Authors: Qi Xu a, Aaron Akah a, Musaed Ghrami a, Abdennour Bourane a, and Ibrahim Abba a
a Research and Development Center, Saudi Aramco, Dhahran, Saudi Arabia, 31311
Source: This paper was published in the 13th International Conference on Fluidized Bed Technology in 2021.
Abstract: Use of machine learning (ML) to aid innovation in the chemical and refining industry is becoming ubiquitous. In this study, machine learning was employed to elucidate the kinetics of olefins production from the direct cracking of crude oil feedstock in a dual-downer high severity fluidized bed system. The reactions considered include both catalytic and thermal cracking chemistries. First, many sets of experimental data were produced in a lab-scale microdowner reactor, which were then used to train the ML algorithm to: (a) enable prediction of olefins and (b) guide retro-synthesis, through innovative combinations of operating conditions, catalyst properties and reactor geometry. Second, predictions from the trained model were compared with those from our conventional 5-lump kinetic model, and they showed good agreement. Finally, using the ML-based kinetic model, results from this study provided deeper insights into the important role of thermal and catalytic cracking as competing mechanisms in the direct cracking of crude oil under high-severity operating conditions. Specifically, the study elucidated the dominance of thermal cracking mechanism at high temperatures representative of high-severity operating conditions. Thermal cracking mechanism is comparatively negligible at lower temperatures representative of conventional FCC conditions, where the catalytic cracking mechanism is dominant.