Simulation of ammonia and hydrogen combustion using transported PDF and artificial neural networks chemistry speedup method
Title: Simulation of ammonia and hydrogen combustion using transported PDF and artificial neural networks chemistry speedup method
SNIC Project: Berzelius-2021-19
Project Type: LiU Berzelius
Principal Investigator: Shijie Xu <shijie.xu@energy.lth.se>
Affiliation: Lunds universitet
Duration: 2021-09-01 – 2022-03-01
Classification: 20304
Homepage: http://www.fm.energy.lth.se
Keywords:

Abstract

This project aims at the simulation of ammonia and hydrogen combustion using Computational Fluid Dynamics (CFD) coupled with artificial neural networks (ANN). Ammonia is a renewable carbon-free fuel made from sun, air, and water that could be a promising energy carrier. However, due to its low flame speed and the fuel-bound nitrogen, practical applications of ammonia in industrial gas turbines suffer from unreliable ignition, high unburned ammonia, and NOx emissions issues. This could be significantly improved by using an ammonia/hydrogen dual-fuel combustion strategy. A large-eddy simulation study will be conducted in this project to investigate the ammonia/hydrogen dual-fuel combustion strategy. In this project, the ANN will be introduced and combined with the Eulerian stochastic field (ESF) to simulate the ammonia/hydrogen dual-fuel combustion. The ESF-based transported probability density function (PDF) method is an ideal method for dual-fuel combustion as it takes the turbulence chemistry interaction into consideration. In the previous project (SNIC 2020/13-107), the ESF-based transported PDF method was developed and validated. This will be published in a journal paper (Xu S et. al. On the element mass conservation in Eulerian stochastic field modeling of turbulent combustion. Combustion and Flame, 2021, Accepted). However, as ammonia/hydrogen dual-fuel combustion involves hundreds of reactions, over 85% of CPU hours are consumed in chemistry integration. To speed up the simulation, an ANN-based chemistry speedup method (Zhang Y and Xu S, et. al. Energy and AI, 2020, 2: 100021) will be combined with the ESF model, which significantly reduces the computational cost. In this model, the chemistry reactions (a set of ordinary differential equations, ODE) are integrated and tabulated in a pretreatment process. The tabulated data have three input dimensions, i.e., mixture fraction, progress variable, temperature, and 60-100 output dimensions, e.g., the species mass fractions and reaction rates. The chemistry table, i.e. tabulated data with a typical size of 100 GB, will be used as input for ANN training. Both ODE integration and ANN training require GPU. Using GPU could significantly reduce the simulation time as ODE integration and ANN training are more efficient in GPU. This project has great practical significance since it addresses the global challenge of sustainable development and the environmentally friendly transport sector in Sweden. The project meets the long-term vision of Sweden to develop a zero greenhouse emission energy system by 2050. The scientific problems addressed in this project are of fundamental and practical importance, i.e., ignitions and multiple mode combustion in advanced combustion strategies, and improving the computational efficiency of the combustion simulation method using artificial intelligence (AI).