relatively smooth streamlines of airflow. It must be noted that in the simulation there is a part of the ambient air is remaining laminar. This aspect has been studied by considering and comparing four different flow speeds (i.e. 0.1 m/s, 0.05 m/s, 0.025 m/s,   0.01 m/s) of the ambient air. The predicted velocity components and temperature show minor difference at various elevations. As a result, the ambient air shows minor influence on the spray combustion characteristics and the steady laminar flamelet combustion model can be applied.

Results and discussion
In this section two typical conditions, cold co-flow condition (300 K, 21% vol O 2 ) and hot co-flow condition (1500 K, 6% vol O 2 ) are compared to the closest conditions from experiment in this section. They represents spray flames under conventional condition and towards-HiTAC condition, respectively. Following that, the influences of O 2 and temperature in the co-flow are discussed.

The HiTAC condition and preliminary validation of models
Among the cases with various co-flow conditions, only several are feasible to be carried out in the experiment. The co-flow in the experiment is generated by the secondary burner located in the middle of the setup (see Fig. 1.). The O 2 concentration and the temperature of the co-flow are dependent on the inlet conditions of this burner. The validation of numerical data is done for the spray flames with hot co-flow and with cold co-flow. The former represents a HiTAC-like combustion regime whereas the latter is a conventional spray flame. In the simulation, 300 K and 21% vol O 2 condition is used for the cold co-flow condition, and 1500 K and 6% vol O 2 for the hot co-flow condition. In both, the co-flow velocity and the mass flow rate of ethanol are kept the same. In the experiment however, it is not possible to maintain the coflow velocity the same and neither the exact matching of the mass flow of ethanol due to the limitation of the mass flow control and influence of high temperature on the atomizer. Besides, as discussed in our previous study [12], the trajectories of the spray need to be analyzed from the experiment in order to obtain good matching of SMD distribution between the experiment and simulation. As a result, the validation in the present study is mainly focused on the flame profile, and the range and trend of SMD at various elevations as discussed below. Fig. 4 shows the experimental flames (LHS) and temperature contours of corresponding numerical solutions (RHS). The conventional flame is luminous and shows experimentally a low frequency oscillation, while the flame in hot co-flow turns into bluish and is very stable. The average temperature field in the present study, for both of the two cases present high similarities between the simulation and the experiment. In hot co-flow, the flame becomes "flameless" and shows the features of the HiTAC condition. Even though the input enthalpy of the hot co-flow is much higher than that of the cold co-flow, the predicted peak temperature is about 100 K lower than in the cold co-flow (1790 K). This can be attributed to the enlarged flame zone produced by the flame with low O 2 concentration co-flow, which results in a more uniform temperature distribution. Since the NO X formation is mainly dependent on the peak temperature zone, the lower peak temperature in hot co-flow leads to a remarkably reduced NO X emission as shown in Fig. 18. Fig. 5 shows the predicted SMD of the drops at different elevations for the conventional flame. The experimentally measured SMD under similar conditions is shown in Fig. 6. In the numerical simulation, the SMD ranges from 5 µm to 40 µm at height z = 10 mm, and gradually changes to the range from 10 µm to 50 µm at height z = 40 mm. The measured data show a range from 12 µm to 38 µm at height z = 10 mm, and it changes to a range from 18 µm to 52 µm at height z = 40 mm. The predicted SMD matches the experimental data in the region of high droplet density, which is approximately in the middle of the range at each elevation. The trend of the SMD at each elevation also shows similarity between the simulation and the experiment. However, the range of radial locations of the SMD in the simulation is narrower than those in the experiment. This is attributed to two reasons. Firstly, as what we observed in the experiment, even at higher elevations, some small droplets still exist in the near center area, while few droplets are found in this region in the simulation. Both the low co-flow velocity of 0.29 m/s in the experiment, and the spray model in the simulation can be the causes of the difference. Secondly, we used an atomizer inner diameter of 0.21 mm (see Table 1) in the simulation according to the nozzle design. As we discussed in [12], this parameter is often larger than the design value and it requires further analysis in a specific case. With an adjusted inner diameter, the predicted SMD is supposed to spread in a wider range of radial position and match better with the measured data. Nonetheless, in the present study we keep the inner diameter as a constant of 0.21 mm for comparative investigation.
