I know it has been a while, but I thought that I would share the alternative solution that I came up with to solve this problem, in case it helps someone in the future, and also in hopes that a more mature version of this could be included as a feature in a future release. I ended up creating a template TurboGrid state file for my blades, and I have a small script to extract the CAD geometry into curve files for each individual blade that I am interested in studying. Because the geometry perturbations are mostly localized to the leading edge and the first ~20–25% of the chord, and are usually small, the same mesh template works for all my cases. Although the pitch is the same, the periodic wall gets reshaped for each blade, which results in fairly large gaps when I try to assemble the model in CFX. I did some preliminary studies with multiple domains and GGIs between boundaries, but I found that this was an unsatisfying solution and the gaps were a source of considerable numerical inaccuracy.
To get around this, I actually didn't stop using TurboGrid. Instead, I made a small MATLAB script that computes displacement functions that map each node on the perturbed periodic boundary onto the nominal periodic boundary. Then, I have a template CCL file that I fill in with the output from the script, which can be loaded into a small CFX model with mesh deformations turned on. CFX lets you drive a Specified Displacement mesh-motion BC with a 3-D interpolation FUNCTION keyed on the initial node coordinates. So I precompute a per-node displacement that moves every node of the damaged passage's periodic wall exactly onto the target (nominal) periodic surface, ship it as an interpolation table, and ramp it in over a fluids-off pseudo-transient. The deformed mesh then has a periodic surface geometrically identical to the nominal one, and a standard automatic mesh connection should then work without any other modifications. Actually, in most cases, the model can use a 1:1 connection.
A minimal example of what this could look like is below:
FUNCTION: High Periodic DX
Option = Interpolation
Argument Units = [m], [m], [m]
Result Units = [m]
INTERPOLATION DATA:
Option = Three Dimensional
Data = x1,y1,z1,dx1, x2,y2,z2,dx2, ... # One table per component, per wall
END
END
...
BOUNDARY:
MESH MOTION:
Option = Specified Displacement
DISPLACEMENT:
Option = Cartesian Components
Displacement X Component = High Periodic DX(Initial X, Initial Y, Initial Z) * min(1.0, t/1[s])
Displacement Y Component = High Periodic DY(Initial X, Initial Y, Initial Z) * min(1.0, t/1[s])
Displacement Z Component = High Periodic DZ(Initial X, Initial Y, Initial Z) * min(1.0, t/1[s])
END
END
END
...
FLOW: Flow Analysis
EXPERT PARAMETERS:
solve emag = f
solve energy = f
solve fluids = f
solve fsd = f
solve masfrc = f
solve mixture fraction = f
solve postproc masfrc = f
solve postproc tvariance = f
solve radiation = f
solve reaction progress = f
solve scalar = f
solve soot = f
solve temperature variance = f
solve turbulence = f
solve volfrc = f
solve wallscale = f
transient initialisation override = t
END
ENDAnother problem that I ran into was getting the displacement right, since a naive nearest-point projection folds the surface mesh. I have two different mapping procedures, depending on the meshes, and the script verifies injectivity in both cases:
- Direct: The exact node-to-node transformation, which works only when the two non-matching walls share the same mesh topology (this should work for all of my cases, since the passage meshes usually come from the same template). There is no interpolation, and I have checks in my code to ensure that the meshes are conformal before actually performing the mapping.
- Floater mean-value: I flatten both walls to the unit square with strictly positive (mean-value) weights. In this way, a piecewise-linear map to a convex domain with positive weights cannot fold. I then lift each perturbed node, via its unit-square image, onto the nominal wall. The result is a wall-to-wall map defined by a composition of two flattening homeomorphisms, Φ = σnominal-1 ∘ σperturbed, and is therefore injective by construction.
To summarize, my workflow from end-to-end looks like this:
- Export the high/low periodic wall node clouds + face connectivity for the nominal and damaged passages (CFX profile layout:
[Data] + [Faces]). - Run the offline script → injective displacement map per wall + autofilled CCL with the six interpolation functions (High/Low × DX/DY/DZ) and the mesh-motion BCs.
- Import the CCL, run a short mesh-only pseudo-transient (fluids/turbulence off,
transient initialisation override) to ramp the morph in. - Use the deformed mesh; the periodic surfaces now match for a 1:1 interface.
It seems like the missing piece is on the pre-processing side. If TurboGrid or CFX-Pre could natively align a passage's periodic surface onto a specified target surface (a neighbour's wall, or an imported one) with a fold-free guarantee, or equivalently let an interface be specified between non-identical passages and build the conforming morph itself, it would make mistuning, erosion, and as-manufactured-geometry studies a lot easier. I also would be happy to share code or CCL files that I am using for this.
Â
You are navigating away from the AIS Discovery experience