Data-Driven Approach to Estimating Soot Distribution Inside Catalytic Filters in Automotive Exhaust Gas Aftertreatment

The performance and the necessary regeneration frequency of catalytic filters (CFs) used in the treatment of automotive exhaust gases depend strongly on the solid matter accumulated within their porous walls.Reliable predictions of solid matter (soot) accumulation are crucial in the development and optimisation of CFs.In this contribution, we exploit the tools of artificial intelligence (AI) to estimate the distribution of soot directly in the porous microstructure of CFs.Specifically, our AI model uses deep neural networks (DNNs) and convolutional autoencoders (CAEs) to predict the soot distribution from information about the microstructure and the initial velocity field.To provide the model with training and validation data, we used our previously developed transient numerical model of particle deposition in the CF walls to calculate soot distribution in a dataset of artificial 2D geometries.The results of the developed AI model are in good agreement with simulation regarding the total amount of accumulated soot.However, the accuracy in the spatial distribution of the soot is not optimal, and consequently, using estimated particle deposits to simulate the pressure drop in the artificial microstructure results in 35 % accuracy.