It is observed that the conventional ethanol spray flame is a diffusion flame, and its main body is attached to the spray by a triple flame composed of a rich premixed flame where the droplets are vaporizing and two lean premixed flames on both sides of the jet, as reported by Ruetsch et al. in [42].
The predicted and measured SMD of the droplets in hot co-flow conditions at various elevations are shown in Figs. 7, and 8,  respectively. At height z = 10 mm, the SMD range and distribution is almost the same as for the conventional flame (i.e. with 21% O 2 and 300 K co-flow conditions). However, in the experiment the small droplets we observed in the conventional flame at higher elevations do not exist in the near-center-line area, but at larger radial locations. This makes the match of simulation results and experimental data of SMD distribution better than the conventional flame. Differently from the conventional conditions, the peak values of SMD at high elevations in hot co-flow do not increase gradually, but they are keeping the same value of approx. 40 µm. Both simulation and experiment show the same phenomena.
In the experiment, the data were measured at elevation z = 10 mm, 20 mm, 30 mm and 35 mm. At higher elevations the droplet number density drops and thus it is more difficult to measure accurately the SMD distribution. However, the less "steep" trend at a higher elevation in hot co-flow conditions shown in the experiment is predicted by the simulation with results at further elevations z = 40 mm and 50 mm. Besides, the left and right ends, which represent small droplets and large droplets, will be discussed in next section with the analysis of droplet number density. It is reasonable for the simulation to predict those small amount of droplets, while in the experiment they are not taken into account due to the Phase Doppler Anemometry (PDA) measuring.
The ethanol spray flame in hot co-flow exhibits a clear conical shape. The relatively larger cone observed in the experiment, compared to the conventional flame, is shown also in the simulation. Based on the obtained results it can be concluded that there is a good match between experimental and numerical data. In further study the ethanol spray flame is investigated under various co-flow conditions. Those results are presented and discussed in the following sections.

Influence of co-flow temperature
In this section, the volume fraction of oxygen in the co-flow is kept constant at 21% and the velocity of the co-flow is kept at 3.5 m/s. The predicted temperature contours of the ethanol spray flames under 300 K, 900 K and 1500 K co-flow conditions are presented in Fig. 9.
In all these conditions, an inner low temperature zone in the flame is observed. With the increase of co-flow temperature from 300 K to 1500 K, the peak temperature rises from 1809 K to 2267 K. The volume of peak temperature zone increases and it extends downstream. As a consequence, the high temperature of co-flow does not lead to a HiTAC condition, but increases the peak temperature and thus the thermal NO X formation. In many other HiTAC applications, a high temperature of combustion air can result in a more uniform temperature distribution and a lower peak temperature. That is due to the increased velocity of combustion air under high temperature conditions. The high velocity leads to a larger amount of entrainment of flue gas, and thus creates a low O 2 concentration "co-flow", leading to a HiTAC-like condition. By increasing the velocity of the fuel, which can enhance the entrainment of flue gas as well, the HiTAC condition is also achievable for gaseous fuel [43]. Fig. 10 shows the axial velocity contours under 300 K and 1500 K co-flow conditions. A higher local velocity of co-flow near the fuel injection is found under 1500 K co-flow condition. This is attributed to increased fuel evaporation rate of the droplets under high co-flow temperatures. The fuel evaporation rate changes can be validated by the droplet number density and SMD distribution below.     11 shows the droplet distribution at various elevations under 300 K, 900 K and 1500 K co-flow conditions. Below the height of z = 20 mm, the droplet concentration is alike for all investigated conditions. Further downstream, it can be noticed that the droplets evaporate faster under high temperature co-flow conditions. In 300 K coflow case some droplets still exist at the height of z = 400 mm and at large radii, while very few droplets can be found already at z = 100 mm under 900 K and 1500 K cases, and no droplet found at radial locations larger than 0.06 m. This is confirmed by the experiment, where it has been observed that some droplets escape from the flame under cold co-flow conditions, while this is not seen under hot coflow conditions. The simulation results show that the droplets present at large radii have the possibility to escape under low temperature coflow conditions. Under 900 K and 1500 K co-flow conditions, the droplets reach the boiling temperature very fast due to the presence of hot co-flow at radii larger than 0.02 m. The evaporation rate is then mainly dependent on the liquid-vapor equilibrium around the droplets. As a result, at radii smaller than 0.03 m, the droplet concentration under 900 K and 1500 K is similar, whereas at larger radii, the droplets evaporate and disappear faster with the presence of 1500 K co-flow.
