An adjoint-based hp-adaptive stabilized finite-element method with shock capturing for turbulent flows
Introduction
After several decades of development, high-order finite-element methods are now being considered for realistic and large-scale Computational Fluid Dynamics (CFD) simulations. This necessitates further studies on utilization of mesh adaptation techniques in order to reach reliable solutions at minimal computational cost. In many engineering applications, a specific scalar objective, like lift or drag coefficient, is of particular interest. The adjoint-based adaptation techniques [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21] are able to detect regions in the mesh where the discretization errors directly affect the accuracy of the computed objective. Among adaptive high-order finite-element methods being developed for compressible flow problems, the Discontinuous-Galerkin (DG) schemes have been arguably the most utilized. However, for elements with low polynomial degrees, a Petrov–Galerkin (PG) scheme requires significantly less degrees of freedom (DOFs) and non-zero matrix entries than a DG scheme for comparable accuracy [22], [23], [24], [25]. This advantage can be further enhanced using adaptation techniques. Although in the CFD community, a considerable amount of research has been conducted on adjoint-based adaptation for DG schemes [8], [11], [12], [13], [18], [20], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], the methodology for PG schemes is still not rigorously established. The objective of the present work is multifold. Here, an adjoint-based adaptation algorithm has been developed for PG schemes such that it can be effectively employed in a wide range of compressible flow applications, including subsonic and transonic turbulent flows. To this end, numerous techniques must be implemented to simulate over these flow regimes. This paper describes these techniques in detail.
Since the adaptation process is navigated by an adjoint-based technique, the quality of adjoint solution is of critical importance. In particular, in presence of Dirichlet boundary conditions, special attention should be given to the boundary conditions and to the definition of the functional output. In the present work, the spatial discretization is based on a Streamline-Upwind Petrov–Galerkin (SUPG) scheme [36], [37], [38]. To obtain a smooth adjoint solution near the no-slip walls, the boundary conditions have been imposed weakly [39], [40], [41] and the definition of the functional outputs have been modified to be compatible with these boundary conditions. The implementation of the boundary conditions is, in essence, based on the Nitsche’s method [42] and its particular formulation has been taken from a Symmetric Interior Penalty Galerkin (SIPG) method [13], [24], [43], [44] that is commonly used in the DG discretizations.
A major concern in the solution of transonic and supersonic flows is the need to stabilize the numerical scheme in vicinity of the shock waves. Although several methods exist to address this problem [45], [46], [47], [48], [49], present study seeks a method that has been previously tested in an adjoint-based hp-adaptation algorithm. To this end, a PDE-based artificial viscosity based on Refs. [48], [49] has been added to the governing equations.
In the context of finite elements, the hp-adaptation is the most effective approach to enhance the accuracy. Such an approach utilizes p-enrichment in regions with smooth solution, and h-refinement in regions with sharp gradients. A common approach in hp-adaptation is to use only one level of refinement for p- and/or h-refinements at each adaptation cycle. In the present work, for p-enrichment, one level of enrichment is used, but, for h-refinement, depending on the smoothness of the solution, either one or two levels of refinement may be used at each adaptation cycle. This approach has been shown to be particularly beneficial in the vicinity of the singularity points.
Due to simplicity, speed, and versatility, a non-conformal mesh adaptation approach has been developed in this study. However, a major difficulty that arises during non-conformal h- and p-adaptation is the generation of hanging nodes. In PG schemes, the discrete solution is required to be continuous over the computational domain. Clearly, hanging nodes can violate this requirement across the interface between refined and unrefined elements. In a previous study [21], a technique known as constrained approximation [50], [51], [52], [53], [54], [55], [56], [57] was employed in which a function value at a hanging node is constrained by the function values at adjacent nodes such that a continuous solution is obtained across all element interfaces. In that study, Lagrange polynomials were used as shape functions. Although the implementation was successful up to cubic elements for two dimensional problems, the extension of the method to three dimensional problems was cumbersome. To alleviate this problem, in the present work, hierarchical polynomials have been employed. As it will be described later, this choice results in a more modular implementation.
To represent the geometries accurately, high-order curved boundary edges have been generated. It should be noted that curving the boundary edges could result in collapsed cells when high aspect ratio elements are present in the viscous boundary layers. To address this problem, the interior meshes have been deformed through a linear elasticity solver.
Finally, it should be mentioned that the present adaptive PG methodology has been developed within a general framework, denoted as FUNSAFE (Fully UNStructured Adaptive Finite Elements) [22], [24], [25], [58], [59], [60], [61], [62], [63].
