CFD Vision 2030 Roadmap Update 2022

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

CFD Vision 2030 Roadmap Progress and Technology Highlights for 2022

CFD Vision 2030 IC Roadmap Subcommittee∗


September 2023

1 Introduction
This document is intended to highlight advancement toward the CFD Vision 2030[1] that occurred or was published
during 2022. This is part of an ongoing campaign by the AIAA CFD Vision 2030 Integration Committee to encourage
continued progress toward reaching the Vision. As part of this activity, key events and publications in each of the
Domains are identified and the progress toward the Roadmap milestones are assessed using a TRL scale to assist in
identifying progress.
The AIAA High Lift Prediction Workshop 4 (HLPW4) was one of the key events in 2022 that demonstrated ad-
vancements toward the objectives of the CFD Vision 2030. Held in January, this workshop demonstrated consistency
in grid-converged RANS results among multiple codes/organizations through the use of mesh adaptation. Beyond
this, it also highlighted a significant increase in the number of scale resolving simulations with over 15 contributions
compared to 2 during High Lift Prediction Workshop 3 in 2017. These results appeared to have much better corre-
lation with the experimental CL,max results than RANS simulations, including good agreement with other metrics
such as pitching moment and pressure distributions. The workshop, delayed from June 2021, also featured the use
of “Technology Focus Groups” (TFGs) that had been holding regular meetings for almost 18 months to increase the
learning that could be extracted from the workshop. The TFGs for HLPW4 included geometry and meshing, RANS,
mesh-adaptation, wall model LES, hybrid RANS/LES, and higher-order methods. Collaboration in each of these
TFGs helped to further advance the objectives of CFD Vision 2030.
Another significant advancement benefiting the Vision was the increase in supercomputing capability toward
exascale as reflected by the May 2022 demonstration of 1.1 exaflops by the Frontier supercomputer at Oak Ridge
National Laboratory. This achievement has significant implications for CFD in both the scale and quantity of results
that it will enable. This work is expected to have strong ramifications in Domains including machine learning for
turbulence models, uncertainty quantification, knowledge extraction, and multi-disciplinary analysis and optimiza-
tion. These implications are described in more detail in the following summaries of progress in the different Roadmap
Domains.

2 Roadmap Domains
Progress during 2022 in each of the six Domains identified in the CFD Vision 2030 Roadmap are described below.
In contrast to previous years, this summary has attempted to focus on approximately a single page of highlights.
This condensation has likely led to omission of significant publications, but is intended to provide a more concise
summary for future researchers and program managers to use as a starting point in identifying research needs and
progress.
∗ The Roadmap Subcommittee authors are Andrew Cary (The Boeing Company), John Chawner (Cadence Design Systems), Earl

Duque (Intelligent Light), Bill Grop (University of Illinois), Eric Nielsen (NASA), Brian Smith (Lockheed Martin Corporation), and
Nathan Wukie (Air Force Research Laboratory).
For more information on this document, including past year summaries, please see www.cfd2030.com/roadmap. Comments and feedback
about the value of this document are desired at aiaacfd2030@gmail.com. Please cite this document as: CFD Vision 2030 IC Roadmap
Subcommittee. “CFD Vision 2030 Roadmap Update 2022.” www.cfd2030.com/roadmap. March 2024.
Copyright © 2024 United States Government as represented by the Administrator of the National Aeronautics and Space Administration;
Lockheed Martin Corporation; The Boeing Company; Intelligent Light; Cadence Design Systems, and The Board of Trustees of the
University of Illinois. All Rights Reserved.

1
2.1 High Performance Computing
The HPC community reached a significant milestone in 2022, as a large team of industry and government researchers
demonstrated FP64 performance exceeding one exaflop for the first time. It was announced in May (2022) that the
new HPE/AMD GPU-based Frontier system located at Oak Ridge National Laboratory had successfully achieved
1.1 exaflops on the well-known High-Performance Linpack benchmark[2]. Equally as impressive, the system drew
just 21 megawatts of power to accomplish the milestone [3] – remarkable energy efficiency that would have been
unheard of 10-15 years ago when the community began to chart an initial course to exascale. Frontier is expected to
host general production jobs beginning in 2023. It is also worth noting that the long-awaited HPE/Intel GPU-based
Aurora system hosted by Argonne National Laboratory is currently being assembled and is targeting two exaflops of
FP64 performance.
Headlined by Frontier, the semi-annual Top500 rankings saw three new systems appear in the top ten over the
past year, with each of the top five systems now capable of well over 100 petaflops of performance[4]. GPU-based
paradigms play a key role, with seven of the top ten systems powered by AMD or NVIDIA GPUs. An overview of
the impact of GPU technologies at NASA and across the broader US aerospace community for both capacity- and
capability-class CFD applications was recently presented in Big Compute 22[5].
The aerospace CFD community continues to embrace the challenge of porting key CFD applications to GPU-
based architectures and bringing them to bear on complex applications. Some recent use-cases include applications
to the CREATE-AV Helios toolset [6, 7], scale-resolving simulations [8, 9, 10, 11, 12, 13, 14, 15, 16], multidisciplinary
applications [17, 18, 19], and design optimization [20]. Other examples are described in [21, 22, 23, 24, 25, 26].
The past year also saw a diverse range of leadership-class computing efforts covered in the literature. In [27],
Larsson describes large-eddy simulations of shock/boundary-layer interactions computed using as many as 65,000
CPU cores of the Theta system located at the Argonne Leadership Computing Facility. De Bruyn Kops and colleagues
performed high-resolution DNS simulations of turbulence using as many as three trillion grid points at the Oak Ridge
Leadership Computing Facility (OLCF) [28, 29]. A team of collaborators studying retropropulsion flows for human-
scale Mars Lander concepts performed finite-rate chemistry simulations using 16,000 GPUs on the Summit system
at OLCF [30]. Researchers also leveraged 145,000 CPU cores on the Japanese Fugaku system to study high-lift flows
using as many as 11 billion degrees of freedom [31, 32]. Homogeneous isotropic turbulence was evaluated using up to
1,024 GPUs in [33] on the OLCF Summit system and the Lassen facility at Lawrence Livermore National Laboratory.
Finally, Priebe and co-workers also used the OLCF Summit system to evaluate open-rotor concepts [34].

2.2 Physical Modeling


Recent progress in the RANS, Hybrid RANS/LES and LES Physical Modeling timelines of the CFD 2030 Roadmap
was summarized by two significant events in 2022. The fourth AIAA CFD High Lift Prediction Workshop (HLPW4)
was held in January. This workshop highlighted results obtained with RANS, Hybrid RANS/LES (HRLES) and
Wall-modeled LES (WMLES) for a swept wing transport aircraft in a high-lift configuration. The NASA Symposium
on Turbulence Modeling: Roadblocks, and the Potential for Machine Learning was held in July. This symposium
reviewed the progress and prospects for RANS turbulence modeling including the potential contributions of machine
learning. The 2020 update of the CFD 2030 Roadmap included a technology milestone, ”Machine Learning for
Complex Flow” on the RANS timeline. The symposium provided a valuable forum for assessing the state of the art
for this milestone.
The results from HLPW4 include contributions from leaders in RANS, HRLES and WMLES modeling and
applications. The high-lift CRM configuration has details and complexity sufficient to assess roadmap milestones for
this review[35]. The accuracy of RANS results did not appreciably improve from HLPW3, held in 2017. There was a
much higher level of consistency among contributors for the verification case compared to the HLPW3. Meshes with
200 million grid points were sufficient to provide near grid resolved results in the linear portions of the CL − α curve,
but meshes of this size were insufficient at maximum lift conditions. Mesh adaptation technology brought much
more consistency to the high-lift results, helping to reduce discretization error as a major confounding factor[36].
Approximately half of the RANS solutions contributed to the workshop did not converge to a steady state for both the
7.05◦ and the 19.57◦ AOA conditions, highlighting a challenge for RANS simulation of complex configurations with
steady-state solution algorithms. There was likely some incremental progress in the development of RANS modeling
methods over the past year. However, based on the workshop, these improvements have had minimal impact on the
capabilities employed by most of the industry for large scale CFD simulations of high-lift configurations.

