2.1 Control equations

Control equations, including the mass conservation equation, momentum conservation equation, energy conservation equation and component mass conservation equation, are listed in (1)-(6):

(1) \begin{equation}\frac{\partial \rho }{\partial t}{\rm +}\frac{\partial \left(\rho u\right)}{\partial x}{\rm +}\frac{\partial \left(\rho v\right)}{\partial y}{\rm +}\frac{\partial \left(\rho w\right)}{\partial z}{\rm =0\ }\dots\end{equation}

(2) \begin{equation}\frac{\partial \left(\rho u\right)}{\partial t}{\rm +div}\left(\rho u{\mathbf u}\right){\rm =-}\frac{\partial p}{\partial x}{\rm +}\frac{\partial {\tau }_{xx}}{\partial x}{\rm +}\frac{\partial {\tau }_{yx}}{\partial y}{\rm +}\frac{\partial {\tau }_{zx}}{\partial z}{\rm +}F_x{\rm \ }\dots\end{equation}

(3) \begin{equation}\frac{\partial \left(\rho v\right)}{\partial t}{\rm +div}\left(\rho v{\mathbf u}\right){\rm =-}\frac{\partial p}{\partial y}{\rm +}\frac{\partial {\tau }_{xy}}{\partial x}{\rm +}\frac{\partial {\tau }_{yy}}{\partial y}{\rm +}\frac{\partial {\tau }_{zy}}{\partial z}{\rm +}F_y\dots\end{equation}

(4) \begin{equation}{\rm \ }\frac{\partial \left(\rho w\right)}{\partial t}{\rm +div}\left(\rho w{\mathbf u}\right){\rm =-}\frac{\partial p}{\partial z}{\rm +}\frac{\partial {\tau }_{xz}}{\partial x}{\rm +}\frac{\partial {\tau }_{yz}}{\partial y}{\rm +}\frac{\partial {\tau }_{zz}}{\partial z}{\rm +}F_z{\rm \ }\dots\end{equation}

(5) \begin{equation}\frac{\partial \left(\rho T\right)}{\partial t}{\rm +div}\left(\rho {\mathbf u}T\right){\rm =div}\left(\frac{k}{c_p}{\rm grad\ }T\right){\rm +}S_T\dots\end{equation}

(6) \begin{equation}\frac{\partial \left(\rho c_s\right)}{\partial t}{\rm +div}\left(\rho {\mathbf u}c_s\right){\rm =div}\left(D_s{\rm \ grad\ }\left(\rho c_s\right)\right){\rm +}S_s\dots\end{equation}

2.2 Turbulence model

The Shear-Stress Transport (SST) $k - \omega $ model has seen fairly wide application in the aerospace industry. The transport equations for the kinetic energy $k$ and specific dissipation rate $\omega $ are

(7) \begin{equation}\frac{\partial \left(\rho k\right)}{\partial t}+\nabla \cdot \left(\rho k\overline{{\mathbf u}}\right)=\nabla \cdot \left[\left(\mu +{\sigma }_k{\mu }_t\right)\nabla k\right]+P_k-\rho {\beta }^*f_{{\beta }^*}\left(\omega k-{\omega }_0k_0\right)+S_k\dots\end{equation}

(8) \begin{equation}\frac{\partial \left(\rho \omega \right)}{\partial t}+\nabla \cdot \left(\rho \omega \overline{{\mathbf u}}\right)=\nabla \cdot \left[\left(\mu +{\sigma }_{\omega }{\mu }_t\right)\nabla \omega \right]+P_{\omega }-\rho \beta f_{\beta }\left({\omega }^2-{\omega }^2_0\right)+S_{\omega }\dots\end{equation}

where $\bar{u}$ is the mean velocity, $\;\mu $ is the dynamic viscosity, ${\sigma _k}$ and ${\sigma _\omega }$ are model coefficients, ${P_k}$ and ${P_\omega }$ are production terms, ${f_{{\beta ^*}}}$ is the free-shear modification factor, ${f_\beta }$ is the vortex-stretching modification factor, ${S_k}$ and ${S_\omega }$ are the user-specified source terms and ${k_0}$ and ${\omega _0}$ are the ambient turbulence values that counteract turbulence decay.

