Surrogate modeling plays a crucial role in connecting high-fidelity data generation techniques with everyday engineering tasks. These surrogate models act as quick substitutes for more expensive, high-fidelity methods, albeit with some loss in accuracy. The success of these models largely hinges on the quality of the data used to train them. This data should offer enough information to educate the surrogate model about the crucial aspects within the domain of interest. However, as problems become more intricate and involve higher dimensions, it becomes increasingly challenging to determine in advance the ideal data distribution necessary to accurately capture the non-linearities in the domain. Consequently, employing adaptive training methods to pinpoint where additional data is required becomes critical for producing a high-quality surrogate model. This effort focuses on demonstrating the application of adaptive training to enhance the underlying data distribution used for training a machine learning-based surrogate model. The primary goal here is to reduce the time required to obtain a solution, rather than aiming for a perfectly optimized data distribution.

The rapid prediction of loads for engineering applications is of high interest in the aerodynamic community. An approach for obtaining these rapid predictions is through the use of surrogate modeling. Surrogate models enable a faster turn-around time for various engineering needs that high-fidelity computational models cannot accommodate. A machine learning (ML) framework to support surrogate modeling of integrated aerodynamic loads predictions for aircraft is investigated in this effort. The ML framework includes core Deep Neural Network (DNN) components built to support surrogate models for both steady and unsteady aerodynamics. A vital aspect of a successful surrogate model is the prediction accuracy when non-linear flow phenomena such as flow separation and transonic effects impact the flowfield. For highly separated flows – such as dynamic stall – a two-step, novel physics-state predictor approach is laid out. The dual framework employs an intermediate physics-state predictor that enables the surrogate model to more accurately model dynamic separated and non-linear flow patterns. Parameter studies are conducted to investigate the impact of different input feature sets on the surrogate model's predictive capabilities. A discussion on the potential benefits of incorporating Convolutional Neural Network (CNN) based DNN architectures for the surrogate model is also included. Initial 2D/3D verification results for steady and unsteady aerodynamic problems are presented. The eventual goal is to develop trusted surrogate models for real-time system assessments supporting Digital Engineering initiatives.

A semi structured technique is presented for surfaces and volumes with mixed and simplicial elements in order to enhance the classical metric based adaptivity. The advantage of the current procedure is the guarantee that mixed elements will be preserved during the adaptivity, provided stronger angle bounds compared to simplices. The volume and 2D axisymmetric analysis is run by the CREATE AV Kestrel solver, while the surface mesh is run by the CREATE AV Helios Strand solver. Various improvements are presented, as well as issues found during the implementation. The boundary layer surface mesh generated in Capstone is not purely semistructured, which may lead to misinterpretations of the columns to be refined on the surface. The semi structured volume is more straightforward as long as multiple normals and tangential adaptivity are not triggered. Due to the semi structured fashion of the algorithm, a special smoothing has to be performed to begin with a consistent initial metric field.

The field of machine learning is broad, covering many different areas and applications. The application of machine learning to expand automated software testing is one of many possibilities. The ability to train a machine learning model to detect issues with aerodynamic simulations enables testing to be performed with more depth than current methods. The focus of the current effort is the development of a machine learning-based automated test system for computational fluid dynamic software assessment. The underlying methodology employed is novelty detection applied through the support vector machine. Augmentation is employed to accommodate the actual data available to train the model. The machine learning-based automated test system is then demonstrated for typical external aerodynamic test cases.

This paper documents continuing activities associated with the extension of the HPCMP CREATE(TM)-AV Kestrel computational fluid dynamics software toward thermochemical generality. Previous publications document the extension of Kestrel to general equations of state, including those associated with both equilibrium and non-equilibrium chemistry. This publication focuses on further extension to thermal non-equilibrium as simulated with a two-temperature model. It also documents a not-yet-released capability for a monolithic (as opposed to segregated) solver option for multispecies flows.

In the context of the HPCMP − CREATETM project, a first attempt to implement metric based anisotropic adaptivity is proposed. Local operators modify the mesh in an iterative fashion. Some issues and remedies that were found during the implementation are highlighted. Numerical results in partnership with various teams of the CREATE project are presented.

Kestrel is an integrating product in the HPCMP CREATE(TM) program that allows crossover between simulation of aerodynamics, thermochemistry, dynamic stability and control, structures, propulsion, and store separation. It provides a robust and accurate multidisciplinary simulation capability targeting fixed-wing aircraft and is being used extensively in the DoD aircraft acquisition process. This paper presents a summary of recent code capabilities that have entered into production via multiple feature releases over the last year as well as emerging features that will be available in future releases.

