Application of a Machine Learning Approach for Selective Catalyst Reduction Catalyst 3D-CFD Modeling: Numerical Method Development and Experimental Validation

Abstract

<div class="section abstract"><div class="htmlview paragraph">Internal combustion engines (ICEs) exhaust emissions, particularly nitrogen oxides (NOx), have become a growing environmental and health concern. The biggest challenge for contemporary ICE industry is the development of clean ICEs, and the use of advanced design tools like Computational Fluid Dynamics (CFD) simulation is paramount to achieve this goal. In particular, the development of aftertreatment systems like Selective Catalyst Reduction (SCR) is a key step to reduce NOx emissions, and accurate and efficient CFD models are essential for its design and optimization. In this work, we propose a novel 3D-CFD methodology, which uses a Machine Learning (ML) approach as a surrogate model for the SCR catalyst chemistry, which aims to enhance accuracy of the simulations with a moderate computational cost. The ML approach is trained on a dataset generated from a set of 1D-CFD simulations of a single channel of an SCR catalyst. The trained model is then applied to the 3D-CFD simulation as a surrogate model for the SCR chemistry, allowing the calculation of the spatial distribution of chemical species at the catalyst’s outlet. The model is finally applied to the calculation of the SCR conversion efficiency and to the simulation of NOx sensors reading on the tailpipe. Experimental validation of the developed 3D-CFD models was performed on a dedicated test bench. An ad-hoc local sampling probe has been used to measure the spatial distribution of the chemical species downstream the SCR system, together with the reading of three commercial NOx sensors. Our findings suggest that the ML-based surrogate model for the SCR is a reliable and efficient methodology for including the SCR chemistry in a 3D-CFD simulation at a moderate computational cost. The developed numerical method could provide valuable insights into the SCR catalyst<b>’</b>s design and optimization and aid in the development of ultra-low NOx emission ICEs.</div></div>