Transforming Energy through Computational Excellence. Exascale Computing: Combustion; Deep Learning for Presumed Probability Density Function (PDF) Models

DOE OSTI · 1825674

Transforming Energy through Computational Excellence. Exascale Computing: Combustion; Deep Learning for Presumed Probability Density Function (PDF) Models

View original source ↗Read source document ↗Export metadata ↓

Abstract

NREL researchers use advanced machine learning techniques to define improved methods using deep learning models to resolve reacting flows in turbulent combustion flows, reducing the computational burden, increasing computational speed, and improving accuracy. These advancements reduce cost and improve fidelity of rapid-turn-around engineering calculations.

Explore related subjects

combustion deep learning DNN exascale computing MATHEMATICS AND COMPUTING probability density function

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

2021-10-07. Transforming Energy through Computational Excellence. Exascale Computing: Combustion; Deep Learning for Presumed Probability Density Function (PDF) Models. https://www.osti.gov/biblio/1825674

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related reports

DOE OSTI2026

Flame Characterization and Flashback Studies of Hydrogen-Enriched Low-Swirl Flames Using High-Speed OH-PLIF

Here, this paper reports the flashback mechanism observed in hydrogen-enriched flames stabilized in a low-swirl burner (LSB) at atmospheric temperature and pressure. The fundamentals of hydrogen-rich stable flames and spatiotemporal information of flashback phenomena were observed experimentally using a high-repetition-rate nanosecond (ns)-duration hydroxyl planar laser-induced fluorescence (OH-PLIF) diagnostic. Testing was conducted for methane and hydrogen (50–90%H 2 , by mole) blends in an optically accessible LSB premixer with measured swirl numbers varying from 0.43 to 0.49. The flashback propensity showed dependence on liftoff length, which was dependent on premixer velocity (𝑉), hydrogen content (𝑋 H 2 ), and equivalence ratio (𝜙) at constant temperature and pressure. High-speed OH-PLIF images revealed that lifted flames were first observed at low 𝜙/𝑋 H 2 conditions, which were changed to an M-shaped flame attached to the burner rim with an increase in 𝜙/𝑋 H 2 . A further increase in 𝜙/𝑋 H 2 triggered the flame flashback into the premixer. Flame 𝜙 at flashback (𝜙 FB ) showed an expected linearly increasing trend with increasing 𝑉 and decreasing 𝑋 H 2 , and the conclusions drawn aligned well with detailed liftoff length investigations. The 𝜙 FB increased with increasing perforated plate hole diameter and increasing swirler vane angle. The vane angle had little effect on flashback tendency at high premixer velocity.

combustion↗

DOE OSTI2024

Adaptive Mesh Refinement Simulations for Turbulent Reacting Flow

With the increased availability of exascale computing hardware, detailed simulations of realistic devices can be performed at practically relevant time and length scales. Insights into the multiscale driving mechanisms in compressible reacting flow systems with complex geometry, such as combustors, can be used for design optimization and technology improvements. However, to effectively perform these simulations, advanced numerical algorithms must be used to maintain solution accuracy without incurring undue computational costs. PeleC, part of the Pele suite of codes, leverages block-structured adaptive mesh refinement (AMR) through the AMReX library to capture fine-scale flow features in compressible reacting flows. In this talk, we discuss recent improvements to the numerical algorithms, particularly in regard to describing flows at complex boundary structures, and PeleC's performance on exascale computing hardware. We will demonstrate that PeleC is well-suited for modern, extreme-scale, heterogenous compute platforms.

combustion↗

DOE OSTI2023

DNS of a Simplified Gas Turbine Premixer: Auto-Ignition and Flame Stabilization

With the increasing need for fuel-flexibility, an understanding of auto-ignition inside gas turbine premixers is crucial. Direct numerical simulation (DNS) is a great tool to reveal the fine grain details of physics inside the premixer and guide the choice of the models to use for less expensive simulations. We present numerical simulations of a simplified and downscaled premixer, with multiple air jets in rich fuel crossflow. The simulations show intense fuel-air mixing and flame-turbulence interaction. We examine flow acceleration by the incoming jets, and the combustion within the device. We observe low-temperature and high-temperature combustion signatures, and compute the flame index to quantify local stabilization modes.

combustion↗