An outline of the paper is as follows. In Section 2, the governing equations for compressible turbulent flows are reviewed. In Section 3, first, the Petrov–Galerkin formulation is presented for a regular computational mesh. This includes description of the weak implementation of Dirichlet boundary conditions as well as the time integration scheme used to drive solution toward a steady state. Next, the implementation of constraint approximation as well as appropriate basis functions are discussed. In Section 4, the details of adjoint-based adaptation including error estimation formulation and decision making mechanisms for grouping elements in h-refinement and p-enrichment groups are discussed. In Section 5, the mesh curving strategy is explained. In Section 6, the numerical test cases are presented and finally, in Section 7, conclusion are made.
Section snippets
Governing equations
The governing equations consist of the compressible Reynolds Averaged Navier–Stokes (RANS) equations coupled with the negative variant of the one-equation Spalart–Allmaras (SA) turbulence model [64]. In the conservative form, these equations can be written as Here, the bold letters denote vector variables due to multiple equations, and indexes the spatial dimension. Also, as seen in the following, denotes a vector in spatial dimensions. The vector of the
Spatial discretization
To start, the strong form of the problem is written as an initial boundary value problem: where, is a bounded domain with Lipschitz-continuous boundary , are the prescribed boundary fluxes through portion of the boundary, and is the Dirichlet boundary condition on the portion of the boundary, and the operator is defined as (see also Refs. [66], [67]) where
Adjoint-based adaptation
In many simulations, a specific functional output, usually defined in an integral form, is of particular interest. Lift and drag coefficients are familiar examples in aeronautical applications. In such cases, output-, or adjoint-based methods perhaps offer the most reliable approaches for mesh adaptation, as they target the chosen functional and try to adapt the mesh such that a prescribed precision is ensured. For this purpose, the sensitivity of the functional with respect to the local
Mesh curving strategy
A common practice in CFD, and particularly in the second-order finite-volume schemes, is that the boundaries of the geometry are represented by a series of linear elements. However, to achieve higher accuracies, an increased conformity is required to account for surface curvature. Therefore, a mechanism is required to project the boundary edges to the exact geometry. However, such projection may generate collapsed elements near the boundary, especially when high aspect ratio elements are used,
Numerical results
In this section, the developed adjoint-based adaptive algorithm is applied to five numerical examples including four subsonic flows, and one transonic flow. For four examples, the NACA0012 airfoil is used and the exact geometry of the airfoil is calculated based on a modified formula from Turbulence Modeling Resource (TMR) website [76], which is supported by NASA Langley Research Center for verifying and validating turbulence models. This formula results in a sharp trailing edge.
Conclusion
In this study, an adjoint-based adaptation algorithm was implemented within a Petrov–Galerkin finite-element method. The mesh modification mechanisms included h- and hp-adaptation which were performed in a non-conformal manner. The constrained approximation method was utilized to retain the continuity of the solution space. The resulting method can be utilized within any continuous Galerkin method and it is particularly beneficial for multidisciplinary applications. For the geometric surfaces,
Acknowledgment
This work was supported by Tennessee Higher Education Commission Center of Excellence for Applied Computational Science and Engineering.
References (89)
- et al.
Adjoint error estimation and grid adaptation for functional outputs: Application to quasi-one-dimensional flow
J. Comput. Phys.
(2000) - et al.
Grid adaptation for functional outputs: Application to two-dimensional inviscid flows
J. Comput. Phys.
(2002) - et al.
Anisotropic grid adaptation for functional outputs: Application to two-dimensional viscous flows
J. Comput. Phys.
(2003) - et al.
Adjoint-based error estimation and grid adaptation for functional outputs: Application to two-dimensional, inviscid, incompressible flows
Comput. & Fluids
(2009) - et al.
Adjoint-based –p adaptive discontinuous Galerkin methods for the 2D compressible Euler equations
J. Comput. Phys.
(2009) - et al.
Output-based space–time mesh adaptation for the compressible Navier–Stokes equations
J. Comput. Phys.
(2011) - et al.
Output-based mesh adaptation for high order Navier–Stokes simulations on deformable domains
J. Comput. Phys.
(2013) - et al.
Petrov–Galerkin and discontinuous-Galerkin methods for time-domain and frequency-domain electromagnetic simulations
J. Comput. Phys.
(2011) - et al.
Discontinuous Galerkin and Petrov Galerkin methods for compressible viscous flows
Comput. Fluids
(2014) - et al.
Adaptive discontinuous Galerkin finite element methods for the compressible Euler equations
J. Comput. Phys.
(2002)
Streamline upwind/Petrov–Galerkin formulations for convection dominated flows with particular emphasis on the incompressible Navier–Stokes equations
Comput. Methods Appl. Mech. Engrg.
Large eddy simulation of turbulent Taylor–Couette flow using isogeometric analysis and the residual-based variational multiscale method
J. Comput. Phys.
Weak imposition of Dirichlet boundary conditions in fluid mechanics
Comput. Fluids
Weak Dirichlet boundary conditions for wall-bounded turbulent flows
Comput. Methods Appl. Mech. Engrg.
Stabilization and shock-capturing parameters in SUPG formulation of compressible flows
Comput. Methods Appl. Mech. Engrg.
