Machine learning-guided catalyst design: optimization by combining targeted screening, intrinsic properties and techno-economic criteria for volatile organic compounds (VOC’s) oxidation

This work uses machine learning (ML) procedures to investigate the effect of the properties of different cobalt-based catalysts on the oxidation of toluene and propane. The hydrocarbon conversion was modeled using 600 ANNs and 8 supervised regression algorithms. An optimization framework for the input variables was subsequently developed, using the best neural networks to minimize both the catalyst costs and the energy consumption for reaching 97.5% conversion. This optimization analysis showed that for toluene and propane oxidation, the best result was coincidental with that reported elsewhere and with the commercial catalyst, respectively. As expected, the optimization analysis selected the cheapest catalyst (practically negligible influence of the energy cost). The optimization analysis in terms of physical properties of the catalyst showed that the best results obtained with the toluene oxidation would approximately correspond to catalyst Co-C2O4 while those obtained with the propane oxidation did not result to be conclusive.