Shape optimisation not converging

elninoft9 · January 23, 2023, 07:42

Hello! I am currently embarking on a specific research on the effects of altering the Mach number on shape optimization for the NACA 0012 airfoil using SU2, based on the "Unconstrained shape design of a transonic inviscid airfoil at a cte. AoA" tutorial. I have not changed anything except from altering the Mach number from 0.8 to 0.88 in the .cfg file. The mesh files and other parameters in the .cfg file are untouched. I then execute the " $ python shape_optimization.py -g CONTINUOUS_ADJOINT -o SLSQP -f inv_NACA0012_basic.cfg" code. My simulations converge till a Mach of 0.88, even though it takes a long time. However, for a Mach of 0.89, the solver does not converge at all. The solver states " Iteration limit reached) and i think the iterations have been set to a limit of 100. Will increasing the iteration limit help? I'm afraid that this step might not aid in convergence and the simulation will just take more time. Any other remedies for this scenario or recommendations on how to approach my research for altering the Mach numbers?

bigfootedrockmidget · January 24, 2023, 15:42

Maybe run a CFD only simulation first with SU2_CFD and plot the residuals to see if it converges for the new Mach number. It might indeed be that it takes more iterations to converge, especially when you are restarting from a solution that was obtained for Ma=0.8

elninoft9 · January 24, 2023, 21:19

Thank you for your advice! I will give it a try. However, what exactly do you mean by residuals?

bigfootedrockmidget · January 25, 2023, 09:03

It is important to understand the basics of CFD before continuing your research.

Besides studying one of the many textbooks on the topic, the youtube channel of CFD101 is a great place to start, here is a lecture on residuals.

https://www.youtube.com/watch?v=v9OnNeYH4Ok

In SU2, the residuals are reported as RMS values to the screen and to the history file. You can open the history file in e.g. matlab, python or paraview to visualize them and see if all residuals go down.

January 23, 2023, 07:42	Shape optimisation not converging	#1
elninoft9 New Member El Nino Join Date: Jan 2023 Posts: 8 Rep Power: 3	Hello! I am currently embarking on a specific research on the effects of altering the Mach number on shape optimization for the NACA 0012 airfoil using SU2, based on the "Unconstrained shape design of a transonic inviscid airfoil at a cte. AoA" tutorial. I have not changed anything except from altering the Mach number from 0.8 to 0.88 in the .cfg file. The mesh files and other parameters in the .cfg file are untouched. I then execute the " $ python shape_optimization.py -g CONTINUOUS_ADJOINT -o SLSQP -f inv_NACA0012_basic.cfg" code. My simulations converge till a Mach of 0.88, even though it takes a long time. However, for a Mach of 0.89, the solver does not converge at all. The solver states " Iteration limit reached) and i think the iterations have been set to a limit of 100. Will increasing the iteration limit help? I'm afraid that this step might not aid in convergence and the simulation will just take more time. Any other remedies for this scenario or recommendations on how to approach my research for altering the Mach numbers?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
adjointOptimisationFoam: difficulty completing second optimisation cycle	Hugh_	OpenFOAM Running, Solving & CFD	2	October 11, 2022 23:17
3D RANS based shape optimisation mesh deformation	nikki_vlk	SU2 Shape Design	0	May 30, 2022 04:36
Spline generation for Adjoint Based Shape Optimisation	radioactivity	OpenFOAM	1	November 1, 2021 00:11
Shape Optimisation Parametrisation	Nick Janssens	OpenFOAM Programming & Development	1	May 22, 2020 09:47
Calculation of the Governing Equations	Mihail	CFX	7	September 7, 2014 06:27

January 24, 2023, 15:42		#2
bigfootedrockmidget Senior Member bigfoot Join Date: Dec 2011 Location: Netherlands Posts: 585 Rep Power: 17	Maybe run a CFD only simulation first with SU2_CFD and plot the residuals to see if it converges for the new Mach number. It might indeed be that it takes more iterations to converge, especially when you are restarting from a solution that was obtained for Ma=0.8

January 24, 2023, 21:19		#3
elninoft9 New Member El Nino Join Date: Jan 2023 Posts: 8 Rep Power: 3	Thank you for your advice! I will give it a try. However, what exactly do you mean by residuals?

January 25, 2023, 09:03		#4
bigfootedrockmidget Senior Member bigfoot Join Date: Dec 2011 Location: Netherlands Posts: 585 Rep Power: 17	It is important to understand the basics of CFD before continuing your research. Besides studying one of the many textbooks on the topic, the youtube channel of CFD101 is a great place to start, here is a lecture on residuals. https://www.youtube.com/watch?v=v9OnNeYH4Ok In SU2, the residuals are reported as RMS values to the screen and to the history file. You can open the history file in e.g. matlab, python or paraview to visualize them and see if all residuals go down.