Application of Design of Experiments and Surrogate Modeling within the NASA Advanced Concepts Office, Earth-to-Orbit Design Process

Zwack, Matthew R.; Dees, Patrick D.; Holt, James B.

doi:10.2514/6.2016-5649

Cited by 6 publications

(5 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These tools are both widely accepted by the launch vehicle design community and utilize direct shooting and direct collocation optimization schemes, respectively. 17 While results from POST and its main competitor OTIS arrive at very similar solutions, 18 POST is used here and in the previous work because it is the tool in use at NASA Marshall Space Flight Center's Advanced Concepts Office 19,20 (ACO).…”

Section: Approachmentioning

confidence: 96%

Augmenting Conceptual Design Trajectory Tradespace Exporation with Graph Theory

et al. 2016

Self Cite

View full text Add to dashboard Cite

Within conceptual design changes occur rapidly due to a combination of uncertainty and shifting requirements. To stay relevant in this fluid time, trade studies must also be performed rapidly. In order to drive down analysis time while improving the information gained by these studies, surrogate models can be created to represent the complex output of a tool or tools within a specified tradespace. In order to create this model however, a large amount of data must be collected in a short amount of time. By this method, the historical approach of relying on subject matter experts to generate the data required is schedule infeasible. However, by implementing automation and distributed analysis the required data can be generated in a fraction of the time. Previous work focused on setting up a tool called multiPOST capable of orchestrating many simultaneous runs of an analysis tool assessing these automated analyses utilizing heuristics gleaned from the best practices of current subject matter experts. In this update to the previous work, elements of graph theory are included to further drive down analysis time by leveraging data previously gathered. It is shown to outperform the previous method in both time required, and the quantity and quality of data produced.

show abstract

Section: Approachmentioning

confidence: 96%

Augmenting Conceptual Design Trajectory Tradespace Exporation with Graph Theory

et al. 2016

Self Cite

View full text Add to dashboard Cite

show abstract

“…Their approach applies the concepts of DOE and response surface metamodel to more exhaustively explore the trade space and supply the customer with additional design information to inform decision making. 53 In the case of nondeterministic trajectory design optimization problems, a very few research have been presented until now. Most of them have been focused on the conceptual design of LV not specifically on the trajectory design problem.…”

Section: Introductionmentioning

confidence: 99%

A novel metamodel management strategy for robust trajectory design of an expendable launch vehicle

Roshanian

Bataleblu

2019

Proceedings of the Institution of Mechanical Engineers, Part G:

View full text Add to dashboard Cite

Uncertainty-based design optimization has been widely acknowledged as an advanced methodology to address competing objectives of aerospace vehicle design, such as reliability and robustness. Despite the usefulness of uncertainty-based design optimization, the computational burden associated with uncertainty propagation and analysis process still remains a major challenge of this field of study. The metamodeling is known as the most promising methodology for significantly reducing the computational cost of the uncertainty propagation process. On the other hand, the nonlinearity of the uncertainty-based design optimization problem's design space with multiple local optima reduces the accuracy and efficiency of the metamodels prediction. In this article, a novel metamodel management strategy, which controls the evolution during the optimization process, is proposed to alleviate these difficulties. For this purpose, a combination of improved Latin hypercube sampling and artificial neural networks are involved. The proposed strategy assesses the created metamodel accuracy and decides when a metamodel needs to be replaced with the real model. The metamodeling and metamodel management strategy are conspired to propose an augmented strategy for robust design optimization problems. The proposed strategy is applied to the multiobjective robust design optimization of an expendable launch vehicle. Finally, based on non-dominated sorting genetic algorithm-II, a compromise between optimality and robustness is illustrated through the Pareto frontier. Results illustrate that the proposed strategy could improve the computational efficiency, accuracy, and globality of optimizer convergence in uncertainty-based design optimization problems.

show abstract

“…16 The percent error distributions of the resulting surrogates can be seen in Figure 7 below. As depicted in the figure, the surrogate from the random repetitions method had a maximum error of 5.2% with a standard deviation around one.…”

Section: Ivd Surrogate Modelingmentioning

confidence: 99%

“…The DOE's for these trades were very similar to those laid out in the previous work and focused on upper stage parameters. 4,16,17 Both sets in this case were run over similar time periods, each using the same parallel computing resources with 80 available threads. The random repetitions data was executed over the course of two and a half days, while the SVM data ran for just under three days.…”

Section: Ivd Surrogate Modelingmentioning

confidence: 99%

Improvement of Automated POST Case Success Rate using Support Vector Machines

Zwack

Dees

2017

AIAA SPACE and Astronautics Forum and Exposition

Self Cite

View full text Add to dashboard Cite

During early conceptual design of complex systems, concept down selection can have a large impact upon program life-cycle cost. Therefore, any concepts selected during early design will inherently commit program costs and affect the overall probability of program success. For this reason it is important to consider as large a design space as possible in order to better inform the down selection process.For conceptual design of launch vehicles, trajectory analysis and optimization often presents the largest obstacle to evaluating large trade spaces. This is due to the sensitivity of the trajectory discipline to changes in all other aspects of the vehicle design. Small deltas in the performance of other subsystems can result in relatively large fluctuations in the ascent trajectory because the solution space is non-linear and multi-modal. 1, 2In order to help capture large design spaces for new launch vehicles, the authors have performed previous work seeking to automate the execution of the industry standard tool, Program to Optimize Simulated Trajectories (POST). This work initially focused on implementation of analyst heuristics to enable closure of cases in an automated fashion, with the goal of applying the concepts of design of experiments (DOE) and surrogate modeling to enable near instantaneous throughput of vehicle cases.3 As noted in [4] work was then completed to improve the DOE process by utilizing a graph theory based approach to connect similar design points.The conclusion of the previous work illustrated the utility of the graph theory approach for completing a DOE through POST. However, this approach was still dependent upon the use of random repetitions to generate seed points for the graph. As noted in, 4 only 8% of these random repetitions resulted in converged trajectories. This ultimately affects the ability of the random reps method to confidently approach the global optima for a given vehicle case in a reasonable amount of time. With only an 8% pass rate, tens or hundreds of thousands of reps may be needed to be confident that the best repetition is at least close to the global optima. However, typical design study time constraints require that fewer repetitions be attempted, sometimes resulting in seed points that have only a handful of successful completions. If a small number of successful repetitions are used to generate a seed point, the graph method may inherit some inaccuracies as it chains DOE cases from the non-global-optimal seed points. This creates inherent noise in the graph data, which can limit the accuracy of the resulting surrogate models.For this reason, the goal of this work is to improve the seed point generation method and ultimately the accuracy of the resulting POST surrogate model. The work focuses on increasing the case pass rate for seed point generation. It is expected that by vastly improving the pass rate, more successful repetitions will be completed in the same time frame, resulting in a higher probability that the global optima is found. This ultimately transl...

show abstract

Application of Design of Experiments and Surrogate Modeling within the NASA Advanced Concepts Office, Earth-to-Orbit Design Process

Cited by 6 publications

References 5 publications

Augmenting Conceptual Design Trajectory Tradespace Exporation with Graph Theory

Augmenting Conceptual Design Trajectory Tradespace Exporation with Graph Theory

A novel metamodel management strategy for robust trajectory design of an expendable launch vehicle

Improvement of Automated POST Case Success Rate using Support Vector Machines

Contact Info

Product

Resources

About