102 research outputs found
Effects of turbulence on variations in early development of hydrogen and iso-octane flame kernels under engine conditions
No full textThe understanding and prediction of the early development of flame kernels are of high practical importance for the robust relight of aviation gas turbines and the control of cycle-to-cycle variations (CCV) of spark-ignition engines. CCV are known to correlate strongly with early flame kernel development and complicate the optimization of such engines in terms of safety, thermal efficiency, and engine emissions. The flame kernel initiated by a spark is initially small, in the very early combustion phase typically smaller than the size of the turbulent integral length scales. Therefore, the development of the flame kernel is dominated by local, intermittent flow fluctuations and can vary under the same nominal conditions. In this study, the effects of turbulence on the early development of premixed iso-octane and hydrogen turbulent flame kernels under realistic engine conditions are investigated through direct numerical simulations. Multiple realizations were simulated under the same nominal conditions for both fuels. Significant variations in flame kernel interactions with turbulence can be identified among different realizations. The fuel consumption rate varies by a factor of two, which is remarkable considering that only statistical differences in the local flow field are present between different realizations. Effects of different flow features of the initial flow fields on the flame kernel development were analyzed. It was found that the flow motion on the scale of the ignition radius, specifically the fluid deformation, which is characterized by the invariants of the strain rate tensor, determines the global shape of the kernel, while the variations of the kernel growth rate are mostly driven by the variations of the smallest turbulent scales. In particular, turbulence influences the flame surface area growth mainly through the tangential strain rate at the flame surface, which is shown to result from the small-scale turbulent motion. Due to differential diffusion effects, hydrogen and iso-octane exhibit significantly different flame responses to curvature, which is comprehensively studied for both fuels. The findings in this study will guide the development of combustion models that are capable to capture variations of the early flame kernels based on the local turbulence dissipation rate.</p
An alternative definition of order dependent dissipation scales
No full textWhile Kolmogorov's similarity hypothesis suggests that velocity structure functions scale with the mean dissipation \left and the viscosity , we find that the even order scales with \left. This implies that there are other cut-off lengths than the Kolmogorov length . These cut-off lengths are smaller than and decrease with increasing order and Reynolds-number. They are compared to a previous definition of order dependent dissipative scales by Schumacher~et.~al\cite{schumacher2007asymptotic}
Dissipative Range Scaling of Higher Order Structure Functions for Velocity and Passive Scalars
No full textDifferently to Kolmogorov's second similarity hypothesis, we find that the 2n-th order velocity and scalar structure functions scale with n-th order moment of the energy dissipation and the scalar dissipation, respectively. The origins of this scaling are analyzed by the transport equations of the fourth order velocity and scalar increment moments and by direct numerical simulations
Statistics and scaling laws of turbulent scalar mixing at high Reynolds numbers
No full textIn this thesis the turbulent mixing of a passive scalar and its Reynolds number dependence is studied by means of highly-resolved direct numerical simulation (DNS). The passive scalar is advected by statistically homogeneous isotropic turbulence and a uniform mean gradient is imposed on the scalar field which induces an anisotropy of scalar statistics. The Taylor microscale based Reynolds number is varied between 88 and 29. It is customary to examine turbulent flows by means of two-point statistics, because they capture the dynamics of the non-local structure that is inherent to turbulence. First, an equation for the even moments of the scalar increment is derived from first principles that generalizes Yaglom's equation to statistically homogeneous but not isotropic scalar fields. This equation is interpreted as a scalar scale-by-scale energy budget equation that incorporates the energy flux through a sphere of radius and is analyzed by means of DNS and in a filtered form in the context of large eddy simulations. Thereby, it is found that modeling the subgrid transport by an eddy-viscosity approach is suitable for statistically homogeneous turbulence to correctly predict the energy transport. The generalized Yaglom equation serves as a starting point for further analysis of the Reynolds number dependence of the flatness of scalar increments. To this end, the evolution equation for the fourth order scalar increment is analyzed. It is shown that the non-universality of higher order moments originates from a coupling of dissipative effects to the inertial subrange. Based on this finding a normalization of the fourth order structure function is proposed that makes the curves of the normalized fourth order structure function collapse independently of Reynolds number. The information about the local structure of turbulent flows is lost by taking ensemble averages over fixed separation distances in the sense of the Kolmogorov-Obhukov-Corrsin theory. This issue may be overcome by an approach that decomposes the turbulent field along a straight line into so-called turbulent line segments. The decomposition is based on the local extremal points of the scalar field so that within each individual line segment the scalar value varies monotonously. By this approach the linear separation distance between adjacent extreme points becomes an intrinsic stochastic quantity that is determined by the turbulent field itself. The decomposition is one-dimensional and can be easily related to conventional two-point statistics. The line segments are parameterized by their length and the scalar difference between the end points, and additionally, by the scalar mean gradient. A statistical analysis of line segments based on these parameters is conducted. The respective probability density functions (pdf) are computed and resulting conditional moments are compared with focusing on the Reynolds number dependence. The marginal length pdf becomes Reynolds number independent when normalized by the mean length. The mean length is shown to scale with the Kolmogorov length and, additionally, this scaling law is derived theoretically. While the marginal pdf of the mean length obeys a quasi-universal distribution, this is not the case for the distribution functions of the scalar difference or the mean gradient. Their tails are exponential or stretched-exponential and the non-Gaussianity becomes more pronounced toward small scales or for rising Reynolds numbers. In a next step, the universality of small scales is examined by line segments and conventional statistics, where Kolmogorov's phenomenology is adapted to the method of line segments. The conditional moments of exhibit a clear inertial subrange scaling. Scaling exponents of the conditional moments are computed in order to analyze intermittency effects. Based on conditional statistics it is shown that an intermediate length scale has a major contribution to the mean gradients of line segments and that a scale similarity between the moments of mean gradients and the moments of the local gradients exists. Using this result, a presumed pdf is proposed to compute gradient statistics based on the principle of decomposition and reconstruction of line segments. Additionally, the method of line segments describes the local structure of turbulence. This understanding leads to a novel description of the physics behind cliff-ramp structures and provides a well-defined estimate for the length scale at which large cliff-like structures occur
Scale-by-scale Statistics of Passive Scalar Mixing with Uniform Mean Gradient in Turbulent Flows
No full text
Dissipative range statistics of turbulent flows with variable viscosity
No full textInternational audienc
Acceleration of complex high-performance computing ensemble simulations with super-resolution-based subfilter models
No full textDeep Learning at Scale for Subgrid Modeling in Turbulent Flows: Regression and Reconstruction
No full text- …