2.3 Near-wall treatment

The near-wall area can be roughly divided into three layers. The innermost layer adjacent to the wall is the viscous sublayer; the outer layer is the log-law layer; between them, is the buffer layer. In STAR-CCM+, the continuous functions which are called blended wall functions (all y+ wall treatment) are used to cover all three sublayers. They represent the buffer layer by appropriately blending the viscous sublayer and the log layer.

2.4 VOF model

The VOF model is a tracking technique for two or more immiscible fluid interfaces on a fixed Eulerian grid. In the calculation equation of the VOF model, each phase fluid shares a system of equations, and the volume fraction of each phase is tracked throughout the computational domain. In each control volume, the sum of the volume fractions of all phases is one. As long as the volume fraction of each phase at each point in the calculation domain is known, the fields of all variables and physical properties are shared by the phases and represent the volume average. Thus, depending on the value of the volume fraction, the variables and physical properties within any unit are either representatives of one phase or representative of a multiphase mixture. By solving the continuity equation for one or more phase volume fractions, the interface between the phases can be tracked. The continuity equation for the q-phase volume fraction is

(9) \begin{equation} \frac{1}{{\rho }_q}\left[\frac{\partial }{\partial t}\left({\alpha }_q{\rho }_q\right)+\nabla \cdot \left({\alpha }_q{\rho }_q\overrightarrow{v_q}\right)=S_{{\alpha }_q}+\sum^n_{p=1}{\left({\dot{m}}_{pq}-{\dot{m}}_{qp}\right)}\right]\dots\end{equation}

where ${\alpha _q}$ is the fluid volume fraction of the q-phase, ${\rho _q}$ is the physical density of the q-phase, $\overrightarrow {{v_q}} $ is the velocity of the q-phase, ${\dot m_{qp}}$ is the mass transfer from phase q to phase p, ${\dot m_{pq}}$ is the mass transfer from phase p to phase q, and ${S_{{\alpha _q}}}$ is the source term with a default value of zero and can also be specified as a constant or user-defined quality source term.

Phase interface tracking is the focus of two-phase flow simulation. An important quality of a system of immiscible phases (for example, air and water) is that the fluids always remain separated by a sharp interface. The High-Resolution Interface Capturing (HRIC) scheme is designed to mimic the convective transport of immiscible fluid components, resulting in a scheme that is suited for tracking sharp interfaces.

2.5 Gridding strategy and refinement

A trimmed grid was generated in the computational domain using the built-in meshing module of STAR-CCM+ software. Since a very fine mesh was required in the VOF model to accurately capture the free boundary surfaces of fluids between different phases, the mesh in the areas where the free boundary surfaces of fluids between different phases exist in the calculation domain was refined. The refinement process of the grid was artificially processed during the calculation process. Both the validation case and the simulation of LJIC at real engine operating condition employed the same gridding strategy and refinement method.

2.6 Model selection

The calculation used an implicit unsteady solver, and the time step was set to be 0.1$\mu s$ (1 × 10^–7s). The turbulence model adopted the SST k–ω double-equation eddy-viscosity model, and the near-wall area adopted all y+ wall treatment. The multiphase model adopted the VOF model to capture the accurate free boundary surface.

3.1 Calculation domain

In the experiment, the crossflow pipe had a rectangle cross-section with dimensions of 25.8mm $ \times $ 28.9mm $ \times $ 100mm (height $ \times $ width $ \times $ length). In the calculation, to save calculation time, the setting of the calculation domain was smaller than the real experimental pipeline, which was 25.8mm $ \times $ 2.54mm $ \times $ 25.4mm (height $ \times $ width $ \times $ length) and is shown in Fig. 2. The axial location of the nozzle was 5.08mm after the crossflow inlet, and the liquid used in the experiment was acetone.

Figure 2.

Schematic of calculation domain.

3.2 Calculation grid

The refined mesh is shown in Fig. 3. After the final refinement, the finest grid size was 0.00635mm, and the size of the grid was approximately 22.1m.

