Bayesian optimization-guided design of composite solvents for selective separation of FCC slurry oil

Fluid catalytic cracking slurry oil (SLO) is an underutilized by-product that contains a high-value polycyclic aromatic hydrocarbon (PAH) fraction but is difficult to separate cleanly from resins and asphaltenes. Here, a multiscale closed-loop framework is established that couples thermodynamic screening, single-stage extraction experiments, and constrained Bayesian optimization to design composite solvents for selective SLO fractionation. Thermodynamic calculations identify n -pentane (C 5 ) as an effective main solvent that enhances discrimination between PAHs and polar heavy species, and methyl ethyl ketone (MEK) as an effective anti-solvent in C 5 -based systems. The resulting C 5 –MEK composite solvent delivers higher aromatic recovery and lower asphaltene contents than other tested solvent combinations. Gaussian-process-based constrained optimization in the two-dimensional space of C 5 mass fraction and solvent-to-oil ratio delineates a narrow feasible region within the tested range, in which stable phase separation and asphaltene contents ( W Asp ) below 1.0 wt% are maintained while high aromatic yields can be attained (up to ∼ 99% in this study; W Asp = 0.49 wt%). Starting from a nine-point Latin hypercube sampling initialization, the sequential optimization converged after four Bayesian optimization-selected experiments once the targeted performance was achieved under the feasibility constraints. More broadly, this workflow illustrates a data-efficient and transferable strategy that integrates thermodynamic screening with constraint Bayesian optimization to accelerate solvent and operating-condition development for complex-feed separations.