The SMD trend at downstream elevations is shown in Fig. 12. The peak in SMD values increases at higher downstream elevations for the 300 K co-flow conditions. This effect is caused by rapid evaporation of small droplets. However, for the high temperature co-flow, at each high elevation (radii above 0.02 m from where co-flow affects effectively) both large and small droplets decrease their size quickly and simultaneously, especially at locations of low droplet number density. This results in the vanishing of small droplets inside the spray cone and large droplets outside of the cone at each high elevations. Compared to the 900 K case, the droplets at high elevations (e.g. z = 90 mm) evaporate faster in 1500 K co-flow, and the SMD values in the middle of the profile, where a relatively large droplet number density exists, decrease from about 34 µm to about 30 µm. The profiles for both 900 and 1500 K are a bit wrinkled since a much lower amount of droplets exists at high elevations (see Fig. 11) and thus the statistical error in the captured samples can cause noticeable fluctuations.
Once the evaporated fuel meets the co-flow, a combustion process occurs. The O 2 concentration contours under 300 K and 1500 K co-flow conditions are presented in Fig. 13. A high temperature of co-flow accelerates the combustion process, whereas the low density of co-flow has a positive effect on enlarging the combustion zone. As a result, although the peak temperature increases adequately to the co-flow temperature, the temperature difference between the peak temperature and the co-flow temperature decreases. The increment of peak temperature from the case with 300 K co-flow to 1500 K, is only 458 K (from 1809 K to 2267 K), much less than the increase of absolute co-flow temperature, i.e. 1200 K (from 300 K to 1500 K).

Influence of O 2 concentration in the co-flow
In the present study, the applied co-flow conditions resemble properties of combustion air mixed with flue gas, i.e. high temperature and low oxygen concentration. The simulation results discussed in the previous section, indicate that although the decreased density of coflow (due to a higher temperature) enlarges the flame zone, the increased enthalpy input still results in a higher peak temperature in the  flame and thus leads to more thermal NO X formation. Additionally, a higher co-flow temperature also accelerates the evaporation of droplets along the spray trajectories, which has negative effect on the "delay" of combustion and creating HiTAC-like conditions. Since various studies of HiTAC on gaseous fuel show that the combination of an increased temperature and a lowered O 2 concentration, which is created by entraining flue gas with combustion air/fuel, leads to HiTAC condition, we further investigate the effect of O 2 concentration in the co-flow on spray combustion in this section. Fig. 14 shows the predicted average temperature and O 2 concentration contours of the ethanol spray flames under 18%, 12% and 6% volume fraction of O 2 co-flow conditions. The co-flow temperature is kept at 1500 K and the co-flow velocity remains 3.5 m/s. The peak temperature decreases from 2175 K to 1705 K and 1689 K with reduction of O 2 concentration from 18% to 12%, and to 6% in the co-flow, respectively. An enlargement of combustion zone is also found. Compared to the co-flow temperature, 1500 K, the temperature difference in the flame has been reduced from 675 K to 205 K, and to 189 K. SMD of the drops and their concentration profiles at different elevations under various co-flow O 2 concentrations (18%, 12% and 6%) are presented in Figs. 15and 16, respectively. According to Fig. 15, with the decrease of O 2 concentration in the co-flow, the SMD profile at z = 90 mm is less wrinkled. As we discussed before, the wrinkled profile is mainly due to the low droplet concentration. Thus more droplets still exist at that elevation with a lower O 2 concentration coflow. This is also visible from Fig. 16. Besides, SMD values at location of high droplet concentration for cases with 1500 K and 21%vol O 2 concentration are generally lower than those comparable data for cases of 1500 K and low O 2 content.. This indicates that the evaporation process has been reduced for the low O 2 concentration co-flow. This