Aero-Structural Discrete Adjoint Sensitivities and Python Wrapper Developments #1750
Aero-Structural Discrete Adjoint Sensitivities and Python Wrapper Developments #1750
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
|
Hi @patelha57 what is the current status? Would be nice to see a couple of older PR's get merged into develop, especially this one. |
No additional code development needed from my POV (please correct me if I'm wrong!). Just need to fix these regression tests failures, add example for the residual solver adjoint, and add documentation. |
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. surface coordinates or displacements at mesh | ||
| * vertices. \return Total sensitivity of the objective function w.r.t. the un-deformed coordinates. | ||
| */ | ||
| vector<vector<passivedouble>> GetObjectiveCoordinatesTotalSensitivities() const; | ||
|
|
||
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. surface coordinates or displacements at a mesh | ||
| * vertex. \param[in] iPoint - Mesh vertex index. \return Total sensitivity of the objective function w.r.t. the | ||
| * un-deformed coordinates. | ||
| */ | ||
| vector<passivedouble> GetObjectiveCoordinatesTotalSensitivities(unsigned long iPoint) const; | ||
|
|
||
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. surface coordinates or displacements at marker | ||
| * vertices. \param[in] iMarker - Marker identifier. \return Total sensitivity of the objective function w.r.t. the | ||
| * surface coordinates. | ||
| */ | ||
| vector<vector<passivedouble>> GetMarkerObjectiveCoordinatesTotalSensitivities(unsigned short iMarker) const; | ||
|
|
||
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. surface coordinates or displacements at a marker | ||
| * vertex. \param[in] iMarker - Marker identifier. \param[in] iVertex - Marker vertex index. \return Total sensitivity | ||
| * of the objective function w.r.t. the surface coordinates. | ||
| */ | ||
| vector<passivedouble> GetMarkerObjectiveCoordinatesTotalSensitivities(unsigned short iMarker, | ||
| unsigned long iVertex) const; | ||
|
|
||
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. the design variables. | ||
| * \return Total sensitivity of the objective function w.r.t. the design variables. | ||
| */ | ||
| vector<vector<passivedouble>> GetObjectiveDesignVariablesTotalSensitivities() const; | ||
|
|
||
| /*! | ||
| * \brief Get total sensitivity of the objective function w.r.t. a design variable. | ||
| * \param[in] iDV - Design Variable index. | ||
| * \return Total sensitivity of the objective function w.r.t. a design variable. | ||
| */ | ||
| vector<passivedouble> GetObjectiveDesignVariablesTotalSensitivities(unsigned short iDV) const; |
There was a problem hiding this comment.
This interface is not adequate, there is a better alternative as I've explained before.
There was a problem hiding this comment.
You are talking about the matrix view implementation, correct?
There was a problem hiding this comment.
Yep, there is a recent example of doing that for another type of container here https://github.com/su2code/SU2/pull/2388/changes
There was a problem hiding this comment.
Thank you so much, I started converting them. Would appreciate a review and any additional comments when you get a chance!
pcarruscag
left a comment
pcarruscag
left a comment
There was a problem hiding this comment.
Thank you, looks good to me, mostly just comments about naming conventions.
| * \param[in] alpha - Angle (degree). | ||
| */ | ||
| void SetFarFieldAoA(passivedouble alpha); | ||
| void SetAngleOfAttack(passivedouble alpha); |
There was a problem hiding this comment.
| void SetAngleOfAttack(passivedouble alpha); | |
| void SetFarFieldAoA(passivedouble alpha); |
| * \param[in] beta - Angle (degree). | ||
| */ | ||
| void SetFarFieldAoS(passivedouble beta); | ||
| void SetAngleOfSideslip(passivedouble beta); |
There was a problem hiding this comment.
| void SetAngleOfSideslip(passivedouble beta); | |
| void SetFarFieldAoS(passivedouble beta); |
| /*! | ||
| * \brief Get the number of conservative state variables. | ||
| * \return Number of conservative state variables. | ||
| */ | ||
| unsigned long GetNumberStateVariables() const; | ||
|
|
||
| /*! | ||
| * \brief Get the number of primitive state variables. | ||
| * \return Number of primitive state variables. | ||
| */ | ||
| unsigned long GetNumberPrimitiveVariables() const; |
There was a problem hiding this comment.
Is this known from the primitive indices return? If not, this info belongs there, or at least together with the corresponding functions.
| /*! | ||
| * \brief Get sensitivity of flow residuals with respect to farfield design variables as a matrix-vector product with | ||
| * the adjoint variable. | ||
| * \return Partial derivative of aerodynamic residuals with respect to farfield design variable. | ||
| */ | ||
| vector<passivedouble> GetResidualsFarfieldVariablesSensitivities() const; | ||
|
|
There was a problem hiding this comment.
