Fluid dynamics characterization of stirred-tank reactors via approximate Bayesian computational (ABC) for parameter estimation and model selection

Abstract

The Approximate Bayesian Computing (ABC) was used in this work to perform the selection and calibration of models in reactor dynamics. Five phenomenologically based hydrodynamic models on dead volume, flow deviation and retro-mix anomalies were evaluated and classified into two groups. The inverse problems were solved from experimental data obtained by the technique of residence time distribution (RTD) in positive step, using the methylene blue tracer in a stirred-tank reactor. The experiments evaluated the influence of volumetric flow, effluent outlet position, agitation position and rotation on the hydrodynamic characterization of the reactor. The results revealed that the most influential parameter in the first group was the dead volume, while in the second group, the retro-mix anomaly was highly sensitive to rotation. The ABC algorithm applied in this work used the Morozov discrepancy principle to select the tolerance in the final population. Tolerances for other intermediate populations were obtained from the median distance of each particle accepted in the previous population. HIGHLIGHTS The application of the Bayesian approach in RTD problems was excellent for the hydrodynamic characterization of reactors. The ABC SMC algorithm proved to be efficient for the calibration and selection of hydrodynamic models of reactors. The most influential anomaly in the flow was the dead volume.