Direct injection (DI) compressed natural gas (CNG) engines are emerging as a promising technology for highly efficient and low-emission engines. However, the design of DI systems for compressible gas is challenging due to supersonic flows and the occurrence of shocks. An outwardly opening poppet-type valve design is widely used for DI-CNG. The formation of a hollow cone gas jet resulting from this configuration, its subsequent collapse, and mixing is challenging to characterize using experimental methods. Therefore, numerical simulations can be helpful to understand the process and later to develop models for engine simulations. In this project, a high-fidelity large-eddy simulation (LES) of a stand-alone injector (Figure 1) is performed to understand the evolution of the hollow cone gas jet, and models are developed for the full cycle engine simulations. 

Figure 1: Initial development of the gas jet in terms of Mach number close to the nozzle exit and classical features of supersonic jets (LES).

The hollow cone gas jet is characterized in terms of several parameters such as axial penetration length, maximum jet width, area of jet, volume of jet, and mixing in terms of the mass-weighted probability density of the injected gas within the jet volume (Figure 2). Different grid resolutions have been used to study the effect on the gas jet behavior as well as mixing. Since Unsteady Reynolds-Averaged Navier-Stokes (URANS) approach is more widely used in industry, applicability of different turbulence models is evaluated. The transient needle opening has been found to impact initial stages of the gas jet formation and is responsible for the linear jet evolution observed in experiments. Moreover, the initial condition within the nozzle has a strong influence on later jet evolution in case of fixed needle simulations [1].

Figure 2: Development of the gas jet and mixing behavior with helium as injected gas (LES).

After gaining detailed insights on the injection process, the next step is to investigate the effects of gaseous direct injection on in-cylinder flow field and mixing. In this work, numerical simulations are supported by Particle-Image Velocimetry (PIV) and Holographic Tomographic Interferometry (HTI) measurements performed on an optically accessible ICE at the Institute of Aerodynamics (AIA). However, simulations of the gas injection processes are very expensive from a numerical point of view due to very small gas passages of the order of micrometers, high Mach number flow, and compressible turbulence. The small length scales accompanied by large velocity scales increase the computational wall time required for the same simulation run time. Additionally, large differences between in-nozzle and in-cylinder length and velocity scales make the simultaneous simulation of both regions computationally expensive. Therefore, time scales of the injection processes need to be decoupled from the in-cylinder flow time scales, which leads to the development of multi-scale models.

A mapped boundary condition (MBC) approach is proposed [2]–[4], where the simulation is split into two parts: nozzle flow and in-cylinder flow. This approach retains the geometrical information of the nozzle and provides accurate boundary conditions for the in-cylinder flow simulation. In this approach, a stand-alone nozzle simulation, either LES or URANS, is carried out for a desired pressure ratio. The nozzle flow and near-nozzle flow reach a steady state within a short time compared to the duration of injection. The injector nozzle is decoupled from the cylinder by slicing it from the computational domain along an arbitrary surface, which is treated as the mapping boundary on the cylinder. Since the flow at the nozzle exit is supersonic due to high-pressure ratio, the downstream flow conditions have no effect on the upstream flow field. The flow variables in the nozzle simulation are recorded at the mapping boundary and then transferred onto the corresponding inflow boundary of the full-scale simulation (Figure 3).

Figure 3: Downstream spatial distribution of Mach number (Ma), temperature (T), and helium mass fraction (Y_He) in Y = 0.0 clip plane and resulting inflow profiles at time t = 75 ms in nozzle simulations for injection pressures 15 bar (left) and 8 bar (right).

The MBC approach is applied in stand-alone LES (Figure 4) as well as URANS simulations of the engine to investigate the in-cylinder velocity and mixing fields (Figure 5). The cycle-to-cycle fluctuations in the measured velocity field are found to be high, presumably due to shocks and their reflections from cylinder walls. The proposed simulation approach can reasonably predict the ensemble-averaged velocity field. The impact of the gas jet from a centrally mounted injector on tumble motion and turbulent kinetic energy was investigated. Gas injection at high injection pressures and lower engine speeds destroys the tumble flow generating high turbulence levels, which benefits the mixing. However, as a result, overall turbulence levels are reduced near the combustion top dead center compared to those without injection, which may lead to slower combustion in a fired engine.

Figure 5: Comparison of velocity magnitude, ensemble-averaged PIV (left) and URANS (right) in Y = 0.0 plane at a crank angle 88° (before SOI), 104°, and 120° ATDC (after SOI); line plots: CA 88°—Z = -10 mm, CA 104°— Z = -15 mm, CA 120°—Z = -15 mm.

The MBC approach provides a considerable reduction of computational wall time, but it is still expensive and dependent on the mesh resolution of the mapped boundary for the accuracy of the mass flow rate. Also, the approach is limited to cases where the cylinder pressure did not vary significantly through the cycle, keeping the pressure ratio across the nozzle largely constant. Therefore, a new fully adaptive model for an outwardly opening poppet-type injector nozzle is developed [5], which is able to capture the in-nozzle flow with reasonable accuracy, reproduce the hollow-cone gas jet behavior, and significantly decrease the computational wall time for full multi-scale simulations. This is achieved by following a quasi-one-dimensional (quasi-1-D) approach for the injector nozzle flow and coupling it to the computational domain through source terms (Figure 6). The model also allows for needle motion (Video 1). In this work, the injector model is developed and validated against LES of nozzle flow for different pressure ratios. The quasi-1-D nozzle-flow model is dynamically coupled to a three-dimensional flow solver through source terms in the governing equations, named as dynamically coupled source model. The dynamically coupled source model is then applied to a temporal gas jet evolution case and a cold flow engine case. The results showed that the dynamically coupled source model can reasonably predict the gas jet behavior in both cases. All simulations using the new model led to reductions of computational wall time by a factor of 5 or higher.

Figure 6: Quasi-1-D modeling of hollow-cone gas injector.