Can this be a matrix/vector view?
| /*! | ||
| * \brief Get sensitivity of objective function with respect to farfield design variables as a partial derivative. | ||
| * \return Partial derivative of aerodynamic function with respect to farfield design variable. | ||
| */ | ||
| vector<passivedouble> GetObjectiveFarfieldVariablesSensitivities() const; |
There was a problem hiding this comment.
What are the farfield design variables? can you document the expected size of the vector?
There was a problem hiding this comment.
Right now the only supported farfield design variables are 1) angle of attack and 2) mach number.
| /*! | ||
| * \brief Get a read-only view of dCoordinates/dCoordinates^T * psi for all mesh nodes. | ||
| * \return CPyWrapperMatrixView of shape (nPoint, nDim). | ||
| */ | ||
| CPyWrapperMatrixView CoordinatesCoordinatesSensitivities() const; | ||
|
|
||
| /*! | ||
| * \brief Get a read-only view of dCoordinates/dDisplacements^T * psi for a marker. | ||
| * \param[in] iMarker - Marker index. | ||
| * \return CPyWrapperMatrixView of shape (nVertex, nDim). | ||
| */ | ||
| CPyWrapperMatrixView MarkerCoordinatesDisplacementsSensitivities(unsigned short iMarker) const; |
There was a problem hiding this comment.
Are these derivatives of deformed coordinates with respect to original coordinates and boundary deformations, respectively? The comments or function names could reflect that deformed/original difference better
There was a problem hiding this comment.
CoordinatesCoordinatesSensitivities() is sensitivity of volume coordinate w.r.t deformed surface coordinates... so dxv/dxa. And MarkerCoordinatesDisplacementsSensitivities() is for deformed surface coordinates w.r.t surface displacements dxa/dua.
I will update function names and documentation accordingly.
| * \brief Get a read-only view of dObjective/dStates for all mesh nodes. | ||
| * \return CPyWrapperMatrixView of shape (nPoint, nVar). | ||
| */ | ||
| CPyWrapperMatrixView ObjectiveStatesSensitivities() const; | ||
|
|
||
| /*! | ||
| * \brief Get a read-only view of dResiduals/dStates^T * psi for all mesh nodes. | ||
| * \return CPyWrapperMatrixView of shape (nPoint, nVar). | ||
| */ | ||
| CPyWrapperMatrixView ResidualsStatesSensitivities() const; | ||
|
|
||
| /*! | ||
| * \brief Get a read-only view of dTractions/dStates^T * psi for all mesh nodes. | ||
| * \return CPyWrapperMatrixView of shape (nPoint, nVar). | ||
| */ | ||
| CPyWrapperMatrixView ForcesStatesSensitivities() const; |
There was a problem hiding this comment.
What is meant by states here? The solution, i.e. conservative variables? If so, please make the term consistent.
There was a problem hiding this comment.
Yes states refers to conservative variables. I will make that explicit... ObjectiveSolutionSensitivities(), ResidualsSolutionSensitivities(), etc...
There was a problem hiding this comment.
Did you mean to add these? They don't seem used.
There was a problem hiding this comment.
Just using this simple box example to debug and verify sens from residual-based solver match fixed-point. Will remove these configs after incorporating the check into the new example script
7 participants
Motivation
Fundamental bottlenecks exist for industrial adoption of high-fidelity physics codes with fully coupled discrete adjoint sensitivity analysis. These bottlenecks include a lack of flexibility, modularity, and robustness of the computational tools, as well as the potential startup development costs needed to implement and verify the MDAO features. The purpose of this work is to facilitate the coupling of SU2 with external structural codes (e.g. NASTRAN, TACS, Airbus structural suite Lagrange) for gradient-based aerodynamic shape and structural sizing optimization using dedicated frameworks (e.g. OpenMDAO).
Startup development costs include the creation of modular tools that are designed to be driven by another framework like OpenMDAO, rather than to drive execution themselves. However, leveraging those MDAO frameworks assumes that software codes and modules to be coupled exist, have appropriate data structures, execution and query APIs, and are wrapped in Python for flexibility and ease of use. The proposed changes were motivated with the goal of making SU2 more modular and flexible, particularly to facilitate its integration into large-scale MDAO frameworks. For more details on the motivation, methodology, and verification & validation results, please refer to our paper from AIAA Aviation 2022.
Proposed Changes
@aa-g and I propose the following code updates:
Implementation of residual-based discrete adjoint solver as
CDiscAdjResidualSolvera. New config option:
KIND_DISC_ADJEnhancements and standardization of SU2 Python API
a. Add pysu2/pysu2ad methods
b. Standardize Python API and function names
c. Overloaded getter/setter methods to make data handling more flexible
Related Work
These efforts are related to Issue #1262, Pull Request #1300, and Discussion #1325.
PR Checklist