2
Both HRLES[37] and WMLES[38] methods gave more accurate predictions at the max-lift condition than RANS.
However, they were less accurate in the linear region of the CL − α curve. This presents a significant challenge for
scale-resolving mesh methods. Until this is resolved, application engineers must determine a priori which approach
is appropriate for a given set of flow conditions to obtain the most accurate prediction. For these wind-tunnel
simulations, computational requirements for HRLES are 10-15 times greater than for a steady state RANS simulation,
while WMLES simulations are reported as 5-10 times more computationally intensive; flight-scale Reynolds numbers
have not been assessed. The computational requirement presents a significant challenge for generating large datasets
for aerodynamic design and S&C development[39]. Both HRLES and WMLES results are highly sensitive to mesh
density, and WMLES is sensitive to cell shape and aspect ratio. Meshes suitable for RANS applications were found
to be not suitable for scale-resolving methods. The sensitivity of these methods to mesh resolution and substandard
accuracy for low angle-of-attack flows are critical considerations in assessing their maturity.
The NASA symposium was important in that it brought leaders in the traditional turbulence modeling and
machine learning communities together[40]. This interaction helped the machine learning community to under-
stand the requirements of their methods for CFD application and the modeling constraints that could be extremely
helpful in the development of their approaches. These interactions also give the turbulence modeling community
an appreciation for the potential and limitations for machine learning in physical modeling. A highlight of the
turbulence-modeling-focused portion of the program is the clear description of fundamental requirements for a valid
and effective turbulence model[41]. Establishing these requirements is helpful for both traditional and machine-
learning-based model development.
For the machine-learning section of the symposium, participants were invited to apply their methods to four
cases: (1) Fully-developed channel flow, (2) Axisymmetric subsonic jet, (3) Wall mounted hump separated flow, and
(4) an airfoil. “The ‘Collaborative Testing Challenge’ was conceived as an integral part of this symposium, with the
idea of getting a group of experts to try to achieve data-driven turbulence models that work well across a fairly wide
range of simple test cases.”[40] Results from seven contributors were presented at the symposium. To be valid, the
same modeling approach had to be applied to all four test cases. The challenge proved difficult for the participants.
While reasonable results were generated for each case, there was a lot of variation in prediction accuracy. In addition,
approaches that gave good predictions on one case often gave inferior predictions on other cases. The results from
this symposium make it clear that the machine-learning methods for turbulence modeling are immature. None of the
models were presented with sufficient clarity and documentation to allow a researcher to implement the model and
duplicate the results. It is not clear at this time whether or not machine-learning methods will provide a significant
advance to RANS modeling. Given the large number of efforts in this area, both from challenge contributors and
from others, it is likely that the usefulness of machine learning for physical modeling will be determined in the next
few years. The significant interest in the workshop indicates that there continues to be a need for RANS models in
industry.

2.3 Algorithms
The Algorithms Domain is divided into two timelines focusing on improving the robustness and efficiency of the
simulation and on developing methods to understand the accuracy and variations in the results.

2.3.1 Convergence/Robustness Algorithms


The Algorithm timeline of the Roadmap is focused on the development of methods that are both faster and more
accurate, with advances in ability to leverage HPC progress and increasing robustness key to obtaining CFD results
that will impact engineering processes. The advances identified in the 4th High Lift Prediction Workshop (HLPW4)
that were previously referenced were partially enabled by the advances in algorithms on the current generation of HPC
computing, including GPUs. Decreasing numerical dissipation to allow for increased accuracy on a given mesh are
key enablers for scale-resolving simulations. During 2022, there were different focuses on how to identify and reduce
the numerical dissipation[42, 43] as well as increasing the understanding of the dissipation in present algorithms[44].
In addition, several high-order methods for both RANS and WMLES debuted in the HLPW series and demonstrated
their potential in reducing the total number of degrees of freedom [45] relative to 2nd-order methods. The Adaptive
RANS TFG of the HLPW4 workshop demonstrated consistent grid convergence for a complex configuration with a
common turbulence model across multiple codes[36], making progress toward the “grid convergence for a complex
configuration” milestone.

3
Beyond generating the fluid flow solution itself, efficiently obtaining and applying sensitivities for different ap-
plications from output-based mesh adaptation to uncertainty quantification and shape optimization is important.
Adjoint methods have been increasingly available for this purpose in multiple flow solvers, but convergence robust-
ness and accuracy remain challenging for many of these problems. Padway and Mavriplis[46] elaborate on a one-shot
strategy for both primal and adjoint solution that is efficient and robust for integration.

2.3.2 Uncertainty Quantification


The Uncertainty Quantification timeline in the Roadmap has initial milestones associated with characterization of
uncertainty in aerospace and propagation of uncertainty using CFD results. In 2022, there has been a significant
number of publications that have leveraged uncertainty techniques such as polynomial chaos and Kriging methods
to reduce the cost of propagation of uncertainty through CFD analysis. The applications in these papers use three-
dimensional Navier-Stokes analyses, including unsteady simulations, to obtain uncertainty estimates for complex
multi-disciplinary problems such as store separation[47] and sonic boom estimation[48, 49]. To further reduce the
cost of UQ analysis, multifidelity techniques are being applied that utilize different levels of CFD analysis and/or
experimental data to generate composite estimates of uncertainty or sensitivity. These techniques are being applied to
look at complicated problems such as shock-fin interactions[50] and predicting aircraft wing pressure distributions[51,
52], as well as identifying sensitivities associated with CFD modeling of engine bleed[53].
Uncertainty quantification tools and techniques applicable to CFD applications are also being developed with
simulation cost being a primary consideration. These range from techniques to leverage sparse uncertainty data[54]
to methods appropriate for optimization under uncertainty[55], as well as intrusive techniques that can significantly
reduce simulation cost[56] and toolboxes that efficiently package these capabilities for application users[57]. Efforts fo-
cused on improving turbulence model predictions through either calibration or improved modeling leveraging machine
learning were described in the physical modeling section (Section 2.2). Similar machine-learning techniques have been
used to increase understanding of the impact of uncertainties in turbulence modeling in specific situations[58, 59, 60],
as well as enhancing other methods for estimating turbulence uncertainty[61].
CFD validation is a necessary prerequisite for many CFD applications, but is challenging to perform for multiple
reasons. A vision for certification based on analysis[62] heavily leveraged the need for CFD validation and the
associated uncertainty estimation that this requires. Taylor[63] identified that there are many aspects of validation
that need to be considered that are context or application specific and called for more definition in these areas. Part
of this process will require improved error estimation using techniques such as the error transport equation[64].

2.4 Geometry Modeling and Mesh Generation


The Geometry Modeling and Mesh Generation Domain is divided into timelines focusing on the geometrical definition
of the simulation domain and on the generation of a suitable-for-purpose computational mesh.

2.4.1 Geometry Modeling


In pursuit of the goals set out in the CFD 2030 vision, geometry modelling environments originating from the research
community continue to develop. Two of the most widely used, Open VSP (sponsored by NASA) and ESP (sponsored
by AFRL) have reported significant updates this year[65, 66]. The capabilities offered by both toolsets have advanced
significantly in recent years. Ease of use has always been a critical factor driving their development. As their user
base has expanded, so has the emphasis placed on the facilities supported by their graphical user interfaces[66],
training, and on-line support[65].
Research continues into novel ways of representing airframe shapes. Of note is the progress reported by the
team at University of Bristol regarding the development of Volume of Solid driven Cellular Automata[67]. Their
approach offers a means of generating smooth two- and three-dimensional bodies, suitable for aerodynamic (and
multi-disciplinary) analyses and topological optimization. Their work is in its early stages, with only results for
two-dimensional optimization being reported thus far.
Development of the concept of Simulation Intent[68] to aid incorporation of computational simulation into the
wider product development lifecycle continues. Researchers at Queens University of Belfast report the development
of a novel hierarchical BREP graph shape representation in conjunction with a convolutional neural network (Hi-
erarchical CADNet)[69]. These appear to outperform other state-of-the-art deep learning architectures on feature
identification contained in MCAD geometry models – a prerequisite for the application of Simulation Intent. The
wider potential of BREP topology to aid the development of Digital thread is also reported[70].

