Progress in Scramjet Design Optimization Under Uncertainty Using Simulations of the HIFiRE Direct Connect Rig

Geraci, Gianluca; Menhorn, Friedrich; Huan, Xun; Safta, Cosmin; Marzouk, Youssef; Najm, Habib N.; Eldred, Michael

doi:10.2514/6.2019-0725

Cited by 3 publications

(3 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These tolerance, cost, and variance values are in line with those observed in MLMC studies, such as [14,15]. The CPU cost is C C = αC G for α ∈ [5,50], which spans the range of common GPU to CPU cost ratios. For example, Lassen, the number 10 supercomputer in the world as of November 2019, has a peak GPU TFLOPS to peak CPU TFLOPS ratio of 26 [16].…”

Section: Example 1: Csupporting

confidence: 80%

“…We demonstrate H-MLMC for two examples, one where C C /C G is the same for all levels and one where it is different on each level. When C C /C G is the same for all levels, H-MLMC takes 9-82% as long as MLMC, depending on the values of C C /C G ∈ [5,50], regardless of the number of levels used. H-MLMC offers more benefit over MLMC for smaller values of C C /C G .…”

Section: Discussionmentioning

confidence: 99%

“…Multilevel Monte Carlo (MLMC) [2], multifidelity Monte Carlo (MFMC) [3], and control variate Monte Carlo [4] combine solutions from multiple models to reduce the variance and thereby accelerate the convergence of the output estimates. Geraci et al [5] and Qian et al [6] extend MLMC and MFMC to estimate not just the mean, but also the variance of the output, and Krumscheid et al [7] introduces a strategy to estimate moments of any order. Giles et al [8] and Krumscheid and Nobile [9] propose variations of MLMC to estimate the full distribution of the output.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Multilevel Monte Carlo Sampling on Heterogeneous Computer Architectures

Adcock

Jofre

et al. 2020

Int. J. UncertaintyQuantification

View full text Add to dashboard Cite

Monte Carlo (MC) sampling is the standard approach for uncertainty propagation in problems with high-dimensional stochastic inputs. Various acceleration techniques have been developed to overcome the slow convergence of MC estimates, such as multilevel Monte Carlo (MLMC). MLMC uses successive approximations computed on levels, models with different levels of accuracy, and computational cost to reduce the estimator variance. MLMC analytically determines the number of samples required on each level to achieve a given accuracy at minimal cost. We propose an extension of the original MLMC theoretical framework for modern, heterogeneous computer architectures in which accelerators (GPUs) are available and, therefore, samples can be distributed on both different levels and different compute units (CPUs and GPUs). We derive the optimal sample allocation for the proposed MLMC extension by solving a convex optimization problem. We apply the MLMC extension to a stochastically heated channel flow to provide insight for a study on the design of concentrated solar energy receivers. We demonstrate for the stochastically heated channel flow that the proposed MLMC extension leads to considerable total cost reduction (up to 86%) compared to MLMC using only GPUs.

show abstract

Section: Example 1: Csupporting

confidence: 80%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Multilevel Monte Carlo Sampling on Heterogeneous Computer Architectures

Adcock

Jofre

et al. 2020

Int. J. UncertaintyQuantification

View full text Add to dashboard Cite

show abstract

Numerical Investigation of the Ramjet and Scramjet Operation Regimes of the HIFiRE-2 Combustion Chamber

Seleznev

2022

Fluid Dyn

View full text Add to dashboard Cite

Optimization of Thrust of a Generic X-51 Hypersonic Vehicle

Choi,

Driscoll

2024

Journal of Propulsion and Power

View full text Add to dashboard Cite

A prune-and-search optimization algorithm was combined with a reduced-order model of the scramjet engine within a generic X-51 vehicle. The goal was to maximize the thrust in order to achieve thrust-to-drag ratios that equal or exceed 1, even for the difficult cases of flight at high altitudes and high Mach numbers up to 10. For the inlet to the engine, the lengths and inclination angles of three wall panels were varied. For the combustor, three parameters were varied: the diameter and number of fuel injector ports and the wall divergence angle. These variables affect both the combustion efficiency and the finite-rate chemistry of the surrogate JP-7 fuel. The optimization algorithm showed that at higher flight Mach numbers an inlet that maximizes static pressure ratio is preferred over one that maximizes the stagnation pressure ratio, because good combustion efficiency requires a sufficient static pressure in the combustor. As the flight Mach number is increased, thrust decreases while the drag increases. Thus, it becomes difficult to achieve thrust that exceeds drag. A solution is shown to be increasing the engine inlet area above a critical value that was computed. Advantages of a reduced-order model over a high-fidelity CFD approach are discussed when thousands of computations of all components within an entire engine are required for optimization.

show abstract

Progress in Scramjet Design Optimization Under Uncertainty Using Simulations of the HIFiRE Direct Connect Rig

Cited by 3 publications

References 35 publications

Multilevel Monte Carlo Sampling on Heterogeneous Computer Architectures

Multilevel Monte Carlo Sampling on Heterogeneous Computer Architectures

Numerical Investigation of the Ramjet and Scramjet Operation Regimes of the HIFiRE-2 Combustion Chamber

Optimization of Thrust of a Generic X-51 Hypersonic Vehicle

Contact Info

Product

Resources

About