3.3 Boundary conditions

Since the computing domain used a smaller width to save computing resources, the wall boundaries on both sides of the computing domain were set to periodic boundaries. The liquid was set to acetone in the simulation to be consistent with the test. The detailed boundary conditions for crossflow and liquid jet are presented in Table 2.

Table 2

Boundary conditions of crossflow and liquid jet

Figure 3.

Mesh at centre plane and zoomed view of mesh refinement.

Since the crossflow velocity distribution in the wind tunnel had been measured in the reference, here the crossflow velocity distribution in the height direction at the air inlet boundary used the fitted experimental data, as shown in Fig. 4. Considering that the width of the calculation domain was small, the crossflow velocity distribution in the width direction was assumed to be uniform. At the same time, as the complete geometry of the nozzle was given, the nozzle was modelled and simulated before the simulation of LJIC, and the detailed velocity distribution at the nozzle exit was obtained and set as the liquid inlet boundary condition. The contour map of the velocity distribution at the nozzle exit is shown in Fig. 5. As can be seen from the figure, the velocity in the central area of the liquid jet was about 8.5m/s, which was higher than the average velocity of 7.1m/s that was calculated based on the volume flux and the nozzle area. Conversely, the liquid velocity near the nozzle wall area was lower than the average velocity.

Figure 4.

Fitted velocity profile at injection plane.

Figure 5.

Contour map of velocity distribution at nozzle exit.

3.4 Results and discussion

The unsteady calculation reached a relatively stable state after about 2.8ms, and then the time-averaged liquid jet trajectory was obtained after a further 1ms of calculation. The contour map of volume fraction of acetone in the centre plane of the calculation domain is shown in Fig. 6. It can be seen from this figure that the liquid jet was uniformly cylindrical at the exit of the nozzle. With its development in the height direction, under the action of crossflow aerodynamics, the jet was gradually deflected downstream.

Figure 6.

Contour map of volume fraction of acetone in centre plane (t = 3.83ms).

It can be seen from Fig. 6 that the penetration height of the liquid jet was only about one-third of the height of the crossflow channel. Considering the vertical velocity distribution tested in the experiment, it can be inferred that the crossflow velocity corresponding to the root of the liquid column was lower than the average value of the crossflow velocity, and the crossflow velocity corresponding to the curved and breakup region at the top of the liquid column was greater than the average value of the crossflow velocity. The distribution of the crossflow velocity along the height direction causes the change in crossflow and liquid column velocity discrepancy, and the discrepancy in crossflow and liquid column velocity was the cause of K–H instability, which directly affected the breakup of the liquid column.

Figure 7 shows the boundaries of the liquid jet (the iso-surface with a volume fraction of liquid of 0.5) at different viewing angles. It can be seen from the figure that the round jet gradually became flat under the aerodynamic effect of crossflow, and the width of the liquid column on the windward side gradually increased in the Y direction. The K–H instability waves developed along the surface of the liquid column until the wavelength corresponded to the width of the liquid column. Because of the blockage of the liquid column, the crossflow decelerated and stagnated on the surface of the liquid column, causing the pressure on the windward surface of the liquid column to be greater than the pressure on the leeward side, as shown in Fig. 8. Under the effect of the air pressure difference, a bag-shaped liquid membrane was formed. As the distance along the liquid column increased, the bag expanded in the crossflow direction, and then gradually began to break up at the tip of the liquid column. The droplets formed by the membrane were very small because the membrane was very thin at the point of breakup. Subsequently, the ring-shaped region with a larger diameter at the edge of the bag broke at the wave node. The two semi-circular liquid columns formed after the disconnection deformed into large droplets under the surface tension, then spread out to each side. The droplets formed by the node itself were also large, and these big droplets flowed downstream along the trajectory of the jet centre.

Figure 7.

Various views of boundaries of liquid jet (t = 3.83ms).

Figure 8.

Contour map of gauge pressure in centre plane (t = 3.83ms).