is confirmed further by the droplet concentration distribution shown in Fig. 16. At elevations z = 10 mm and 20 mm in Fig. 16, the droplet concentration is similar for various O 2 co-flow conditions. From z = 30 mm, it is clear that more droplets survive in a lower O 2 concentration co-flow. As a result, when the evaporated fuel mixes with the co-flow containing a lower O 2 concentration (18%, 12% and 6% co- flow O 2 %vol), this leads to a lower peak temperature in the flame and a reduced consumption of fuel. This combined with a lowered temperature in return slows down the evaporation process of droplets. Thus the volume of combustion zone increases creating a HiTAC-like condition. Based on previous observation, i.e. that a high temperature co-flow leads to a high peak temperature in the flame, we concluded that in the combination of a high temperature and a low O 2 concentration in the co-flow, which can lead to HiTAC-like conditions, reducing the O 2 concentration is the dominant factor for decreasing the peak temperature in the flame. However, the combustion process is continuously consuming O 2 . If no sufficient O 2 in the low O 2 co-flow can be provided for the combustion process, some other process like secondary cracking can occur and soot formation can be increased as well. For gaseous fuels, velocity components or locations of fuel injection can be tuned in a relatively simple way to mix fuel with low O 2 concentration co-flow (mixture of combustion air and flue gas). Spray flames have a fuel-rich zone along or confined by the spray, which makes the mixing more difficult to be optimized towards HiTAC conditions. The optimization needs to be carried out based on the analysis of a specific fuel in order to create a HiTAC-like condition. Fig. 17 shows the peak temperatures in the cases with different O 2 concentration at various co-flow temperature conditions. Each point represents one test case. The condition with a temperature of 300 K and 6%vol O 2 in the co-flow is beyond the lean flammability limit. Under constant co-flow temperature regime, the peak temperature is reduced when the O 2 concentration in the co-flow decreases. Besides, the temperature difference between the peak and co-flow temperature drops with the reduced O 2 concentration in the co-flow. For the same co-flow O 2 concentration, the peak temperature increases with an increase of co-flow temperature. However, the temperature difference between the peak and co-flow temperatures decreases. As a result, among the investigated cases, the one with 6%vol O 2 concentration and 1500 K co-flow condition has the most uniform temperature distribution.
NO average mass fraction contours for conventional (300 K, 21%vol O 2 concentration) and HiTAC (1500 K, 6%vol O 2 concentration) flames are shown in Fig. 18. The former has a peak temperature of 1809 K and the latter of 1689 K according to the simulation results. The NO formation occurs mainly in the region with peak temperature, therefore it is suppressed under HiTAC regime. Considering the extra enthalpy input from the 1500 K co-flow, the NO formation is reduced considerably, since the peak temperature in the regular spray flame under co-flow temperature equal to that of the HiTAC, i.e. 1500 K and 21%vol O 2 concentration is 2267 K.
The peak temperature under various co-flow conditions is related to the flame volume. The flame volume can be explained with support of oxidation mixture ratio [44] in form of Eq. (23): with m the mass fraction, n the stoichiometric ratio, M the molar mass, and index O, F, C representing oxygen, fuel and flue gas respectively. Following [44] we assume that R 0 = 0.99 is representative for the external boundary of the flame and R 0 = 0.01 is representative for the internal boundary. Fig. 19 shows the calculated flame volumes under various co-flow conditions. With the same co-flow temperature, when the O 2 concentration decreases in the co-flow, the ethanol spray flame volume is enlarged, especially in case when the O 2 concentration is lower than 12% vol. Since for the same co-flow temperature, the heat dissipated by the flame with a bigger volume (co-flow with reduced O 2 concentration) is absorbed via a larger combustion zone, the peak temperature in the flame decreases.
