Inquiry about Data File Format - Multiple Independent Variables (Mach, Re, AoA, AR)

[OscarEspinosaBarcenas]

Dear Aviary Development Team,

I'm working on integrating an aerodynamic model with OpenMDAO using Aviary's interpolation capabilities. I'm seeking clarification on the correct data file format and the handling of multiple independent variables.

My aerodynamic model calculates the following aerodynamic coefficients directly from:

• Mach number (Ma)
• Reynolds number (Re)
• Angle of Attack (aoa)
• Aspect Ratio (AR)

The model does not use a traditional drag polar decomposition (Cd_0 + k*Cl^2). It directly outputs the total coefficients. The coefficients calculated are:

• CL (Lift Coefficient)
• CD (Drag Coefficient)
• CLalpha (Lift Curve Slope)
• CDalpha (Drag Curve Slope)
• CM (Pitching Moment Coefficient)
• CMalpha (Pitching Moment Derivative)

Based on my understanding of the Aviary documentation, specifically the sections on "External Data Files" and "User Specified Tabular Drag Polars," I believe the most efficient approach is to use a single CSV data file with a structured grid format. The proposed structure is:

# Aerodynamic Data
# Independent Variables: Mach, Re, aoa, AR
# Dependent Variables: CL, CD, CLalpha, CDalpha, CM, CMalpha

Mach, Re, aoa, AR, CL, CD, CLalpha, CDalpha, CM, CMalpha
<Mach1>, <Re1>, <aoa1>, <AR1>, <CL1>, <CD1>, <CLalpha1>, <CDalpha1>, <CM1>, <CMalpha1>
<Mach1>, <Re1>, <aoa1>, <AR2>, <CL2>, <CD2>, <CLalpha2>, <CDalpha2>, <CM2>, <CMalpha2>
... (varying all independent variables)

The data would represent a multi-dimensional grid covering my flight envelope, with discrete values for Mach, Reynolds number, angle of attack, and aspect ratio. I would then use build_data_interpolator with structured=True, passing the path to this single file, and defining interpolator_outputs as follows:

interpolator_outputs = {
    "CL": "unitless",
    "CD": "unitless",
    "CLalpha": "1/rad",  # Or "1/deg" if appropriate
    "CDalpha": "1/rad",  # Or "1/deg"
    "CM": "unitless",
    "CMalpha": "1/rad", # Or "1/deg"
}

Can you please confirm that this single-file, structured grid approach, with four independent variables (Mach, Re, aoa, AR), is correct and supported by Aviary for my described use case? Specifically, I want to ensure that build_data_interpolator can handle more than two independent variables when using the MetaModelStructuredComp. Are there any restrictions, changes, or suggestions regarding the number of independent variables or the overall approach?

Thank you for your time and assistance.

Sincerely,

Oscar

[Kenneth-T-Moore]

Hey, Oscar

Thank you for the question. I think you are on the right track here, in that build_data_interpolator should let you choose an arbitrary set of input and output variables corresponding to your columns in the CSV file. We have used it for a few different tables including one with flap deflection as an input. You can see the main ones here:

https://github.com/OpenMDAO/Aviary/blob/main/aviary/subsystems/aerodynamics/gasp_based/table_based.py

Note that the Height Energy equations of motion for mission do not supply an Angle of Attack, but you can compute it with a solver that varies alpha to drive lift to equal weight. We have a built-in group that does that, which might be a good starting point for yours:

https://github.com/OpenMDAO/Aviary/blob/main/aviary/subsystems/aerodynamics/flops_based/tabular_aero_group.py

There is also an example of a very simple builder that replaces the aviary aero with a custom calculation.

https://github.com/OpenMDAO/Aviary/blob/main/aviary/examples/external_subsystems/custom_aero/custom_aero_builder.py
https://github.com/OpenMDAO/Aviary/blob/main/aviary/examples/external_subsystems/custom_aero/run_simple_aero.py

I think these are all good starting points, and I should be able to provide further assistance if you run into anything tricky, like getting variables to connect or promote where they are supposed to.

