Bayesian Optimization of Wet-Impregnated Co-Mo/Al2O3 Catalyst for Maximizing the Yield of Carbon Nanotube Synthesis

Abstract

Multimetallic catalysts have demonstrated their high potential for the controlled synthesis of carbon nanotubes (CNTs), but their development requires a more complicated optimization than that of monometallic catalysts. Here, we employed Bayesian optimization (BO) to optimize the preparation of Co-Mo/Al2O3 catalyst using wet impregnation, with the goal of maximizing carbon yield in the chemical vapor deposition (CVD) synthesis of CNTs. In the catalyst preparation process, we selected four parameters to optimize: the weight percentage of metal, the ratio of Co to Mo in the catalyst, the drying temperature, and the calcination temperature. We ran two parallel BO processes to compare the performance of two types of acquisitions: expected improvement (EI), which does not consider noise, and one-shot knowledge gradient (OKG), which takes noise into account. As a result, both acquisition functions successfully optimized the carbon yield with similar performance. The result suggests that the use of EI, which has a lower computational load, is acceptable if the system has sufficient robustness. The investigation of the contour plots showed that the addition of Mo has a negative effect on carbon yield.