Scholarlink

The transition to carbon–neutral fuels is crucial considering the alarming threats of climate change and global warming. Though hydrogen (H2) is the best carbon–neutral fuel owing to its highest calorific value, its storage, transportation, and the possibility of embrittlement pose many challenges for its usage. Hence, due to its high gravimetric hydrogen, ammonia (NH3) is currently being exploited as the H2 carrier. However, the usage of NH3 directly as a fuel is not effective owing to its low laminar flame speed, low flame characteristics and high nitrogen oxide (NOx) emissions. Thus, it is generally blended with H2 and carbon fuels to improve its flame characteristics. The main aim of the study is thus to find the optimum blend of NH3, natural gas (CH4), n-dodecane and H2 that minimizes NH3 slip along with minimization of CO and NOx emissions. The combustion mechanism of this blend fuel involves 242 species and 1769 reactions including both NOx and soot sub-mechanisms. Considering such a mechanism in a Computational Fluid Dynamics (CFD) framework, first a high-fidelity model is constructed to understand the several species behaviour under the combustion process. However, finding the optimum parameters using the high-fidelity combustor model is computationally very expensive due to the iterative nature of optimization modules. Therefore, this study focuses on developing Artificial Neural Network (ANN) models that emulate the CFD model and finally solving the afore-mentioned optimization formulation using the ANN surrogates. The data points required for developing the ANN model were generated through SOBOL sampling using four parameters, viz., wattage, NH3 composition in fuel, pressure and air preheat temperature, where CFD simulations were performed. Using the wattage, NH3 composition, CH4 composition in fuel, pressure and air preheat temperature as inputs, and compositions of CO, NH3 and NOx in the flue gas as outputs, three ANN surrogates were developed. Using these surrogates, the optimization problem was solved using the Non-dominated Sorting Genetic Algorithm (NSGA II). Based on the requirement, the fuel blend with high NH3 composition and low CH4 composition or the one with low NH3 composition and high CH4 composition can thus be selected from the obtained Pareto solutions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Decoding the Challenge of Ammonia Combustion: A Hybrid Approach Considering CFD and AI

  • 章节

  • 发表日期：

    2026年

查看全文
  • Aswitha Tadepalli,
  • Richie Shaji Mathew,
  • Raja Banerjee,
  • Kishalay Mitra

摘要

The transition to carbon–neutral fuels is crucial considering the alarming threats of climate change and global warming. Though hydrogen (H2) is the best carbon–neutral fuel owing to its highest calorific value, its storage, transportation, and the possibility of embrittlement pose many challenges for its usage. Hence, due to its high gravimetric hydrogen, ammonia (NH3) is currently being exploited as the H2 carrier. However, the usage of NH3 directly as a fuel is not effective owing to its low laminar flame speed, low flame characteristics and high nitrogen oxide (NOx) emissions. Thus, it is generally blended with H2 and carbon fuels to improve its flame characteristics. The main aim of the study is thus to find the optimum blend of NH3, natural gas (CH4), n-dodecane and H2 that minimizes NH3 slip along with minimization of CO and NOx emissions. The combustion mechanism of this blend fuel involves 242 species and 1769 reactions including both NOx and soot sub-mechanisms. Considering such a mechanism in a Computational Fluid Dynamics (CFD) framework, first a high-fidelity model is constructed to understand the several species behaviour under the combustion process. However, finding the optimum parameters using the high-fidelity combustor model is computationally very expensive due to the iterative nature of optimization modules. Therefore, this study focuses on developing Artificial Neural Network (ANN) models that emulate the CFD model and finally solving the afore-mentioned optimization formulation using the ANN surrogates. The data points required for developing the ANN model were generated through SOBOL sampling using four parameters, viz., wattage, NH3 composition in fuel, pressure and air preheat temperature, where CFD simulations were performed. Using the wattage, NH3 composition, CH4 composition in fuel, pressure and air preheat temperature as inputs, and compositions of CO, NH3 and NOx in the flue gas as outputs, three ANN surrogates were developed. Using these surrogates, the optimization problem was solved using the Non-dominated Sorting Genetic Algorithm (NSGA II). Based on the requirement, the fuel blend with high NH3 composition and low CH4 composition or the one with low NH3 composition and high CH4 composition can thus be selected from the obtained Pareto solutions.