4
The ISO WG12 T1 Geometry and Topology Group is in the process of developing a hybrid geometric modeling
capability for the next edition of ISO 10303 (STEP) AP 242 (ed 4). Hybrid modeling is the utilization of various forms
of geometric data with a single underlying BREP topological model. Commercially this is referred to as “Convergent
Modeling” by Siemens in their Parasolid, NX, and Solid Edge products[71, 72, 73, 74], “Mixed Modeling” by PTC
in onshape[75, 76] (as an implementation of Convergent Modeling through Siemens’ Parasolid kernel), and “CGM
Polyhedra” by Dassault Systèmes in CGM Modeler, CATIA, Spatial products, etc.[77, 78]

2.4.2 Mesh Generation


The mesh adaptation technical focus group of the 4th AIAA High-Lift Prediction Workshop[35] demonstrated that
mesh adaptation controlling interpolation error or output error clearly improved the consistency of RANS-based
maximum lift calculations over more expensive expert-crafted mesh systems for a transport aircraft model with
deployed high-lift devices. The improvement in consistency is expected to contribute to the understanding of modeling
errors in RANS but did not lead to direct improvements in the comparisons of simulation to experiment. Questions
remain whether or not RANS modeling can improve when the ambiguities of iterative and discretization errors are
controlled or if mesh adaptation applications to scale-resolving simulations can simultaneously reduce modeling and
discretization errors[36].
2022 has seen notable collaborative work between groups from academia and meshing software vendors like MIT
and Cadence[79]. In this work, it is not necessarily the result that is important but the longer-term outcomes of
new cross collaborations between groups that have historically worked separately on topics such mesh adaptation or
baseline meshes for RANS.

2.5 Knowledge Extraction


The Knowledge Extraction Domain consists of two timelines: Integrated Databases and Visualization. There has
been activity within the community that contribute towards the technology milestones and demonstrations.

2.5.1 Integrated Databases - Invited Panel on Physics Based Model Improvement and Uncertainty
Quantification for the Digital Engineering Transformation
In 2018, the Department of Defense (DoD) published its Digital Engineering Strategy which laid out the path how
Physics-Based models (i.e. Computational Fluid Dynamics, Computational Structural Dynamics, Finite Element
Methods) would be used in the design and acquisition of new vehicles, machines, and components[80]. One definition
of Digital Engineering can be found at the Defense Acquisition University:
Digital Engineering: An integrated digital approach that uses authoritative sources of systems’ data and
models as a continuum across disciplines to support lifecycle activities from concept through disposal [81].
The use of Digital Engineering practices will be put to task via the recent announcement by the Army of the
Future Long-Range Assault Aircraft (FLRAA) contract. The objective of this program is to replace the “Blackhawk”
UH-60 helicopter which was developed in the early 1970’s and has been in service since 1979. Major General Robert
Barrie from the the Army’s Program Executive Officer for Aviation was quoted as saying that Bell Flight, the
winner of the competition, will deliver their preliminary design of the aircraft as “virtual prototypes of a potentially
model-based system.”[82]
The DoD strategy promotes the use of digital representations of systems and components and the use of digital
artifacts to design and sustain national defense systems. The DoD identified five Goals and Focus Areas[80]. Of
these goals, Goal 3, “Incorporate Technological Innovation to Improve the Engineering Practice,” identified many of
the Technology Innovations that were highlighted in the CFD 2030 Vision Report. To achieve these goals, the DoD
formed Focus Areas where for Goal 3 these focus areas are to “Establish an end-to-end digital engineering enterprise”
and to “Use technological innovations to improve the digital engineering practice.” The benefits of Physics-Based
Models and Digital Engineering practices to an acquisition program was highlighted by the successful deployment of
multi-fidelity models applied to the Engine Gas Re-ingestion problem of the CH-53K program.[83]
The DoD strategy and the CFD 2030 Vision intersect in the areas of Physics-Based Models, Computing Technol-
ogy, Data Visualization, Artificial Intelligence, and Big Data Analytics. The CFD 2030 Vision Report specifically
identified research and development needs and associated roadmaps that address the DoD Digital Engineering Goals

5
by enabling the AIAA community to “derive useful information from the simulations,” and to perform “smart knowl-
edge extraction from large-scale databases and simulations, automate the process of sifting through large amounts
of information, often at a number of different geographic locations, and extracting patterns and actionable design
decisions.”
The CFD 2030 Integration Committee recognized the intersection between the DoD Digital Engineering Strategy
and the CFD 2030 Vision and hosted an invited panel session at the AIAA AVIATION 2022 conference that addressed
“How can we use Physics-Based Simulations to create models that enable decision makers to make confident informed
choices in diverse engineering disciplines such as aerodynamics, structures, stability, guidance, and control during a
vehicle design cycle?” The session convened the AIAA community with a panel of leading experts from government,
academia, and industry in the areas of High Fidelity Physics-Based Simulations, Reduced Order/Surrogate Models,
Machine Learning/Artificial Intelligence, Uncertainty Quantification, and Data Extraction and Management. The
presentations and discussions that ensued highlighted the areas of Digital Engineering, surrogate modeling, AI/ML,
UQ, and effective co-processing with data management. It showed that the community requires collaboration across
industry, government labs, and individual companies to achieve the desired Digital Engineering and CFD 2030 goals.
The panel is summarized by Duque et al.[84]
The panel identified several gaps in Knowledge Extraction technology whereby actionable information can be
easily gained from large databases via co-processing and traditional visualization methods. These gaps overlap with
the Databases and Visualization swimlanes. In particular, as HPC systems enable ever larger databases, extracting
knowledge, capturing and managing that information remains a problem. The large systems are making it easier to
create data than to consume it for useful information. Furthermore, cybersecurity concerns need to be addressed
particularly as cloud/remote access further complicates securing data. As has been the case for a long time, solver
codes are not changing fast enough to assist knowledge capture. Most solver codes still rely upon traditional writing
of large data files which is not optimal for the endpoint engineering consumer. Even though co-processing technology
exists, its adoption has been slow. Some may consider data compression as a way out, however its utility is limited.
As databases grow to 1000’s of unsteady CFD simulations plus test data, data provenance and associated metadata
needs to be captured as part of the co-processing so that one can find the data - ”So many files, so much knowledge. . .
where is it?” In addition, traditional visualization techniques will become less useful because classical images and
animations will put too much burden on subjective visual interpretation by a human. This gap provides opportunities
for new innovation in comparative and ML-assisted visualization.

2.5.2 Visualization - Large Compute enables Larger Simulations


As outlined in Sec. 2.1, the highest-performing HPC systems in the world push the limits on the size of data that
can be generated and then processed to extract knowledge. These cases continue to challenge knowledge-extraction
methods and require co-processing methods to enable data extraction for further analysis and visualization.
On the Fugaku system, Prof. Kawai at Tohoku University continues to scale up their simulations. Figure 1
presents a recent simulation using a Wall-Modeled LES of a full aircraft configuration. The solver uses a hierarchical
Cartesian grid and the wall-modeling on non-body conforming boundaries. Figure 1 is from a 12 billion cell case
and, to date, his team has performed simulations on upwards of 49.2 Billion cells.[85]

2.6 MDAO
Milestones in the Multidisciplinary Design Analysis and Optimization domain can be broadly categorized under mul-
tidisciplinary coupling, transient and chaotic design sensitivities, full-vehicle MDAO, and uncertainly quantification
in MDAO. The following sections detail recent progress in those areas.

2.6.1 Multidisciplinary coupling: frameworks and standards


There has been significant activity in the development of coupling frameworks and techniques for a variety of multi-
disciplinary models. Mphys[87] is a project built on OpenMDAO developed with the intent of standardizing certain
multiphysics problems for the purpose of design optimization. Mphys found application this year by Jacobson and
Stanford[88] for flutter-constrained design optimization, by Thelen et al.[89] for multifidelity aeroelastic design op-
timization, as well as by Lamkin et al.[90] who developed a coupled aeropropulsion model for optimization of a
high-bypass turbofan engine. This year, the first Mphys workshop was held in conjunction with the OpenMDAO
workshop to communicate progress and plan future developments. FUNtoFEM[91] is another MDAO framework tar-
geted towards static and transient aeroelastic and aerothermoelastic design optimization that progressed this year.

