Bayesian Optimization for Automobile Catalyst Development

In this study, we propose an efficient computational methodology for developing Selective catalytic reduction (SCR) with high NOx conversion and resistance to hydrothermal aging, using Bayesian optimization (BO). In order to focus on the catalytic performance at low temperature, Cu-Fe bimetallic catalyst supported by SSZ- 13 (Si/Al = 12) is targeted. An initial surrogate model is constructed by referring experimental data from previously published papers. The next sampling points are determined from the Bayesian optimization algorithm. NOx conversion is observed under fresh condition and hydrothermally aged condition after manufacturing a catalyst sample consisting of suggested metal compositions. We also consider the catalytic activity after hydrothermal aging in the air of 900 °C containing 10 % water for 16 hours. In this way, the optimal composition for bimetallic SCR catalyst is discovered, maximizing NOx conversion and hydrothermal resistance in only a few steps of experimentation. The proposed SCR catalyst can reduce 95.86 % of nitrogen oxides at 250 °C. After hydrothermal aging, it can eliminate 88.83 % of nitrogen oxides at the same temperature.