[OscarEspinosaBarcenas]

Hi @Kenneth-T-Moore,

Thank you for the hints on this previously. I’ve restarted my work on integrating a custom aerodynamics model and would appreciate your guidance.

I am using Aviary v0.9.10. My goal is to use my custom aerodynamics model for a specific mission (I'm trying only one mission segment for the moment), so in my phase_info, I have set subsystem_options to {'method': 'external'} and added my CustomAeroBuilder to the external_subsystems list for that phase.

As required by this new Aviary version, I am setting a global AERODYNAMICS_METHOD in my main aircraft data file. This is where I'm running into a problem during prob.setup().

Depending on the global setting, I get one of two errors:

• If I set inputs.set_val(Settings.AERODYNAMICS_METHOD, LegacyCode.GASP), I receive:

• ValueError: Invalid parameter in phase traj.phases.cruise_1. Parameter 'interference_independent_of_shielded_area' has invalid target(s).

• If I set inputs.set_val(Settings.AERODYNAMICS_METHOD, LegacyCode.FLOPS), I receive:

• ValueError: Unable to find the source 'drag' in the ODE.

In both cases, it seems that even though I've specified an external model for the cruise phase, Aviary is still trying to connect variables from the default GASP or FLOPS models, which don't exist in my custom component.

Could you advise on the correct way to configure this?

Thank you for your help. This is a key step for my PhD research, and any insight would be greatly appreciated.

[Kenneth-T-Moore]

Hey, @OscarEspinosaBarcenas

Is there any way that you can package up the code for me to take a look at? There might be an edge case here that has fallen through the cracks, but I would need to debug it a little to know for sure.

[OscarEspinosaBarcenas]

Hi @Kenneth-T-Moore,

Thank you for offering to look at the code. I've made significant progress since last time. I've resolved the original ValueError by correcting how I was handling the subsystem builder and variable promotions.

However, I am now facing a RuntimeError when the solver runs (you can find the error below).

I've attached a screenshot of the N2 diagram for the external aerodynamics component and have a couple of specific observations I was hoping you could clarify:

1. I noticed the implicit variable angle_of_attack is highlighted in yellow. Does this indicate a special connection status that I need to handle differently?
2. From the RuntimeError, my current hypothesis is that there might be an issue with how the lift variable is promoted or connected back to the solver.

I've packaged the code that reproduces this error. The zip file also contains trimmed_custom_aero_v02.py , which is a standalone test I used to confirm the core component works correctly outside of the full Aviary mission.

S45_HE_MF_AE_PMU_COL_NOM.ToKenneth.zip

I'll continue to debug, but any insight you have on this would be a great help.

Thank you again for your time.

PS:

RuntimeError: Jacobian in 'traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero' is not full rank. The following set of states/residuals contains one or more equations that is a linear combination of the others: 
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 1.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 2.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 3.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 4.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 5.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 6.
 'angle_of_attack' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.balance_aoa.angle_of_attack') index 7.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 1.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 2.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 3.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 4.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 5.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 6.
 'lift' ('traj.phases.cruise_1.rhs_all.external_subsystems.custom_aero.forces_calc.lift') index 7.

[jkirk5]

The yellow box indicates it is the implicit variable in the BalanceComp (basically the variable that gets varied so lift == weight), so that is all good. Part of your issue might be the ordering of your components - try flipping the order so the balance comp is after the aero calculations, making AoA the feedback instead of Lift. I think you need the implicit variable as the feedback in this case, although I'm not sure about the technicalities of why.

Here's what an aero component plus the AoA balance comp looks like in a working Aviary problem:

[jkirk5]

The build_data_interpolator() function should be able to handle any arbitrary number of dependent & independent variables - I don't think we've explicitly tested more than three independent variables off the top of my head, but there shouldn't be any reason why you can't. I agree that a single data table is probably most efficient, as long as you are ok with using the same interpolation scheme for each dependent variable (e.g. linear, lagrange).

Another thing to keep in mind is that Aviary's flight physics wants overall forces on the vehicle (so for aerodynamics that is total drag). You will want to add a component to calculate that, the SimpleDrag component here may be helpful.