6
Figure 1: Wall-modeled large-eddy simulation of turbulent flow around full aircraft configuration
with high-lift and landing devices at flight Reynolds number Rec ≈ 4.0 × 107 using fully automated
Cartesian-grid-based compressible flow solver FFVHC-ACE. Isosurfaces of the Q-criterion colored by
streamwise velocity. [86]

Halim et al.[92] demonstrated high-fidelity aerothermoelastic design optimization of stiffened thin-walled structures
using FUNtoFEM. In the work of Thelen et at.[89] FUNtoFEM was utilized to facilitate the high-fidelity aeroelastic
model in their multi-fidelity aeroelastic design optimization study. Rumpfkeil and Beran[93] also used FUNtoFEM
in their studies on optimization of an efficient supersonic air vehicle configuration. An additional development in the
area of MDAO frameworks was detailed by Backhaus et al.[94] who developed a gradient-based, high-fidelity static
aeroelastic MDO infrastructure based on the integration of their FlowSimulator capability with OpenMDAO.

2.6.2 Transient and chaotic sensitivities


The area of transient and chaotic sensitivities has progressed over the past several years. Previous work by Bhatia
and Makhija[95] and Bhatia and Taoudi[96] that focused on approaches for computing sensitivities of chaotic physical
systems were leveraged this year by Stanford[97] for gradient-based optimization of a panel exhibiting chaotic flutter
physics. The work included a demonstrated extension to two-dimensions as well as a utilization of the chaotic
sensitivities for design optimization. Furthermore, Donello et al.[98] developed an approach for computing sensitivities
of transient systems that takes the form of a set of forward evolution equations that are a low-rank approximation of
the sensitivity equations. This approach contrasts with adjoint-based methods by avoiding I/O required for forward-
backward solves and is not tied to a particular objective function. While fundamental in nature, the work has
potential to address multiple barriers for optimization of transient systems.

2.6.3 Towards full-vehicle MDAO


Towards the milestone of full-vehicle multidisciplinary design optimization, there have been some efforts in the past
year targeting better handling of geometry parameterization for the purpose of design, as well as efforts targeting
integration of various subsystem components within coupled vehicle design optimization processes. Wu et al.[99]
proposed an approach for constructing an orthogonal set of design variables for shape optimization. The new design
variable set along with a proposed automatic scaling method for that set of shape variables addresses the challenges

7
of constructing a well-behaved parameterization and associated scaling for optimization. Recovery of rapid quasi-
Newton convergence was demonstrated in contrast to the reference design variable set and scaling. Lupp et al.[100]
developed an approach for coupled optimization of vehicle mission and shape parameters for a HALE aircraft subject
to transient power and thermal constraints. They used the recently developed Dymos package to solve the transient
trajectory and control optimization. While the physical models driving aerodynamics and subsystem responses
were low-fidelity, this effort demonstrates the feasibility of coupling mission, shape, and subsystem parameters for
conceptual and preliminary design activities.

2.6.4 Uncertainty quantification in MDAO


Jofre and Doostan[101] extended a stochastic gradient-descent approach for aerodynamic shape optimization subject
to uncertainties in operating conditions and turbulence-model parameters. In their demonstration, the approach
exhibited a significant reduction in drag for a wide range of operating conditions while requiring a relatively small
number of random samples at each optimization step. Stanford et al.[102] demonstrated the use of low and high-
fidelity frequency-domain methods for several reliability analysis methods focusing on low-probability transonic
flutter. Such approaches are important for injection of flutter-based uncertainty quantification in future design
processes. Finally, Chaudhuri et al.[103] propose an approach to design under uncertainty that they term certifiable
risk-based design optimization (CRIBDO). The first condition of certifiability they put forward is based on data-
informed conservativeness and risk measures that account for magnitude of failure rather that hard-thresholding.
The second condition is that the risk measure preserves underlying convexity. The framework represents a useful
formalism for incorporating uncertainty in design optimization but they note the importance of efficient underlying
approaches for computing risk measures.

3 Overall Observations
The authors of this summary, representing technical leaders from across aerospace industry and government, ac-
knowledge that there may be inadvertent omissions from this summary that is largely focused on achievements
documented within AIAA. CFD progress is being made in other industries, but the present effort focuses on the
solutions and capabilities that are being applied in aerospace analysis. Based on this review, we have documented
progress on each of the milestones identified in the Roadmap and see advances that have occurred in integrated mod-
eling of turbulence transition, machine learning for complex flow simulations, simulations of complex unsteady high
Reynolds number flows, and computations of coupled analytic sensitivities for chaotic systems. As noted in the 2021
summary, the significance of other advancements may not yet be recognized because they have not had the time to
prove themselves. The authors welcome additional comments and can be reached through aiaacfd2030@gmail.com.
Among the most significant accomplishments identified in 2022 were High Lift Prediction Workshop 4 and its
associated activities covering multiple facets of CFD, as well as the NASA Symposium on Turbulence Modeling:
Roadblocks, and the Potential for Machine Learning. These events brought the community together across multiple
skill sets and promoted significant discussion leading to advancements. The imminent potential of increased HPC
capacity as demonstrated by the progress in leadership class computing on Frontier and Aurora supercomputers is
also a significant step forward for CFD because of the additional capabilities that it enables.
As the year 2030 approaches, there are a number of the Domains that are not seeing sufficient progress to meet
the Vision objectives. There are several underlying issues that the authors have suggested may be contributing to
this lack of advancement, focused around aerospace work force challenges and HPC. While advances in computing
capability are enabling changes in CFD, they are also making changes across engineering. Some of these fields are
enticing leading engineers and researchers out of Aerospace as other fields, including AI/ML, are providing additional
exciting and lucrative alternatives. With the many new challenges associated with privatization of space travel and
urban air mobility, as well as the many unresolved challenges in traditional aerospace vehicles, it is important to
maintain interest in CFD and highlight the meaningful challenges that still remain to be solved. To reach the potential
of CFD Vision 2030, we must continue to engage the highest quality of talent in the aerospace field. Furthermore,
this talent needs to be armed with adequate computing capability to meet the needs for both research and practice.
Leadership class computing is vital to demonstrate breakthroughs, but it must be supplemented by the large amount
of production computing to perform detailed engineering. There is presently a gap in the availability of this class
of computing for the aerospace industry and growth in adjacent domains like AI/ML is providing competition for
available compute resources. Despite the increased computing at leadership facilities, computing resource shortages
may slow the progress toward the Vision. Reflecting the significance and potential of CFD, private companies

8
are increasingly pushing a larger proportion of major CFD advances. This development represents a shift from
traditional CFD practices as more of these advances are held as proprietary. It remains to be seen how this will
impact industry-wide growth in CFD. Efforts should be made to encourage leading-edge development and open
research from disparate sources to benefit from a diversity of opinions and accelerate the development toward the
Vision.

Acknowledgments
The AIAA CFD Vision 2030 Roadmap subcommittee compiled this report with primary authors Andrew Cary, John
Chawner, Earl Duque, Bill Gropp, Eric Nielsen, Brian Smith, and Nathan Wukie. This team leveraged a much wider
network for inputs, including:

• Algorithm contributors; Dimiti Mavriplis, Z. J. Wang


• Geometry contributors: Mike Park, Adrien Loseille, Nigel Taylor, Ben Urick, with editing by Steve Karman
• Knowledge extraction contributors: Steve M. Legensky, Atsushi Toyoda, Soshi Kawai

• MDAO contributors: Bret Stanford, Sean Wakayama, Geojoe Kuruvila, Daniel L. Clark
The section on MDAO has been cleared for public release under case number AFRL-2023-4770.

Appendix: Milestones and TRL Rating History


Each year, the Roadmap Subcommittee reviews progress related to each of the milestones on the Roadmap and
assesses progress following a Technology Readiness Level scale described in Cary, et al.[104] The intent of this
assessment is to identify technology items that are at risk of not meeting target milestone dates and to track ongoing
development progress. Figure 2 contains a summary of these reviews.

