|
Estimating lift change using sensitivity
I would like to use the sensitivity output in surface_adjoint.dat to predict the change in lift coefficient on a 2D airfoil being rotated.
SU2 version 4.1.2 was used with the NACA 0012 demo of the QuickStart directory. Here are the steps followed: 1. Freestream conditions changed: Mach number = 0.3 and angle of attack = 0 degree. 2. SU2_CFD was first run in DIRECT MODE to obtain an inviscid solution (PHYSICAL_PROBLEM=EULER), then it was run using the CONTINUOUS_ADJOINT approach, with OBJECTIVE_FUNCTION changed to LIFT and OPT_OBJECTIVE to LIFT*0.001. The sensitivity dCL/dni on the surface of the airfoil in the direction normal to the surface was output in surface_adjoint.dat for the NPTS points. 3. A rotation of 2 degrees counterclockwise was applied to the surface points making the airfoil. The center of rotation was at the leading edge (0,0). This provided the values of the displacements of the points ΔXion the airfoil surface. 4. The lift increment ΔCL was estimated by the following expression: ΔCL = ∑ dCL/d ni Ni ● ΔXi ΔSi The summation is from IDEB to NPTS - IDEB points were not taken to remove high sensitivity values near the sharp trailing edge. For IDEB = 5, ΔCL = -1.8 and for IDEB = 10, ΔCL = -0.81. For 2 degrees change in angle of attack, this would give a change of -0.2, for a 2D airfoil. I have the following questions: 1. Was it the basic mistake being done? 2. In a previous e-mail in this forum, it was suggested to identify sharp edges and filter out the sensitivity near them. The settings recommended are SENS_REMOVE_SHARP (set to true to use the sharp edge filter), REF_SHARP_EDGES and SHARP_EDGES_COEF (control how far away from a sharp edge the sensitivities are removed, default value = 3), as defined in confg_structure.cpp. Are those coefficients used when using SU2_CFD or only when using SU2_DOT? 3. Would the use of the DISCRETE_ADJOINT approach remove those high sensitivity values at the trailing edge? |
| All times are GMT -4. The time now is 17:33. |