There are two fundamentally distinct approaches to the computational prediction of turbulent flows. First, the ‘direct’ route is to compute all the details of the time-evolving turbulent eddies and motions within the flow. This is referred to as Turbulence Simulation. Weather prediction is a variant of this direct route. Because the range of eddy dimensions within any practically relevant flow typically spans 3-5 orders of magnitude, this approach requires the use of extremely fine numerical grids, containing billions of nodes, and hence billions of coupled equations. This route thus tends to be very expensive, requiring the use of powerful supercomputers.
The alternative route is to adopt a statistical formulation, wherein the effects of the turbulence on the gross flow characteristics – rather than the turbulence structure itself – are predicted by means of a statistical model. This allows the fundamental time-dependent evolution equations – the Navier-Stokes equations - to be simplified to a set of steady-state equations which represent the statistical (time-averaged) state of the flow. This route, referred to as Turbulence Modelling, is far more economical than simulation, because it does not require the multi-scale field of eddies to be resolved. However, the penalty is that the predicted flow is only approximate, the errors often being difficult to quantify, and the method also requires the formulation and computational implementation of elaborate statistical closure models, which consist of complex partial differential equations. One of several textbooks dealing with this highly challenging aspect of computational fluid dynamics is that shown on the RHS of this text.
91桃色 on modelling and simulating turbulent flows spans a huge range of topics, from fundamental studies on the physics of turbulence and its computational characterisation, to the application of complete algorithms to real-world flows, including those around aircraft, cars and trains, within a large number of engineering components and systems (e.g., jet engines, wind/water/steam turbines, pumps, ducts, vessels, combustors, buildings…) and many environmental flows (e.g., weather systems, oceans, lakes, rivers, ground-water systems, oil and gas wells, CO2 sequestration, magma flows, avalanches….).
Some of the past research of the writer is summarised under 91桃色 overview. The emphasis of thi research has been on the simulation of flows via Large Eddy Simulation (LES), Direct Numerical Simulation (DNS) and hybrid LES-RANS methods, in which “RANS” stands for “Reynolds-averaged Navier Stokes” and is, essentially, equivalent to turbulence modelling outlined above. LES is a cheaper variant of DNS, in that it only resolves the large-to-medium-sized eddies, while smaller (less influential) eddies are represented by a statistical approximation. Hybrid LES-RANS schemes are used in circumstances in which a full LES is deemed to be too costly. In this case, a turbulence model is applied to the near-wall layer, in which the turbulence structure is dominated by small-scale eddies, and this is coupled to the LES away from the wall.
After more than a century of research on RANS modelling, efforts to develop new model forms and improve existing models have declined markedly after about 2005. As a consequence of rapid advances in computer technology and thus steep rise in computing power, the proportion of practical flows simulated with LES has increased steadily. However, RANS models remain extremely important, because the large majority of engineering flows continue to be routinely computed with such models.
Fully-resolving turbulence simulation (DNS) is pursued, on the one hand, in order to gain insight into complex turbulence phenomena in a variety of generic flows and, on the other hand, to identify the capabilities and limitations of RANS-based techniques, in particular, when applied to complex, industrially-relevant flows.
In recent years, significant research efforts have targeted topics relevant to the control of near-wall flows, either to avoid separation, with pulsed synthetic jets, or to achieve drag reduction. These topics are especially pertinent to civil aviation under the general heading of "green technologies". In the area of drag reduction, particular attention has focused on turbulence mechanisms and benefits arising from the imposition of transverse oscillatory wall motion (an "active" method) and the use of wavy surfaces (a "passive" method). As part of this research area, gaining insight into the influence of large-scale coherent turbulent structures in the outer part of the near-wall layer on the near-wall region, and thus on the drag-reduction effectiveness, is the objective of ongoing efforts by the flow-control community.
Much of the recent research outlined under 91桃色 overview on flow and turbulence control has involved the use of DNS performed on large national (UK) supercomputer facilities - HECTOR and ARCHER, in particular - with simulations typically run on thousands of CPUs in parallel, generating many terabytes of data. In addition, DNS databases generated at extremely high cost elsewhere (e.g. in Spain and the USA) on highly turbulent flows in channels have been exploited in an effort to understand key fundamental turbulence mechanisms that impact on friction drag and thus on the ability of active and passive control strategies to reduce the drag.
Examples of DNS studies relevant to flow control. Left: pulsed jets discharged into a cross flow designed to increase turbulence in an effort to avoid flow separation at the upper wall; right: the streaky structure in a flow (moving to the right) close to a rigids wall oscillating in the spanwise (vertical) direction, designed to reduce friction drag.