A comparison of the simulation result of the liquid jet trajectory and penetration with the experimental photo is illustrated in Fig. 9. Although the simulation accurately described the main characteristics of the bag breakup, including the fluctuation of the surface of the liquid column, the initial formation and deformation of the bag, and the breakup of the wave node, compared with the experimental photo, the simulation still cannot capture the detailed features of the membrane near the deformation limit and breakup.

Figure 9.

Comparison of simulation result with experiment photo (t = 3.83ms).

Wang et al.^{(Reference Wang, Huang, Wang and Liu19)} investigated the bag breakup process of round liquid jets in crossflows. The formation and breakup process of bags in a water jet is shown in Fig. 10. The figure shows that, although in the early stage of bag development, the membrane on the tip of the bag was very thin. At this position, however, in the current study’s numerical simulations, when the thickness of the membrane was less than the minimum mesh size, for the VOF model, the liquid phase was no longer continuous from a mathematical perspective. If the liquid phase is not continuous, the force balance between the pressure divergence and surface tension cannot be built up, hence the membrane cannot be formed. Furthermore, it was very difficult to accurately capture the precise phase boundary and find the right normal direction to establish the force balance of the membrane in a turbulent transient simulation. How to simulate all the details in bag breakup still needs further investigation.

Figure 10.

Bag breakup process of water jet (Wang et al.^{(Reference Wang, Huang, Wang and Liu19)}).

An intuitive comparison of liquid jet trajectory is shown in Fig. 11. It can be seen from the figure that when accurately simulating the velocity distribution of liquid jet and crossflow under the same experimental conditions, the numerical method used in this article can precisely simulate the trajectory and penetration of LJIC. In the range of $x/{d_j} < 20$, the error between the simulated LJIC trajectory and the experimental measurement was less than ±5%.

Figure 11.

Comparison of liquid jet trajectory (experiment data from Stenzler et al.^{(Reference Stenzler, Lee and Santavicca18)}).

The results show that the CFD is valid for predicting jet trajectory and main characteristics of the breakup but not for predicting the finer flow features associated with jet breakup.

4.1 Calculation domain

Although the main stage passage was actually an annular channel, to show the details of the flow conveniently, the sector calculation domain was transformed into a rectangular calculation domain. The size of the final calculation domain was 10mm $ \times $ 5.1mm $ \times $ 6mm (length $ \times $ height $ \times $ width), as shown in Fig. 13. The axial position of the nozzle was 5mm after the crossflow inlet.

Figure 13.

Schematic of calculation domain.

4.2 Calculation grid

It is generally believed that when droplet size is less than 0.01mm, the droplet tends to follow the airflow because of its small mass, which has little influence on the main breakup characteristics of the liquid jet. Considering that the main objects studied in this paper are liquid jet trajectory and primary breakup characteristics, the minimum grid size is set to 0.01mm here. The refined mesh is shown in Fig. 14. After the final refinement, the size of the grid was approximately 13.7m.

Figure 14.

Mesh at centre plane and zoomed view of mesh refinement.

4.3 Boundary conditions

The detailed boundary conditions for crossflow and liquid jet are shown in Table 3.

Table 3

Boundary conditions of crossflow and liquid jet

The crossflow inlet velocity distribution used a fully developed turbulent velocity distribution, which was obtained from the simulation of a channel 200 times longer than the height of the passage, as shown in Fig. 15. The geometry of the nozzle used a structure similar to that of the validation case, and the velocity distribution at the nozzle exit is shown in Fig. 16.

Figure 15.

Velocity profile at injection plane.

Figure 16.

Contour map of velocity distribution at nozzle exit.

4.4 Results and discussion

The calculation was performed in 0.19ms.

4.4.1 Flow-field structure

Figures 17 and 18 show the velocity distribution at the centre plane and Y = 1.25 d _j plane, respectively.

Figure 17.

Velocity distribution at centre plane (t = 0.19ms).

Figure 18.

Velocity distribution at Y = 1.25 d _j plane (t = 0.19ms).

