Enhancing\nthe Bitumen Partial Upgrading Process with\nFe<sub>3</sub>O<sub>4</sub>‑Coated Cenospheres and Artificial\nNeural Network-Driven Process Optimization

This study introduces a new approach to partially upgrade\noil sand\nbitumen utilizing waste fly ash cenospheres coated with an Fe<sub>3</sub>O<sub>4</sub> layer. Through the use of a combination of Fe<sup>2+</sup> and Fe<sup>3+</sup> precursors, the cenospheres were covered\nwith a layer of Fe<sub>3</sub>O<sub>4</sub> to form a new catalyst\ndenoted as (Fe-Ceno), which was subsequently characterized in detail\nusing scanning electron microscopy, energy-dispersive X-ray spectroscopy,\nX-ray diffraction, and X-ray photoelectron spectroscopy. The characterization\nresults investigated the catalyst’s morphology, microstructure,\ncrystalline structure, and surface chemistry, confirming that the\nFe<sub>3</sub>O<sub>4</sub> layer was successfully applied. During\nthe catalytic bitumen upgrading, the Fe-Ceno catalyst was dispersed\nin a liquid hydrogen donor solution to facilitate the transformation\nof oil sand bitumen into a partially upgraded liquid oil product.\nThe results of the upgrading process demonstrated that the introduction\nof only 1 wt % of Fe-Ceno led to a significant enhancement in the\nquality of the upgraded oil, noted by the reduced olefin content to\nbelow 1 wt %, the improved phase stability, and the significant reduction\nin the oil’s viscosity and density to values below 300 cP and\n940 kg/m<sup>3</sup>, respectively, to satisfy the pipeline transportation\nspecifications. Additionally, this study builds beyond the experimental\napproach and develops a tailored artificial neural network (ANN) model\nthat can accurately predict the rheological properties of the upgraded\nbitumen without the need to perform additional upgrading experiments.\nThe developed artificial intelligence model was able to successfully\npredict values, such as viscosity and density, for the upgraded oil\nsamples under different catalytic operating conditions, with a coefficient\nof determination (<i>R</i><sup>2</sup>) of >0.99, an\naverage\nabsolute deviation (AAD) of <0.1%, and a root mean square error\n(RMSE) of <0.2. With the further leverage of more extensive data\nsets and the improvement of generalization, this methodology will\nexhibit the promising potential of ANN models to accelerate advancements\nin catalyst discovery and optimization to enhance upgrading process\nefficiency.