I have seen a lot of papers where people use LES to predict the noise generated like in trailing edge of airfoils.It made me a bit curious.Is it possible for the filtering technique to filter some acoustic data as the acoustic variables are usually of a much less order of magnitude compared to mean flow.Is it possible to do something in the context of LES to overcome this problem.

-Harish

Harish, A suggestion - manually 'cut' your data into the range where you are expecting to find the information you are looking for. This may require a little experimentation before you locate the correct range. Vibration levels are generally, as you say, at a very much smaller scale than bulk flow information. Target range say -1e-5 to +1e-5 & smaller.

Folks will try to tell you that this is 'numeric noise'. I don't believe that this is the case, as the vibration signatures are too perfect for pure numeric noise.

I was just asking out of curiosity since I have been reading about LES being used as a tool.I do not use it though.In fact the magnitude of noise in the fluid dynamic simulations are the usual range of magnitudes in most of the acoustic simulations.

-H

Harish:

In fact the magnitude of noise in the fluid dynamic simulations are the usual range of magnitudes in most of the acoustic simulations.

diaw:

So I've been told.

How could you be absolutely certain about that? Would the values change with different boundary-conditions, different numeric precision etc?

How to 'measure' the real noise & numeric noise correctly? What if you were measuring at the incorrect temporal, or spatial frequencies? What *are* the correct measurement ranges?

diaw...

How could you be absolutely certain about that? Would the values change with different boundary-conditions, different numeric precision etc?

The maximum noise level measured in the take off of supersonic jets are of the order of 140db which gives a pressure magnitude of 1/1000th of atmospheric pressure.BC do not exist if the flow is external.Most boundary conditions are artificial to restrict the size of the domain.Yes aeroacoustic calculations requires proper artificial boundary conditions to terminate the domain or else energy can be reflected back into the domain.

How to 'measure' the real noise & numeric noise correctly? What if you were measuring at the incorrect temporal, or spatial frequencies? What *are* the correct measurement ranges? It is not easy in many cases to separate them.Also most noise generations are from turbulent flows and hence a wide range of temporal and spatial scales have to be covered.

I do not know about the measurement ranges.

How did your research with the eigen values and vectors go.

-H

Based on what you've said, Harish, at the speeds you guys are working at, distinguishing between the real noise vs numeric noise frequencies could certainly become a problem. I was refering to much slower flows than you appear to work at... My work is directly influenced by the near boundaries.

Harish:

How did your research with the eigen values and vectors go.

diaw:

Thanks for all your help in answering my previous questions. The research is going well. I basically read up on the eigenvalue & eigenvector concepts & correlated it with a few other approaches. Most approaches are variations of a similar theme. N-S shows up some interesting things when viewed from an eigenvalue/eigenvalue macro-perspective.

I then decided to use a new strategy to reveal the flow regions in a more direct way. Voila, the N-S suddenly reveal their true colours. The flow domain basically sorts itself out to correspond to the prevailing boundary conditions. I use this approach to look directly into what the flow is doing. When seen on a computer monitor, under very different solvers, many arguments regarding the form of the N-S parabolic, elliptic, hyperbolic etc, are answered.

More to come...

diaw... (Des Aubery)

Based on what you've said, Harish, at the speeds you guys are working at, distinguishing between the real noise vs numeric noise frequencies could certainly become a problem. I was refering to much slower flows than you appear to work at... My work is directly influenced by the near boundaries.

At the lower speeds it is possible to split the flow solution into a mean flow part and an acoustic part.The mean flow is assumed incompressible and solved using traditional CFD techniques.The acoustic part is obtained by subtracting incompressible N-S from compressible N-S and the resulting equation is solved using higher order schemes.Both of the methods use different grid sizes and an interpolation is used for forcing the acoustic part using mean flow.

Regarding Eigen vectors Try solving the unsteady incompressible N-S and steady incompressible N-S you might observe a lot.Former has an elliptic parabolic parabolic character and the latter has an elliptic hyperbolic character I think.

diaw:

Thanks for the observations on the bulk (mean) & acoustic components. I have developed a slightly modified way to do this, & to let all components be solved together. It does end up with rather a few simultaneous equations in the end, but with no approximations whatsoever - all non-linearities are retained. I'll get down to coding it up someday.

Harish:

Regarding Eigen vectors Try solving the unsteady incompressible N-S and steady incompressible N-S you might observe a lot.Former has an elliptic parabolic parabolic character and the latter has an elliptic hyperbolic character I think.

diaw:

The big problem is that if a characteristic equation is derived for the full N-S, then a formally parabolic result is obtained. This concurs with the work of other authors. This seems to be natural to the 'shape' & apparent symmetry of the N-S. The result is the prediction of a single 'real steady-state characteristic' at each point in the x-y plane - if it exists. There are additional related phenomena... more to follow...

The problem is what happens to this characteristic in the flow domain itself. In some regions, definite boundary-layer parabolic behaviour is clearly evident, in other areas, a very different 'parabolic' behaviour is observed. It has a different 'nature'.

The formal 'parabolic' label does not do these additional behaviours justice. Perhaps the Mathematically gifted amongst our community can shed some further insights on this phenomenon.

diaw...

After re-reading my last posting & your 'elliptic parabolic', 'elliptic hyperbolic' references, I got to thinking... The reason has now clicked in my head at to why I'm observing the different 'parabolic' nature. The spatial description goes part of the way, the remainder lies with the temporal term. Good...

diaw...

and...

A sound pressure level of 94dB represents a pretty load noise - load enough to make you shout to be heard at close range. In pressure terms this is 1Pa (RMS). Not a lot.

I am sure that someone may need to do more than shout over 94dB. That limit is used as the upper noise threshold in some countries for vehicular noise (perhpas lower by now). A motorcycle emitting that sound level is a noisy beast indeed. Ask me, I was fined for such an infraction when young...

I think Ipod generated a much higher noise level and apple was forced to change it by patching the maximum allowed noise level.