9
Milestone
2020 2021 2022
year
HPC
CFD on massively parallel systems
Demonstrate implementation of CFD algorithms for extreme parallelism in CFD codes (e.g.,
2019 6 6 6
FUN3D) TRL key:
Demonstrate efficiently scaled CFD simulation capability on an exascale system 2024 0 0 0 0
30 exaFLOPS, unsteady, maneuvering flight, full engine simulation (with combustion) 2030 0 0 0 1
CFD on Revolutionary Systems (Quantum, Bio, etc.) 2
Demonstrate solution of a representative model problem 2023 2 2 2 3
Demonstrate solution of a representative model problem 2027 0 0 0 4
Physical Modeling 5
RANS 6
Improved RST models in CFD codes 2016 7 7 7 7
Integrated transition prediction (Tollmein-Schlichting modeling) 2017 6 6 6 8
Integrated transition prediction (non-TS) 2027 3 3 4 9
Highly accurate RST models for flow separation 2025 2 2 2
Demonstration of machine learning to simulation of complex flow regime 2025 1 3 4
Hybrid RANS/LES
Integrated transition prediction 2025 2 2 2
Unsteady, complex geometry, separated flow at flight Reynolds number (e.g., high lift) 2023 4 4 5
LES
Integrated transition prediction 2025 2 2 2
WMLES/WRLES for complex 3D flows at appropriate Re 2023 5 5 5
Unsteady, 3D geometry, separated flow (e.g., rotating turbomachinery with reactions) 2027 3 3 3
Combustion
Chemical kinetics calculation speedup 2017 3 3 3
Chemical kinetics in LES 2021 4 4 4
Multiregime turbulence-chemistry interaction model 2025 3 3 3
Unsteady, 3D geometry, separated flow (e.g., rotating turbomachinery with reactions) 2027 3 3 3

Figure 2: Milestone summary with TRL assessment (continued).

10
Milestone
2020 2021 2022
year
Algorithms
Convergence/Robustness
Automated robust solvers 2022 7 7 7 TRL key:
Unsteady, complex geometry, separated flow at flight Reynolds number (e.g., high lift) 2023 7 7 8 0
Scalable optimal solvers 2021 6 6 6 1
Improved discretizations for scale-resolving methods (low-dissipation, HO,…) 2024 5 5 6 2
Accurate and robust methods for long time integration 2026 2 2 2 3
Production scalable entropy-stable solvers 2029 3 3 3 4
Uncertainty Quantification (UQ) 5
Characterization of UQ in aerospace 2023 4 4 4 6
Reliable error estimates in CFD codes 2025 5 5 5 7
Uncertainty propagation capabilities in CFD 2022 4 4 4 8
Identification of tail events/probabilities from CFD codes 2027 3 3 3 9
Large scale stochastic capabilities in CFD 2030 0 0 0
Geometry Modeling and Mesh Generation
Geometry Modeling
Quantified, reversible data transfer demonstrated between opaque and open geometry model
2023 5 5 5
representations.
Associative equivalence demonstrated for OML manipulation schemes. 2025 0 3 3
Distributed, open geometry representation platform established 2027 0 2 2
Robust, quantifiable multidisciplinary data exchange supported by open data standard. 2029 0 2 2
HPC Meshing
Large-scale parallel mesh generation 2020 5 5 5
Generate a 100 billion cell, fit-for-purpose volume mesh. 2025 1 1 1
Generate a 1 trillion cell, fit-for-purpose volume mesh. 2030 1 1 1
Fixed Meshing
Tighter CAD coupling 2015 9 9 9
CAD coupling available in commercial grid generation 2023 5 5 5
Automatic generation of suitable mesh on complex geometry on 1st attempt. 2021 4 4 4
Automatic generation of a family of meshes about a complex configuration. 2023 4 4 4
Adaptive Grid
Production AMR in CFD codes 2016 5 5 5
Adaptive meshing techniques will accept typical assembly tolerance levels and unfavorable B-
2023 0 4 4
Rep topologies to accept a pragmatic interpretation of geometry.
Adaptive curved meshing to support higher-order solvers will be available from multiple
2026 0 3 3
implementations.
Accurate CFD solutions are verified by asymptotic convergence rate demonstration or low
2028 0 3 3
variation between independent implementations.
Adaptive mesh computations displace fixed meshes as the default and practitioners will rarely
2030 0 2 2
visualize the mesh directly.
Knowledge Extraction
Integrated Databases
Simplified data representation 2017 3 3 3
Accepted data fusion techniques 2026 3 3 3
Creation of real-time multifidelity database: 1000 unsteady CFD simulations plus test data with
2025 2 3 3
complete UQ of all data sources
Visualization
On demand analysis/visualization of a 10B point unsteady CFD simulation 2022 4 4 4
On demand analysis/visualization of a 100B point unsteady CFD simulation 2025 2 2 2
MDAO
Define standard for coupling to other disciplines 2016 4 4 4
High fidelity coupling techniques/frameworks 2017 4 4 4
Robust CFD for complex MDAs 2019 4 4 4
Incorporation of UQ for MDAO 2025 2 2 2
MDAO simulation of an entire aircraft (e.g., aeroacoustics) 2027 3 3 3
UQ-enabled MDAO 2030 1 1 1
Full vehicle coupled analytic sensitivities, including geometric and subsystems 2025 4 4 4
Full Vehicle coupled analytic sensitivites for chaotic systems 2030 0 0 1

Figure 2: Milestone summary with TRL assessment (concluded).

11
References
[1] Jeffrey Slotnick, Abdollah Khodadoust, Juan Alonso, David Darmofal, William Gropp, Elizabeth Lurie, and
Dimitri Mavriplis. CFD Vision 2030 study: A path to revolutionary computational aerosciences. NASA/CR-
2014-218178, March 2014. URL https://ntrs.nasa.gov/citations/20140003093.
[2] S. Shoemaker. Frontier Supercomputer Debuts as World’s Fastest, Breaking Exascale Barrier, May 2022. URL
https://www.ornl.gov/news/frontier-supercomputer-debuts-worlds-fastest-breaking-exascale-
barrier.
[3] T. Trader. From Exasperation to Exascale: HPE’s Nic Dubé on Frontier’s Untold Story, December 2022. URL
https://www.hpcwire.com/2022/12/02/at-sc22-hpes-nic-dube-on-frontiers-untold-story/.
[4] E. Strohmaier, J. Dongarra, H. Simon, and M. Meuer. The Top500, 2022. URL https://top500.org.
[5] E. Nielsen. Big Compute 22: The Impact of Emerging Architectures on Engineering and Leadership-Class
Computational Fluid Dynamics, 2022. URL https://www.bigcompute.org/big-compute-22/the-impact-
of-emerging-architectures-on-engineering-and-leadership-class-computational-fluid-dynamics.
[6] D. Jude, J. Sitaraman, and A. Wissink. An octree-based, Cartesian Navier-Stokes solver for modern cluster
architectures. Journal of Supercomputing, 78:11409–11440, 2022. doi: 10.1007/s11227-022-04324-7.
[7] Jay Sitaraman, Dylan Jude, Beatrice Roget, Steven A. Tran, Jennifer Abras, Jonhoon Bin, and Jeremy Ship-
man. Algorithmic improvements and capability enhancements to PUNDIT. AIAA Paper 2022-1175, January
2022. doi: 10.2514/6.2022-1175.
[8] Z. J. Wang. High order wall-modeled large-eddy simulation on mixed unstructured meshes. AIAA Journal, 60
(12):6881–6896, 2022. doi: 10.2514/1.J061641.
[9] Salman K. Rahmani and Z. J. Wang. Large eddy simulation of the Sandia axisymmetric transonic hump using
a high-order method. AIAA Paper 2022-1534, January 2022. doi: 10.2514/6.2022-1534.
[10] Konrad Goc, Sanjeeb T. Bose, and Parviz Moin. Large eddy simulation of the NASA high-lift common research
model. AIAA Paper 2022-1556, January 2022. doi: 10.2514/6.2022-1556.
[11] Guillaume A. Brès, Sanjeeb T. Bose, Christopher B. Ivey, Michael Emory, and Frank Ham. GPU-accelerated
large-eddy simulations of supersonic jets from twin rectangular nozzle. AIAA Paper 2022-3001, 2022. doi:
10.2514/6.2022-3001.
[12] Michael Laufer, Steven H. Frankel, and David Greenblatt. GPU-accelerated implicit large eddy simulation of
a naca 0018 airfoil with active flow control. AIAA Paper 2022-0471, January 2022. doi: 10.2514/6.2022-0471.
[13] Hussain Ali Abid, Annabel P. Markesteijn, Vasily Gryazev, Sergey A. Karabasov, Hasan Kamliya Jawahar,
and Mahdi Azarpeyvand. Jet installation noise modelling informed by GPU LES. AIAA Paper 2022-2906,
June 2022. doi: 10.2514/6.2022-2906.
[14] Vasily Gryazev, Annabel P. Markesteijn, and Sergey A. Karabasov. Low-order modelling of NASA three-
stream jets using acoustic analogy informed by large eddy simulations. AIAA Paper 2022-3000, June 2022.
doi: 10.2514/6.2022-3000.
[15] Lidia Caros, Oliver Buxton, Tsuyoshi Shigeta, Takayuki Nagata, Taku Nonomura, Keisuke Asai, and Peter
Vincent. Direct numerical simulation of flow over a triangular airfoil under martian conditions. AIAA Journal,
60(7):3961–3972, 2022. doi: 10.2514/1.J061454.
[16] Ngoc Cuong Nguyen, Sebastien Terrana, and Jaime Peraire. Large-eddy simulation of transonic buffet using
matrix-free discontinuous Galerkin method. AIAA Journal, 60(5):3060–3077, 2022. doi: 10.2514/1.J060459.
[17] David A. Kessler, Andrew M. Hess, Keith Obenschain, David C. Eder, Alice Koniges, Anthony Knutson, Gra-
ham V. Candler, Heath Johnson, Spencer Starr, Joel Bretheim, Kevin Roe, Eric J. Nielsen, Aaron Walden,
Gabriel Nastac, Kevin Jacobson, Roy Campbell, David R. McDaniel, and Ryan B. Bond. Performance of
coupled physics solvers for multidisciplinary hypersonic flow simulations on several classes of computer archi-
tectures. AIAA Paper 2022-0973, January 2022. doi: 10.2514/6.2022-0973.