With the same constant O 2 concentration in the co-flow, the change in the flame volume caused by a high temperature co-flow is not sufficient to cover the local enthalpy increase in the system and thus the peak temperature is enhanced. For the co-flow with 6%vol and 9%vol O 2 concentrations, when the co-flow temperature increases, the flame volume is enlarged much more than in other O 2 concentration conditions. Since the co-flow velocity is the same for all investigated cases, the increment of enthalpy input due to co-flow with a higher temperature in all O 2 concentration conditions is almost the same. As a result, the ethanol spray flames in the co-flow with 6%vol and 9%vol O 2 concentrations have greater potential than other cases to reach HiTAC conditions. The HiTAC is related to the uniformity of the temperature distribution mainly, and also to the ignition temperature and adiabatic temperature for a specific fuel. Since the 1500 K and 6%vol O 2 concentration case has a peak temperature of 1689 K, with only a difference of 189 K from the co-flow temperature, and shows a "flameless" feature in the experiment, it is then considered to be the HiTAC condition.

Conclusions
We presented results of the numerical investigation of ethanol spray combustion under various co-flow conditions. The different temperature and O 2 concentration content in the co-flow were applied to mimic the mixing conditions of combustion air and flue gas. Their influence on the spray combustion was numerically studied in order to achieve HiTAC-like conditions with uniform temperature distributions and correspondingly low NO X emissions. The present study was based on the analysis of the Delft Spray-in-Hot-Coflow (DSHC) experimental setup and the operating conditions. The experiment was carried out in open space to allow for optical measurement. Thus our defined computational domain also contains the air flow of ambient air, although it was further found that the ambient air showed minor influence on the flame zone.
The numerical results of the cold co-flow case (300 K and 21%vol O 2 concentration) and the hot co-flow case (1500 K and 6%vol O 2 concentration) were compared with the experimental data under the similar co-flow conditions. Good match was shown on the flame profile, and the range and trend of SMD at various elevations. Some deviations were discussed and attributed to limitations of either the experiment or models used in simulation. Generally, good agreement between numerical and experimental data was achieved.
The simulation results showed that although the increased temperature leads to a lowered density of co-flow which then enlarges the flame zone, the increased enthalpy input still results in a high peak temperature in the flame and thus leads to more thermal NO X formation. Moreover, a high co-flow temperature also accelerates the evaporation of droplets along the spray trajectories, which has negative effect on the "delay" of combustion and creation of HiTAC-like conditions. However, the temperature difference between the peak temperature and the co-flow temperature decreases with the increased co- flow temperature.
In some HiTAC applications with gaseous fuels, a high temperature of combustion air can result in a more uniform temperature distribution and a lower peak temperature. This is due to the increased velocity of combustion air under high temperature conditions, with which more flue gas is entrained and it creates a lower O 2 concentration "co-flow", leading to a HiTAC-like condition. According to previous research [43], increasing the velocity of the gas fuel can create the same situations and thus supports achieving the HiTAC conditions. However, if a larger amount of entrainment does not contain a relatively low O 2 concentration, it may strengthen the combustion process rather than reducing the peak temperature and thermal NO X formation.
For ethanol spray flames, a lower O 2 concentration leads to a lower peak temperature in the flame and reduced consumption of fuel. Both in return slow down the evaporation process of droplets. Thus the combustion zone volume increases creating a HiTAC-like condition. The calculated flame volumes in various co-flow conditions were compared and discussed. The ethanol spray flames in the co-flow with 6%vol and 9%vol O 2 concentrations resembles the HiTAC flames mainly due to the considerable enlargement of combustion zone (the enthalpy input from high temperature co-flow remain similar). Since the 1500 K and 6%vol O 2 concentration case has a peak temperature of 1689 K, with only a difference of 189 K from the co-flow temperature, and shows a "flameless" feature in the experiment, it is then considered as a HiTAC condition.
For heavy fuel oils, the mixing process is more difficult to be optimized towards HiTAC conditions than light fuel oils. On one hand, the peak temperature can be reduced in a low O 2 concentration co-flow condition in spray flames; on the other hand, the even lower O 2 concentration condition created after combustion may lead to problems of secondary cracking, soot formation and flame extinction. The optimization needs to be carried out based on the analysis of a specific fuel in order to create a HiTAC-like condition.