Computation of Turbulent Flows
In aeroacoustics, turbulence is a principal source of broadband noise. Therefore, research and development of turbulence modeling and turbulence simulation are an integral part of CAA.
Because of the availability of more and more powerful and faster and faster computers, turbulence, nowadays, becomes a favorite activity of large-scale computation. It is known that direct numerical simulation (DNS) of high Reynolds number turbulent flows requires an exceedingly large number of mesh points and long CPU time. On account of such requirements, DNS is presently not considered feasible for solving practical CAA problems. Recently, attention has turned to LES. However, owing to the three-dimensional nature of turbulence, realistically, LES can be carried out only in relatively small computational domains. Simple estimates of mesh and computer requirements would convince even the most ardent proponents of LES that it would be sometime in the future, when much more powerful and faster computers become available, before LES would become a design tool in CAA.
The appeal of DNS and LES is that they can, in principle, compute the entire or a large part of the turbulence spectrum. But, for noise prediction, it is highly plausible that it is not necessary to know everything about turbulence or the entire turbulence spectrum. How much does one really need to know about turbulence before one can calculate turbulence noise is an open question. It appears that if one’s primary concern is on the dominant part of the noise spectrum, it is very likely that only the resolution of the most energetic part of the turbulence spectrum is necessary.
Calculating the mean flow velocity profile and other mean quantities of turbulent flows is, in general, important for engineering applications. For mean flow calculation, a practical way is to use the RANS equations with a two-equation turbulence model (e. g., the к – є or the к – ы model) or, if desired, a more advanced model.
The question pertinent to CAA is whether models of a similar level of sophistication could be developed for noise calculation. There is no clear answer to this question. It is a matter of intense research at this time.