12
[18] Bret Stanford, Kevin Jacobson, and Pawel Chwalowski. Ongoing aeroelastic prediction and validation activities
at NASA Langley Research Center. AIAA Paper 2022-1557, January 2022. doi: 10.2514/6.2022-1557.
[19] Fabio Casagrande Hirono, Antonio Torija Martinez, Andrew Elliott, James Taylor, Samuel Grimshaw, and
Demetrios Lefas. Aeroacoustic design and optimisation of an all-electric ducted fan propulsion module for
low-noise impact. AIAA Paper 2022-3034, June 2022. doi: 10.2514/6.2022-3034.
[20] Thomas R. Wainwright, Daniel J. Poole, and Christian B. Allen. GPU-accelerated aerodynamic shape optimi-
sation framework for large turbine blades. AIAA Paper 2022-1292, January 2022. doi: 10.2514/6.2022-1292.
[21] Yoshitaka Nakashima, Yuya Ando, Seiichi Taguchi, Takao Itami, and Tomohiro Irie. RANS solutions on high
lift common research model with scFLOW, a polyhedral finite-volume solver. AIAA Paper 2022-3522, June
2022. doi: 10.2514/6.2022-3522.
[22] C. Jackson. OVERFLOW on GPUs: Progress and lessons learned. In 15th Symposium on Overset Composite
Grids and Solution Technology, November 2022.
[23] Laurence Kedward and Christian B. Allen. Summary of investigations into finite volume methods on GPUs,
January 2022. doi: 10.2514/6.2022-0028.
[24] Bumseok Lee, Yong Su Jung, and James D. Baeder. Validation of SA-γ-Reθt and SA-γ turbulence/transition
model. AIAA Paper 2022-0910, January 2022. doi: 10.2514/6.2022-0910.
[25] Ganesh Nampelly, Ananth Sivaramakrishnan Malathi, Nagabhushana Rao Vadlamani, Sriram Rengarajan, and
Konstantinos Kontis. Surface roughness benefits in open cavity flows. AIAA Paper 2022-0473, January 2022.
doi: 10.2514/6.2022-0473.
[26] Tao Tang, Zhenguo Wang, Yuhui Huang, Mingbo Sun, and Hongbo Wang. Investigation of combustion structure
and flame stabilization in an axisymmetric scramjet. AIAA Journal, 61(2):585–601, 2023. doi: 10 . 2514 / 1 .
J062192.
[27] Johan Larsson, Vedant Kumar, Nikhil Oberoi, Mario Di Renzo, and Sergio Pirozzoli. Large-eddy simulations
of idealized shock/boundary-layer interactions with crossflow. AIAA Journal, 60(5):2767–2779, 2022. doi:
10.2514/1.J061060.
[28] Kedar Prashant Shete, David J. Boucher, James J. Riley, and Stephen M. de Bruyn Kops. Effect of viscous-
convective subrange on passive scalar statistics at high Reynolds number. Phys. Rev. Fluids, 7:024601, February
2022. doi: 10.1103/PhysRevFluids.7.024601.
[29] Miles M. P. Couchman, Stephen M. de Bruyn Kops, and Colm-cille P. Caulfield. Transport across stable
density interfaces in forced stratified turbulence, 2022.
[30] Ashley M. Korzun, Gabriel Nastac, Aaron Walden, Eric J. Nielsen, William T. Jones, and Patrick Moran.
Application of a detached eddy simulation approach with finite-rate chemistry to mars-relevant retropropulsion
operating environments. AIAA Paper 2022-2298, January 2022. doi: 10.2514/6.2022-2298.
[31] H. Asada and S. Kawai. Fully-automated high-fidelity LES around high-lift aircraft configuration near stall.
In Eleventh International Conference on Computational Fluid Dynamics, 2022.
[32] Yoshiharu Tamaki, Hiroyuki Asada, Ryoji Takaki, and Soshi Kawai. Wall-modeled LES around the CRM-HL
using fully-automated Cartesian-grid-based flow solver FFVHC-ACE. AIAA Paper 2022-3435, June 2022. doi:
10.2514/6.2022-3435.
[33] Youngdae Kim, Debojyoti Ghosh, Emil M. Constantinescu, and Ramesh Balakrishnan. GPU-accelerated dns
of compressible turbulent flows, 2022.
[34] Stephan Priebe, Trevor Wood, Junsok Yi, and Arash Mousavi. Large eddy simulation of an open rotor fan
blade. In Volume 10A: Turbomachinery - Axial Flow Fan and Compressor Aerodynamics, Turbo Expo: Power
for Land, Sea, and Air. ASME, June 2022. doi: 10.1115/GT2022-80538.
[35] Christopher L. Rumsey, Jeffrey P. Slotnick, and Carolyn D. Woeber. HLPW-4/GMGW-3: Overview and
workshop summary. AIAA Paper 2022–3295, June 2022. doi: 10.2514/6.2022-3295.

13
[36] Michael A. Park, Frédéric Alauzet, and Todd Michal. HLPW-4/GMGW-3: Mesh adaptation for RANS tech-
nology focus group workshop summary. AIAA Paper 2022–3210, June 2022. doi: 10.2514/6.2022-3210.
[37] Neil Ashton, Paul Batten, Andrew Cary, and Kevin Holst. HLPW-4/GMGW-3: Hybrid RANS/LES technology
focus group workshop summary. AIAA Paper 2022–3293, June 2022. doi: 10.2514/6.2022-3293.

[38] Cetin Kiris, A. Ghate, Oliver Browne, Jeff Slotnick, and Johan Larsson. HLPW-4/GMGW-3: Wall-modeled
LES and lattice-boltzmann technology focus group workshop summary. AIAA Paper 2022–3294, June 2022.
doi: 10.2514/6.2022-3294.
[39] Brian Smith and M. A. McWaters. Aerodynamic database requirements for the detailed design of tactical
aircraft: Implications for the expanded application of CFD. NATO MP-AVT-366-02, September 2022. doi:
10.14339/STO-MP-AVT-366.
[40] Chris Rumsey and Gary Coleman. NASA symposium on turbulence modeling: Roadblocks, and the potential
for machine learning. NASA/TM 2022001595, November 2022.
[41] Philippe Spalart. An Old-Fashioned Framework for Machine Learning in Turbulence Modeling, 2022. Keynote
lecture at NASA 2022 Symposium on Turbulence Modeling: Roadblocks, and the Potential for Machine Learn-
ing.
[42] Emmett Padway and Hiroaki Nishikawa. Resolving confusions over third-order accuracy of unstructured
MUSCL. AIAA Journal, 60(3):1415–1439, 2022. doi: 10.2514/1.J060773.
[43] Yoonpyo Hong, Kwanjung Yee, and Soo Hyung Park. Assessment of high-order schemes for vortex-dominated
flowfield using truncation error analysis. AIAA Paper 2022-4121, June 2022. doi: 10.2514/6.2022-4121.
[44] Yoshikatsu Furusawa and Keiichi Kitamura. Roles of multi-dimensional velocity components in all-speed
numerical flux SLAU. AIAA Paper 2022-4033, June 2022. doi: 10.2514/6.2022-4033.
[45] M. Galbraith and S. Karman. Hlpw-4/gmgw-3: High order discretization technology focus group workshop
summary. AIAA Paper 2022-3292, January 2022. doi: 10.2514/6.2022-3292.