A new turbomachinery simulation capability is currently being developed under the HPCMP CREATE-AV program. The end goal of this effort is to provide full-annulus turbomachinery simulations including the aircraft inlet system and blade structural deformation using highly parallel analysis software. This paper describes the status of the unstructured grid turbomachinery computational fluid dynamics (CFD) solver being developed as part of this capability. The new capability was shown to agree well with data for single-blade row and single-stage simulations. The paper also includes an example fullannulus rotor/inlet calculation. This new capability will serve to improve the fidelity of aircraft numerical simulations and can be used as a design tool or diagnostic technique to improve the quality and reduce the cost of future weapon systems. The CFD capability is available with the release of HPCMP CREATE-AV Kestrel v5.0.

ed, object-based approach to the input definitions, it was fairly easy to remove this explicit behavior and automatically determine the components needed in a simulation based on the job characteristics, motion types, etc. Finally, it is worth noting that Kestrel performs the same validation checks on the input data at run time that were accomplished during the job setup, protecting the user against incorrect manual modifications to the inputs. This has proven to be helpful to users since the complexity of the input file increases rapidly with more involved cases. B. Input Management and Consolidation One of the more important behind-the-scenes pieces of the Kestrel code base is the shared utility code that collects and “normalizes” all of the disparate user inputs collected into the hierarchical XML input file into a format that each of the execution components can easily utilize – termed the “body hierarchy” (inspired by Ref. 6). The body hierarchy has two main purposes. First, it collects all of the inputs and presents them to each of the components in a manner such that the component can only extract the inputs it cares about (e.g., the control surface motion component on a particular compute rank only cares about the control surface definitions on the body being computed on the local rank). The implementation includes a number of helper methods to assist components with extracting the desired component-specific inputs from the rich body hierarchy data structure. Second, the body hierarchy converts all input data into a user-specified unit system, a user-specified scale, and an internal coordinate system so that individual components do not have to worry about those types of conversions. The XML input file represents each section of the user inputs in its own specific coordinate and unit system as requested by the user. At run time, the body hierarchy reconciles various input contexts and resolves inputs into the common unit system, scale, and coordinate system. C. Setup Validation Even with the measures taken to try to reduce errors associated with the simulation inputs, it is likely that problems will still exist due to a misunderstanding of a particular input definition, user mistakes, etc. Therefore, Kestrel includes a utility called “jobview” that allows users to visualize their complex simulation setups and see how Kestrel will interpret the job inputs. Jobview displays the surface meshes of each of the bodies in their assembled state. The user is then free to interact with the view by panning, rotating, zooming, and selecting additional view options that decorate the scene with icons that illustrate how the inputs are interpreted. Examples of inputs that can be shown are boundary condition on each patch, center of mass for each body, arrows indicating the location, direction, and relative magnitude of external forces, bounds of the Cartesian mesh and size of the coarsest cell (Cartesian off-body paradigm), structural nodes, and tap points. Jobview performs the same operations that Kestrel does when it prepares a job for simulation. This includes converting to the internal coordinate system, scaling to match the user-specified unit system and simulation scale, and rotating/translating into the assembled position according to the body relationships. When setting up a simulation that entails multiple bodies and/or motions of any kind, jobview is indispensible and makes mistakes with the various interconnected inputs very clear. Figure 4 shows a screenshot of the jobview display for an assembled mesh system and some selected inputs.

This paper documents the third version of an integrating product that allows cross-over between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. Computational efficiency will also be improved by targeting the next generation peta-flop architectures envisioned for the 2010+ time frame. The need for Kestrel developed from the fact that existing computational resources (hardware and CSE software) are insufficient to generate decision data in a timely enough way to impact early-phase and even many sustainment phase acquisition processes. Kestrel is also targeted to the need of simulating multi-disciplinary physics such as fluid-structure interactions, inclusion of propulsion effects, moving control surfaces, and coupled flight control systems. The Kestrel software product is to address these needs for fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Version 3.0 adds a relative motion capability for multiple bodies (e.g. aircraft store separation, aircraft cargo release). The relative motion capability utilizes PUNDIT, an implicit hole cutting domain connectivity component to calculate donor and receptor cells, as well as interpolation weights for data transfer between the respective meshes. Preliminary multi-mesh calculations are presented, as well as solutions to demonstrate improvements of the kAVUS solver for steady and unsteady simulations.

