Artificial Intelligence for Simulation of Soot Distribution Inside Porous Filter Walls

Abstract

The ability to estimate the influence of the accumulated solid matter on the performance of catalytic filters (CFs) in automotive exhaust gas aftertreatment leads to the ability to estimate the required filter regeneration frequency. The ability to perform such estimates fast and only from CFs microstructural data would allow for CFs microstructure optimization. In this work, we present an approach to estimate the distribution of soot deposited in the porous structure of CFs walls. The approach leverages methods of artificial intelligence (AI) and is based on convolutional autoencoders and deep neural networks. The resulting method is trained and tested on an artificial dataset that corresponds to a single pore in the CF wall. The dataset is prepared using our previously developed transient pore-scale model of particle deposit formation in the 3D microstructure of the catalytic filter wall. The developed AI model yields good results in terms of total amount of accumulated soot, but is less accurate in its distribution. As a result, using the estimated particle deposits to calculate the pressure drop and filtration efficiency of the artificial pore allows to estimate these two CF performance indicators with 33.6 % accuracy.