[46] Emmett Padway and Dimitri J. Mavriplis. Inexact linearization of fixed point iterations in the piggy-back
iterations of the one-shot adjoint. AIAA Paper 2022-1613, January 2022. doi: 10.2514/6.2022-1613.
[47] Haimin Huang and Matteo Giacobello. Uncertainty quantification in store separation analysis using Kestrel,
design of experiments and surrogate modelling. AIAA Paper 2022-1318, January 2022. doi: 10.2514/6.2022-
1318.
[48] Huseyin Emre Tekaslan, Sihmehmet Yildiz, Yusuf Demiroglu, and Melike Nikbay. Implementation of multidis-
ciplinary and multifidelity uncertainty quantification methods for sonic boom prediction. Journal of Aircraft,
60(2):410–422, 2023. doi: 10.2514/1.C036962.
[49] Makoto Endo and Ben D. Phillips. Uncertainty quantification of CFD model assumptions against sonic boom
noise prediction of a commercial supersonic transport. AIAA Paper 2022-0401, January 2022. doi: 10.2514/6.
2022-0401.
[50] Ragini Acharya, Kevin Hartman, and Venkateswaran Narayanaswamy. Multifidelity reduced-dimensional mod-
eling for shock-fin interaction. AIAA Paper 2022-3479, June 2022. doi: 10.2514/6.2022-3479.

[51] Kai Li, Jiaqing Kou, and Weiwei Zhang. Deep learning for multifidelity aerodynamic distribution modeling
from experimental and simulation data. AIAA Journal, 60(7):4413–4427, 2022. doi: 10.2514/1.J061330.
[52] Mehdi Anhichem, Sebastian Timme, Jony Castagna, Andrew Peace, and Moira Maina. Multifidelity data fusion
applied to aircraft wing pressure distribution. AIAA Paper 2022-3536, June 2022. doi: 10.2514/6.2022-3526.
[53] Matthew J. Schwartz, Datta V. Gaitonde, and John W. Slater. Uncertainty and sensitivity analysis of bleed
modeling in shock/turbulent interactions. Journal of Propulsion and Power, 39(1):106–120, 2023. doi: 10 .
2514/1.B38785.

14
[54] John A. Schaefer, Andrew W. Cary, Mori Mani, Thomas A. Grandine, Christopher J. Roy, and Heng Xiao.
Uncertainty quantification across design space using spatially accurate polynomial chaos. AIAA Journal, 60
(3):1482–1504, 2022. doi: 10.2514/1.J060333.
[55] Jolan Wauters, Joris Degroote, Ivo Couckuyt, and Guillaume Crevecoeur. SAMURAI: A new asynchronous
Bayesian optimization technique for optimization-under-uncertainty. AIAA Journal, 60(11):6133–6156, 2022.
doi: 10.2514/1.J061112.
[56] Bingran Wang, Mark Sperry, Victor E. Gandarillas, and John T. Hwang. Efficient uncertainty propagation
through computational graph modification and automatic code generation. AIAA Paper 2022-3997, June 2022.
doi: 10.2514/6.2022-3997.
[57] Philipp Bekemeyer, Anna Bertram, Derrick A. Hines Chaves, Mateus Dias Ribeiro, Andrea Garbo, Anna
Kiener, Christian Sabater, Mario Stradtner, Simon Wassing, Markus Widhalm, Stefan Goertz, Florian Jaeckel,
Robert Hoppe, and Nils Hoffmann. Data-driven aerodynamic modeling using the DLR SMARTy toolbox.
AIAA Paper 2022-3899, June 2022. doi: 10.2514/6.2022-3899.
[58] Jin-ping Li, Fan-Zhi Zeng, Zhen-Hua Jiang, Yao Li, and Chao Yan. Investigations on turbulence model
uncertainty for hypersonic shock-wave/boundary-layer interaction flows. AIAA Journal, 60(8):4509–4522, 2022.
doi: 10.2514/1.J061355.
[59] Fanzhi Zeng, Wei Zhang, Jinping Li, Tianxin Zhang, and Chao Yan. Adaptive model refinement approach
for Bayesian uncertainty quantification in turbulence model. AIAA Journal, 60(6):3502–3516, 2022. doi:
10.2514/1.J060889.
[60] Owen J. Williams, Hariprasad Annamalai, Thomas A. Ozoroski, Christopher J. Roy, and Todd Lowe. Compar-
ison of hill-type geometries for the validation and advancement of turbulence models. AIAA Paper 2022-1032,
January 2022. doi: 10.2514/6.2022-1032.
[61] Marcel Matha, Karsten Kucharczyk, and Christian Morsbach. Assessment of data-driven Reynolds stress tensor
perturbations for uncertainty quantification of RANS turbulence models. AIAA Paper 2022-3767, June 2022.
doi: 10.2514/6.2022-3767.
[62] Timothy Mauery, Juan J. Alonso, Andrew W. Cary, Vincent Lee, Robert Malecki, Dimitri J. Mavriplis, Gorazd
Medic, John Schaefer, and Jeffrey P. Slotnick. A 20-year vision for flight and engine certification by analysis.
AIAA Paper 2022-1553, January 2022. doi: 10.2514/6.2022-1553.
[63] Nigel J. Taylor. CFD validation: what is it and how do we effect it? AIAA Paper 2022-3768, June 2022. doi:
10.2514/6.2022-3768.
[64] William A. Jordan, Hongyu Wang, and Christopher J. Roy. Application of error transport equations to unsteady
flows. AIAA Paper 2022-1572, January 2022. doi: 10.2514/6.2022-1572.
[65] R. A. McDonald and J. R. Gloudemans. Open Vehicle Sketch Pad: An open source parametric geometry and
analysis tool for conceptual aircraft design. AIAA Paper 2022–0004, January 2022. doi: 10.2514/6.2022-0004.
[66] P. Mokotoff and J. Dannenhoffer. GLOVES: A graphical layout of vehicle systems for the Engineering Sketch
Pad. AIAA Paper 2022–3493, June 2022. doi: 10.2514/6.2022-3493.
[67] Maximillian Wood, T. Rendall, Christian B. Allen, Laurence Kedward, Nigel J. Taylor, James Fincham, and
Nick E. Leppard. Constructing geometry for aerodynamic shape optimisation using volume of solid driven
cellular automata. AIAA Paper 2022–1861, January 2022. doi: 10.2514/6.2022-1861.
[68] D. Nolan. Defining simulation intent. Computer-Aided Design, 59:50–63, February 2015. doi: 10.1016/j.cad.
2014.08.030.
[69] Andrew R. Colligan, Trevor T. Robinson, Declan C. Nolan, Yang Hua, and Weijuan Cao. Hierarchical CADNet:
Learning from B-Reps for machining feature. Computer-Aided Design, 147, May 2022. doi: 10.1016/j.cad.
2022.103226.
[70] Nigel J. Taylor. Digital engineering: Recognizing and honing our 6th sense with respect to digital thread.
AIAA Paper 2022–3234, June 2022. doi: 10.2514/6.2022-3234.