This paper documents the second version of a new integrating product that allows cross-over between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. Computational efficiency will also be improved by targeting the next generation peta-flop architectures envisioned for the 2010+ time frame. The need for Kestrel developed from the fact that existing computational resources (hardware and CSE software) are insufficient to generate decision data in a timely enough way to impact early-phase and even many sustainment phase acquisition processes. Kestrel is also targeted to the need of simulating multi-disciplinary physics such as fluid-structure interactions, inclusion of propulsion effects, moving control surfaces, and coupled flight control systems. The Kestrel software product is to address these needs for fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Version 2.0 adds a 6 Degree of Freedom (6DoF) component to provide a predictive motion capability and a moving control surface capability to allow more realistic aircraft simulations. Version 2.0 also provides parallel scalability performance improvements for moving and deforming mesh use cases. Results of a C-17 cargo aircraft and a GBU-38 Joint Direct Attack Munition (JDAM) are provided for static rigid body conditions. Results are also presented for a NACA0015 wing with a moving flap control surface and a 6DOF predictive motion of a MK-82 gravity bomb in freefall. Finally, parallel performance is demonstrated with a 24 million cell F-16 simulation with up to 2000 cores for static and moving conditions.

Kestrel is an integrating software product in the CREATE program that allows crossover between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. It is written in a modular form with a Python infrastructure to allow growth to additional capabilities as needed. The second version of the Kestrel product includes the addition of two new components that enable both static and dynamic movement of aerodynamic control surfaces to be included in a fixed-wing simulation. This paper documents the design and implementation of these new components as well as their integration into the overall Kestrel environment. Both components are fully parallel and capable of handling any number of conventional control surfaces in a single unstructured mesh. Results from the Kestrel v2.0 testing involving a NACA 0015 wing with a single trailing edge control surface are presented.

This paper documents the second version of a new integrating product that allows crossover between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. Computational efficiency will also be improved by targeting the next generation peta-flop architectures envisioned for the 2010+ time frame. The need for Kestrel developed from the fact that existing computational resources (hardware and CSE software) are insufficient to generate decision data in a timely enough way to impact early-phase and even many sustainment phase acquisition processes. Kestrel is also targeted to the need of simulating multi-disciplinary physics such as fluid-structure interactions, inclusion of propulsion effects, moving control surfaces, and coupled flight control systems. The Kestrel software product is to address these needs for fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Version 2.0 adds a 6 Degree of Freedom (6DoF) component to provide a predictive motion capability and a moving control surface capability to allow more realistic aircraft simulations. Version 2.0 also provides parallel scalability performance improvements for moving and deforming mesh use cases.

This paper documents results from the first version of a new integrating product that allows cross-over between simulation of aerodynamics, dynamic stability and control, and structures in the first version, as well as propulsion and store separation in later versions. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. The Kestrel software product has been designed from the ground up to address fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Kestrel v1.0 has three simulation capabilities, static rigid body aircraft, rigid body maneuvering aircraft, and aeroelastic wings. Results for two F-16C configurations at two Mach numbers and the Joint Stand Off Weapon (JSOW) are provided for static rigid body aircraft. Results are presented for an F-16C in a dynamic pitch maneuver and an F-22 in a wind up turn both compared to Lockheed performance data. Results are also presented for an AGARD 445.6 aeroelastic wing and compared to experiment and other researcher’s data.

This paper documents the second version of a new integrating product that allows crossover between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. Computational efficiency will also be improved by targeting the next generation peta-flop architectures envisioned for the 2010+ time frame. The need for Kestrel developed from the fact that existing computational resources (hardware and CSE software) are insufficient to generate decision data in a timely enough way to impact early-phase and even many sustainment phase acquisition processes. Kestrel is also targeted to the need of simulating multi-disciplinary physics such as fluid-structure interactions, inclusion of propulsion effects, moving control surfaces, and coupled flight control systems. The Kestrel software product is to address these needs for fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Version 2.0 adds a 6 Degree of Freedom (6DoF) component to provide a predictive motion capability and a moving control surface capability to allow more realistic aircraft simulations. Version 2.0 also provides parallel scalability performance improvements for moving and deforming mesh use cases.

This paper documents a new integrating product that allows cross-over between simulation of aerodynamics, dynamic stability and control, structures, propulsion, and store separation. The Kestrel software product is an integrating product written in modular form with a Python infrastructure to allow growth to additional capabilities as needed. Computational efficiency will also be improved by targeting the next generation peta-flop architectures envisioned for the 2010+ timeframe. Kestrel is also targeted to the need of simulating multidisciplinary physics, such as fluid-structure interactions, inclusion of propulsion effects, moving control surfaces, and coupled flight control systems. The Kestrel software product is to address these needs for fixed-wing aircraft in flight regimes ranging from subsonic through supersonic flight, including maneuvers, multi-aircraft configurations, and operational conditions. Preliminary results of the F-16C with comparison to experiments are provided. Parallel scalability analysis of the initial version of Kestrel is also presented.