Preparation and assembly of discrete adjoint CFD codes

Jones, Dominic; Müller, Jens-Dominik; Christakopoulos, Faidon

doi:10.1016/j.compfluid.2011.01.042

Cited by 26 publications

(21 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…After the primal has been reorganised for AD, the process can be automated through Makefiles. Further details on applying AD to modern CFD codes are presented in the companion paper [7].…”

Section: Application Of Admentioning

confidence: 99%

Pseudo-timestepping and verification for automatic differentiation derived CFD codes

2011

Self Cite

View full text Add to dashboard Cite

“…After the primal has been reorganised for AD, the process can be automated through Makefiles. Further details on applying AD to modern CFD codes are presented in the companion paper [7].…”

Section: Application Of Admentioning

confidence: 99%

Pseudo-timestepping and verification for automatic differentiation derived CFD codes

2011

Self Cite

View full text Add to dashboard Cite

“…A technique to tackle the problem of deriving the discrete adjoint in such complex cases is AD, in which the adjoint code to evaluate the gradients is obtained by directly manipulating the original CFD code, as in the work of Mohammadi [23,24]. This approach has grown in popularity and was later pursued by other research groups [7,8,9,25,26]. AD may be obtained by means of source-transformation tools or via operator overloading in programming languages such as FORTRAN 90 and C++.…”

Section: Discrete Approachmentioning

confidence: 99%

“…A more realistic goal for AD is in assisting the derivation of the discrete adjoint by hand-differentiation, by automatically differentiating, and by transposing individual source code functions. This approach was adopted, for instance, in Mader et al [8] and in Jones et al [9] and was also followed here.…”

Section: Discrete Approachmentioning

confidence: 99%

“…As followed in the works of Mader et al [8] and Jones et al [9], the individual functions of the CFD solver have been automatically differentiated and assembled afterwards to build the necessary terms in the sensitivity equation, both in tangent and adjoint forms. The linear system associated to the sensitivity problem is solved using a matrix-free version of the fully implicit fixed-point iteration scheme of the base flow solver.…”

Section: Introductionmentioning

confidence: 99%

“…Nonetheless, this approach can be tedious if the flow equations involve complex terms, like upwinding terms for the inviscid fluxes or source terms of turbulence models. Recent advances in automatic differentiation (AD) tools [4,5,6], however, enable to produce the computer code for the differentials of these complex terms directly from the source code of the CFD solver [7,8,9].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Fully Implicit Discrete-Adjoint Methods for Rotorcraft Applications

2016

View full text Add to dashboard Cite

This paper presents the development of a fully implicit, low-memory, discrete adjoint method by means of automatic source-code differentiation applied to the Helicopter Multi-Block computational fluid dynamics solver. The method is suitable for applications in flight mechanics as well as shape optimization, and is demonstrated in this paper for popular flow cases reported in the literature. In particular, adjoint CFD computations were undertaken for airfoils, wings and rotor blade cases, and the obtained results were found to agree well with published solutions and with finite differences of flow derivatives. The method has been demonstrated for inviscid and viscous cases and results suggest that the current implementation is robust and efficient. The cost of the adjoint computations is relatively low due to the employed source-code differentiation and most of the times it is no more than the cost of a steady-state flow solution.

show abstract

CAD‐based shape optimisation with CFD using a discrete adjoint

Jahn²,

Müller

2013

Numerical Methods in Fluids

Self Cite

View full text Add to dashboard Cite

SUMMARY One of the major challenges of shape optimisation in practical industrial cases is to generically parametrise the wide range of complex shapes. A novel approach is presented, which takes CAD descriptions as input and produces the optimal shape in CAD form using the control points of the Non‐Uniform Rational B‐Splines (NURBS) boundary representation as design variables. An implementation of the NURBS equations in source allows to include the CAD‐based shape deformation inside the design loop and evaluate its sensitivities efficiently and robustly. In order to maintain or establish the required level of geometric continuity across patch interfaces, geometric constraints are imposed on the control point displacements. The paper discusses the discrete adjoint flow solver used and the computation of the complete sensitivities of the design loop by differentiating all components using automatic differentiation tools. The resulting rich but smooth deformation space is demonstrated on the optimisation of a vehicle climate duct. Copyright © 2013 John Wiley & Sons, Ltd.

show abstract

Preparation and assembly of discrete adjoint CFD codes

Cited by 26 publications

References 12 publications

Pseudo-timestepping and verification for automatic differentiation derived CFD codes

Pseudo-timestepping and verification for automatic differentiation derived CFD codes

Fully Implicit Discrete-Adjoint Methods for Rotorcraft Applications

CAD‐based shape optimisation with CFD using a discrete adjoint

Contact Info

Product

Resources

About