adjointShapeOptimizationFoam pointers

NickG · August 4, 2011, 17:15

Hi

Could someone give me some pointers for the values of alphaMax and lambda in the transportProperties for the adjointShapeOptimizationFoam solver in OF2.0.

Are they related to the inlet velocity? Unfortunately I don't have access to the aiaa paper referred to in another thread so can't read that. I can see the equation in the .c source file but don't understand it fully.

I've tried varying the values for the pitzDaily example and get similar results within a certain range but am not sure which are more valid/accurate. Should I try to find the midpoint of the range or does the better result lie closer to the unbalanced result...

Answers on a post... please

Nick

NickG · August 5, 2011, 18:19

Okay so alpha essentially relates to a porosity coefficient.Is lambda the second coefficient of viscosity, the eigenvalue of one of the matrices or something else?

boger · October 20, 2011, 15:23

Hi Nick,

As I also mentioned here, lambda refers to the step size taken in the steepest descent optimization, which helps (along with the under-relaxation factor that is applied) to control how rapidly alpha changes.

As far as the values, I suspect lambda is based on trial-and-error, and probably alphaMax is too. The porosity (alpha) just needs to grow to be "large enough" to slow the velocity in those cells to near-zero by way of the drag term in the momentum equation, and at some point, it probably becomes a question of diminishing returns in terms of convergence to let alpha grow indefinitely.

Hope that helps,
David

NickG · October 21, 2011, 07:34

Thanks David. I've been playing around with it and getting seemingly sensible results but it's nice to know why!

Nick

giuli@ · November 16, 2011, 07:21

Hi all..
my question is: does the geometry actually changes or the plots of alpha just give an indication of where to optimize the shape?
Both in the tutorial and in another case made by me no shape is optimized o.O

could you please give some directions..?

cheers

boger · November 16, 2011, 08:32

The interface between alpha=1 and alpha=0 indicates the location where the optimal boundary should be, but the geometry/mesh is not changed directly. Go back to the reference listed in the source code for adjointShapeOptimizationFoam.C.

August 4, 2011, 17:15	adjointShapeOptimizationFoam pointers	#1
NickG Member Nick Gardiner Join Date: Apr 2009 Location: Chichester, UK Posts: 94 Rep Power: 17	Hi Could someone give me some pointers for the values of alphaMax and lambda in the transportProperties for the adjointShapeOptimizationFoam solver in OF2.0. Are they related to the inlet velocity? Unfortunately I don't have access to the aiaa paper referred to in another thread so can't read that. I can see the equation in the .c source file but don't understand it fully. I've tried varying the values for the pitzDaily example and get similar results within a certain range but am not sure which are more valid/accurate. Should I try to find the midpoint of the range or does the better result lie closer to the unbalanced result... Answers on a post... please Nick

August 5, 2011, 18:19		#2
NickG Member Nick Gardiner Join Date: Apr 2009 Location: Chichester, UK Posts: 94 Rep Power: 17	Okay so alpha essentially relates to a porosity coefficient.Is lambda the second coefficient of viscosity, the eigenvalue of one of the matrices or something else? Last edited by NickG; August 25, 2011 at 11:10.

October 20, 2011, 15:23		#3
boger Senior Member David Boger Join Date: Mar 2009 Location: Penn State Applied Research Laboratory Posts: 146 Rep Power: 17	Hi Nick, As I also mentioned here, lambda refers to the step size taken in the steepest descent optimization, which helps (along with the under-relaxation factor that is applied) to control how rapidly alpha changes. As far as the values, I suspect lambda is based on trial-and-error, and probably alphaMax is too. The porosity (alpha) just needs to grow to be "large enough" to slow the velocity in those cells to near-zero by way of the drag term in the momentum equation, and at some point, it probably becomes a question of diminishing returns in terms of convergence to let alpha grow indefinitely. Hope that helps, David __________________ David A. Boger

November 16, 2011, 08:32		#6
boger Senior Member David Boger Join Date: Mar 2009 Location: Penn State Applied Research Laboratory Posts: 146 Rep Power: 17	The interface between alpha=1 and alpha=0 indicates the location where the optimal boundary should be, but the geometry/mesh is not changed directly. Go back to the reference listed in the source code for adjointShapeOptimizationFoam.C. __________________ David A. Boger

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search
Display Modes
Linear Mode Switch to Hybrid Mode Switch to Threaded Mode

October 21, 2011, 07:34		#4
NickG Member Nick Gardiner Join Date: Apr 2009 Location: Chichester, UK Posts: 94 Rep Power: 17	Thanks David. I've been playing around with it and getting seemingly sensible results but it's nice to know why! Nick

November 16, 2011, 07:21		#5
giuli@ New Member FlowLy Join Date: Aug 2010 Location: Germany Posts: 18 Rep Power: 16	Hi all.. my question is: does the geometry actually changes or the plots of alpha just give an indication of where to optimize the shape? Both in the tutorial and in another case made by me no shape is optimized o.O could you please give some directions..? cheers

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Pasing Pointers to functions solve and linearInterpolate	arkangel	OpenFOAM Programming & Development	0	July 7, 2011 10:05
Multiple inheritance, function pointers or?	ngj	OpenFOAM	0	March 20, 2009 11:36
pointers and threads	Andrew Garrard	FLUENT	3	March 16, 2004 13:57
Pointers to wall adjacent cells in CFX 5	Fabian	CFX	2	November 21, 2002 12:14
Pointers --- (CFX4-3)	Alex	CFX	0	April 15, 2002 16:03