It can be seen from the figures that, because of the blocking effect of the liquid jet, the crossflow decelerated and stagnated on the windward side of the liquid jet, and the static pressure increased. The crossflow formed a small corner recirculation zone (circle in Fig. 17) at the front side of the liquid jet root. The stagnated crossflow accelerated along the windward side of the liquid jet; one part flowed up and one part flowed around the liquid column. A large area of low speed was formed downstream of the liquid column, which was full of irregular vortices of various sizes.

Figure 19 shows the temperature distribution at the centre plane. Because the temperature of the liquid jet was low and the temperature of the crossflow was high, the interaction between the liquid jet and the crossflow was accompanied by heat exchange. But since the calculated time scale was very small, the degree of heat exchange was relatively low, and the temperature distribution also showed the distribution of the liquid spray in the flow field.

Figure 19.

Temperature distribution at centre plane (t = 0.19ms).

4.4.2 Breakup mechanism

Figure 20 shows the boundaries of the liquid jet at different viewing angles. It can be seen that, at real engine operating conditions, the breakup mode of LJIC was surface shear breakup. The liquid column became flat under the aerodynamic force of the crossflow and bent downstream. The liquid column showed a small amount of deformation in the width direction. As the airflow accelerated on the surface of the liquid column, the velocity difference between the liquid column and the airflow on the gas–liquid boundary gradually increased, which caused the liquid surface to fluctuate because of K–H instability. The waves developed along the surface of the liquid column upwards and to both sides, forming a thin liquid film on the sides of the liquid column, and finally breaking into small droplets under the action of shear force. The droplets formed on the sides of the liquid column were very small (less than 0.01mm in diameter), and the current grid size cannot show the subsequent movement of such small droplets in detail.

Figure 20.

Various views of boundaries of liquid jet (t = 0.19ms).

The contour map of the volume fraction of fuel in the centre plane is illustrated in Fig. 21. The time interval of each figure was 0.01ms. This figure shows the full development period of a surface wave, from the initial formation to the breakup at the tip of the liquid column. The initial wave was small, and the disturbance gradually increased as the airflow accelerated on the surface of the liquid. However, because of the surface tension of the liquid, the turbulent structure of the liquid surface eventually formed a vortex structure. The airflow decelerated and stagnated within the vortex, the static pressure increased and a pressure difference was formed on the liquid vortex. The pressure differential drove the vortex to develop rapidly on the liquid surface, and finally broke into large droplets at the top of the liquid column, moved downstream along the edge of the liquid plume, and further broke up into small droplets under the action of aerodynamic force. The observation of the development of surface waves can provide references for subsequent experimental measurements, such as that an appropriate high-speed camera response frequency is needed to capture the characteristics of surface waves.

Figure 21.

Contour map of volume fraction of fuel in centre plane.

4.4.3 Trajectory of the LJIC

Figure 22 shows a comparison between the simulation result and the correlation predictions of the liquid jet trajectory.

Figure 22.

Comparison between simulation result and correlation prediction of liquid jet trajectory.

At position $x/{d_j}=12$ downstream of the nozzle, the penetration height of the liquid jet was about two-thirds of the height of the annular channel. Because the range of experimental conditions corresponding to different correlations and the boundary conditions of the velocity distribution of crossflow and liquid jets are very different, the results obtained by calculating the working conditions of this study with different correlations also show a big variation. This is not to say that the accuracy of these correlations is not good, but that it may not be suitable for the working conditions of this study. Compared with many correlations for predicting the trajectory of liquid jets, the simulation result was closest to the correlation of Ragucci et al.^{(Reference Ragucci, Bellofiore and Cavaliere6)}, and the accuracy of the simulation result is yet to be verified by subsequent experiments. Table 4 presents a comparison of the operating range provided by Ragucci and the simulation of this research. The trajectory of LJIC is affected by various parameters as presented in Table 4, as well as the velocity profile of both airflow and liquid jet. It can be seen from the comparison in Table 4 that the magnitude of $q$ and T is closer, and there is a great difference of $W{e_g}$ and P. It can be preliminarily concluded that $q$ and T have a greater impact on the trajectory of the liquid jet under typical aero-engine working conditions. But this is only a preliminary conclusion, which needs to be verified by experiments.$\;$

Table 4

Comparison of operating range