SUPG finite element computation of inviscid supersonic flows with YZ shock-capturing
Comput. & Fluids
Shock capturing with PDE-based artificial viscosity for DGFEM: Part I. Formulation
J. Comput. Phys.
Aspects of an adaptive hp-finite element method: Adaptive strategy, conforming approximation and efficient solvers
Comput. Methods Appl. Mech. Engrg.
Analysis of the equilibrated residual method for a posteriori error estimation on meshes with hanging nodes
Comput. Methods Appl. Mech. Engrg.
Arbitrary-level hanging nodes and automatic adaptivity in the hp-FEM
Math. Comput. Simul.
Hierarchical basis for stabilized finite element methods for compressible flows
Comput. Methods Appl. Mech. Engrg.
A new finite element formulation for computational fluid dynamics: VIII. The Galerkin/least-squares method for advective-diffusive equations
Comput. Methods Appl. Mech. Engrg.
The variational multiscale method—a paradigm for computational mechanics
Comput. Methods Appl. Mech. Engrg.
Variational multiscale stabilization of high-order spectral elements for the advection–diffusion equation
J. Comput. Phys.
Positivity of flux vector splitting schemes
J. Comput. Phys.
A new finite element formulation for computational fluid dynamics: X. The compressible Euler and Navier–Stokes equations
Comput. Methods Appl. Mech. Engrg.
Approximate Riemann solvers, parameter vectors, and difference schemes
J. Comput. Phys.
Adjoint consistency analysis of residual-based variational multiscale methods
J. Comput. Phys.
An implicit upwind algorithm for computing turbulent flows on unstructured grids
Comput. Fluids
Implicit/multigrid algorithms for incompressible turbulent flows on unstructured grids
J. Comput. Phys.
Adjoint recovery of superconvergent functionals from PDE approximations
SIAM Rev.
An optimal control approach to a posteriori error estimation in finite element methods
Acta Numer.
Adjoint-based, three-dimensional error prediction and grid adaptation
AIAA J.
Comparison of adjoint-based and feature-based grid adaptation for functional outputs
Internat. J. Numer. Methods Fluids
Techniques for high-order adaptive discontinuous Galerkin discretizations in fluid dynamics
An Adaptive Discontinuous Galerkin Solver for Aerodynamic Flows
Review of output-based error estimation and mesh adaptation in computational fluid dynamics
AIAA J.
A robust hp-adaptation method for discontinuous Galerkin discretizations applied to aerodynamic flows
Cited by (16)
Linearization errors in discrete goal-oriented error estimation
2023, Computer Methods in Applied Mechanics and EngineeringA new paradigm of dissipation-adjustable, multi-scale resolving schemes for compressible flows
2022, Journal of Computational PhysicsCitation Excerpt :Recent works include monotonicity-preserving (MP) WENO [25,26], compact-reconstruction WENO (CRWENO) [27] and targeted ENO (TENO) [28,29] et al. It is worth mentioning that many shock-capturing schemes described above have also been extended to other high order discretization framework and implemented as limiting strategies for high speed compressible flow simulations [30–33]. As indicated in [34], the numerical dissipation of current shock-capturing schemes mainly comes from the inherent dissipation errors in upwind-biased interpolations, and extra numerical errors introduced by limiting processes which are designed to suppress numerical oscillations.
Output-based error estimation and mesh adaptation for unsteady turbulent flow simulations
2022, Computer Methods in Applied Mechanics and EngineeringCitation Excerpt :The present work considers a posteriori output-based error estimates, as these are the most useful for engineering applications that demand accuracy in output scalars and efficient meshes for their prediction. Much previous work exists in this area, mostly for steady [3–10] and deterministic unsteady [11–21] simulations. However, current high-fidelity analyses are often unsteady, e.g. large-eddy and other scale-resolving simulations of turbulent flow [2,22].
An adaptive enriched semi-Lagrangian finite element method for coupled flow-transport problems
2022, Computers and FluidsAn implicit block ILU smoother for preconditioning of Newton–Krylov solvers with application in high-order stabilized finite-element methods
2020, Computer Methods in Applied Mechanics and EngineeringA scalable solution strategy for high-order stabilized finite-element solvers using an implicit line preconditioner
2018, Computer Methods in Applied Mechanics and EngineeringCitation Excerpt :Therefore, PG schemes do not require curved meshes for second-order-accurate Navier–Stokes solutions. In view of these advantages, PG schemes have been employed in several flow solvers such as GGNS [6], PHASTA [7–10], FUNSAFE [2,11–15], FUN3D [16,17], and the COFFE [18–21]. A major incentive for developing finite-element methods in CFD is to benefit from high-order discretizations and to reach higher accuracies at lower computational costs.
- 1
Present address: Computational AeroSciences branch, NASA Langley Research Center, Hampton, VA 23681, United States.