While flow separations are easy to identify in 2D domains, it is much more complicated to spot them out in 3D domains. For instance, in turbine passages, one often encounter secondary flows and vortexes. It is possible to have negative axial velocity without having separations. Could anyone suggest a criteria to identify separations? Cheers

I can't really speak for turbine passages, but for my dissertation I studied the 3-D flow over a rotating disc in forward flight, which revealed no separation on the advancing side of the disc although the flow was reversed.

You can download my dissertation from my website,

www.microcfd.com/papers.htm

One way to identify 3-D separation is to plot the local traction, which is the local viscous stress tensor dotted with the local outward unit normal to the surface (see p.70, Eq. 4.55). The resulting lines, when scaled to a common length, are called 'tufts' and indicate the local viscous or 'shearing' force (read Section 5.3, p.75).

On the non-rotating disc (p.106), the line of flow separation is easy to identify, because these tufts suddenly change their direction.

For the rotating disc (spin ratio SR = omega * R / V = 1) flow separation can be identified on the receding rim by the crossing of tufts (p.144). Surprisingly, there is no separation on the advancing rim, where it would normally be expected, since the flow is opposite to the free stream. Flow reversal takes place away from the surface (p.136).

Also, in a highly unsteady compressible flow, the traction is not always tangential to the surface (read bottom of p. 70). This is due to the 'stickiness' components of the viscous stress tensor, which act normal to the surface (similar to pressure, but much smaller in magnitude). Notice that I never refer to it as the 'shear' stress tensor, because shear implies tangency.

My suggestion: Plot the local traction over your turbine blades, and see what they look like. If you see a crossing of tufts or sudden reversal in direction, this is where you have separation.

Hi Peter,

Maybe, instead of looking at the velocity field, you can start from the basic definition of separation, i.e., the location where the surface stress changes its sign. Therefore, you should look for surface contours of zero shear.

I hope this helps,

Rami

>Maybe, instead of looking at the velocity field, you can start from the basic definition of separation, i.e., the location where the surface stress changes its sign. Therefore, you should look for surface contours of zero shear.

I don't think you have tried to visualize this whole concept. This is exactly Peter's problem, that your 2-D method does not always apply in 3-D. In 3-D your stress is not always lined up with the major axes, thus a sign change can be a meaningless quantity. The method I described reveals sudden changes in 3-D surface stress along any direction, which clearly identifies separation.

Also, I was not looking at the velocity field (I don't think you carefully read my post). The relative velocity at the surface is zero. However, the traction gives a clear indication of what the velocity field is just above the surface. Thus the comparison to tufts. What is plotted in my dissertation is traction, or 3-D surface stress, if you want to call it that way. In 3-D, the surface stress is a vector. It has both magnitude and direction. Trying to identify zero-vectors is a difficult task, which is why I plotted all tufts the same length, after I was searching for a line of separation within a bunch of dots first.

You can look up papers dealing with surface topology. For example, try the NASA or AIAA paper databases and search for three-dimensional separation, or authors such as Delery, from which you can get references to all the other classical papers dealing with 3D separation. The separation in 3D, at least using the topological definition, is the location where stream surfaces diverge.

Axel,

My idea is actually not very different from yours. I suggested to look at the surface shear stress (2 components in 3D problem). This comprises a local surface vector, i.e, it has a magnitude and an angle to some arbitrary fixed direction (e.g., the direction of the main flow).

I agree that the vector may not become zero (if both components are not zero simultaneously), so separation may be indicated by abrupt change of the vector direction - again, quite similar to your suggestion.