15
[71] 2022. URL https://www.plm.automation.siemens.com/global/en/our-story/glossary/convergent-
modeling/24461. Accessed: September 6, 2023.
[72] 2022. URL https://oneplm.com/wp-content/uploads/2020/05/siemens_parasolid_with_convergent_
modeling_fact_sheet_tcm27-70396.pdf. Accessed: September 6, 2023.
[73] 2022. URL https://solidedge.siemens.com/en/solutions/products/3d- design/next- generation-
design/convergent-modeling/. Accessed: September 6, 2023.
[74] 2022. URL https : / / www . engineering . com / story / mesh - modeling - part - 2 - siemens - convergent -
modeling-for-meshes-and-nurbs-based-geometry. Accessed: September 6, 2023.
[75] 2022. URL https://www.onshape.com/en/resource-center/what-is-new/mixed-modeling-replicate-
tool-in-context-editing. Accessed: September 6, 2023.
[76] 2022. URL https://cad.onshape.com/help/Content/mixedmodeling.htm. Accessed: September 6, 2023.
[77] 2022. URL https://www.spatial.com/products/cgm_polyhedra. Accessed: September 6, 2023.
[78] 2022. URL https://cdn2.hubspot.net/hubfs/10956/_DATASHEETS/2017/2017SPATIAL_CGM_Polyhedra.
pdf. Accessed: September 6, 2023.
[79] Steven R. Allmaras, Marshall C. Galbraith, and Nicholas J. Wyman. Output-based mesh adaptation using
commercial mesh generation software. AIAA Paper 2022–4084, June 2022. doi: 10.2514/6.2022-4084.
[80] M. D. Griffin, K. Baldwin, J. Stanley, R. H. Kewley, and W. Bray. Department of Defense Digital Engineering
Strategy, 2018. URL https://ac.cto.mil/wp-content/uploads/2019/06/USA001603-18-DSD.pdf.
[81] DAU. Definition of digital engineering, 2022. URL https : / / www . dau . edu / glossary / Pages / Glossary.
Accessed: December 12, 2022.
[82] Ashley Roque. Bell textron’s valor wins army’s flraa competition to replace black hawk, December 2022. URL
https://breakingdefense-com.cdn.ampproject.org/c/s/breakingdefense.com/2022/12/bells-valor-
wins-armys-future-long-range-assault-aircraft-competition-to-replace-black-hawk/?amp=1.
[83] Shyam Neerarambam, Patrick O. Bowles, Byung-Young Min, Donald Lamb, Alexander F. Dunn, Jonathan
Frydman, Greg Harrington, Chenzhou Lian, Mike Kazlauskas, Brian E. Wake, Noopur Joshi, Noah Becker,
James R. Forsythe, Russell W. Powers, Zachary Stratton, Craig Collins, John Spyropoulos, Buvana Jayaraman,
Joseph Simonetti, Chris Foti, and Jim Axtell. An overview of the exhaust gas reingestion challenges on the
CH-53K King Stallion. AIAA Paper 2021-0028, January 2021. doi: 10.2514/6.2021-0028.
[84] E. P. N. Duque, S. A. Morton, A. Wissink, S. W. Stone, W. Spotz, D. J. Caraway, and S. M. Legensky.
Summary of the CFD 2030 integration committee invited panel on physics based model improvement and
uncertainty quantification for the digital engineering transformation. AIAA Paper 2023-1200, January 2023.
doi: 10.2514/6.2023-1200.
[85] S. Kawai. personal communication, December 2022.
[86] H. Asada, Y. Tamaki, Y. Tataki, T. Yumitori, S. Tamura, K. Hatanaka, K. Imai, H. Maeyama, and S. Kawai.
FFVHC-ACE: Fully automated Cartesian-grid-based solver for compressible large-eddy simulation. AIAA
Journal, April 2023. doi: 10.2514/1.J062593.
[87] A. Yildirim, Kevin E. Jacobson, Bret K. Stanford, Justin S. Gray, C. A. Mader, J. R. R. A. Martins, and
G. J. Kennedy. MPhys: A modular multiphysics simulation package using the OpenMDAO framework. AIAA
Aviation Forum, virtual presentation, 2020. URL https://github.com/OpenMDAO/mphys. Accessed: July 7,
2023.
[88] Kevin Jacobson and Bret Stanford. Flutter-constrained optimization with the linearized frequency-domain
approach. AIAA Paper 2022-2242, January 2022. doi: 10.2514/6.2022-2242.
[89] Andrew S. Thelen, Dean E. Bryson, Bret K. Stanford, and Philip S. Beran. Multi-fidelity gradient-based
optimization for high-dimensional aeroelastic configurations. Algorithms, 15(4), 2022. ISSN 1999-4893. doi:
10.3390/a15040131.

16
[90] A. H. R. Lamkin, A. Yildirim, and J. R. R. A. Martins. Coupled aeropropulsive analysis and optimization of a
high bypass turbofan engine. In Congress of the International Council of the Aeronautical Sciences, September
2022.
[91] Kevin Jacobson, Jan F. Kiviaho, Marilyn J. Smith, and Graeme Kennedy. An aeroelastic coupling framework
for time-accurate analysis and optimization. AIAA Paper 2018-0100, 2018. doi: 10.2514/6.2018-0100.

[92] Lenard J. Halim, Sejal Sahu, Graeme Kennedy, and Marilyn J. Smith. Aerothermoelastic analysis and opti-
mization of stiffened thin-walled structures. AIAA Paper 2022-1612, January 2022. doi: 10.2514/6.2022-1612.
[93] Markus P. Rumpfkeil and Philip S. Beran. Aeroelastic analysis and optimization using FUNtoFEM of an
efficient supersonic air vehicle. AIAA Paper 2022-2094, January 2022. doi: 10.2514/6.2022-2094.

[94] Thomas Backhaus, Sebastian Gottfried, Andrei Merle, John T. Hwang, and Arthur Stueck. Modularization of
high-fidelity static aeroelastic MDO enabling a framework-based optimization approach for HPC. AIAA Paper
2021-1236, January 2022. doi: 10.2514/6.2021-1236.
[95] Manav Bhatia and David Makhija. Sensitivity analysis of time-averaged quantities of chaotic systems. AIAA
Journal, 57(5):2088–2099, 2019. doi: 10.2514/1.J057522.

[96] Manav Bhatia and Lamiae Taoudi. Assessment of stabilized sensitivity analysis approach for high-dimensional
chaotic systems. AIAA Paper 2019-0170, January 2019. doi: 10.2514/6.2019-0170.
[97] Bret Stanford. Gradient based optimization of chaotic panel flutter. AIAA Paper 2022-3786, June 2022. doi:
10.2514/6.2022-3786.

[98] Michael Donello, Mark H. Carpenter, and Hessam Babaee. Computing sensitivities in evolutionary systems:
A real-time reduced order modeling strategy. SIAM Journal on Scientific Computing, 44(1):A128–A149, 2022.
doi: 10.1137/20M1388565.
[99] Neil Wu, Charles Mader, and Joaquim R. Martins. Sensitivity-based geometric parameterization for aerody-
namic shape optimization. AIAA Paper 2022-3931, June 2022. doi: 10.2514/6.2022-3931.

[100] Christopher A. Lupp, Daniel L. Clark, Christopher T. Aksland, and Andrew G. Alleyne. Mission and shape
optimization of a hale aircraft including transient power and thermal constraints. AIAA Paper 2022-3935, June
2022. doi: 10.2514/6.2022-3935.
[101] L. Jofre and A. Doostan. Rapid aerodynamic shape optimization under uncertainty using a stochastic gradient
approach. Structural and Multidiscipinary Optimization, 65, June 2022.
[102] Bret Stanford, Annie Sauer, Kevin Jacobson, and James Warner. Gradient-enhanced reliability analysis of
transonic aeroelastic flutter. AIAA Paper 2022-0632, January 2022. doi: 10.2514/6.2022-0632.
[103] Anirban Chaudhuri, Boris Kramer, Matthew Norton, Johannes O. Royset, and Karen Willcox. Certifiable
risk-based engineering design optimization. AIAA Journal, 60(2):551–565, 2022. doi: 10.2514/1.J060539.

[104] Andrew Cary, John Chawner, Earl Duque, William Gropp, Bil Kleb, Ray Kolonay, Eric Nielsen, and Brian
Smith. CFD Vision 2030 Road Map: Progress and Perspectives. AIAA Paper 2021–2726, August 2021. doi:
10.2514/6.2021-2726.

17

You might also like