Density Functional Theory (DFT) Based Machine & Deep Learning Strategies for Rational Screening and Directed Design of Catalytic Promoter Materials and Supported Nanoparticle Interfaces for Renewable Fuel Applications Conference: Southwest Process Technology ConferenceYear: 2017Proceeding: 9th AIChE Southwest Process Technology ConferenceGroup: Southwest Process Technology ConferenceSession: Meet The Industry Poster Reception Time: Friday, October 13, 2017 - 3:00pm-6:00pm Authors: Kasiraju, S., University of Houston Grabow, L. C., University of Houston The high oxygen content (35-40 wt. %) of bio-oil produced from fast pyrolysis of biomass reduces its heating value and stability, and limits its subsequent use as a transportation fuel. Bio-oil may be upgraded to renewable bio-fuels via catalytic hydrodeoxygenation (HDO) and MoO3 is a promising catalyst candidate for this reaction with high reactivity and selectivity.  HDO requires the initial creation of an oxygen vacancy site where the feed molecule can adsorb. However, the oxygen vacancy formation on MoO3(010) is slow (high activation barrier) and highly endothermic. We investigated transition metal promotion (Fe, Co, Ni, Cu and Zn) on MoO3 and determined that the transition metal promotion facilitates oxygen vacancy formation. Therefore, we expanded our search for the promoters across all transition, alkali and alkaline earth metals based on machine learning techniques. With this approach, we aim to predict ab-initio/experimental reactivity trends, by using just the intrinsic physical properties of promoter materials used in the screening study. Previous experimental work observed the presence of molybdenum carbide or an oxy-carbide phase along with the oxide with potential implications to the HDO mechanistic chemistry and overall activity. To understand the individual role of the oxide and the carbide phase for HDO chemistry, molybdenum carbide supported on the oxide was modeled based on earlier HRTEM characterization. A novel two-step approach was implemented using Deep Learning techniques such as artificial neutral networks (ANN) to account for the absence of detailed structural and phase composition information, and also to reduce the computational tediousness of sifting through all possible combinations of nano-particle support interfaces using ab-initio DFT calculations. Preliminary results suggest a drastic reduction in the total computational time required to predict the structure and thermodynamic feasibility of the interface with this approach. Oxygen vacancy formation, and H2 spillover at the interface were also investigated. The ability to rapidly screen promoters for MoO3 and rationally design mixed oxy-carbide interfaces will ultimately lead the way to rationally designing novel HDO catalysts for the commercial upgrade of biomass to chemicals and fuels. (1) Prasomsri, T.; Nimmanwudipong, T.; Román-Leshkov, Y. Energy Environ. Sci. 2013, 6 (6), 1732.(2) Wang, H.; Liu, S.; Smith, K. J. Energy & Fuels 2016, 30 (7), 6039.(3) Delporte, P.; Meunier, F.; Pham-Huu, C.; Vennegues, P.; Ledoux, M. J.; Guille, J. Catal. Today 1995, 23 (3), 251.